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     if (server.ds_enabled) {
 417         cacheForcePointInTime();
 418         return dsRdbSave(filename);
 419     }
 420
 421     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
 422     fp = fopen(tmpfile,"w");
 423     if (!fp) {
 424         redisLog(REDIS_WARNING, "Failed opening .rdb for saving: %s",
 425             strerror(errno));
 426         return REDIS_ERR;
 427     }
 428     if (fwrite("REDIS0002",9,1,fp) == 0) goto werr;
 429     for (j = 0; j < server.dbnum; j++) {
 430         redisDb *db = server.db+j;
 431         dict *d = db->dict;
 432         if (dictSize(d) == 0) continue;
 433         di = dictGetSafeIterator(d);
 434         if (!di) {
 435             fclose(fp);
 436             return REDIS_ERR;
 437         }
 438
 439         /* Write the SELECT DB opcode */
 440         if (rdbSaveType(fp,REDIS_SELECTDB) == -1) goto werr;
 441         if (rdbSaveLen(fp,j) == -1) goto werr;
 442
 443         /* Iterate this DB writing every entry */
 444         while((de = dictNext(di)) != NULL) {
 445             sds keystr = dictGetEntryKey(de);
 446             robj key, *o = dictGetEntryVal(de);
 447             time_t expire;
 448
 449             initStaticStringObject(key,keystr);
 450             expire = getExpire(db,&key);
 451             if (rdbSaveKeyValuePair(fp,&key,o,expire,now) == -1) goto werr;
 452         }
 453         dictReleaseIterator(di);
 454     }
 455     /* EOF opcode */
 456     if (rdbSaveType(fp,REDIS_EOF) == -1) goto werr;
 457
 458     /* Make sure data will not remain on the OS's output buffers */
 459     fflush(fp);
 460     fsync(fileno(fp));
 461     fclose(fp);
 462
 463     /* Use RENAME to make sure the DB file is changed atomically only
 464      * if the generate DB file is ok. */
 465     if (rename(tmpfile,filename) == -1) {
 466         redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
 467         unlink(tmpfile);
 468         return REDIS_ERR;
 469     }
 470     redisLog(REDIS_NOTICE,"DB saved on disk");
 471     server.dirty = 0;
 472     server.lastsave = time(NULL);
 473     return REDIS_OK;
 474
 475 werr:
 476     fclose(fp);
 477     unlink(tmpfile);
 478     redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
 479     if (di) dictReleaseIterator(di);
 480     return REDIS_ERR;
 481 }
 482
 483 int rdbSaveBackground(char *filename) {
 484     pid_t childpid;
 485     long long start;
 486
 487     if (server.bgsavechildpid != -1 ||
 488         server.bgsavethread != (pthread_t) -1) return REDIS_ERR;
 489
 490     server.dirty_before_bgsave = server.dirty;
 491
 492     if (server.ds_enabled) {
 493         cacheForcePointInTime();
 494         return dsRdbSaveBackground(filename);
 495     }
 496
 497     start = ustime();
 498     if ((childpid = fork()) == 0) {
 499         int retval;
 500
 501         /* Child */
 502         if (server.ipfd > 0) close(server.ipfd);
 503         if (server.sofd > 0) close(server.sofd);
 504         retval = rdbSave(filename);
 505         _exit((retval == REDIS_OK) ? 0 : 1);
 506     } else {
 507         /* Parent */
 508         server.stat_fork_time = ustime()-start;
 509         if (childpid == -1) {
 510             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
 511                 strerror(errno));
 512             return REDIS_ERR;
 513         }
 514         redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
 515         server.bgsavechildpid = childpid;
 516         updateDictResizePolicy();
 517         return REDIS_OK;
 518     }
 519     return REDIS_OK; /* unreached */
 520 }
 521
 522 void rdbRemoveTempFile(pid_t childpid) {
 523     char tmpfile[256];
 524
 525     snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
 526     unlink(tmpfile);
 527 }
 528
 529 int rdbLoadType(FILE *fp) {
 530     unsigned char type;
 531     if (fread(&type,1,1,fp) == 0) return -1;
 532     return type;
 533 }
 534
 535 time_t rdbLoadTime(FILE *fp) {
 536     int32_t t32;
 537     if (fread(&t32,4,1,fp) == 0) return -1;
 538     return (time_t) t32;
 539 }
 540
 541 /* Load an encoded length from the DB, see the REDIS_RDB_* defines on the top
 542  * of this file for a description of how this are stored on disk.
 543  *
 544  * isencoded is set to 1 if the readed length is not actually a length but
 545  * an "encoding type", check the above comments for more info */
 546 uint32_t rdbLoadLen(FILE *fp, int *isencoded) {
 547     unsigned char buf[2];
 548     uint32_t len;
 549     int type;
 550
 551     if (isencoded) *isencoded = 0;
 552     if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR;
 553     type = (buf[0]&0xC0)>>6;
 554     if (type == REDIS_RDB_6BITLEN) {
 555         /* Read a 6 bit len */
 556         return buf[0]&0x3F;
 557     } else if (type == REDIS_RDB_ENCVAL) {
 558         /* Read a 6 bit len encoding type */
 559         if (isencoded) *isencoded = 1;
 560         return buf[0]&0x3F;
 561     } else if (type == REDIS_RDB_14BITLEN) {
 562         /* Read a 14 bit len */
 563         if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR;
 564         return ((buf[0]&0x3F)<<8)|buf[1];
 565     } else {
 566         /* Read a 32 bit len */
 567         if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR;
 568         return ntohl(len);
 569     }
 570 }
 571
 572 /* Load an integer-encoded object from file 'fp', with the specified
 573  * encoding type 'enctype'. If encode is true the function may return
 574  * an integer-encoded object as reply, otherwise the returned object
 575  * will always be encoded as a raw string. */
 576 robj *rdbLoadIntegerObject(FILE *fp, int enctype, int encode) {
 577     unsigned char enc[4];
 578     long long val;
 579
 580     if (enctype == REDIS_RDB_ENC_INT8) {
 581         if (fread(enc,1,1,fp) == 0) return NULL;
 582         val = (signed char)enc[0];
 583     } else if (enctype == REDIS_RDB_ENC_INT16) {
 584         uint16_t v;
 585         if (fread(enc,2,1,fp) == 0) return NULL;
 586         v = enc[0]|(enc[1]<<8);
 587         val = (int16_t)v;
 588     } else if (enctype == REDIS_RDB_ENC_INT32) {
 589         uint32_t v;
 590         if (fread(enc,4,1,fp) == 0) return NULL;
 591         v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
 592         val = (int32_t)v;
 593     } else {
 594         val = 0; /* anti-warning */
 595         redisPanic("Unknown RDB integer encoding type");
 596     }
 597     if (encode)
 598         return createStringObjectFromLongLong(val);
 599     else
 600         return createObject(REDIS_STRING,sdsfromlonglong(val));
 601 }
 602
 603 robj *rdbLoadLzfStringObject(FILE*fp) {
 604     unsigned int len, clen;
 605     unsigned char *c = NULL;
 606     sds val = NULL;
 607
 608     if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 609     if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 610     if ((c = zmalloc(clen)) == NULL) goto err;
 611     if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
 612     if (fread(c,clen,1,fp) == 0) goto err;
 613     if (lzf_decompress(c,clen,val,len) == 0) goto err;
 614     zfree(c);
 615     return createObject(REDIS_STRING,val);
 616 err:
 617     zfree(c);
 618     sdsfree(val);
 619     return NULL;
 620 }
 621
 622 robj *rdbGenericLoadStringObject(FILE*fp, int encode) {
 623     int isencoded;
 624     uint32_t len;
 625     sds val;
 626
 627     len = rdbLoadLen(fp,&isencoded);
 628     if (isencoded) {
 629         switch(len) {
 630         case REDIS_RDB_ENC_INT8:
 631         case REDIS_RDB_ENC_INT16:
 632         case REDIS_RDB_ENC_INT32:
 633             return rdbLoadIntegerObject(fp,len,encode);
 634         case REDIS_RDB_ENC_LZF:
 635             return rdbLoadLzfStringObject(fp);
 636         default:
 637             redisPanic("Unknown RDB encoding type");
 638         }
 639     }
 640
 641     if (len == REDIS_RDB_LENERR) return NULL;
 642     val = sdsnewlen(NULL,len);
 643     if (len && fread(val,len,1,fp) == 0) {
 644         sdsfree(val);
 645         return NULL;
 646     }
 647     return createObject(REDIS_STRING,val);
 648 }
 649
 650 robj *rdbLoadStringObject(FILE *fp) {
 651     return rdbGenericLoadStringObject(fp,0);
 652 }
 653
 654 robj *rdbLoadEncodedStringObject(FILE *fp) {
 655     return rdbGenericLoadStringObject(fp,1);
 656 }
 657
 658 /* For information about double serialization check rdbSaveDoubleValue() */
 659 int rdbLoadDoubleValue(FILE *fp, double *val) {
 660     char buf[128];
 661     unsigned char len;
 662
 663     if (fread(&len,1,1,fp) == 0) return -1;
 664     switch(len) {
 665     case 255: *val = R_NegInf; return 0;
 666     case 254: *val = R_PosInf; return 0;
 667     case 253: *val = R_Nan; return 0;
 668     default:
 669         if (fread(buf,len,1,fp) == 0) return -1;
 670         buf[len] = '\0';
 671         sscanf(buf, "%lg", val);
 672         return 0;
 673     }
 674 }
 675
 676 /* Load a Redis object of the specified type from the specified file.
 677  * On success a newly allocated object is returned, otherwise NULL. */
 678 robj *rdbLoadObject(int type, FILE *fp) {
 679     robj *o, *ele, *dec;
 680     size_t len;
 681     unsigned int i;
 682
 683     redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
 684     if (type == REDIS_STRING) {
 685         /* Read string value */
 686         if ((o = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 687         o = tryObjectEncoding(o);
 688     } else if (type == REDIS_LIST) {
 689         /* Read list value */
 690         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 691
 692         /* Use a real list when there are too many entries */
 693         if (len > server.list_max_ziplist_entries) {
 694             o = createListObject();
 695         } else {
 696             o = createZiplistObject();
 697         }
 698
 699         /* Load every single element of the list */
 700         while(len--) {
 701             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 702
 703             /* If we are using a ziplist and the value is too big, convert
 704              * the object to a real list. */
 705             if (o->encoding == REDIS_ENCODING_ZIPLIST &&
 706                 ele->encoding == REDIS_ENCODING_RAW &&
 707                 sdslen(ele->ptr) > server.list_max_ziplist_value)
 708                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 709
 710             if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 711                 dec = getDecodedObject(ele);
 712                 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
 713                 decrRefCount(dec);
 714                 decrRefCount(ele);
 715             } else {
 716                 ele = tryObjectEncoding(ele);
 717                 listAddNodeTail(o->ptr,ele);
 718             }
 719         }
 720     } else if (type == REDIS_SET) {
 721         /* Read list/set value */
 722         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 723
 724         /* Use a regular set when there are too many entries. */
 725         if (len > server.set_max_intset_entries) {
 726             o = createSetObject();
 727             /* It's faster to expand the dict to the right size asap in order
 728              * to avoid rehashing */
 729             if (len > DICT_HT_INITIAL_SIZE)
 730                 dictExpand(o->ptr,len);
 731         } else {
 732             o = createIntsetObject();
 733         }
 734
 735         /* Load every single element of the list/set */
 736         for (i = 0; i < len; i++) {
 737             long long llval;
 738             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 739             ele = tryObjectEncoding(ele);
 740
 741             if (o->encoding == REDIS_ENCODING_INTSET) {
 742                 /* Fetch integer value from element */
 743                 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
 744                     o->ptr = intsetAdd(o->ptr,llval,NULL);
 745                 } else {
 746                     setTypeConvert(o,REDIS_ENCODING_HT);
 747                     dictExpand(o->ptr,len);
 748                 }
 749             }
 750
 751             /* This will also be called when the set was just converted
 752              * to regular hashtable encoded set */
 753             if (o->encoding == REDIS_ENCODING_HT) {
 754                 dictAdd((dict*)o->ptr,ele,NULL);
 755             } else {
 756                 decrRefCount(ele);
 757             }
 758         }
 759     } else if (type == REDIS_ZSET) {
 760         /* Read list/set value */
 761         size_t zsetlen;
 762         size_t maxelelen = 0;
 763         zset *zs;
 764
 765         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 766         o = createZsetObject();
 767         zs = o->ptr;
 768
 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
 779             /* Don't care about integer-encoded strings. */
 780             if (ele->encoding == REDIS_ENCODING_RAW &&
 781                 sdslen(ele->ptr) > maxelelen)
 782                     maxelelen = sdslen(ele->ptr);
 783
 784             znode = zslInsert(zs->zsl,score,ele);
 785             dictAdd(zs->dict,ele,&znode->score);
 786             incrRefCount(ele); /* added to skiplist */
 787         }
 788
 789         /* Convert *after* loading, since sorted sets are not stored ordered. */
 790         if (zsetLength(o) <= server.zset_max_ziplist_entries &&
 791             maxelelen <= server.zset_max_ziplist_value)
 792                 zsetConvert(o,REDIS_ENCODING_ZIPLIST);
 793     } else if (type == REDIS_HASH) {
 794         size_t hashlen;
 795
 796         if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 797         o = createHashObject();
 798         /* Too many entries? Use an hash table. */
 799         if (hashlen > server.hash_max_zipmap_entries)
 800             convertToRealHash(o);
 801         /* Load every key/value, then set it into the zipmap or hash
 802          * table, as needed. */
 803         while(hashlen--) {
 804             robj *key, *val;
 805
 806             if ((key = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 807             if ((val = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 808             /* If we are using a zipmap and there are too big values
 809              * the object is converted to real hash table encoding. */
 810             if (o->encoding != REDIS_ENCODING_HT &&
 811                ((key->encoding == REDIS_ENCODING_RAW &&
 812                 sdslen(key->ptr) > server.hash_max_zipmap_value) ||
 813                 (val->encoding == REDIS_ENCODING_RAW &&
 814                 sdslen(val->ptr) > server.hash_max_zipmap_value)))
 815             {
 816                     convertToRealHash(o);
 817             }
 818
 819             if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 820                 unsigned char *zm = o->ptr;
 821                 robj *deckey, *decval;
 822
 823                 /* We need raw string objects to add them to the zipmap */
 824                 deckey = getDecodedObject(key);
 825                 decval = getDecodedObject(val);
 826                 zm = zipmapSet(zm,deckey->ptr,sdslen(deckey->ptr),
 827                                   decval->ptr,sdslen(decval->ptr),NULL);
 828                 o->ptr = zm;
 829                 decrRefCount(deckey);
 830                 decrRefCount(decval);
 831                 decrRefCount(key);
 832                 decrRefCount(val);
 833             } else {
 834                 key = tryObjectEncoding(key);
 835                 val = tryObjectEncoding(val);
 836                 dictAdd((dict*)o->ptr,key,val);
 837             }
 838         }
 839     } else if (type == REDIS_HASH_ZIPMAP ||
 840                type == REDIS_LIST_ZIPLIST ||
 841                type == REDIS_SET_INTSET ||
 842                type == REDIS_ZSET_ZIPLIST)
 843     {
 844         robj *aux = rdbLoadStringObject(fp);
 845
 846         if (aux == NULL) return NULL;
 847         o = createObject(REDIS_STRING,NULL); /* string is just placeholder */
 848         o->ptr = zmalloc(sdslen(aux->ptr));
 849         memcpy(o->ptr,aux->ptr,sdslen(aux->ptr));
 850         decrRefCount(aux);
 851
 852         /* Fix the object encoding, and make sure to convert the encoded
 853          * data type into the base type if accordingly to the current
 854          * configuration there are too many elements in the encoded data
 855          * type. Note that we only check the length and not max element
 856          * size as this is an O(N) scan. Eventually everything will get
 857          * converted. */
 858         switch(type) {
 859             case REDIS_HASH_ZIPMAP:
 860                 o->type = REDIS_HASH;
 861                 o->encoding = REDIS_ENCODING_ZIPMAP;
 862                 if (zipmapLen(o->ptr) > server.hash_max_zipmap_entries)
 863                     convertToRealHash(o);
 864                 break;
 865             case REDIS_LIST_ZIPLIST:
 866                 o->type = REDIS_LIST;
 867                 o->encoding = REDIS_ENCODING_ZIPLIST;
 868                 if (ziplistLen(o->ptr) > server.list_max_ziplist_entries)
 869                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 870                 break;
 871             case REDIS_SET_INTSET:
 872                 o->type = REDIS_SET;
 873                 o->encoding = REDIS_ENCODING_INTSET;
 874                 if (intsetLen(o->ptr) > server.set_max_intset_entries)
 875                     setTypeConvert(o,REDIS_ENCODING_HT);
 876                 break;
 877             case REDIS_ZSET_ZIPLIST:
 878                 o->type = REDIS_ZSET;
 879                 o->encoding = REDIS_ENCODING_ZIPLIST;
 880                 if (zsetLength(o) > server.zset_max_ziplist_entries)
 881                     zsetConvert(o,REDIS_ENCODING_SKIPLIST);
 882                 break;
 883             default:
 884                 redisPanic("Unknown encoding");
 885                 break;
 886         }
 887     } else {
 888         redisPanic("Unknown object type");
 889     }
 890     return o;
 891 }
 892
 893 /* Mark that we are loading in the global state and setup the fields
 894  * needed to provide loading stats. */
 895 void startLoading(FILE *fp) {
 896     struct stat sb;
 897
 898     /* Load the DB */
 899     server.loading = 1;
 900     server.loading_start_time = time(NULL);
 901     if (fstat(fileno(fp), &sb) == -1) {
 902         server.loading_total_bytes = 1; /* just to avoid division by zero */
 903     } else {
 904         server.loading_total_bytes = sb.st_size;
 905     }
 906 }
 907
 908 /* Refresh the loading progress info */
 909 void loadingProgress(off_t pos) {
 910     server.loading_loaded_bytes = pos;
 911 }
 912
 913 /* Loading finished */
 914 void stopLoading(void) {
 915     server.loading = 0;
 916 }
 917
 918 int rdbLoad(char *filename) {
 919     FILE *fp;
 920     uint32_t dbid;
 921     int type, rdbver;
 922     redisDb *db = server.db+0;
 923     char buf[1024];
 924     time_t expiretime, now = time(NULL);
 925     long loops = 0;
 926
 927     fp = fopen(filename,"r");
 928     if (!fp) return REDIS_ERR;
 929     if (fread(buf,9,1,fp) == 0) goto eoferr;
 930     buf[9] = '\0';
 931     if (memcmp(buf,"REDIS",5) != 0) {
 932         fclose(fp);
 933         redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
 934         return REDIS_ERR;
 935     }
 936     rdbver = atoi(buf+5);
 937     if (rdbver < 1 || rdbver > 2) {
 938         fclose(fp);
 939         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
 940         return REDIS_ERR;
 941     }
 942
 943     startLoading(fp);
 944     while(1) {
 945         robj *key, *val;
 946         expiretime = -1;
 947
 948         /* Serve the clients from time to time */
 949         if (!(loops++ % 1000)) {
 950             loadingProgress(ftello(fp));
 951             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 952         }
 953
 954         /* Read type. */
 955         if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 956         if (type == REDIS_EXPIRETIME) {
 957             if ((expiretime = rdbLoadTime(fp)) == -1) goto eoferr;
 958             /* We read the time so we need to read the object type again */
 959             if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 960         }
 961         if (type == REDIS_EOF) break;
 962         /* Handle SELECT DB opcode as a special case */
 963         if (type == REDIS_SELECTDB) {
 964             if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR)
 965                 goto eoferr;
 966             if (dbid >= (unsigned)server.dbnum) {
 967                 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
 968                 exit(1);
 969             }
 970             db = server.db+dbid;
 971             continue;
 972         }
 973         /* Read key */
 974         if ((key = rdbLoadStringObject(fp)) == NULL) goto eoferr;
 975         /* Read value */
 976         if ((val = rdbLoadObject(type,fp)) == NULL) goto eoferr;
 977         /* Check if the key already expired */
 978         if (expiretime != -1 && expiretime < now) {
 979             decrRefCount(key);
 980             decrRefCount(val);
 981             continue;
 982         }
 983         /* Add the new object in the hash table */
 984         dbAdd(db,key,val);
 985
 986         /* Set the expire time if needed */
 987         if (expiretime != -1) setExpire(db,key,expiretime);
 988
 989         decrRefCount(key);
 990     }
 991     fclose(fp);
 992     stopLoading();
 993     return REDIS_OK;
 994
 995 eoferr: /* unexpected end of file is handled here with a fatal exit */
 996     redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
 997     exit(1);
 998     return REDIS_ERR; /* Just to avoid warning */
 999 }
1000
1001 /* A background saving child (BGSAVE) terminated its work. Handle this. */
1002 void backgroundSaveDoneHandler(int exitcode, int bysignal) {
1003     if (!bysignal && exitcode == 0) {
1004         redisLog(REDIS_NOTICE,
1005             "Background saving terminated with success");
1006         server.dirty = server.dirty - server.dirty_before_bgsave;
1007         server.lastsave = time(NULL);
1008     } else if (!bysignal && exitcode != 0) {
1009         redisLog(REDIS_WARNING, "Background saving error");
1010     } else {
1011         redisLog(REDIS_WARNING,
1012             "Background saving terminated by signal %d", bysignal);
1013         rdbRemoveTempFile(server.bgsavechildpid);
1014     }
1015     server.bgsavechildpid = -1;
1016     server.bgsavethread = (pthread_t) -1;
1017     server.bgsavethread_state = REDIS_BGSAVE_THREAD_UNACTIVE;
1018     /* Possibly there are slaves waiting for a BGSAVE in order to be served
1019      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
1020     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
1021 }
1022
1023 void saveCommand(redisClient *c) {
1024     if (server.bgsavechildpid != -1 || server.bgsavethread != (pthread_t)-1) {
1025         addReplyError(c,"Background save already in progress");
1026         return;
1027     }
1028     if (rdbSave(server.dbfilename) == REDIS_OK) {
1029         addReply(c,shared.ok);
1030     } else {
1031         addReply(c,shared.err);
1032     }
1033 }
1034
1035 void bgsaveCommand(redisClient *c) {
1036     if (server.bgsavechildpid != -1 || server.bgsavethread != (pthread_t)-1) {
1037         addReplyError(c,"Background save already in progress");
1038     } else if (server.bgrewritechildpid != -1) {
1039         addReplyError(c,"Can't BGSAVE while AOF log rewriting is in progress");
1040     } else if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
1041         addReplyStatus(c,"Background saving started");
1042     } else {
1043         addReply(c,shared.err);
1044     }
1045 }