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 int rdbSaveType(FILE *fp, unsigned char type) {
  13     if (fwrite(&type,1,1,fp) == 0) return -1;
  14     return 0;
  15 }
  16
  17 int rdbSaveTime(FILE *fp, time_t t) {
  18     int32_t t32 = (int32_t) t;
  19     if (fwrite(&t32,4,1,fp) == 0) return -1;
  20     return 0;
  21 }
  22
  23 /* check rdbLoadLen() comments for more info */
  24 int rdbSaveLen(FILE *fp, uint32_t len) {
  25     unsigned char buf[2];
  26
  27     if (len < (1<<6)) {
  28         /* Save a 6 bit len */
  29         buf[0] = (len&0xFF)|(REDIS_RDB_6BITLEN<<6);
  30         if (fwrite(buf,1,1,fp) == 0) return -1;
  31     } else if (len < (1<<14)) {
  32         /* Save a 14 bit len */
  33         buf[0] = ((len>>8)&0xFF)|(REDIS_RDB_14BITLEN<<6);
  34         buf[1] = len&0xFF;
  35         if (fwrite(buf,2,1,fp) == 0) return -1;
  36     } else {
  37         /* Save a 32 bit len */
  38         buf[0] = (REDIS_RDB_32BITLEN<<6);
  39         if (fwrite(buf,1,1,fp) == 0) return -1;
  40         len = htonl(len);
  41         if (fwrite(&len,4,1,fp) == 0) return -1;
  42     }
  43     return 0;
  44 }
  45
  46 /* Encode 'value' as an integer if possible (if integer will fit the
  47  * supported range). If the function sucessful encoded the integer
  48  * then the (up to 5 bytes) encoded representation is written in the
  49  * string pointed by 'enc' and the length is returned. Otherwise
  50  * 0 is returned. */
  51 int rdbEncodeInteger(long long value, unsigned char *enc) {
  52     /* Finally check if it fits in our ranges */
  53     if (value >= -(1<<7) && value <= (1<<7)-1) {
  54         enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT8;
  55         enc[1] = value&0xFF;
  56         return 2;
  57     } else if (value >= -(1<<15) && value <= (1<<15)-1) {
  58         enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT16;
  59         enc[1] = value&0xFF;
  60         enc[2] = (value>>8)&0xFF;
  61         return 3;
  62     } else if (value >= -((long long)1<<31) && value <= ((long long)1<<31)-1) {
  63         enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT32;
  64         enc[1] = value&0xFF;
  65         enc[2] = (value>>8)&0xFF;
  66         enc[3] = (value>>16)&0xFF;
  67         enc[4] = (value>>24)&0xFF;
  68         return 5;
  69     } else {
  70         return 0;
  71     }
  72 }
  73
  74 /* String objects in the form "2391" "-100" without any space and with a
  75  * range of values that can fit in an 8, 16 or 32 bit signed value can be
  76  * encoded as integers to save space */
  77 int rdbTryIntegerEncoding(char *s, size_t len, unsigned char *enc) {
  78     long long value;
  79     char *endptr, buf[32];
  80
  81     /* Check if it's possible to encode this value as a number */
  82     value = strtoll(s, &endptr, 10);
  83     if (endptr[0] != '\0') return 0;
  84     ll2string(buf,32,value);
  85
  86     /* If the number converted back into a string is not identical
  87      * then it's not possible to encode the string as integer */
  88     if (strlen(buf) != len || memcmp(buf,s,len)) return 0;
  89
  90     return rdbEncodeInteger(value,enc);
  91 }
  92
  93 int rdbSaveLzfStringObject(FILE *fp, unsigned char *s, size_t len) {
  94     size_t comprlen, outlen;
  95     unsigned char byte;
  96     void *out;
  97
  98     /* We require at least four bytes compression for this to be worth it */
  99     if (len <= 4) return 0;
 100     outlen = len-4;
 101     if ((out = zmalloc(outlen+1)) == NULL) return 0;
 102     comprlen = lzf_compress(s, len, out, outlen);
 103     if (comprlen == 0) {
 104         zfree(out);
 105         return 0;
 106     }
 107     /* Data compressed! Let's save it on disk */
 108     byte = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_LZF;
 109     if (fwrite(&byte,1,1,fp) == 0) goto writeerr;
 110     if (rdbSaveLen(fp,comprlen) == -1) goto writeerr;
 111     if (rdbSaveLen(fp,len) == -1) goto writeerr;
 112     if (fwrite(out,comprlen,1,fp) == 0) goto writeerr;
 113     zfree(out);
 114     return comprlen;
 115
 116 writeerr:
 117     zfree(out);
 118     return -1;
 119 }
 120
 121 /* Save a string objet as [len][data] on disk. If the object is a string
 122  * representation of an integer value we try to safe it in a special form */
 123 int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) {
 124     int enclen;
 125
 126     /* Try integer encoding */
 127     if (len <= 11) {
 128         unsigned char buf[5];
 129         if ((enclen = rdbTryIntegerEncoding((char*)s,len,buf)) > 0) {
 130             if (fwrite(buf,enclen,1,fp) == 0) return -1;
 131             return 0;
 132         }
 133     }
 134
 135     /* Try LZF compression - under 20 bytes it's unable to compress even
 136      * aaaaaaaaaaaaaaaaaa so skip it */
 137     if (server.rdbcompression && len > 20) {
 138         int retval;
 139
 140         retval = rdbSaveLzfStringObject(fp,s,len);
 141         if (retval == -1) return -1;
 142         if (retval > 0) return 0;
 143         /* retval == 0 means data can't be compressed, save the old way */
 144     }
 145
 146     /* Store verbatim */
 147     if (rdbSaveLen(fp,len) == -1) return -1;
 148     if (len && fwrite(s,len,1,fp) == 0) return -1;
 149     return 0;
 150 }
 151
 152 /* Save a long long value as either an encoded string or a string. */
 153 int rdbSaveLongLongAsStringObject(FILE *fp, long long value) {
 154     unsigned char buf[32];
 155     int enclen = rdbEncodeInteger(value,buf);
 156     if (enclen > 0) {
 157         if (fwrite(buf,enclen,1,fp) == 0) return -1;
 158     } else {
 159         /* Encode as string */
 160         enclen = ll2string((char*)buf,32,value);
 161         redisAssert(enclen < 32);
 162         if (rdbSaveLen(fp,enclen) == -1) return -1;
 163         if (fwrite(buf,enclen,1,fp) == 0) return -1;
 164     }
 165     return 0;
 166 }
 167
 168 /* Like rdbSaveStringObjectRaw() but handle encoded objects */
 169 int rdbSaveStringObject(FILE *fp, robj *obj) {
 170     /* Avoid to decode the object, then encode it again, if the
 171      * object is alrady integer encoded. */
 172     if (obj->encoding == REDIS_ENCODING_INT) {
 173         return rdbSaveLongLongAsStringObject(fp,(long)obj->ptr);
 174     } else {
 175         redisAssert(obj->encoding == REDIS_ENCODING_RAW);
 176         return rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr));
 177     }
 178 }
 179
 180 /* Save a double value. Doubles are saved as strings prefixed by an unsigned
 181  * 8 bit integer specifing the length of the representation.
 182  * This 8 bit integer has special values in order to specify the following
 183  * conditions:
 184  * 253: not a number
 185  * 254: + inf
 186  * 255: - inf
 187  */
 188 int rdbSaveDoubleValue(FILE *fp, double val) {
 189     unsigned char buf[128];
 190     int len;
 191
 192     if (isnan(val)) {
 193         buf[0] = 253;
 194         len = 1;
 195     } else if (!isfinite(val)) {
 196         len = 1;
 197         buf[0] = (val < 0) ? 255 : 254;
 198     } else {
 199 #if (DBL_MANT_DIG >= 52) && (LLONG_MAX == 0x7fffffffffffffffLL)
 200         /* Check if the float is in a safe range to be casted into a
 201          * long long. We are assuming that long long is 64 bit here.
 202          * Also we are assuming that there are no implementations around where
 203          * double has precision < 52 bit.
 204          *
 205          * Under this assumptions we test if a double is inside an interval
 206          * where casting to long long is safe. Then using two castings we
 207          * make sure the decimal part is zero. If all this is true we use
 208          * integer printing function that is much faster. */
 209         double min = -4503599627370495; /* (2^52)-1 */
 210         double max = 4503599627370496; /* -(2^52) */
 211         if (val > min && val < max && val == ((double)((long long)val)))
 212             ll2string((char*)buf+1,sizeof(buf),(long long)val);
 213         else
 214 #endif
 215             snprintf((char*)buf+1,sizeof(buf)-1,"%.17g",val);
 216         buf[0] = strlen((char*)buf+1);
 217         len = buf[0]+1;
 218     }
 219     if (fwrite(buf,len,1,fp) == 0) return -1;
 220     return 0;
 221 }
 222
 223 /* Save a Redis object. */
 224 int rdbSaveObject(FILE *fp, robj *o) {
 225     if (o->type == REDIS_STRING) {
 226         /* Save a string value */
 227         if (rdbSaveStringObject(fp,o) == -1) return -1;
 228     } else if (o->type == REDIS_LIST) {
 229         /* Save a list value */
 230         if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 231             unsigned char *p;
 232             unsigned char *vstr;
 233             unsigned int vlen;
 234             long long vlong;
 235
 236             if (rdbSaveLen(fp,ziplistLen(o->ptr)) == -1) return -1;
 237             p = ziplistIndex(o->ptr,0);
 238             while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 239                 if (vstr) {
 240                     if (rdbSaveRawString(fp,vstr,vlen) == -1)
 241                         return -1;
 242                 } else {
 243                     if (rdbSaveLongLongAsStringObject(fp,vlong) == -1)
 244                         return -1;
 245                 }
 246                 p = ziplistNext(o->ptr,p);
 247             }
 248         } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 249             list *list = o->ptr;
 250             listIter li;
 251             listNode *ln;
 252
 253             if (rdbSaveLen(fp,listLength(list)) == -1) return -1;
 254             listRewind(list,&li);
 255             while((ln = listNext(&li))) {
 256                 robj *eleobj = listNodeValue(ln);
 257                 if (rdbSaveStringObject(fp,eleobj) == -1) return -1;
 258             }
 259         } else {
 260             redisPanic("Unknown list encoding");
 261         }
 262     } else if (o->type == REDIS_SET) {
 263         /* Save a set value */
 264         if (o->encoding == REDIS_ENCODING_HT) {
 265             dict *set = o->ptr;
 266             dictIterator *di = dictGetIterator(set);
 267             dictEntry *de;
 268
 269             if (rdbSaveLen(fp,dictSize(set)) == -1) return -1;
 270             while((de = dictNext(di)) != NULL) {
 271                 robj *eleobj = dictGetEntryKey(de);
 272                 if (rdbSaveStringObject(fp,eleobj) == -1) return -1;
 273             }
 274             dictReleaseIterator(di);
 275         } else if (o->encoding == REDIS_ENCODING_INTSET) {
 276             intset *is = o->ptr;
 277             int64_t llval;
 278             int i = 0;
 279
 280             if (rdbSaveLen(fp,intsetLen(is)) == -1) return -1;
 281             while(intsetGet(is,i++,&llval)) {
 282                 if (rdbSaveLongLongAsStringObject(fp,llval) == -1) return -1;
 283             }
 284         } else {
 285             redisPanic("Unknown set encoding");
 286         }
 287     } else if (o->type == REDIS_ZSET) {
 288         /* Save a set value */
 289         zset *zs = o->ptr;
 290         dictIterator *di = dictGetIterator(zs->dict);
 291         dictEntry *de;
 292
 293         if (rdbSaveLen(fp,dictSize(zs->dict)) == -1) return -1;
 294         while((de = dictNext(di)) != NULL) {
 295             robj *eleobj = dictGetEntryKey(de);
 296             double *score = dictGetEntryVal(de);
 297
 298             if (rdbSaveStringObject(fp,eleobj) == -1) return -1;
 299             if (rdbSaveDoubleValue(fp,*score) == -1) return -1;
 300         }
 301         dictReleaseIterator(di);
 302     } else if (o->type == REDIS_HASH) {
 303         /* Save a hash value */
 304         if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 305             unsigned char *p = zipmapRewind(o->ptr);
 306             unsigned int count = zipmapLen(o->ptr);
 307             unsigned char *key, *val;
 308             unsigned int klen, vlen;
 309
 310             if (rdbSaveLen(fp,count) == -1) return -1;
 311             while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) {
 312                 if (rdbSaveRawString(fp,key,klen) == -1) return -1;
 313                 if (rdbSaveRawString(fp,val,vlen) == -1) return -1;
 314             }
 315         } else {
 316             dictIterator *di = dictGetIterator(o->ptr);
 317             dictEntry *de;
 318
 319             if (rdbSaveLen(fp,dictSize((dict*)o->ptr)) == -1) return -1;
 320             while((de = dictNext(di)) != NULL) {
 321                 robj *key = dictGetEntryKey(de);
 322                 robj *val = dictGetEntryVal(de);
 323
 324                 if (rdbSaveStringObject(fp,key) == -1) return -1;
 325                 if (rdbSaveStringObject(fp,val) == -1) return -1;
 326             }
 327             dictReleaseIterator(di);
 328         }
 329     } else {
 330         redisPanic("Unknown object type");
 331     }
 332     return 0;
 333 }
 334
 335 /* Return the length the object will have on disk if saved with
 336  * the rdbSaveObject() function. Currently we use a trick to get
 337  * this length with very little changes to the code. In the future
 338  * we could switch to a faster solution. */
 339 off_t rdbSavedObjectLen(robj *o, FILE *fp) {
 340     if (fp == NULL) fp = server.devnull;
 341     rewind(fp);
 342     redisAssert(rdbSaveObject(fp,o) != 1);
 343     return ftello(fp);
 344 }
 345
 346 /* Return the number of pages required to save this object in the swap file */
 347 off_t rdbSavedObjectPages(robj *o, FILE *fp) {
 348     off_t bytes = rdbSavedObjectLen(o,fp);
 349
 350     return (bytes+(server.vm_page_size-1))/server.vm_page_size;
 351 }
 352
 353 /* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
 354 int rdbSave(char *filename) {
 355     dictIterator *di = NULL;
 356     dictEntry *de;
 357     FILE *fp;
 358     char tmpfile[256];
 359     int j;
 360     time_t now = time(NULL);
 361
 362     /* Wait for I/O therads to terminate, just in case this is a
 363      * foreground-saving, to avoid seeking the swap file descriptor at the
 364      * same time. */
 365     if (server.vm_enabled)
 366         waitEmptyIOJobsQueue();
 367
 368     snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
 369     fp = fopen(tmpfile,"w");
 370     if (!fp) {
 371         redisLog(REDIS_WARNING, "Failed saving the DB: %s", strerror(errno));
 372         return REDIS_ERR;
 373     }
 374     if (fwrite("REDIS0001",9,1,fp) == 0) goto werr;
 375     for (j = 0; j < server.dbnum; j++) {
 376         redisDb *db = server.db+j;
 377         dict *d = db->dict;
 378         if (dictSize(d) == 0) continue;
 379         di = dictGetIterator(d);
 380         if (!di) {
 381             fclose(fp);
 382             return REDIS_ERR;
 383         }
 384
 385         /* Write the SELECT DB opcode */
 386         if (rdbSaveType(fp,REDIS_SELECTDB) == -1) goto werr;
 387         if (rdbSaveLen(fp,j) == -1) goto werr;
 388
 389         /* Iterate this DB writing every entry */
 390         while((de = dictNext(di)) != NULL) {
 391             sds keystr = dictGetEntryKey(de);
 392             robj key, *o = dictGetEntryVal(de);
 393             time_t expiretime;
 394
 395             initStaticStringObject(key,keystr);
 396             expiretime = getExpire(db,&key);
 397
 398             /* Save the expire time */
 399             if (expiretime != -1) {
 400                 /* If this key is already expired skip it */
 401                 if (expiretime < now) continue;
 402                 if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) goto werr;
 403                 if (rdbSaveTime(fp,expiretime) == -1) goto werr;
 404             }
 405             /* Save the key and associated value. This requires special
 406              * handling if the value is swapped out. */
 407             if (!server.vm_enabled || o->storage == REDIS_VM_MEMORY ||
 408                                       o->storage == REDIS_VM_SWAPPING) {
 409                 /* Save type, key, value */
 410                 if (rdbSaveType(fp,o->type) == -1) goto werr;
 411                 if (rdbSaveStringObject(fp,&key) == -1) goto werr;
 412                 if (rdbSaveObject(fp,o) == -1) goto werr;
 413             } else {
 414                 /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
 415                 robj *po;
 416                 /* Get a preview of the object in memory */
 417                 po = vmPreviewObject(o);
 418                 /* Save type, key, value */
 419                 if (rdbSaveType(fp,po->type) == -1) goto werr;
 420                 if (rdbSaveStringObject(fp,&key) == -1) goto werr;
 421                 if (rdbSaveObject(fp,po) == -1) goto werr;
 422                 /* Remove the loaded object from memory */
 423                 decrRefCount(po);
 424             }
 425         }
 426         dictReleaseIterator(di);
 427     }
 428     /* EOF opcode */
 429     if (rdbSaveType(fp,REDIS_EOF) == -1) goto werr;
 430
 431     /* Make sure data will not remain on the OS's output buffers */
 432     fflush(fp);
 433     fsync(fileno(fp));
 434     fclose(fp);
 435
 436     /* Use RENAME to make sure the DB file is changed atomically only
 437      * if the generate DB file is ok. */
 438     if (rename(tmpfile,filename) == -1) {
 439         redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
 440         unlink(tmpfile);
 441         return REDIS_ERR;
 442     }
 443     redisLog(REDIS_NOTICE,"DB saved on disk");
 444     server.dirty = 0;
 445     server.lastsave = time(NULL);
 446     return REDIS_OK;
 447
 448 werr:
 449     fclose(fp);
 450     unlink(tmpfile);
 451     redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
 452     if (di) dictReleaseIterator(di);
 453     return REDIS_ERR;
 454 }
 455
 456 int rdbSaveBackground(char *filename) {
 457     pid_t childpid;
 458
 459     if (server.bgsavechildpid != -1) return REDIS_ERR;
 460     if (server.vm_enabled) waitEmptyIOJobsQueue();
 461     server.dirty_before_bgsave = server.dirty;
 462     if ((childpid = fork()) == 0) {
 463         /* Child */
 464         if (server.vm_enabled) vmReopenSwapFile();
 465         if (server.ipfd > 0) close(server.ipfd);
 466         if (server.sofd > 0) close(server.sofd);
 467         if (rdbSave(filename) == REDIS_OK) {
 468             _exit(0);
 469         } else {
 470             _exit(1);
 471         }
 472     } else {
 473         /* Parent */
 474         if (childpid == -1) {
 475             redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
 476                 strerror(errno));
 477             return REDIS_ERR;
 478         }
 479         redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
 480         server.bgsavechildpid = childpid;
 481         updateDictResizePolicy();
 482         return REDIS_OK;
 483     }
 484     return REDIS_OK; /* unreached */
 485 }
 486
 487 void rdbRemoveTempFile(pid_t childpid) {
 488     char tmpfile[256];
 489
 490     snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
 491     unlink(tmpfile);
 492 }
 493
 494 int rdbLoadType(FILE *fp) {
 495     unsigned char type;
 496     if (fread(&type,1,1,fp) == 0) return -1;
 497     return type;
 498 }
 499
 500 time_t rdbLoadTime(FILE *fp) {
 501     int32_t t32;
 502     if (fread(&t32,4,1,fp) == 0) return -1;
 503     return (time_t) t32;
 504 }
 505
 506 /* Load an encoded length from the DB, see the REDIS_RDB_* defines on the top
 507  * of this file for a description of how this are stored on disk.
 508  *
 509  * isencoded is set to 1 if the readed length is not actually a length but
 510  * an "encoding type", check the above comments for more info */
 511 uint32_t rdbLoadLen(FILE *fp, int *isencoded) {
 512     unsigned char buf[2];
 513     uint32_t len;
 514     int type;
 515
 516     if (isencoded) *isencoded = 0;
 517     if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR;
 518     type = (buf[0]&0xC0)>>6;
 519     if (type == REDIS_RDB_6BITLEN) {
 520         /* Read a 6 bit len */
 521         return buf[0]&0x3F;
 522     } else if (type == REDIS_RDB_ENCVAL) {
 523         /* Read a 6 bit len encoding type */
 524         if (isencoded) *isencoded = 1;
 525         return buf[0]&0x3F;
 526     } else if (type == REDIS_RDB_14BITLEN) {
 527         /* Read a 14 bit len */
 528         if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR;
 529         return ((buf[0]&0x3F)<<8)|buf[1];
 530     } else {
 531         /* Read a 32 bit len */
 532         if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR;
 533         return ntohl(len);
 534     }
 535 }
 536
 537 /* Load an integer-encoded object from file 'fp', with the specified
 538  * encoding type 'enctype'. If encode is true the function may return
 539  * an integer-encoded object as reply, otherwise the returned object
 540  * will always be encoded as a raw string. */
 541 robj *rdbLoadIntegerObject(FILE *fp, int enctype, int encode) {
 542     unsigned char enc[4];
 543     long long val;
 544
 545     if (enctype == REDIS_RDB_ENC_INT8) {
 546         if (fread(enc,1,1,fp) == 0) return NULL;
 547         val = (signed char)enc[0];
 548     } else if (enctype == REDIS_RDB_ENC_INT16) {
 549         uint16_t v;
 550         if (fread(enc,2,1,fp) == 0) return NULL;
 551         v = enc[0]|(enc[1]<<8);
 552         val = (int16_t)v;
 553     } else if (enctype == REDIS_RDB_ENC_INT32) {
 554         uint32_t v;
 555         if (fread(enc,4,1,fp) == 0) return NULL;
 556         v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
 557         val = (int32_t)v;
 558     } else {
 559         val = 0; /* anti-warning */
 560         redisPanic("Unknown RDB integer encoding type");
 561     }
 562     if (encode)
 563         return createStringObjectFromLongLong(val);
 564     else
 565         return createObject(REDIS_STRING,sdsfromlonglong(val));
 566 }
 567
 568 robj *rdbLoadLzfStringObject(FILE*fp) {
 569     unsigned int len, clen;
 570     unsigned char *c = NULL;
 571     sds val = NULL;
 572
 573     if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 574     if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 575     if ((c = zmalloc(clen)) == NULL) goto err;
 576     if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
 577     if (fread(c,clen,1,fp) == 0) goto err;
 578     if (lzf_decompress(c,clen,val,len) == 0) goto err;
 579     zfree(c);
 580     return createObject(REDIS_STRING,val);
 581 err:
 582     zfree(c);
 583     sdsfree(val);
 584     return NULL;
 585 }
 586
 587 robj *rdbGenericLoadStringObject(FILE*fp, int encode) {
 588     int isencoded;
 589     uint32_t len;
 590     sds val;
 591
 592     len = rdbLoadLen(fp,&isencoded);
 593     if (isencoded) {
 594         switch(len) {
 595         case REDIS_RDB_ENC_INT8:
 596         case REDIS_RDB_ENC_INT16:
 597         case REDIS_RDB_ENC_INT32:
 598             return rdbLoadIntegerObject(fp,len,encode);
 599         case REDIS_RDB_ENC_LZF:
 600             return rdbLoadLzfStringObject(fp);
 601         default:
 602             redisPanic("Unknown RDB encoding type");
 603         }
 604     }
 605
 606     if (len == REDIS_RDB_LENERR) return NULL;
 607     val = sdsnewlen(NULL,len);
 608     if (len && fread(val,len,1,fp) == 0) {
 609         sdsfree(val);
 610         return NULL;
 611     }
 612     return createObject(REDIS_STRING,val);
 613 }
 614
 615 robj *rdbLoadStringObject(FILE *fp) {
 616     return rdbGenericLoadStringObject(fp,0);
 617 }
 618
 619 robj *rdbLoadEncodedStringObject(FILE *fp) {
 620     return rdbGenericLoadStringObject(fp,1);
 621 }
 622
 623 /* For information about double serialization check rdbSaveDoubleValue() */
 624 int rdbLoadDoubleValue(FILE *fp, double *val) {
 625     char buf[128];
 626     unsigned char len;
 627
 628     if (fread(&len,1,1,fp) == 0) return -1;
 629     switch(len) {
 630     case 255: *val = R_NegInf; return 0;
 631     case 254: *val = R_PosInf; return 0;
 632     case 253: *val = R_Nan; return 0;
 633     default:
 634         if (fread(buf,len,1,fp) == 0) return -1;
 635         buf[len] = '\0';
 636         sscanf(buf, "%lg", val);
 637         return 0;
 638     }
 639 }
 640
 641 /* Load a Redis object of the specified type from the specified file.
 642  * On success a newly allocated object is returned, otherwise NULL. */
 643 robj *rdbLoadObject(int type, FILE *fp) {
 644     robj *o, *ele, *dec;
 645     size_t len;
 646     unsigned int i;
 647
 648     redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
 649     if (type == REDIS_STRING) {
 650         /* Read string value */
 651         if ((o = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 652         o = tryObjectEncoding(o);
 653     } else if (type == REDIS_LIST) {
 654         /* Read list value */
 655         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 656
 657         /* Use a real list when there are too many entries */
 658         if (len > server.list_max_ziplist_entries) {
 659             o = createListObject();
 660         } else {
 661             o = createZiplistObject();
 662         }
 663
 664         /* Load every single element of the list */
 665         while(len--) {
 666             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 667
 668             /* If we are using a ziplist and the value is too big, convert
 669              * the object to a real list. */
 670             if (o->encoding == REDIS_ENCODING_ZIPLIST &&
 671                 ele->encoding == REDIS_ENCODING_RAW &&
 672                 sdslen(ele->ptr) > server.list_max_ziplist_value)
 673                     listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
 674
 675             if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 676                 dec = getDecodedObject(ele);
 677                 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
 678                 decrRefCount(dec);
 679                 decrRefCount(ele);
 680             } else {
 681                 ele = tryObjectEncoding(ele);
 682                 listAddNodeTail(o->ptr,ele);
 683             }
 684         }
 685     } else if (type == REDIS_SET) {
 686         /* Read list/set value */
 687         if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 688
 689         /* Use a regular set when there are too many entries. */
 690         if (len > server.set_max_intset_entries) {
 691             o = createSetObject();
 692             /* It's faster to expand the dict to the right size asap in order
 693              * to avoid rehashing */
 694             if (len > DICT_HT_INITIAL_SIZE)
 695                 dictExpand(o->ptr,len);
 696         } else {
 697             o = createIntsetObject();
 698         }
 699
 700         /* Load every single element of the list/set */
 701         for (i = 0; i < len; i++) {
 702             long long llval;
 703             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 704             ele = tryObjectEncoding(ele);
 705
 706             if (o->encoding == REDIS_ENCODING_INTSET) {
 707                 /* Fetch integer value from element */
 708                 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
 709                     o->ptr = intsetAdd(o->ptr,llval,NULL);
 710                 } else {
 711                     setTypeConvert(o,REDIS_ENCODING_HT);
 712                     dictExpand(o->ptr,len);
 713                 }
 714             }
 715
 716             /* This will also be called when the set was just converted
 717              * to regular hashtable encoded set */
 718             if (o->encoding == REDIS_ENCODING_HT) {
 719                 dictAdd((dict*)o->ptr,ele,NULL);
 720             } else {
 721                 decrRefCount(ele);
 722             }
 723         }
 724     } else if (type == REDIS_ZSET) {
 725         /* Read list/set value */
 726         size_t zsetlen;
 727         zset *zs;
 728
 729         if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 730         o = createZsetObject();
 731         zs = o->ptr;
 732         /* Load every single element of the list/set */
 733         while(zsetlen--) {
 734             robj *ele;
 735             double score;
 736             zskiplistNode *znode;
 737
 738             if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 739             ele = tryObjectEncoding(ele);
 740             if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;
 741             znode = zslInsert(zs->zsl,score,ele);
 742             dictAdd(zs->dict,ele,&znode->score);
 743             incrRefCount(ele); /* added to skiplist */
 744         }
 745     } else if (type == REDIS_HASH) {
 746         size_t hashlen;
 747
 748         if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
 749         o = createHashObject();
 750         /* Too many entries? Use an hash table. */
 751         if (hashlen > server.hash_max_zipmap_entries)
 752             convertToRealHash(o);
 753         /* Load every key/value, then set it into the zipmap or hash
 754          * table, as needed. */
 755         while(hashlen--) {
 756             robj *key, *val;
 757
 758             if ((key = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 759             if ((val = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
 760             /* If we are using a zipmap and there are too big values
 761              * the object is converted to real hash table encoding. */
 762             if (o->encoding != REDIS_ENCODING_HT &&
 763                ((key->encoding == REDIS_ENCODING_RAW &&
 764                 sdslen(key->ptr) > server.hash_max_zipmap_value) ||
 765                 (val->encoding == REDIS_ENCODING_RAW &&
 766                 sdslen(val->ptr) > server.hash_max_zipmap_value)))
 767             {
 768                     convertToRealHash(o);
 769             }
 770
 771             if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 772                 unsigned char *zm = o->ptr;
 773                 robj *deckey, *decval;
 774
 775                 /* We need raw string objects to add them to the zipmap */
 776                 deckey = getDecodedObject(key);
 777                 decval = getDecodedObject(val);
 778                 zm = zipmapSet(zm,deckey->ptr,sdslen(deckey->ptr),
 779                                   decval->ptr,sdslen(decval->ptr),NULL);
 780                 o->ptr = zm;
 781                 decrRefCount(deckey);
 782                 decrRefCount(decval);
 783                 decrRefCount(key);
 784                 decrRefCount(val);
 785             } else {
 786                 key = tryObjectEncoding(key);
 787                 val = tryObjectEncoding(val);
 788                 dictAdd((dict*)o->ptr,key,val);
 789             }
 790         }
 791     } else {
 792         redisPanic("Unknown object type");
 793     }
 794     return o;
 795 }
 796
 797 /* Mark that we are loading in the global state and setup the fields
 798  * needed to provide loading stats. */
 799 void startLoading(FILE *fp) {
 800     struct stat sb;
 801
 802     /* Load the DB */
 803     server.loading = 1;
 804     server.loading_start_time = time(NULL);
 805     if (fstat(fileno(fp), &sb) == -1) {
 806         server.loading_total_bytes = 1; /* just to avoid division by zero */
 807     } else {
 808         server.loading_total_bytes = sb.st_size;
 809     }
 810 }
 811
 812 /* Refresh the loading progress info */
 813 void loadingProgress(off_t pos) {
 814     server.loading_loaded_bytes = pos;
 815 }
 816
 817 /* Loading finished */
 818 void stopLoading(void) {
 819     server.loading = 0;
 820 }
 821
 822 int rdbLoad(char *filename) {
 823     FILE *fp;
 824     uint32_t dbid;
 825     int type, retval, rdbver;
 826     int swap_all_values = 0;
 827     redisDb *db = server.db+0;
 828     char buf[1024];
 829     time_t expiretime, now = time(NULL);
 830     long loops = 0;
 831
 832     fp = fopen(filename,"r");
 833     if (!fp) return REDIS_ERR;
 834     if (fread(buf,9,1,fp) == 0) goto eoferr;
 835     buf[9] = '\0';
 836     if (memcmp(buf,"REDIS",5) != 0) {
 837         fclose(fp);
 838         redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
 839         return REDIS_ERR;
 840     }
 841     rdbver = atoi(buf+5);
 842     if (rdbver != 1) {
 843         fclose(fp);
 844         redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
 845         return REDIS_ERR;
 846     }
 847
 848     startLoading(fp);
 849     while(1) {
 850         robj *key, *val;
 851         int force_swapout;
 852
 853         expiretime = -1;
 854
 855         /* Serve the clients from time to time */
 856         if (!(loops++ % 1000)) {
 857             loadingProgress(ftello(fp));
 858             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 859         }
 860
 861         /* Read type. */
 862         if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 863         if (type == REDIS_EXPIRETIME) {
 864             if ((expiretime = rdbLoadTime(fp)) == -1) goto eoferr;
 865             /* We read the time so we need to read the object type again */
 866             if ((type = rdbLoadType(fp)) == -1) goto eoferr;
 867         }
 868         if (type == REDIS_EOF) break;
 869         /* Handle SELECT DB opcode as a special case */
 870         if (type == REDIS_SELECTDB) {
 871             if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR)
 872                 goto eoferr;
 873             if (dbid >= (unsigned)server.dbnum) {
 874                 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
 875                 exit(1);
 876             }
 877             db = server.db+dbid;
 878             continue;
 879         }
 880         /* Read key */
 881         if ((key = rdbLoadStringObject(fp)) == NULL) goto eoferr;
 882         /* Read value */
 883         if ((val = rdbLoadObject(type,fp)) == NULL) goto eoferr;
 884         /* Check if the key already expired */
 885         if (expiretime != -1 && expiretime < now) {
 886             decrRefCount(key);
 887             decrRefCount(val);
 888             continue;
 889         }
 890         /* Add the new object in the hash table */
 891         retval = dbAdd(db,key,val);
 892         if (retval == REDIS_ERR) {
 893             redisLog(REDIS_WARNING,"Loading DB, duplicated key (%s) found! Unrecoverable error, exiting now.", key->ptr);
 894             exit(1);
 895         }
 896         /* Set the expire time if needed */
 897         if (expiretime != -1) setExpire(db,key,expiretime);
 898
 899         /* Handle swapping while loading big datasets when VM is on */
 900
 901         /* If we detecter we are hopeless about fitting something in memory
 902          * we just swap every new key on disk. Directly...
 903          * Note that's important to check for this condition before resorting
 904          * to random sampling, otherwise we may try to swap already
 905          * swapped keys. */
 906         if (swap_all_values) {
 907             dictEntry *de = dictFind(db->dict,key->ptr);
 908
 909             /* de may be NULL since the key already expired */
 910             if (de) {
 911                 vmpointer *vp;
 912                 val = dictGetEntryVal(de);
 913
 914                 if (val->refcount == 1 &&
 915                     (vp = vmSwapObjectBlocking(val)) != NULL)
 916                     dictGetEntryVal(de) = vp;
 917             }
 918             decrRefCount(key);
 919             continue;
 920         }
 921         decrRefCount(key);
 922
 923         /* Flush data on disk once 32 MB of additional RAM are used... */
 924         force_swapout = 0;
 925         if ((zmalloc_used_memory() - server.vm_max_memory) > 1024*1024*32)
 926             force_swapout = 1;
 927
 928         /* If we have still some hope of having some value fitting memory
 929          * then we try random sampling. */
 930         if (!swap_all_values && server.vm_enabled && force_swapout) {
 931             while (zmalloc_used_memory() > server.vm_max_memory) {
 932                 if (vmSwapOneObjectBlocking() == REDIS_ERR) break;
 933             }
 934             if (zmalloc_used_memory() > server.vm_max_memory)
 935                 swap_all_values = 1; /* We are already using too much mem */
 936         }
 937     }
 938     fclose(fp);
 939     stopLoading();
 940     return REDIS_OK;
 941
 942 eoferr: /* unexpected end of file is handled here with a fatal exit */
 943     redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
 944     exit(1);
 945     return REDIS_ERR; /* Just to avoid warning */
 946 }
 947
 948 /* A background saving child (BGSAVE) terminated its work. Handle this. */
 949 void backgroundSaveDoneHandler(int statloc) {
 950     int exitcode = WEXITSTATUS(statloc);
 951     int bysignal = WIFSIGNALED(statloc);
 952
 953     if (!bysignal && exitcode == 0) {
 954         redisLog(REDIS_NOTICE,
 955             "Background saving terminated with success");
 956         server.dirty = server.dirty - server.dirty_before_bgsave;
 957         server.lastsave = time(NULL);
 958     } else if (!bysignal && exitcode != 0) {
 959         redisLog(REDIS_WARNING, "Background saving error");
 960     } else {
 961         redisLog(REDIS_WARNING,
 962             "Background saving terminated by signal %d", WTERMSIG(statloc));
 963         rdbRemoveTempFile(server.bgsavechildpid);
 964     }
 965     server.bgsavechildpid = -1;
 966     /* Possibly there are slaves waiting for a BGSAVE in order to be served
 967      * (the first stage of SYNC is a bulk transfer of dump.rdb) */
 968     updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
 969 }