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