src/rdb.c

   1 #include "redis.h"
   2 #include "lzf.h"    /* LZF compression library */
   3
   4 #include <math.h>
   5 #include <sys/types.h>
   6 #include <sys/time.h>
   7 #include <sys/resource.h>
   8 #include <sys/wait.h>
   9 #include <arpa/inet.h>
  10 #include <sys/stat.h>
  11
  12 /* Convenience wrapper around fwrite, that returns the number of bytes written
  13  * to the file instead of the number of objects (see fwrite(3)) and -1 in the
  14  * case of an error. It also supports a NULL *fp to skip writing altogether
  15  * instead of writing to /dev/null. */
  16 static int rdbWriteRaw(FILE *fp, void *p, size_t len) {
  17     if (fp != NULL && fwrite(p,len,1,fp) == 0) return -1;
  18     return len;
  19 }
  20
  21 int rdbSaveType(FILE *fp, unsigned char type) {
  22     return rdbWriteRaw(fp,&type,1);
  23 }
  24
  25 int rdbSaveTime(FILE *fp, time_t t) {
  26     int32_t t32 = (int32_t) t;
  27     return rdbWriteRaw(fp,&t32,4);
  28 }
  29
  30 /* check rdbLoadLen() comments for more info */
  31 int rdbSaveLen(FILE *fp, uint32_t len) {
  32     unsigned char buf[2];
  33     int nwritten;
  34
  35     if (len < (1<<6)) {
  36         /* Save a 6 bit len */
  37         buf[0] = (len&0xFF)|(REDIS_RDB_6BITLEN<<6);
  38         if (rdbWriteRaw(fp,buf,1) == -1) return -1;
  39         nwritten = 1;
  40     } else if (len < (1<<14)) {
  41         /* Save a 14 bit len */
  42         buf[0] = ((len>>8)&0xFF)|(REDIS_RDB_14BITLEN<<6);
  43         buf[1] = len&0xFF;
  44         if (rdbWriteRaw(fp,buf,2) == -1) return -1;
  45         nwritten = 2;
  46     } else {
  47         /* Save a 32 bit len */
  48         buf[0] = (REDIS_RDB_32BITLEN<<6);
  49         if (rdbWriteRaw(fp,buf,1) == -1) return -1;
  50         len = htonl(len);
  51         if (rdbWriteRaw(fp,&len,4) == -1) return -1;
  52         nwritten = 1+4;
  53     }
  54     return nwritten;
  55 }
  56
  57 /* Encode 'value' as an integer if possible (if integer will fit the
  58  * supported range). If the function sucessful encoded the integer
  59  * then the (up to 5 bytes) encoded representation is written in the
  60  * string pointed by 'enc' and the length is returned. Otherwise
  61  * 0 is returned. */
  62 int rdbEncodeInteger(long long value, unsigned char *enc) {
  63     /* Finally check if it fits in our ranges */
  64     if (value >= -(1<<7) && value <= (1<<7)-1) {
  65         enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT8;
  66         enc[1] = value&0xFF;
  67         return 2;
  68     } else if (value >= -(1<<15) && value <= (1<<15)-1) {
  69         enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT16;
  70         enc[1] = value&0xFF;
  71         enc[2] = (value>>8)&0xFF;
  72         return 3;
  73     } else if (value >= -((long long)1<<31) && value <= ((long long)1<<31)-1) {
  74         enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT32;
  75         enc[1] = value&0xFF;
  76         enc[2] = (value>>8)&0xFF;
  77         enc[3] = (value>>16)&0xFF;
  78         enc[4] = (value>>24)&0xFF;
  79         return 5;
  80     } else {
  81         return 0;
  82     }
  83 }
  84
  85 /* String objects in the form "2391" "-100" without any space and with a
  86  * range of values that can fit in an 8, 16 or 32 bit signed value can be
  87  * encoded as integers to save space */
  88 int rdbTryIntegerEncoding(char *s, size_t len, unsigned char *enc) {
  89     long long value;
  90     char *endptr, buf[32];
  91
  92     /* Check if it's possible to encode this value as a number */
  93     value = strtoll(s, &endptr, 10);
  94     if (endptr[0] != '\0') return 0;
  95     ll2string(buf,32,value);
  96
  97     /* If the number converted back into a string is not identical
  98      * then it's not possible to encode the string as integer */
  99     if (strlen(buf) != len || memcmp(buf,s,len)) return 0;
 100
 101     return rdbEncodeInteger(value,enc);
 102 }
 103
 104 int rdbSaveLzfStringObject(FILE *fp, unsigned char *s, size_t len) {
 105     size_t comprlen, outlen;
 106     unsigned char byte;
 107     int n, nwritten = 0;
 108     void *out;
 109
 110     /* We require at least four bytes compression for this to be worth it */
 111     if (len <= 4) return 0;
 112     outlen = len-4;
 113     if ((out = zmalloc(outlen+1)) == NULL) return 0;
 114     comprlen = lzf_compress(s, len, out, outlen);
 115     if (comprlen == 0) {
 116         zfree(out);
 117         return 0;
 118     }
 119     /* Data compressed! Let's save it on disk */
 120     byte = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_LZF;
 121     if ((n = rdbWriteRaw(fp,&byte,1)) == -1) goto writeerr;
 122     nwritten += n;
 123
 124     if ((n = rdbSaveLen(fp,comprlen)) == -1) goto writeerr;
 125     nwritten += n;
 126
 127     if ((n = rdbSaveLen(fp,len)) == -1) goto writeerr;
 128     nwritten += n;
 129
 130     if ((n = rdbWriteRaw(fp,out,comprlen)) == -1) goto writeerr;
 131     nwritten += n;
 132
 133     zfree(out);
 134     return nwritten;
 135
 136 writeerr:
 137     zfree(out);
 138     return -1;
 139 }
 140
 141 /* Save a string objet as [len][data] on disk. If the object is a string
 142  * representation of an integer value we try to save it in a special form */
 143 int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) {
 144     int enclen;
 145     int n, nwritten = 0;
 146
 147     /* Try integer encoding */
 148     if (len <= 11) {
 149         unsigned char buf[5];
 150         if ((enclen = rdbTryIntegerEncoding((char*)s,len,buf)) > 0) {
 151             if (rdbWriteRaw(fp,buf,enclen) == -1) return -1;
 152             return enclen;
 153         }
 154     }
 155
 156     /* Try LZF compression - under 20 bytes it's unable to compress even
 157      * aaaaaaaaaaaaaaaaaa so skip it */
 158     if (server.rdbcompression && len > 20) {
 159         n = rdbSaveLzfStringObject(fp,s,len);
 160         if (n == -1) return -1;
 161         if (n > 0) return n;
 162         /* Return value of 0 means data can't be compressed, save the old way */
 163     }
 164
 165     /* Store verbatim */
 166     if ((n = rdbSaveLen(fp,len)) == -1) return -1;
 167     nwritten += n;
 168     if (len > 0) {
 169         if (rdbWriteRaw(fp,s,len) == -1) return -1;
 170         nwritten += len;
 171     }
 172     return nwritten;
 173 }
 174
 175 /* Save a long long value as either an encoded string or a string. */
 176 int rdbSaveLongLongAsStringObject(FILE *fp, long long value) {
 177     unsigned char buf[32];
 178     int n, nwritten = 0;
 179     int enclen = rdbEncodeInteger(value,buf);
 180     if (enclen > 0) {
 181         return rdbWriteRaw(fp,buf,enclen);
 182     } else {
 183         /* Encode as string */
 184         enclen = ll2string((char*)buf,32,value);
 185         redisAssert(enclen < 32);
 186         if ((n = rdbSaveLen(fp,enclen)) == -1) return -1;
 187         nwritten += n;
 188         if ((n = rdbWriteRaw(fp,buf,enclen)) == -1) return -1;
 189         nwritten += n;
 190     }
 191     return nwritten;
 192 }
 193
 194 /* Like rdbSaveStringObjectRaw() but handle encoded objects */
 195 int rdbSaveStringObject(FILE *fp, robj *obj) {
 196     /* Avoid to decode the object, then encode it again, if the
 197      * object is alrady integer encoded. */
 198     if (obj->encoding == REDIS_ENCODING_INT) {
 199         return rdbSaveLongLongAsStringObject(fp,(long)obj->ptr);
 200     } else {
 201         redisAssert(obj->encoding == REDIS_ENCODING_RAW);
 202         return rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr));
 203     }
 204 }
 205
 206 /* Save a double value. Doubles are saved as strings prefixed by an unsigned
 207  * 8 bit integer specifing the length of the representation.
 208  * This 8 bit integer has special values in order to specify the following
 209  * conditions:
 210  * 253: not a number
 211  * 254: + inf
 212  * 255: - inf
 213  */
 214 int rdbSaveDoubleValue(FILE *fp, double val) {
 215     unsigned char buf[128];
 216     int len;
 217
 218     if (isnan(val)) {
 219         buf[0] = 253;
 220         len = 1;
 221     } else if (!isfinite(val)) {
 222         len = 1;
 223         buf[0] = (val < 0) ? 255 : 254;
 224     } else {
 225 #if (DBL_MANT_DIG >= 52) && (LLONG_MAX == 0x7fffffffffffffffLL)
 226         /* Check if the float is in a safe range to be casted into a
 227          * long long. We are assuming that long long is 64 bit here.
 228          * Also we are assuming that there are no implementations around where
 229          * double has precision < 52 bit.
 230          *
 231          * Under this assumptions we test if a double is inside an interval
 232          * where casting to long long is safe. Then using two castings we
 233          * make sure the decimal part is zero. If all this is true we use
 234          * integer printing function that is much faster. */
 235         double min = -4503599627370495; /* (2^52)-1 */
 236         double max = 4503599627370496; /* -(2^52) */
 237         if (val > min && val < max && val == ((double)((long long)val)))
 238             ll2string((char*)buf+1,sizeof(buf),(long long)val);
 239         else
 240 #endif
 241             snprintf((char*)buf+1,sizeof(buf)-1,"%.17g",val);
 242         buf[0] = strlen((char*)buf+1);
 243         len = buf[0]+1;
 244     }
 245     return rdbWriteRaw(fp,buf,len);
 246 }
 247
 248 /* Save a Redis object. Returns -1 on error, 0 on success. */
 249 int rdbSaveObject(FILE *fp, robj *o) {
 250     int n, nwritten = 0;
 251
 252     if (o->type == REDIS_STRING) {
 253         /* Save a string value */
 254         if ((n = rdbSaveStringObject(fp,o)) == -1) return -1;
 255         nwritten += n;
 256     } else if (o->type == REDIS_LIST) {
 257         /* Save a list value */
 258         if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 259             size_t l = ziplistBlobLen((unsigned char*)o->ptr);
 260
 261             if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;
 262             nwritten += n;
 263         } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 264             list *list = o->ptr;
 265             listIter li;
 266             listNode *ln;
 267
 268             if ((n = rdbSaveLen(fp,listLength(list))) == -1) return -1;
 269             nwritten += n;
 270
 271             listRewind(list,&li);
 272             while((ln = listNext(&li))) {
 273                 robj *eleobj = listNodeValue(ln);
 274                 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 275                 nwritten += n;
 276             }
 277         } else {
 278             redisPanic("Unknown list encoding");
 279         }
 280     } else if (o->type == REDIS_SET) {
 281         /* Save a set value */
 282         if (o->encoding == REDIS_ENCODING_HT) {
 283             dict *set = o->ptr;
 284             dictIterator *di = dictGetIterator(set);
 285             dictEntry *de;
 286
 287             if ((n = rdbSaveLen(fp,dictSize(set))) == -1) return -1;
 288             nwritten += n;
 289
 290             while((de = dictNext(di)) != NULL) {
 291                 robj *eleobj = dictGetEntryKey(de);
 292                 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 293                 nwritten += n;
 294             }
 295             dictReleaseIterator(di);
 296         } else if (o->encoding == REDIS_ENCODING_INTSET) {
 297             size_t l = intsetBlobLen((intset*)o->ptr);
 298
 299             if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;
 300             nwritten += n;
 301         } else {
 302             redisPanic("Unknown set encoding");
 303         }
 304     } else if (o->type == REDIS_ZSET) {
 305         /* Save a sorted set value */
 306         if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 307             size_t l = ziplistBlobLen((unsigned char*)o->ptr);
 308
 309             if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;
 310             nwritten += n;
 311         } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
 312             zset *zs = o->ptr;
 313             dictIterator *di = dictGetIterator(zs->dict);
 314             dictEntry *de;
 315
 316             if ((n = rdbSaveLen(fp,dictSize(zs->dict))) == -1) return -1;
 317             nwritten += n;
 318
 319             while((de = dictNext(di)) != NULL) {
 320                 robj *eleobj = dictGetEntryKey(de);
 321                 double *score = dictGetEntryVal(de);
 322
 323                 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 324                 nwritten += n;
 325                 if ((n = rdbSaveDoubleValue(fp,*score)) == -1) return -1;
 326                 nwritten += n;
 327             }
 328             dictReleaseIterator(di);
 329         } else {
 330             redisPanic("Unknown sorted set encoding");
 331         }
 332     } else if (o->type == REDIS_HASH) {
 333         /* Save a hash value */
 334         if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 335             size_t l = zipmapBlobLen((unsigned char*)o->ptr);
 336
 337             if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;
 338             nwritten += n;
 339         } else {
 340             dictIterator *di = dictGetIterator(o->ptr);
 341             dictEntry *de;
 342
 343             if ((n = rdbSaveLen(fp,dictSize((dict*)o->ptr))) == -1) return -1;
 344             nwritten += n;
 345
 346             while((de = dictNext(di)) != NULL) {
 347                 robj *key = dictGetEntryKey(de);
 348                 robj *val = dictGetEntryVal(de);
 349
 350                 if ((n = rdbSaveStringObject(fp,key)) == -1) return -1;
 351                 nwritten += n;
 352                 if ((n = rdbSaveStringObject(fp,val)) == -1) return -1;
 353                 nwritten += n;
 354             }
 355             dictReleaseIterator(di);
 356         }
 357     } else {
 358         redisPanic("Unknown object type");
 359     }
 360     return nwritten;
 361 }
 362
 363 /* Return the length the object will have on disk if saved with
 364  * the rdbSaveObject() function. Currently we use a trick to get
 365  * this length with very little changes to the code. In the future
 366  * we could switch to a faster solution. */
 367 off_t rdbSavedObjectLen(robj *o) {
 368     int len = rdbSaveObject(NULL,o);
 369     redisAssert(len != -1);
 370     return len;
 371 }
 372
 373 /* Save a key-value pair, with expire time, type, key, value.
 374  * On error -1 is returned.
 375  * On success if the key was actaully saved 1 is returned, otherwise 0
 376  * is returned (the key was already expired). */
 377 int rdbSaveKeyValuePair(FILE *fp, robj *key, robj *val,
 378                         time_t expiretime, time_t now)
 379 {
 380     int vtype;
 381
 382     /* Save the expire time */
 383     if (expiretime != -1) {
 384         /* If this key is already expired skip it */
 385         if (expiretime < now) return 0;
 386         if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) return -1;
 387         if (rdbSaveTime(fp,expiretime) == -1) return -1;
 388     }
 389     /* Fix the object type if needed, to support saving zipmaps, ziplists,
 390      * and intsets, directly as blobs of bytes: they are already serialized. */
 391     vtype = val->type;
 392     if (vtype == REDIS_HASH && val->encoding == REDIS_ENCODING_ZIPMAP)
 393         vtype = REDIS_HASH_ZIPMAP;
 394     else if (vtype == REDIS_LIST && val->encoding == REDIS_ENCODING_ZIPLIST)
 395         vtype = REDIS_LIST_ZIPLIST;
 396     else if (vtype == REDIS_SET && val->encoding == REDIS_ENCODING_INTSET)
 397         vtype = REDIS_SET_INTSET;
 398     else if (vtype == REDIS_ZSET && val->encoding == REDIS_ENCODING_ZIPLIST)
 399         vtype = REDIS_ZSET_ZIPLIST;
 400     /* Save type, key, value */
 401     if (rdbSaveType(fp,vtype) == -1) return -1;
 402     if (rdbSaveStringObject(fp,key) == -1) return -1;
 403     if (rdbSaveObject(fp,val) == -1) return -1;
 404     return 1;
 405 }
 406
 407 /* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
 408 int rdbSave(char *filename) {
 409     dictIterator *di = NULL;
 410     dictEntry *de;
 411     FILE *fp;
 412     char tmpfile[256];
 413     int j;
 414     time_t now = time(NULL);
 415
 416     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
 417     fp = fopen(tmpfile,"w");
 418     if (!fp) {
 419         redisLog(REDIS_WARNING, "Failed opening .rdb for saving: %s",
 420             strerror(errno));
 421         return REDIS_ERR;
 422     }
 423     if (fwrite("REDIS0002",9,1,fp) == 0) goto werr;
 424     for (j = 0; j < server.dbnum; j++) {
 425         redisDb *db = server.db+j;
 426         dict *d = db->dict;
 427         if (dictSize(d) == 0) continue;
 428         di = dictGetSafeIterator(d);
 429         if (!di) {
 430             fclose(fp);
 431             return REDIS_ERR;
 432         }
 433
 434         /* Write the SELECT DB opcode */
 435         if (rdbSaveType(fp,REDIS_SELECTDB) == -1) goto werr;
 436         if (rdbSaveLen(fp,j) == -1) goto werr;
 437
 438         /* Iterate this DB writing every entry */
 439         while((de = dictNext(di)) != NULL) {
 440             sds keystr = dictGetEntryKey(de);
 441             robj key, *o = dictGetEntryVal(de);
 442             time_t expire;
 443
 444             initStaticStringObject(key,keystr);
 445             expire = getExpire(db,&key);
 446             if (rdbSaveKeyValuePair(fp,&key,o,expire,now) == -1) goto werr;
 447         }
 448         dictReleaseIterator(di);
 449     }
 450     /* EOF opcode */
 451     if (rdbSaveType(fp,REDIS_EOF) == -1) goto werr;
 452
 453     /* Make sure data will not remain on the OS's output buffers */
 454     fflush(fp);
 455     fsync(fileno(fp));
 456     fclose(fp);
 457
 458     /* Use RENAME to make sure the DB file is changed atomically only
 459      * if the generate DB file is ok. */
 460     if (rename(tmpfile,filename) == -1) {
 461         redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
 462         unlink(tmpfile);
 463         return REDIS_ERR;
 464     }
 465     redisLog(REDIS_NOTICE,"DB saved on disk");
 466     server.dirty = 0;
 467     server.lastsave = time(NULL);
 468     return REDIS_OK;
 469
 470 werr:
 471     fclose(fp);
 472     unlink(tmpfile);
 473     redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
 474     if (di) dictReleaseIterator(di);
 475     return REDIS_ERR;
 476 }
 477
 478 int rdbSaveBackground(char *filename) {
 479     pid_t childpid;
 480     long long start;
 481
 482     if (server.bgsavechildpid != -1) return REDIS_ERR;
 483
 484     server.dirty_before_bgsave = server.dirty;
 485
 486     start = ustime();
 487     if ((childpid = fork()) == 0) {
 488         int retval;
 489
 490         /* Child */
 491         if (server.ipfd > 0) close(server.ipfd);
 492         if (server.sofd > 0) close(server.sofd);
 493         retval = rdbSave(filename);
 494         _exit((retval == REDIS_OK) ? 0 : 1);
 495     } else {
 496         /* Parent */
 497         server.stat_fork_time = ustime()-start;
 498         if (childpid == -1) {
 499             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
 500                 strerror(errno));
 501             return REDIS_ERR;
 502         }
 503         redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
 504         server.bgsavechildpid = childpid;
 505         updateDictResizePolicy();
 506         return REDIS_OK;
 507     }
 508     return REDIS_OK; /* unreached */
 509 }
 510
 511 void rdbRemoveTempFile(pid_t childpid) {
 512     char tmpfile[256];
 513
 514     snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
 515     unlink(tmpfile);
 516 }
 517
 518 int rdbLoadType(FILE *fp) {
 519     unsigned char type;
 520     if (fread(&type,1,1,fp) == 0) return -1;
 521     return type;
 522 }
 523
 524 time_t rdbLoadTime(FILE *fp) {
 525     int32_t t32;
 526     if (fread(&t32,4,1,fp) == 0) return -1;
 527     return (time_t) t32;
 528 }
 529
 530 /* Load an encoded length from the DB, see the REDIS_RDB_* defines on the top
 531  * of this file for a description of how this are stored on disk.
 532  *
 533  * isencoded is set to 1 if the readed length is not actually a length but
 534  * an "encoding type", check the above comments for more info */
 535 uint32_t rdbLoadLen(FILE *fp, int *isencoded) {
 536     unsigned char buf[2];
 537     uint32_t len;
 538     int type;
 539
 540     if (isencoded) *isencoded = 0;
 541     if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR;
 542     type = (buf[0]&0xC0)>>6;
 543     if (type == REDIS_RDB_6BITLEN) {
 544         /* Read a 6 bit len */
 545         return buf[0]&0x3F;
 546     } else if (type == REDIS_RDB_ENCVAL) {
 547         /* Read a 6 bit len encoding type */
 548         if (isencoded) *isencoded = 1;
 549         return buf[0]&0x3F;
 550     } else if (type == REDIS_RDB_14BITLEN) {
 551         /* Read a 14 bit len */
 552         if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR;
 553         return ((buf[0]&0x3F)<<8)|buf[1];
 554     } else {
 555         /* Read a 32 bit len */
 556         if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR;
 557         return ntohl(len);
 558     }
 559 }
 560
 561 /* Load an integer-encoded object from file 'fp', with the specified
 562  * encoding type 'enctype'. If encode is true the function may return
 563  * an integer-encoded object as reply, otherwise the returned object
 564  * will always be encoded as a raw string. */
 565 robj *rdbLoadIntegerObject(FILE *fp, int enctype, int encode) {
 566     unsigned char enc[4];
 567     long long val;
 568
 569     if (enctype == REDIS_RDB_ENC_INT8) {
 570         if (fread(enc,1,1,fp) == 0) return NULL;
 571         val = (signed char)enc[0];
 572     } else if (enctype == REDIS_RDB_ENC_INT16) {
 573         uint16_t v;
 574         if (fread(enc,2,1,fp) == 0) return NULL;
 575         v = enc[0]|(enc[1]<<8);
 576         val = (int16_t)v;
 577     } else if (enctype == REDIS_RDB_ENC_INT32) {
 578         uint32_t v;
 579         if (fread(enc,4,1,fp) == 0) return NULL;
 580         v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
 581         val = (int32_t)v;
 582     } else {
 583         val = 0; /* anti-warning */
 584         redisPanic("Unknown RDB integer encoding type");
 585     }
 586     if (encode)
 587         return createStringObjectFromLongLong(val);
 588     else
 589         return createObject(REDIS_STRING,sdsfromlonglong(val));
 590 }
 591
 592 robj *rdbLoadLzfStringObject(FILE*fp) {
 593     unsigned int len, clen;
 594     unsigned char *c = NULL;
 595     sds val = NULL;
 596
 597     if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 598     if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 599     if ((c = zmalloc(clen)) == NULL) goto err;
 600     if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
 601     if (fread(c,clen,1,fp) == 0) goto err;
 602     if (lzf_decompress(c,clen,val,len) == 0) goto err;
 603     zfree(c);
 604     return createObject(REDIS_STRING,val);
 605 err:
 606     zfree(c);
 607     sdsfree(val);
 608     return NULL;
 609 }
 610
 611 robj *rdbGenericLoadStringObject(FILE*fp, int encode) {
 612     int isencoded;
 613     uint32_t len;
 614     sds val;
 615
 616     len = rdbLoadLen(fp,&isencoded);
 617     if (isencoded) {
 618         switch(len) {
 619         case REDIS_RDB_ENC_INT8:
 620         case REDIS_RDB_ENC_INT16:
 621         case REDIS_RDB_ENC_INT32:
 622             return rdbLoadIntegerObject(fp,len,encode);
 623         case REDIS_RDB_ENC_LZF:
 624             return rdbLoadLzfStringObject(fp);
 625         default:
 626             redisPanic("Unknown RDB encoding type");
 627         }
 628     }
 629
 630     if (len == REDIS_RDB_LENERR) return NULL;
 631     val = sdsnewlen(NULL,len);
 632     if (len && fread(val,len,1,fp) == 0) {
 633         sdsfree(val);
 634         return NULL;
 635     }
 636     return createObject(REDIS_STRING,val);
 637 }
 638
 639 robj *rdbLoadStringObject(FILE *fp) {
 640     return rdbGenericLoadStringObject(fp,0);
 641 }
 642
 643 robj *rdbLoadEncodedStringObject(FILE *fp) {
 644     return rdbGenericLoadStringObject(fp,1);
 645 }
 646
 647 /* For information about double serialization check rdbSaveDoubleValue() */
 648 int rdbLoadDoubleValue(FILE *fp, double *val) {
 649     char buf[128];
 650     unsigned char len;
 651
 652     if (fread(&len,1,1,fp) == 0) return -1;
 653     switch(len) {
 654     case 255: *val = R_NegInf; return 0;
 655     case 254: *val = R_PosInf; return 0;
 656     case 253: *val = R_Nan; return 0;
 657     default:
 658         if (fread(buf,len,1,fp) == 0) return -1;
 659         buf[len] = '\0';
 660         sscanf(buf, "%lg", val);
 661         return 0;
 662     }
 663 }
 664
 665 /* Load a Redis object of the specified type from the specified file.
 666  * On success a newly allocated object is returned, otherwise NULL. */
 667 robj *rdbLoadObject(int type, FILE *fp) {
 668     robj *o, *ele, *dec;
 669     size_t len;
 670     unsigned int i;
 671
 672     redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
 673     if (type == REDIS_STRING) {
 674         /* Read string value */
 675         if ((o = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 676         o = tryObjectEncoding(o);
 677     } else if (type == REDIS_LIST) {
 678         /* Read list value */
 679         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 680
 681         /* Use a real list when there are too many entries */
 682         if (len > server.list_max_ziplist_entries) {
 683             o = createListObject();
 684         } else {
 685             o = createZiplistObject();
 686         }
 687
 688         /* Load every single element of the list */
 689         while(len--) {
 690             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 691
 692             /* If we are using a ziplist and the value is too big, convert
 693              * the object to a real list. */
 694             if (o->encoding == REDIS_ENCODING_ZIPLIST &&
 695                 ele->encoding == REDIS_ENCODING_RAW &&
 696                 sdslen(ele->ptr) > server.list_max_ziplist_value)
 697                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 698
 699             if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 700                 dec = getDecodedObject(ele);
 701                 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
 702                 decrRefCount(dec);
 703                 decrRefCount(ele);
 704             } else {
 705                 ele = tryObjectEncoding(ele);
 706                 listAddNodeTail(o->ptr,ele);
 707             }
 708         }
 709     } else if (type == REDIS_SET) {
 710         /* Read list/set value */
 711         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 712
 713         /* Use a regular set when there are too many entries. */
 714         if (len > server.set_max_intset_entries) {
 715             o = createSetObject();
 716             /* It's faster to expand the dict to the right size asap in order
 717              * to avoid rehashing */
 718             if (len > DICT_HT_INITIAL_SIZE)
 719                 dictExpand(o->ptr,len);
 720         } else {
 721             o = createIntsetObject();
 722         }
 723
 724         /* Load every single element of the list/set */
 725         for (i = 0; i < len; i++) {
 726             long long llval;
 727             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 728             ele = tryObjectEncoding(ele);
 729
 730             if (o->encoding == REDIS_ENCODING_INTSET) {
 731                 /* Fetch integer value from element */
 732                 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
 733                     o->ptr = intsetAdd(o->ptr,llval,NULL);
 734                 } else {
 735                     setTypeConvert(o,REDIS_ENCODING_HT);
 736                     dictExpand(o->ptr,len);
 737                 }
 738             }
 739
 740             /* This will also be called when the set was just converted
 741              * to regular hashtable encoded set */
 742             if (o->encoding == REDIS_ENCODING_HT) {
 743                 dictAdd((dict*)o->ptr,ele,NULL);
 744             } else {
 745                 decrRefCount(ele);
 746             }
 747         }
 748     } else if (type == REDIS_ZSET) {
 749         /* Read list/set value */
 750         size_t zsetlen;
 751         size_t maxelelen = 0;
 752         zset *zs;
 753
 754         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 755         o = createZsetObject();
 756         zs = o->ptr;
 757
 758         /* Load every single element of the list/set */
 759         while(zsetlen--) {
 760             robj *ele;
 761             double score;
 762             zskiplistNode *znode;
 763
 764             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 765             ele = tryObjectEncoding(ele);
 766             if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;
 767
 768             /* Don't care about integer-encoded strings. */
 769             if (ele->encoding == REDIS_ENCODING_RAW &&
 770                 sdslen(ele->ptr) > maxelelen)
 771                     maxelelen = sdslen(ele->ptr);
 772
 773             znode = zslInsert(zs->zsl,score,ele);
 774             dictAdd(zs->dict,ele,&znode->score);
 775             incrRefCount(ele); /* added to skiplist */
 776         }
 777
 778         /* Convert *after* loading, since sorted sets are not stored ordered. */
 779         if (zsetLength(o) <= server.zset_max_ziplist_entries &&
 780             maxelelen <= server.zset_max_ziplist_value)
 781                 zsetConvert(o,REDIS_ENCODING_ZIPLIST);
 782     } else if (type == REDIS_HASH) {
 783         size_t hashlen;
 784
 785         if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 786         o = createHashObject();
 787         /* Too many entries? Use an hash table. */
 788         if (hashlen > server.hash_max_zipmap_entries)
 789             convertToRealHash(o);
 790         /* Load every key/value, then set it into the zipmap or hash
 791          * table, as needed. */
 792         while(hashlen--) {
 793             robj *key, *val;
 794
 795             if ((key = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 796             if ((val = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 797             /* If we are using a zipmap and there are too big values
 798              * the object is converted to real hash table encoding. */
 799             if (o->encoding != REDIS_ENCODING_HT &&
 800                ((key->encoding == REDIS_ENCODING_RAW &&
 801                 sdslen(key->ptr) > server.hash_max_zipmap_value) ||
 802                 (val->encoding == REDIS_ENCODING_RAW &&
 803                 sdslen(val->ptr) > server.hash_max_zipmap_value)))
 804             {
 805                     convertToRealHash(o);
 806             }
 807
 808             if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 809                 unsigned char *zm = o->ptr;
 810                 robj *deckey, *decval;
 811
 812                 /* We need raw string objects to add them to the zipmap */
 813                 deckey = getDecodedObject(key);
 814                 decval = getDecodedObject(val);
 815                 zm = zipmapSet(zm,deckey->ptr,sdslen(deckey->ptr),
 816                                   decval->ptr,sdslen(decval->ptr),NULL);
 817                 o->ptr = zm;
 818                 decrRefCount(deckey);
 819                 decrRefCount(decval);
 820                 decrRefCount(key);
 821                 decrRefCount(val);
 822             } else {
 823                 key = tryObjectEncoding(key);
 824                 val = tryObjectEncoding(val);
 825                 dictAdd((dict*)o->ptr,key,val);
 826             }
 827         }
 828     } else if (type == REDIS_HASH_ZIPMAP ||
 829                type == REDIS_LIST_ZIPLIST ||
 830                type == REDIS_SET_INTSET ||
 831                type == REDIS_ZSET_ZIPLIST)
 832     {
 833         robj *aux = rdbLoadStringObject(fp);
 834
 835         if (aux == NULL) return NULL;
 836         o = createObject(REDIS_STRING,NULL); /* string is just placeholder */
 837         o->ptr = zmalloc(sdslen(aux->ptr));
 838         memcpy(o->ptr,aux->ptr,sdslen(aux->ptr));
 839         decrRefCount(aux);
 840
 841         /* Fix the object encoding, and make sure to convert the encoded
 842          * data type into the base type if accordingly to the current
 843          * configuration there are too many elements in the encoded data
 844          * type. Note that we only check the length and not max element
 845          * size as this is an O(N) scan. Eventually everything will get
 846          * converted. */
 847         switch(type) {
 848             case REDIS_HASH_ZIPMAP:
 849                 o->type = REDIS_HASH;
 850                 o->encoding = REDIS_ENCODING_ZIPMAP;
 851                 if (zipmapLen(o->ptr) > server.hash_max_zipmap_entries)
 852                     convertToRealHash(o);
 853                 break;
 854             case REDIS_LIST_ZIPLIST:
 855                 o->type = REDIS_LIST;
 856                 o->encoding = REDIS_ENCODING_ZIPLIST;
 857                 if (ziplistLen(o->ptr) > server.list_max_ziplist_entries)
 858                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 859                 break;
 860             case REDIS_SET_INTSET:
 861                 o->type = REDIS_SET;
 862                 o->encoding = REDIS_ENCODING_INTSET;
 863                 if (intsetLen(o->ptr) > server.set_max_intset_entries)
 864                     setTypeConvert(o,REDIS_ENCODING_HT);
 865                 break;
 866             case REDIS_ZSET_ZIPLIST:
 867                 o->type = REDIS_ZSET;
 868                 o->encoding = REDIS_ENCODING_ZIPLIST;
 869                 if (zsetLength(o) > server.zset_max_ziplist_entries)
 870                     zsetConvert(o,REDIS_ENCODING_SKIPLIST);
 871                 break;
 872             default:
 873                 redisPanic("Unknown encoding");
 874                 break;
 875         }
 876     } else {
 877         redisPanic("Unknown object type");
 878     }
 879     return o;
 880 }
 881
 882 /* Mark that we are loading in the global state and setup the fields
 883  * needed to provide loading stats. */
 884 void startLoading(FILE *fp) {
 885     struct stat sb;
 886
 887     /* Load the DB */
 888     server.loading = 1;
 889     server.loading_start_time = time(NULL);
 890     if (fstat(fileno(fp), &sb) == -1) {
 891         server.loading_total_bytes = 1; /* just to avoid division by zero */
 892     } else {
 893         server.loading_total_bytes = sb.st_size;
 894     }
 895 }
 896
 897 /* Refresh the loading progress info */
 898 void loadingProgress(off_t pos) {
 899     server.loading_loaded_bytes = pos;
 900 }
 901
 902 /* Loading finished */
 903 void stopLoading(void) {
 904     server.loading = 0;
 905 }
 906
 907 int rdbLoad(char *filename) {
 908     FILE *fp;
 909     uint32_t dbid;
 910     int type, rdbver;
 911     redisDb *db = server.db+0;
 912     char buf[1024];
 913     time_t expiretime, now = time(NULL);
 914     long loops = 0;
 915
 916     fp = fopen(filename,"r");
 917     if (!fp) return REDIS_ERR;
 918     if (fread(buf,9,1,fp) == 0) goto eoferr;
 919     buf[9] = '\0';
 920     if (memcmp(buf,"REDIS",5) != 0) {
 921         fclose(fp);
 922         redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
 923         return REDIS_ERR;
 924     }
 925     rdbver = atoi(buf+5);
 926     if (rdbver < 1 || rdbver > 2) {
 927         fclose(fp);
 928         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
 929         return REDIS_ERR;
 930     }
 931
 932     startLoading(fp);
 933     while(1) {
 934         robj *key, *val;
 935         expiretime = -1;
 936
 937         /* Serve the clients from time to time */
 938         if (!(loops++ % 1000)) {
 939             loadingProgress(ftello(fp));
 940             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 941         }
 942
 943         /* Read type. */
 944         if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 945         if (type == REDIS_EXPIRETIME) {
 946             if ((expiretime = rdbLoadTime(fp)) == -1) goto eoferr;
 947             /* We read the time so we need to read the object type again */
 948             if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 949         }
 950         if (type == REDIS_EOF) break;
 951         /* Handle SELECT DB opcode as a special case */
 952         if (type == REDIS_SELECTDB) {
 953             if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR)
 954                 goto eoferr;
 955             if (dbid >= (unsigned)server.dbnum) {
 956                 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
 957                 exit(1);
 958             }
 959             db = server.db+dbid;
 960             continue;
 961         }
 962         /* Read key */
 963         if ((key = rdbLoadStringObject(fp)) == NULL) goto eoferr;
 964         /* Read value */
 965         if ((val = rdbLoadObject(type,fp)) == NULL) goto eoferr;
 966         /* Check if the key already expired */
 967         if (expiretime != -1 && expiretime < now) {
 968             decrRefCount(key);
 969             decrRefCount(val);
 970             continue;
 971         }
 972         /* Add the new object in the hash table */
 973         dbAdd(db,key,val);
 974
 975         /* Set the expire time if needed */
 976         if (expiretime != -1) setExpire(db,key,expiretime);
 977
 978         decrRefCount(key);
 979     }
 980     fclose(fp);
 981     stopLoading();
 982     return REDIS_OK;
 983
 984 eoferr: /* unexpected end of file is handled here with a fatal exit */
 985     redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
 986     exit(1);
 987     return REDIS_ERR; /* Just to avoid warning */
 988 }
 989
 990 /* A background saving child (BGSAVE) terminated its work. Handle this. */
 991 void backgroundSaveDoneHandler(int exitcode, int bysignal) {
 992     if (!bysignal && exitcode == 0) {
 993         redisLog(REDIS_NOTICE,
 994             "Background saving terminated with success");
 995         server.dirty = server.dirty - server.dirty_before_bgsave;
 996         server.lastsave = time(NULL);
 997     } else if (!bysignal && exitcode != 0) {
 998         redisLog(REDIS_WARNING, "Background saving error");
 999     } else {
1000         redisLog(REDIS_WARNING,
1001             "Background saving terminated by signal %d", bysignal);
1002         rdbRemoveTempFile(server.bgsavechildpid);
1003     }
1004     server.bgsavechildpid = -1;
1005     /* Possibly there are slaves waiting for a BGSAVE in order to be served
1006      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
1007     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
1008 }
1009
1010 void saveCommand(redisClient *c) {
1011     if (server.bgsavechildpid != -1) {
1012         addReplyError(c,"Background save already in progress");
1013         return;
1014     }
1015     if (rdbSave(server.dbfilename) == REDIS_OK) {
1016         addReply(c,shared.ok);
1017     } else {
1018         addReply(c,shared.err);
1019     }
1020 }
1021
1022 void bgsaveCommand(redisClient *c) {
1023     if (server.bgsavechildpid != -1) {
1024         addReplyError(c,"Background save already in progress");
1025     } else if (server.bgrewritechildpid != -1) {
1026         addReplyError(c,"Can't BGSAVE while AOF log rewriting is in progress");
1027     } else if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
1028         addReplyStatus(c,"Background saving started");
1029     } else {
1030         addReply(c,shared.err);
1031     }
1032 }