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 safe 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             unsigned char *p;
 260             unsigned char *vstr;
 261             unsigned int vlen;
 262             long long vlong;
 263
 264             if ((n = rdbSaveLen(fp,ziplistLen(o->ptr))) == -1) return -1;
 265             nwritten += n;
 266
 267             p = ziplistIndex(o->ptr,0);
 268             while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 269                 if (vstr) {
 270                     if ((n = rdbSaveRawString(fp,vstr,vlen)) == -1)
 271                         return -1;
 272                     nwritten += n;
 273                 } else {
 274                     if ((n = rdbSaveLongLongAsStringObject(fp,vlong)) == -1)
 275                         return -1;
 276                     nwritten += n;
 277                 }
 278                 p = ziplistNext(o->ptr,p);
 279             }
 280         } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 281             list *list = o->ptr;
 282             listIter li;
 283             listNode *ln;
 284
 285             if ((n = rdbSaveLen(fp,listLength(list))) == -1) return -1;
 286             nwritten += n;
 287
 288             listRewind(list,&li);
 289             while((ln = listNext(&li))) {
 290                 robj *eleobj = listNodeValue(ln);
 291                 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 292                 nwritten += n;
 293             }
 294         } else {
 295             redisPanic("Unknown list encoding");
 296         }
 297     } else if (o->type == REDIS_SET) {
 298         /* Save a set value */
 299         if (o->encoding == REDIS_ENCODING_HT) {
 300             dict *set = o->ptr;
 301             dictIterator *di = dictGetIterator(set);
 302             dictEntry *de;
 303
 304             if ((n = rdbSaveLen(fp,dictSize(set))) == -1) return -1;
 305             nwritten += n;
 306
 307             while((de = dictNext(di)) != NULL) {
 308                 robj *eleobj = dictGetEntryKey(de);
 309                 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 310                 nwritten += n;
 311             }
 312             dictReleaseIterator(di);
 313         } else if (o->encoding == REDIS_ENCODING_INTSET) {
 314             intset *is = o->ptr;
 315             int64_t llval;
 316             int i = 0;
 317
 318             if ((n = rdbSaveLen(fp,intsetLen(is))) == -1) return -1;
 319             nwritten += n;
 320
 321             while(intsetGet(is,i++,&llval)) {
 322                 if ((n = rdbSaveLongLongAsStringObject(fp,llval)) == -1) return -1;
 323                 nwritten += n;
 324             }
 325         } else {
 326             redisPanic("Unknown set encoding");
 327         }
 328     } else if (o->type == REDIS_ZSET) {
 329         /* Save a set value */
 330         zset *zs = o->ptr;
 331         dictIterator *di = dictGetIterator(zs->dict);
 332         dictEntry *de;
 333
 334         if ((n = rdbSaveLen(fp,dictSize(zs->dict))) == -1) return -1;
 335         nwritten += n;
 336
 337         while((de = dictNext(di)) != NULL) {
 338             robj *eleobj = dictGetEntryKey(de);
 339             double *score = dictGetEntryVal(de);
 340
 341             if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
 342             nwritten += n;
 343             if ((n = rdbSaveDoubleValue(fp,*score)) == -1) return -1;
 344             nwritten += n;
 345         }
 346         dictReleaseIterator(di);
 347     } else if (o->type == REDIS_HASH) {
 348         /* Save a hash value */
 349         if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 350             unsigned char *p = zipmapRewind(o->ptr);
 351             unsigned int count = zipmapLen(o->ptr);
 352             unsigned char *key, *val;
 353             unsigned int klen, vlen;
 354
 355             if ((n = rdbSaveLen(fp,count)) == -1) return -1;
 356             nwritten += n;
 357
 358             while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) {
 359                 if ((n = rdbSaveRawString(fp,key,klen)) == -1) return -1;
 360                 nwritten += n;
 361                 if ((n = rdbSaveRawString(fp,val,vlen)) == -1) return -1;
 362                 nwritten += n;
 363             }
 364         } else {
 365             dictIterator *di = dictGetIterator(o->ptr);
 366             dictEntry *de;
 367
 368             if ((n = rdbSaveLen(fp,dictSize((dict*)o->ptr))) == -1) return -1;
 369             nwritten += n;
 370
 371             while((de = dictNext(di)) != NULL) {
 372                 robj *key = dictGetEntryKey(de);
 373                 robj *val = dictGetEntryVal(de);
 374
 375                 if ((n = rdbSaveStringObject(fp,key)) == -1) return -1;
 376                 nwritten += n;
 377                 if ((n = rdbSaveStringObject(fp,val)) == -1) return -1;
 378                 nwritten += n;
 379             }
 380             dictReleaseIterator(di);
 381         }
 382     } else {
 383         redisPanic("Unknown object type");
 384     }
 385     return nwritten;
 386 }
 387
 388 /* Return the length the object will have on disk if saved with
 389  * the rdbSaveObject() function. Currently we use a trick to get
 390  * this length with very little changes to the code. In the future
 391  * we could switch to a faster solution. */
 392 off_t rdbSavedObjectLen(robj *o) {
 393     int len = rdbSaveObject(NULL,o);
 394     redisAssert(len != -1);
 395     return len;
 396 }
 397
 398 /* Save a key-value pair, with expire time, type, key, value.
 399  * On error -1 is returned.
 400  * On success if the key was actaully saved 1 is returned, otherwise 0
 401  * is returned (the key was already expired). */
 402 int rdbSaveKeyValuePair(FILE *fp, redisDb *db, robj *key, robj *val,
 403                         time_t now)
 404 {
 405     time_t expiretime;
 406
 407     expiretime = getExpire(db,key);
 408
 409     /* Save the expire time */
 410     if (expiretime != -1) {
 411         /* If this key is already expired skip it */
 412         if (expiretime < now) return 0;
 413         if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) return -1;
 414         if (rdbSaveTime(fp,expiretime) == -1) return -1;
 415     }
 416     /* Save type, key, value */
 417     if (rdbSaveType(fp,val->type) == -1) return -1;
 418     if (rdbSaveStringObject(fp,key) == -1) return -1;
 419     if (rdbSaveObject(fp,val) == -1) return -1;
 420     return 1;
 421 }
 422
 423 /* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
 424 int rdbSave(char *filename) {
 425     dictIterator *di = NULL;
 426     dictEntry *de;
 427     FILE *fp;
 428     char tmpfile[256];
 429     int j;
 430     time_t now = time(NULL);
 431
 432     /* FIXME: implement .rdb save for disk store properly */
 433     redisAssert(server.ds_enabled == 0);
 434
 435     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
 436     fp = fopen(tmpfile,"w");
 437     if (!fp) {
 438         redisLog(REDIS_WARNING, "Failed saving the DB: %s", strerror(errno));
 439         return REDIS_ERR;
 440     }
 441     if (fwrite("REDIS0001",9,1,fp) == 0) goto werr;
 442     for (j = 0; j < server.dbnum; j++) {
 443         redisDb *db = server.db+j;
 444         dict *d = db->dict;
 445         if (dictSize(d) == 0) continue;
 446         di = dictGetIterator(d);
 447         if (!di) {
 448             fclose(fp);
 449             return REDIS_ERR;
 450         }
 451
 452         /* Write the SELECT DB opcode */
 453         if (rdbSaveType(fp,REDIS_SELECTDB) == -1) goto werr;
 454         if (rdbSaveLen(fp,j) == -1) goto werr;
 455
 456         /* Iterate this DB writing every entry */
 457         while((de = dictNext(di)) != NULL) {
 458             sds keystr = dictGetEntryKey(de);
 459             robj key, *o = dictGetEntryVal(de);
 460
 461             initStaticStringObject(key,keystr);
 462             if (rdbSaveKeyValuePair(fp,db,&key,o,now) == -1) goto werr;
 463         }
 464         dictReleaseIterator(di);
 465     }
 466     /* EOF opcode */
 467     if (rdbSaveType(fp,REDIS_EOF) == -1) goto werr;
 468
 469     /* Make sure data will not remain on the OS's output buffers */
 470     fflush(fp);
 471     fsync(fileno(fp));
 472     fclose(fp);
 473
 474     /* Use RENAME to make sure the DB file is changed atomically only
 475      * if the generate DB file is ok. */
 476     if (rename(tmpfile,filename) == -1) {
 477         redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
 478         unlink(tmpfile);
 479         return REDIS_ERR;
 480     }
 481     redisLog(REDIS_NOTICE,"DB saved on disk");
 482     server.dirty = 0;
 483     server.lastsave = time(NULL);
 484     return REDIS_OK;
 485
 486 werr:
 487     fclose(fp);
 488     unlink(tmpfile);
 489     redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
 490     if (di) dictReleaseIterator(di);
 491     return REDIS_ERR;
 492 }
 493
 494 int rdbSaveBackground(char *filename) {
 495     pid_t childpid;
 496
 497     if (server.bgsavechildpid != -1) return REDIS_ERR;
 498     redisAssert(server.ds_enabled == 0);
 499     server.dirty_before_bgsave = server.dirty;
 500     if ((childpid = fork()) == 0) {
 501         /* Child */
 502         if (server.ipfd > 0) close(server.ipfd);
 503         if (server.sofd > 0) close(server.sofd);
 504         if (rdbSave(filename) == REDIS_OK) {
 505             _exit(0);
 506         } else {
 507             _exit(1);
 508         }
 509     } else {
 510         /* Parent */
 511         if (childpid == -1) {
 512             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
 513                 strerror(errno));
 514             return REDIS_ERR;
 515         }
 516         redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
 517         server.bgsavechildpid = childpid;
 518         updateDictResizePolicy();
 519         return REDIS_OK;
 520     }
 521     return REDIS_OK; /* unreached */
 522 }
 523
 524 void rdbRemoveTempFile(pid_t childpid) {
 525     char tmpfile[256];
 526
 527     snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
 528     unlink(tmpfile);
 529 }
 530
 531 int rdbLoadType(FILE *fp) {
 532     unsigned char type;
 533     if (fread(&type,1,1,fp) == 0) return -1;
 534     return type;
 535 }
 536
 537 time_t rdbLoadTime(FILE *fp) {
 538     int32_t t32;
 539     if (fread(&t32,4,1,fp) == 0) return -1;
 540     return (time_t) t32;
 541 }
 542
 543 /* Load an encoded length from the DB, see the REDIS_RDB_* defines on the top
 544  * of this file for a description of how this are stored on disk.
 545  *
 546  * isencoded is set to 1 if the readed length is not actually a length but
 547  * an "encoding type", check the above comments for more info */
 548 uint32_t rdbLoadLen(FILE *fp, int *isencoded) {
 549     unsigned char buf[2];
 550     uint32_t len;
 551     int type;
 552
 553     if (isencoded) *isencoded = 0;
 554     if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR;
 555     type = (buf[0]&0xC0)>>6;
 556     if (type == REDIS_RDB_6BITLEN) {
 557         /* Read a 6 bit len */
 558         return buf[0]&0x3F;
 559     } else if (type == REDIS_RDB_ENCVAL) {
 560         /* Read a 6 bit len encoding type */
 561         if (isencoded) *isencoded = 1;
 562         return buf[0]&0x3F;
 563     } else if (type == REDIS_RDB_14BITLEN) {
 564         /* Read a 14 bit len */
 565         if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR;
 566         return ((buf[0]&0x3F)<<8)|buf[1];
 567     } else {
 568         /* Read a 32 bit len */
 569         if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR;
 570         return ntohl(len);
 571     }
 572 }
 573
 574 /* Load an integer-encoded object from file 'fp', with the specified
 575  * encoding type 'enctype'. If encode is true the function may return
 576  * an integer-encoded object as reply, otherwise the returned object
 577  * will always be encoded as a raw string. */
 578 robj *rdbLoadIntegerObject(FILE *fp, int enctype, int encode) {
 579     unsigned char enc[4];
 580     long long val;
 581
 582     if (enctype == REDIS_RDB_ENC_INT8) {
 583         if (fread(enc,1,1,fp) == 0) return NULL;
 584         val = (signed char)enc[0];
 585     } else if (enctype == REDIS_RDB_ENC_INT16) {
 586         uint16_t v;
 587         if (fread(enc,2,1,fp) == 0) return NULL;
 588         v = enc[0]|(enc[1]<<8);
 589         val = (int16_t)v;
 590     } else if (enctype == REDIS_RDB_ENC_INT32) {
 591         uint32_t v;
 592         if (fread(enc,4,1,fp) == 0) return NULL;
 593         v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
 594         val = (int32_t)v;
 595     } else {
 596         val = 0; /* anti-warning */
 597         redisPanic("Unknown RDB integer encoding type");
 598     }
 599     if (encode)
 600         return createStringObjectFromLongLong(val);
 601     else
 602         return createObject(REDIS_STRING,sdsfromlonglong(val));
 603 }
 604
 605 robj *rdbLoadLzfStringObject(FILE*fp) {
 606     unsigned int len, clen;
 607     unsigned char *c = NULL;
 608     sds val = NULL;
 609
 610     if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 611     if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 612     if ((c = zmalloc(clen)) == NULL) goto err;
 613     if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
 614     if (fread(c,clen,1,fp) == 0) goto err;
 615     if (lzf_decompress(c,clen,val,len) == 0) goto err;
 616     zfree(c);
 617     return createObject(REDIS_STRING,val);
 618 err:
 619     zfree(c);
 620     sdsfree(val);
 621     return NULL;
 622 }
 623
 624 robj *rdbGenericLoadStringObject(FILE*fp, int encode) {
 625     int isencoded;
 626     uint32_t len;
 627     sds val;
 628
 629     len = rdbLoadLen(fp,&isencoded);
 630     if (isencoded) {
 631         switch(len) {
 632         case REDIS_RDB_ENC_INT8:
 633         case REDIS_RDB_ENC_INT16:
 634         case REDIS_RDB_ENC_INT32:
 635             return rdbLoadIntegerObject(fp,len,encode);
 636         case REDIS_RDB_ENC_LZF:
 637             return rdbLoadLzfStringObject(fp);
 638         default:
 639             redisPanic("Unknown RDB encoding type");
 640         }
 641     }
 642
 643     if (len == REDIS_RDB_LENERR) return NULL;
 644     val = sdsnewlen(NULL,len);
 645     if (len && fread(val,len,1,fp) == 0) {
 646         sdsfree(val);
 647         return NULL;
 648     }
 649     return createObject(REDIS_STRING,val);
 650 }
 651
 652 robj *rdbLoadStringObject(FILE *fp) {
 653     return rdbGenericLoadStringObject(fp,0);
 654 }
 655
 656 robj *rdbLoadEncodedStringObject(FILE *fp) {
 657     return rdbGenericLoadStringObject(fp,1);
 658 }
 659
 660 /* For information about double serialization check rdbSaveDoubleValue() */
 661 int rdbLoadDoubleValue(FILE *fp, double *val) {
 662     char buf[128];
 663     unsigned char len;
 664
 665     if (fread(&len,1,1,fp) == 0) return -1;
 666     switch(len) {
 667     case 255: *val = R_NegInf; return 0;
 668     case 254: *val = R_PosInf; return 0;
 669     case 253: *val = R_Nan; return 0;
 670     default:
 671         if (fread(buf,len,1,fp) == 0) return -1;
 672         buf[len] = '\0';
 673         sscanf(buf, "%lg", val);
 674         return 0;
 675     }
 676 }
 677
 678 /* Load a Redis object of the specified type from the specified file.
 679  * On success a newly allocated object is returned, otherwise NULL. */
 680 robj *rdbLoadObject(int type, FILE *fp) {
 681     robj *o, *ele, *dec;
 682     size_t len;
 683     unsigned int i;
 684
 685     redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
 686     if (type == REDIS_STRING) {
 687         /* Read string value */
 688         if ((o = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 689         o = tryObjectEncoding(o);
 690     } else if (type == REDIS_LIST) {
 691         /* Read list value */
 692         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 693
 694         /* Use a real list when there are too many entries */
 695         if (len > server.list_max_ziplist_entries) {
 696             o = createListObject();
 697         } else {
 698             o = createZiplistObject();
 699         }
 700
 701         /* Load every single element of the list */
 702         while(len--) {
 703             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 704
 705             /* If we are using a ziplist and the value is too big, convert
 706              * the object to a real list. */
 707             if (o->encoding == REDIS_ENCODING_ZIPLIST &&
 708                 ele->encoding == REDIS_ENCODING_RAW &&
 709                 sdslen(ele->ptr) > server.list_max_ziplist_value)
 710                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 711
 712             if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 713                 dec = getDecodedObject(ele);
 714                 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
 715                 decrRefCount(dec);
 716                 decrRefCount(ele);
 717             } else {
 718                 ele = tryObjectEncoding(ele);
 719                 listAddNodeTail(o->ptr,ele);
 720             }
 721         }
 722     } else if (type == REDIS_SET) {
 723         /* Read list/set value */
 724         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 725
 726         /* Use a regular set when there are too many entries. */
 727         if (len > server.set_max_intset_entries) {
 728             o = createSetObject();
 729             /* It's faster to expand the dict to the right size asap in order
 730              * to avoid rehashing */
 731             if (len > DICT_HT_INITIAL_SIZE)
 732                 dictExpand(o->ptr,len);
 733         } else {
 734             o = createIntsetObject();
 735         }
 736
 737         /* Load every single element of the list/set */
 738         for (i = 0; i < len; i++) {
 739             long long llval;
 740             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 741             ele = tryObjectEncoding(ele);
 742
 743             if (o->encoding == REDIS_ENCODING_INTSET) {
 744                 /* Fetch integer value from element */
 745                 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
 746                     o->ptr = intsetAdd(o->ptr,llval,NULL);
 747                 } else {
 748                     setTypeConvert(o,REDIS_ENCODING_HT);
 749                     dictExpand(o->ptr,len);
 750                 }
 751             }
 752
 753             /* This will also be called when the set was just converted
 754              * to regular hashtable encoded set */
 755             if (o->encoding == REDIS_ENCODING_HT) {
 756                 dictAdd((dict*)o->ptr,ele,NULL);
 757             } else {
 758                 decrRefCount(ele);
 759             }
 760         }
 761     } else if (type == REDIS_ZSET) {
 762         /* Read list/set value */
 763         size_t zsetlen;
 764         zset *zs;
 765
 766         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 767         o = createZsetObject();
 768         zs = o->ptr;
 769         /* Load every single element of the list/set */
 770         while(zsetlen--) {
 771             robj *ele;
 772             double score;
 773             zskiplistNode *znode;
 774
 775             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 776             ele = tryObjectEncoding(ele);
 777             if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;
 778             znode = zslInsert(zs->zsl,score,ele);
 779             dictAdd(zs->dict,ele,&znode->score);
 780             incrRefCount(ele); /* added to skiplist */
 781         }
 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 {
 829         redisPanic("Unknown object type");
 830     }
 831     return o;
 832 }
 833
 834 /* Mark that we are loading in the global state and setup the fields
 835  * needed to provide loading stats. */
 836 void startLoading(FILE *fp) {
 837     struct stat sb;
 838
 839     /* Load the DB */
 840     server.loading = 1;
 841     server.loading_start_time = time(NULL);
 842     if (fstat(fileno(fp), &sb) == -1) {
 843         server.loading_total_bytes = 1; /* just to avoid division by zero */
 844     } else {
 845         server.loading_total_bytes = sb.st_size;
 846     }
 847 }
 848
 849 /* Refresh the loading progress info */
 850 void loadingProgress(off_t pos) {
 851     server.loading_loaded_bytes = pos;
 852 }
 853
 854 /* Loading finished */
 855 void stopLoading(void) {
 856     server.loading = 0;
 857 }
 858
 859 int rdbLoad(char *filename) {
 860     FILE *fp;
 861     uint32_t dbid;
 862     int type, retval, rdbver;
 863     redisDb *db = server.db+0;
 864     char buf[1024];
 865     time_t expiretime, now = time(NULL);
 866     long loops = 0;
 867
 868     fp = fopen(filename,"r");
 869     if (!fp) return REDIS_ERR;
 870     if (fread(buf,9,1,fp) == 0) goto eoferr;
 871     buf[9] = '\0';
 872     if (memcmp(buf,"REDIS",5) != 0) {
 873         fclose(fp);
 874         redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
 875         return REDIS_ERR;
 876     }
 877     rdbver = atoi(buf+5);
 878     if (rdbver != 1) {
 879         fclose(fp);
 880         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
 881         return REDIS_ERR;
 882     }
 883
 884     startLoading(fp);
 885     while(1) {
 886         robj *key, *val;
 887         expiretime = -1;
 888
 889         /* Serve the clients from time to time */
 890         if (!(loops++ % 1000)) {
 891             loadingProgress(ftello(fp));
 892             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 893         }
 894
 895         /* Read type. */
 896         if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 897         if (type == REDIS_EXPIRETIME) {
 898             if ((expiretime = rdbLoadTime(fp)) == -1) goto eoferr;
 899             /* We read the time so we need to read the object type again */
 900             if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 901         }
 902         if (type == REDIS_EOF) break;
 903         /* Handle SELECT DB opcode as a special case */
 904         if (type == REDIS_SELECTDB) {
 905             if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR)
 906                 goto eoferr;
 907             if (dbid >= (unsigned)server.dbnum) {
 908                 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
 909                 exit(1);
 910             }
 911             db = server.db+dbid;
 912             continue;
 913         }
 914         /* Read key */
 915         if ((key = rdbLoadStringObject(fp)) == NULL) goto eoferr;
 916         /* Read value */
 917         if ((val = rdbLoadObject(type,fp)) == NULL) goto eoferr;
 918         /* Check if the key already expired */
 919         if (expiretime != -1 && expiretime < now) {
 920             decrRefCount(key);
 921             decrRefCount(val);
 922             continue;
 923         }
 924         /* Add the new object in the hash table */
 925         retval = dbAdd(db,key,val);
 926         if (retval == REDIS_ERR) {
 927             redisLog(REDIS_WARNING,"Loading DB, duplicated key (%s) found! Unrecoverable error, exiting now.", key->ptr);
 928             exit(1);
 929         }
 930         /* Set the expire time if needed */
 931         if (expiretime != -1) setExpire(db,key,expiretime);
 932
 933         decrRefCount(key);
 934     }
 935     fclose(fp);
 936     stopLoading();
 937     return REDIS_OK;
 938
 939 eoferr: /* unexpected end of file is handled here with a fatal exit */
 940     redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
 941     exit(1);
 942     return REDIS_ERR; /* Just to avoid warning */
 943 }
 944
 945 /* A background saving child (BGSAVE) terminated its work. Handle this. */
 946 void backgroundSaveDoneHandler(int statloc) {
 947     int exitcode = WEXITSTATUS(statloc);
 948     int bysignal = WIFSIGNALED(statloc);
 949
 950     if (!bysignal && exitcode == 0) {
 951         redisLog(REDIS_NOTICE,
 952             "Background saving terminated with success");
 953         server.dirty = server.dirty - server.dirty_before_bgsave;
 954         server.lastsave = time(NULL);
 955     } else if (!bysignal && exitcode != 0) {
 956         redisLog(REDIS_WARNING, "Background saving error");
 957     } else {
 958         redisLog(REDIS_WARNING,
 959             "Background saving terminated by signal %d", WTERMSIG(statloc));
 960         rdbRemoveTempFile(server.bgsavechildpid);
 961     }
 962     server.bgsavechildpid = -1;
 963     /* Possibly there are slaves waiting for a BGSAVE in order to be served
 964      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
 965     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
 966 }