src/rdb.c

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