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. */
 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 set value */
 306         zset *zs = o->ptr;
 307         dictIterator *di = dictGetIterator(zs->dict);
 308         dictEntry *de;
 309
 310         if ((n = rdbSaveLen(fp,dictSize(zs->dict))) == -1) return -1;
 311         nwritten += n;
 312
 313         while((de = dictNext(di)) != NULL) {
 314             robj *eleobj = dictGetEntryKey(de);
 315             double *score = dictGetEntryVal(de);
 316
 317             if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 318             nwritten += n;
 319             if ((n = rdbSaveDoubleValue(fp,*score)) == -1) return -1;
 320             nwritten += n;
 321         }
 322         dictReleaseIterator(di);
 323     } else if (o->type == REDIS_HASH) {
 324         /* Save a hash value */
 325         if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 326             size_t l = zipmapBlobLen((unsigned char*)o->ptr);
 327
 328             if ((n = rdbSaveRawString(fp,o->ptr,l)) == -1) return -1;
 329             nwritten += n;
 330         } else {
 331             dictIterator *di = dictGetIterator(o->ptr);
 332             dictEntry *de;
 333
 334             if ((n = rdbSaveLen(fp,dictSize((dict*)o->ptr))) == -1) return -1;
 335             nwritten += n;
 336
 337             while((de = dictNext(di)) != NULL) {
 338                 robj *key = dictGetEntryKey(de);
 339                 robj *val = dictGetEntryVal(de);
 340
 341                 if ((n = rdbSaveStringObject(fp,key)) == -1) return -1;
 342                 nwritten += n;
 343                 if ((n = rdbSaveStringObject(fp,val)) == -1) return -1;
 344                 nwritten += n;
 345             }
 346             dictReleaseIterator(di);
 347         }
 348     } else {
 349         redisPanic("Unknown object type");
 350     }
 351     return nwritten;
 352 }
 353
 354 /* Return the length the object will have on disk if saved with
 355  * the rdbSaveObject() function. Currently we use a trick to get
 356  * this length with very little changes to the code. In the future
 357  * we could switch to a faster solution. */
 358 off_t rdbSavedObjectLen(robj *o) {
 359     int len = rdbSaveObject(NULL,o);
 360     redisAssert(len != -1);
 361     return len;
 362 }
 363
 364 /* Save a key-value pair, with expire time, type, key, value.
 365  * On error -1 is returned.
 366  * On success if the key was actaully saved 1 is returned, otherwise 0
 367  * is returned (the key was already expired). */
 368 int rdbSaveKeyValuePair(FILE *fp, robj *key, robj *val,
 369                         time_t expiretime, time_t now)
 370 {
 371     int vtype;
 372
 373     /* Save the expire time */
 374     if (expiretime != -1) {
 375         /* If this key is already expired skip it */
 376         if (expiretime < now) return 0;
 377         if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) return -1;
 378         if (rdbSaveTime(fp,expiretime) == -1) return -1;
 379     }
 380     /* Fix the object type if needed, to support saving zipmaps, ziplists,
 381      * and intsets, directly as blobs of bytes: they are already serialized. */
 382     vtype = val->type;
 383     if (vtype == REDIS_HASH && val->encoding == REDIS_ENCODING_ZIPMAP)
 384         vtype = REDIS_HASH_ZIPMAP;
 385     else if (vtype == REDIS_LIST && val->encoding == REDIS_ENCODING_ZIPLIST)
 386         vtype = REDIS_LIST_ZIPLIST;
 387     /* Save type, key, value */
 388     if (rdbSaveType(fp,vtype) == -1) return -1;
 389     if (rdbSaveStringObject(fp,key) == -1) return -1;
 390     if (rdbSaveObject(fp,val) == -1) return -1;
 391     return 1;
 392 }
 393
 394 /* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
 395 int rdbSave(char *filename) {
 396     dictIterator *di = NULL;
 397     dictEntry *de;
 398     FILE *fp;
 399     char tmpfile[256];
 400     int j;
 401     time_t now = time(NULL);
 402
 403     if (server.ds_enabled) {
 404         cacheForcePointInTime();
 405         return dsRdbSave(filename);
 406     }
 407
 408     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
 409     fp = fopen(tmpfile,"w");
 410     if (!fp) {
 411         redisLog(REDIS_WARNING, "Failed opening .rdb for saving: %s",
 412             strerror(errno));
 413         return REDIS_ERR;
 414     }
 415     if (fwrite("REDIS0001",9,1,fp) == 0) goto werr;
 416     for (j = 0; j < server.dbnum; j++) {
 417         redisDb *db = server.db+j;
 418         dict *d = db->dict;
 419         if (dictSize(d) == 0) continue;
 420         di = dictGetIterator(d);
 421         if (!di) {
 422             fclose(fp);
 423             return REDIS_ERR;
 424         }
 425
 426         /* Write the SELECT DB opcode */
 427         if (rdbSaveType(fp,REDIS_SELECTDB) == -1) goto werr;
 428         if (rdbSaveLen(fp,j) == -1) goto werr;
 429
 430         /* Iterate this DB writing every entry */
 431         while((de = dictNext(di)) != NULL) {
 432             sds keystr = dictGetEntryKey(de);
 433             robj key, *o = dictGetEntryVal(de);
 434             time_t expire;
 435
 436             initStaticStringObject(key,keystr);
 437             expire = getExpire(db,&key);
 438             if (rdbSaveKeyValuePair(fp,&key,o,expire,now) == -1) goto werr;
 439         }
 440         dictReleaseIterator(di);
 441     }
 442     /* EOF opcode */
 443     if (rdbSaveType(fp,REDIS_EOF) == -1) goto werr;
 444
 445     /* Make sure data will not remain on the OS's output buffers */
 446     fflush(fp);
 447     fsync(fileno(fp));
 448     fclose(fp);
 449
 450     /* Use RENAME to make sure the DB file is changed atomically only
 451      * if the generate DB file is ok. */
 452     if (rename(tmpfile,filename) == -1) {
 453         redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
 454         unlink(tmpfile);
 455         return REDIS_ERR;
 456     }
 457     redisLog(REDIS_NOTICE,"DB saved on disk");
 458     server.dirty = 0;
 459     server.lastsave = time(NULL);
 460     return REDIS_OK;
 461
 462 werr:
 463     fclose(fp);
 464     unlink(tmpfile);
 465     redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
 466     if (di) dictReleaseIterator(di);
 467     return REDIS_ERR;
 468 }
 469
 470 int rdbSaveBackground(char *filename) {
 471     pid_t childpid;
 472
 473     if (server.bgsavechildpid != -1 ||
 474         server.bgsavethread != (pthread_t) -1) return REDIS_ERR;
 475
 476     server.dirty_before_bgsave = server.dirty;
 477
 478     if (server.ds_enabled) {
 479         cacheForcePointInTime();
 480         return dsRdbSaveBackground(filename);
 481     }
 482
 483     if ((childpid = fork()) == 0) {
 484         int retval;
 485
 486         /* Child */
 487         if (server.ipfd > 0) close(server.ipfd);
 488         if (server.sofd > 0) close(server.sofd);
 489         retval = rdbSave(filename);
 490         _exit((retval == REDIS_OK) ? 0 : 1);
 491     } else {
 492         /* Parent */
 493         if (childpid == -1) {
 494             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
 495                 strerror(errno));
 496             return REDIS_ERR;
 497         }
 498         redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
 499         server.bgsavechildpid = childpid;
 500         updateDictResizePolicy();
 501         return REDIS_OK;
 502     }
 503     return REDIS_OK; /* unreached */
 504 }
 505
 506 void rdbRemoveTempFile(pid_t childpid) {
 507     char tmpfile[256];
 508
 509     snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
 510     unlink(tmpfile);
 511 }
 512
 513 int rdbLoadType(FILE *fp) {
 514     unsigned char type;
 515     if (fread(&type,1,1,fp) == 0) return -1;
 516     return type;
 517 }
 518
 519 time_t rdbLoadTime(FILE *fp) {
 520     int32_t t32;
 521     if (fread(&t32,4,1,fp) == 0) return -1;
 522     return (time_t) t32;
 523 }
 524
 525 /* Load an encoded length from the DB, see the REDIS_RDB_* defines on the top
 526  * of this file for a description of how this are stored on disk.
 527  *
 528  * isencoded is set to 1 if the readed length is not actually a length but
 529  * an "encoding type", check the above comments for more info */
 530 uint32_t rdbLoadLen(FILE *fp, int *isencoded) {
 531     unsigned char buf[2];
 532     uint32_t len;
 533     int type;
 534
 535     if (isencoded) *isencoded = 0;
 536     if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR;
 537     type = (buf[0]&0xC0)>>6;
 538     if (type == REDIS_RDB_6BITLEN) {
 539         /* Read a 6 bit len */
 540         return buf[0]&0x3F;
 541     } else if (type == REDIS_RDB_ENCVAL) {
 542         /* Read a 6 bit len encoding type */
 543         if (isencoded) *isencoded = 1;
 544         return buf[0]&0x3F;
 545     } else if (type == REDIS_RDB_14BITLEN) {
 546         /* Read a 14 bit len */
 547         if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR;
 548         return ((buf[0]&0x3F)<<8)|buf[1];
 549     } else {
 550         /* Read a 32 bit len */
 551         if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR;
 552         return ntohl(len);
 553     }
 554 }
 555
 556 /* Load an integer-encoded object from file 'fp', with the specified
 557  * encoding type 'enctype'. If encode is true the function may return
 558  * an integer-encoded object as reply, otherwise the returned object
 559  * will always be encoded as a raw string. */
 560 robj *rdbLoadIntegerObject(FILE *fp, int enctype, int encode) {
 561     unsigned char enc[4];
 562     long long val;
 563
 564     if (enctype == REDIS_RDB_ENC_INT8) {
 565         if (fread(enc,1,1,fp) == 0) return NULL;
 566         val = (signed char)enc[0];
 567     } else if (enctype == REDIS_RDB_ENC_INT16) {
 568         uint16_t v;
 569         if (fread(enc,2,1,fp) == 0) return NULL;
 570         v = enc[0]|(enc[1]<<8);
 571         val = (int16_t)v;
 572     } else if (enctype == REDIS_RDB_ENC_INT32) {
 573         uint32_t v;
 574         if (fread(enc,4,1,fp) == 0) return NULL;
 575         v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
 576         val = (int32_t)v;
 577     } else {
 578         val = 0; /* anti-warning */
 579         redisPanic("Unknown RDB integer encoding type");
 580     }
 581     if (encode)
 582         return createStringObjectFromLongLong(val);
 583     else
 584         return createObject(REDIS_STRING,sdsfromlonglong(val));
 585 }
 586
 587 robj *rdbLoadLzfStringObject(FILE*fp) {
 588     unsigned int len, clen;
 589     unsigned char *c = NULL;
 590     sds val = NULL;
 591
 592     if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 593     if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 594     if ((c = zmalloc(clen)) == NULL) goto err;
 595     if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
 596     if (fread(c,clen,1,fp) == 0) goto err;
 597     if (lzf_decompress(c,clen,val,len) == 0) goto err;
 598     zfree(c);
 599     return createObject(REDIS_STRING,val);
 600 err:
 601     zfree(c);
 602     sdsfree(val);
 603     return NULL;
 604 }
 605
 606 robj *rdbGenericLoadStringObject(FILE*fp, int encode) {
 607     int isencoded;
 608     uint32_t len;
 609     sds val;
 610
 611     len = rdbLoadLen(fp,&isencoded);
 612     if (isencoded) {
 613         switch(len) {
 614         case REDIS_RDB_ENC_INT8:
 615         case REDIS_RDB_ENC_INT16:
 616         case REDIS_RDB_ENC_INT32:
 617             return rdbLoadIntegerObject(fp,len,encode);
 618         case REDIS_RDB_ENC_LZF:
 619             return rdbLoadLzfStringObject(fp);
 620         default:
 621             redisPanic("Unknown RDB encoding type");
 622         }
 623     }
 624
 625     if (len == REDIS_RDB_LENERR) return NULL;
 626     val = sdsnewlen(NULL,len);
 627     if (len && fread(val,len,1,fp) == 0) {
 628         sdsfree(val);
 629         return NULL;
 630     }
 631     return createObject(REDIS_STRING,val);
 632 }
 633
 634 robj *rdbLoadStringObject(FILE *fp) {
 635     return rdbGenericLoadStringObject(fp,0);
 636 }
 637
 638 robj *rdbLoadEncodedStringObject(FILE *fp) {
 639     return rdbGenericLoadStringObject(fp,1);
 640 }
 641
 642 /* For information about double serialization check rdbSaveDoubleValue() */
 643 int rdbLoadDoubleValue(FILE *fp, double *val) {
 644     char buf[128];
 645     unsigned char len;
 646
 647     if (fread(&len,1,1,fp) == 0) return -1;
 648     switch(len) {
 649     case 255: *val = R_NegInf; return 0;
 650     case 254: *val = R_PosInf; return 0;
 651     case 253: *val = R_Nan; return 0;
 652     default:
 653         if (fread(buf,len,1,fp) == 0) return -1;
 654         buf[len] = '\0';
 655         sscanf(buf, "%lg", val);
 656         return 0;
 657     }
 658 }
 659
 660 /* Load a Redis object of the specified type from the specified file.
 661  * On success a newly allocated object is returned, otherwise NULL. */
 662 robj *rdbLoadObject(int type, FILE *fp) {
 663     robj *o, *ele, *dec;
 664     size_t len;
 665     unsigned int i;
 666
 667     redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
 668     if (type == REDIS_STRING) {
 669         /* Read string value */
 670         if ((o = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 671         o = tryObjectEncoding(o);
 672     } else if (type == REDIS_LIST) {
 673         /* Read list value */
 674         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 675
 676         /* Use a real list when there are too many entries */
 677         if (len > server.list_max_ziplist_entries) {
 678             o = createListObject();
 679         } else {
 680             o = createZiplistObject();
 681         }
 682
 683         /* Load every single element of the list */
 684         while(len--) {
 685             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 686
 687             /* If we are using a ziplist and the value is too big, convert
 688              * the object to a real list. */
 689             if (o->encoding == REDIS_ENCODING_ZIPLIST &&
 690                 ele->encoding == REDIS_ENCODING_RAW &&
 691                 sdslen(ele->ptr) > server.list_max_ziplist_value)
 692                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 693
 694             if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 695                 dec = getDecodedObject(ele);
 696                 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
 697                 decrRefCount(dec);
 698                 decrRefCount(ele);
 699             } else {
 700                 ele = tryObjectEncoding(ele);
 701                 listAddNodeTail(o->ptr,ele);
 702             }
 703         }
 704     } else if (type == REDIS_SET) {
 705         /* Read list/set value */
 706         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 707
 708         /* Use a regular set when there are too many entries. */
 709         if (len > server.set_max_intset_entries) {
 710             o = createSetObject();
 711             /* It's faster to expand the dict to the right size asap in order
 712              * to avoid rehashing */
 713             if (len > DICT_HT_INITIAL_SIZE)
 714                 dictExpand(o->ptr,len);
 715         } else {
 716             o = createIntsetObject();
 717         }
 718
 719         /* Load every single element of the list/set */
 720         for (i = 0; i < len; i++) {
 721             long long llval;
 722             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 723             ele = tryObjectEncoding(ele);
 724
 725             if (o->encoding == REDIS_ENCODING_INTSET) {
 726                 /* Fetch integer value from element */
 727                 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
 728                     o->ptr = intsetAdd(o->ptr,llval,NULL);
 729                 } else {
 730                     setTypeConvert(o,REDIS_ENCODING_HT);
 731                     dictExpand(o->ptr,len);
 732                 }
 733             }
 734
 735             /* This will also be called when the set was just converted
 736              * to regular hashtable encoded set */
 737             if (o->encoding == REDIS_ENCODING_HT) {
 738                 dictAdd((dict*)o->ptr,ele,NULL);
 739             } else {
 740                 decrRefCount(ele);
 741             }
 742         }
 743     } else if (type == REDIS_ZSET) {
 744         /* Read list/set value */
 745         size_t zsetlen;
 746         zset *zs;
 747
 748         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 749         o = createZsetObject();
 750         zs = o->ptr;
 751         /* Load every single element of the list/set */
 752         while(zsetlen--) {
 753             robj *ele;
 754             double score;
 755             zskiplistNode *znode;
 756
 757             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 758             ele = tryObjectEncoding(ele);
 759             if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;
 760             znode = zslInsert(zs->zsl,score,ele);
 761             dictAdd(zs->dict,ele,&znode->score);
 762             incrRefCount(ele); /* added to skiplist */
 763         }
 764     } else if (type == REDIS_HASH) {
 765         size_t hashlen;
 766
 767         if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 768         o = createHashObject();
 769         /* Too many entries? Use an hash table. */
 770         if (hashlen > server.hash_max_zipmap_entries)
 771             convertToRealHash(o);
 772         /* Load every key/value, then set it into the zipmap or hash
 773          * table, as needed. */
 774         while(hashlen--) {
 775             robj *key, *val;
 776
 777             if ((key = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 778             if ((val = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 779             /* If we are using a zipmap and there are too big values
 780              * the object is converted to real hash table encoding. */
 781             if (o->encoding != REDIS_ENCODING_HT &&
 782                ((key->encoding == REDIS_ENCODING_RAW &&
 783                 sdslen(key->ptr) > server.hash_max_zipmap_value) ||
 784                 (val->encoding == REDIS_ENCODING_RAW &&
 785                 sdslen(val->ptr) > server.hash_max_zipmap_value)))
 786             {
 787                     convertToRealHash(o);
 788             }
 789
 790             if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 791                 unsigned char *zm = o->ptr;
 792                 robj *deckey, *decval;
 793
 794                 /* We need raw string objects to add them to the zipmap */
 795                 deckey = getDecodedObject(key);
 796                 decval = getDecodedObject(val);
 797                 zm = zipmapSet(zm,deckey->ptr,sdslen(deckey->ptr),
 798                                   decval->ptr,sdslen(decval->ptr),NULL);
 799                 o->ptr = zm;
 800                 decrRefCount(deckey);
 801                 decrRefCount(decval);
 802                 decrRefCount(key);
 803                 decrRefCount(val);
 804             } else {
 805                 key = tryObjectEncoding(key);
 806                 val = tryObjectEncoding(val);
 807                 dictAdd((dict*)o->ptr,key,val);
 808             }
 809         }
 810     } else if (type == REDIS_HASH_ZIPMAP ||
 811                type == REDIS_LIST_ZIPLIST ||
 812                type == REDIS_SET_INTSET)
 813     {
 814         robj *aux = rdbLoadStringObject(fp);
 815
 816         if (aux == NULL) return NULL;
 817         o = createObject(REDIS_STRING,NULL); /* string is just placeholder */
 818         o->ptr = zmalloc(sdslen(aux->ptr));
 819         memcpy(o->ptr,aux->ptr,sdslen(aux->ptr));
 820         decrRefCount(aux);
 821
 822         /* Fix the object encoding, and make sure to convert the encoded
 823          * data type into the base type if accordingly to the current
 824          * configuration there are too many elements in the encoded data
 825          * type. Note that we only check the length and not max element
 826          * size as this is an O(N) scan. Eventually everything will get
 827          * converted. */
 828         switch(type) {
 829             case REDIS_HASH_ZIPMAP:
 830                 o->type = REDIS_HASH;
 831                 o->encoding = REDIS_ENCODING_ZIPMAP;
 832                 if (zipmapLen(o->ptr) > server.hash_max_zipmap_entries)
 833                     convertToRealHash(o);
 834                 break;
 835             case REDIS_LIST_ZIPLIST:
 836                 o->type = REDIS_LIST;
 837                 o->encoding = REDIS_ENCODING_ZIPLIST;
 838                 if (ziplistLen(o->ptr) > server.list_max_ziplist_entries)
 839                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 840                 break;
 841             case REDIS_SET_INTSET:
 842                 o->type = REDIS_SET;
 843                 o->encoding = REDIS_ENCODING_INTSET;
 844                 if (intsetLen(o->ptr) > server.set_max_intset_entries)
 845                     setTypeConvert(o,REDIS_ENCODING_HT);
 846                 break;
 847             default:
 848                 redisPanic("Unknown enoding");
 849                 break;
 850         }
 851     } else {
 852         redisPanic("Unknown object type");
 853     }
 854     return o;
 855 }
 856
 857 /* Mark that we are loading in the global state and setup the fields
 858  * needed to provide loading stats. */
 859 void startLoading(FILE *fp) {
 860     struct stat sb;
 861
 862     /* Load the DB */
 863     server.loading = 1;
 864     server.loading_start_time = time(NULL);
 865     if (fstat(fileno(fp), &sb) == -1) {
 866         server.loading_total_bytes = 1; /* just to avoid division by zero */
 867     } else {
 868         server.loading_total_bytes = sb.st_size;
 869     }
 870 }
 871
 872 /* Refresh the loading progress info */
 873 void loadingProgress(off_t pos) {
 874     server.loading_loaded_bytes = pos;
 875 }
 876
 877 /* Loading finished */
 878 void stopLoading(void) {
 879     server.loading = 0;
 880 }
 881
 882 int rdbLoad(char *filename) {
 883     FILE *fp;
 884     uint32_t dbid;
 885     int type, retval, rdbver;
 886     redisDb *db = server.db+0;
 887     char buf[1024];
 888     time_t expiretime, now = time(NULL);
 889     long loops = 0;
 890
 891     fp = fopen(filename,"r");
 892     if (!fp) return REDIS_ERR;
 893     if (fread(buf,9,1,fp) == 0) goto eoferr;
 894     buf[9] = '\0';
 895     if (memcmp(buf,"REDIS",5) != 0) {
 896         fclose(fp);
 897         redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
 898         return REDIS_ERR;
 899     }
 900     rdbver = atoi(buf+5);
 901     if (rdbver != 1) {
 902         fclose(fp);
 903         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
 904         return REDIS_ERR;
 905     }
 906
 907     startLoading(fp);
 908     while(1) {
 909         robj *key, *val;
 910         expiretime = -1;
 911
 912         /* Serve the clients from time to time */
 913         if (!(loops++ % 1000)) {
 914             loadingProgress(ftello(fp));
 915             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 916         }
 917
 918         /* Read type. */
 919         if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 920         if (type == REDIS_EXPIRETIME) {
 921             if ((expiretime = rdbLoadTime(fp)) == -1) goto eoferr;
 922             /* We read the time so we need to read the object type again */
 923             if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 924         }
 925         if (type == REDIS_EOF) break;
 926         /* Handle SELECT DB opcode as a special case */
 927         if (type == REDIS_SELECTDB) {
 928             if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR)
 929                 goto eoferr;
 930             if (dbid >= (unsigned)server.dbnum) {
 931                 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
 932                 exit(1);
 933             }
 934             db = server.db+dbid;
 935             continue;
 936         }
 937         /* Read key */
 938         if ((key = rdbLoadStringObject(fp)) == NULL) goto eoferr;
 939         /* Read value */
 940         if ((val = rdbLoadObject(type,fp)) == NULL) goto eoferr;
 941         /* Check if the key already expired */
 942         if (expiretime != -1 && expiretime < now) {
 943             decrRefCount(key);
 944             decrRefCount(val);
 945             continue;
 946         }
 947         /* Add the new object in the hash table */
 948         retval = dbAdd(db,key,val);
 949         if (retval == REDIS_ERR) {
 950             redisLog(REDIS_WARNING,"Loading DB, duplicated key (%s) found! Unrecoverable error, exiting now.", key->ptr);
 951             exit(1);
 952         }
 953         /* Set the expire time if needed */
 954         if (expiretime != -1) setExpire(db,key,expiretime);
 955
 956         decrRefCount(key);
 957     }
 958     fclose(fp);
 959     stopLoading();
 960     return REDIS_OK;
 961
 962 eoferr: /* unexpected end of file is handled here with a fatal exit */
 963     redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
 964     exit(1);
 965     return REDIS_ERR; /* Just to avoid warning */
 966 }
 967
 968 /* A background saving child (BGSAVE) terminated its work. Handle this. */
 969 void backgroundSaveDoneHandler(int exitcode, int bysignal) {
 970     if (!bysignal && exitcode == 0) {
 971         redisLog(REDIS_NOTICE,
 972             "Background saving terminated with success");
 973         server.dirty = server.dirty - server.dirty_before_bgsave;
 974         server.lastsave = time(NULL);
 975     } else if (!bysignal && exitcode != 0) {
 976         redisLog(REDIS_WARNING, "Background saving error");
 977     } else {
 978         redisLog(REDIS_WARNING,
 979             "Background saving terminated by signal %d", bysignal);
 980         rdbRemoveTempFile(server.bgsavechildpid);
 981     }
 982     server.bgsavechildpid = -1;
 983     server.bgsavethread = (pthread_t) -1;
 984     server.bgsavethread_state = REDIS_BGSAVE_THREAD_UNACTIVE;
 985     /* Possibly there are slaves waiting for a BGSAVE in order to be served
 986      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
 987     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
 988 }
 989
 990 void saveCommand(redisClient *c) {
 991     if (server.bgsavechildpid != -1 || server.bgsavethread != (pthread_t)-1) {
 992         addReplyError(c,"Background save already in progress");
 993         return;
 994     }
 995     if (rdbSave(server.dbfilename) == REDIS_OK) {
 996         addReply(c,shared.ok);
 997     } else {
 998         addReply(c,shared.err);
 999     }
1000 }
1001
1002 void bgsaveCommand(redisClient *c) {
1003     if (server.bgsavechildpid != -1 || server.bgsavethread != (pthread_t)-1) {
1004         addReplyError(c,"Background save already in progress");
1005         return;
1006     }
1007     if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
1008         addReplyStatus(c,"Background saving started");
1009     } else {
1010         addReply(c,shared.err);
1011     }
1012 }