src/aof.c

   1 #include "redis.h"
   2
   3 #include <signal.h>
   4 #include <fcntl.h>
   5 #include <sys/stat.h>
   6 #include <sys/types.h>
   7 #include <sys/time.h>
   8 #include <sys/resource.h>
   9 #include <sys/wait.h>
  10
  11 /* Called when the user switches from "appendonly yes" to "appendonly no"
  12  * at runtime using the CONFIG command. */
  13 void stopAppendOnly(void) {
  14     flushAppendOnlyFile();
  15     aof_fsync(server.appendfd);
  16     close(server.appendfd);
  17
  18     server.appendfd = -1;
  19     server.appendseldb = -1;
  20     server.appendonly = 0;
  21     /* rewrite operation in progress? kill it, wait child exit */
  22     if (server.bgsavechildpid != -1) {
  23         int statloc;
  24
  25         if (kill(server.bgsavechildpid,SIGKILL) != -1)
  26             wait3(&statloc,0,NULL);
  27         /* reset the buffer accumulating changes while the child saves */
  28         sdsfree(server.bgrewritebuf);
  29         server.bgrewritebuf = sdsempty();
  30         server.bgsavechildpid = -1;
  31     }
  32 }
  33
  34 /* Called when the user switches from "appendonly no" to "appendonly yes"
  35  * at runtime using the CONFIG command. */
  36 int startAppendOnly(void) {
  37     server.appendonly = 1;
  38     server.lastfsync = time(NULL);
  39     server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
  40     if (server.appendfd == -1) {
  41         redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, but I can't open the AOF file: %s",strerror(errno));
  42         return REDIS_ERR;
  43     }
  44     if (rewriteAppendOnlyFileBackground() == REDIS_ERR) {
  45         server.appendonly = 0;
  46         close(server.appendfd);
  47         redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, I can't trigger a background AOF rewrite operation. Check the above logs for more info about the error.",strerror(errno));
  48         return REDIS_ERR;
  49     }
  50     return REDIS_OK;
  51 }
  52
  53 /* Write the append only file buffer on disk.
  54  *
  55  * Since we are required to write the AOF before replying to the client,
  56  * and the only way the client socket can get a write is entering when the
  57  * the event loop, we accumulate all the AOF writes in a memory
  58  * buffer and write it on disk using this function just before entering
  59  * the event loop again. */
  60 void flushAppendOnlyFile(void) {
  61     time_t now;
  62     ssize_t nwritten;
  63
  64     if (sdslen(server.aofbuf) == 0) return;
  65
  66     /* We want to perform a single write. This should be guaranteed atomic
  67      * at least if the filesystem we are writing is a real physical one.
  68      * While this will save us against the server being killed I don't think
  69      * there is much to do about the whole server stopping for power problems
  70      * or alike */
  71      nwritten = write(server.appendfd,server.aofbuf,sdslen(server.aofbuf));
  72      if (nwritten != (signed)sdslen(server.aofbuf)) {
  73         /* Ooops, we are in troubles. The best thing to do for now is
  74          * aborting instead of giving the illusion that everything is
  75          * working as expected. */
  76          if (nwritten == -1) {
  77             redisLog(REDIS_WARNING,"Exiting on error writing to the append-only file: %s",strerror(errno));
  78          } else {
  79             redisLog(REDIS_WARNING,"Exiting on short write while writing to the append-only file: %s",strerror(errno));
  80          }
  81          exit(1);
  82     }
  83     sdsfree(server.aofbuf);
  84     server.aofbuf = sdsempty();
  85
  86     /* Don't Fsync if no-appendfsync-on-rewrite is set to yes and we have
  87      * childs performing heavy I/O on disk. */
  88     if (server.no_appendfsync_on_rewrite &&
  89         (server.bgrewritechildpid != -1 || server.bgsavechildpid != -1))
  90             return;
  91     /* Fsync if needed */
  92     now = time(NULL);
  93     if (server.appendfsync == APPENDFSYNC_ALWAYS ||
  94         (server.appendfsync == APPENDFSYNC_EVERYSEC &&
  95          now-server.lastfsync > 1))
  96     {
  97         /* aof_fsync is defined as fdatasync() for Linux in order to avoid
  98          * flushing metadata. */
  99         aof_fsync(server.appendfd); /* Let's try to get this data on the disk */
 100         server.lastfsync = now;
 101     }
 102 }
 103
 104 sds catAppendOnlyGenericCommand(sds buf, int argc, robj **argv) {
 105     int j;
 106     buf = sdscatprintf(buf,"*%d\r\n",argc);
 107     for (j = 0; j < argc; j++) {
 108         robj *o = getDecodedObject(argv[j]);
 109         buf = sdscatprintf(buf,"$%lu\r\n",(unsigned long)sdslen(o->ptr));
 110         buf = sdscatlen(buf,o->ptr,sdslen(o->ptr));
 111         buf = sdscatlen(buf,"\r\n",2);
 112         decrRefCount(o);
 113     }
 114     return buf;
 115 }
 116
 117 sds catAppendOnlyExpireAtCommand(sds buf, robj *key, robj *seconds) {
 118     int argc = 3;
 119     long when;
 120     robj *argv[3];
 121
 122     /* Make sure we can use strtol */
 123     seconds = getDecodedObject(seconds);
 124     when = time(NULL)+strtol(seconds->ptr,NULL,10);
 125     decrRefCount(seconds);
 126
 127     argv[0] = createStringObject("EXPIREAT",8);
 128     argv[1] = key;
 129     argv[2] = createObject(REDIS_STRING,
 130         sdscatprintf(sdsempty(),"%ld",when));
 131     buf = catAppendOnlyGenericCommand(buf, argc, argv);
 132     decrRefCount(argv[0]);
 133     decrRefCount(argv[2]);
 134     return buf;
 135 }
 136
 137 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
 138     sds buf = sdsempty();
 139     robj *tmpargv[3];
 140
 141     /* The DB this command was targetting is not the same as the last command
 142      * we appendend. To issue a SELECT command is needed. */
 143     if (dictid != server.appendseldb) {
 144         char seldb[64];
 145
 146         snprintf(seldb,sizeof(seldb),"%d",dictid);
 147         buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n",
 148             (unsigned long)strlen(seldb),seldb);
 149         server.appendseldb = dictid;
 150     }
 151
 152     if (cmd->proc == expireCommand) {
 153         /* Translate EXPIRE into EXPIREAT */
 154         buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
 155     } else if (cmd->proc == setexCommand) {
 156         /* Translate SETEX to SET and EXPIREAT */
 157         tmpargv[0] = createStringObject("SET",3);
 158         tmpargv[1] = argv[1];
 159         tmpargv[2] = argv[3];
 160         buf = catAppendOnlyGenericCommand(buf,3,tmpargv);
 161         decrRefCount(tmpargv[0]);
 162         buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
 163     } else {
 164         buf = catAppendOnlyGenericCommand(buf,argc,argv);
 165     }
 166
 167     /* Append to the AOF buffer. This will be flushed on disk just before
 168      * of re-entering the event loop, so before the client will get a
 169      * positive reply about the operation performed. */
 170     server.aofbuf = sdscatlen(server.aofbuf,buf,sdslen(buf));
 171
 172     /* If a background append only file rewriting is in progress we want to
 173      * accumulate the differences between the child DB and the current one
 174      * in a buffer, so that when the child process will do its work we
 175      * can append the differences to the new append only file. */
 176     if (server.bgrewritechildpid != -1)
 177         server.bgrewritebuf = sdscatlen(server.bgrewritebuf,buf,sdslen(buf));
 178
 179     sdsfree(buf);
 180 }
 181
 182 /* In Redis commands are always executed in the context of a client, so in
 183  * order to load the append only file we need to create a fake client. */
 184 struct redisClient *createFakeClient(void) {
 185     struct redisClient *c = zmalloc(sizeof(*c));
 186
 187     selectDb(c,0);
 188     c->fd = -1;
 189     c->querybuf = sdsempty();
 190     c->argc = 0;
 191     c->argv = NULL;
 192     c->bufpos = 0;
 193     c->flags = 0;
 194     /* We set the fake client as a slave waiting for the synchronization
 195      * so that Redis will not try to send replies to this client. */
 196     c->replstate = REDIS_REPL_WAIT_BGSAVE_START;
 197     c->reply = listCreate();
 198     c->watched_keys = listCreate();
 199     listSetFreeMethod(c->reply,decrRefCount);
 200     listSetDupMethod(c->reply,dupClientReplyValue);
 201     initClientMultiState(c);
 202     return c;
 203 }
 204
 205 void freeFakeClient(struct redisClient *c) {
 206     sdsfree(c->querybuf);
 207     listRelease(c->reply);
 208     listRelease(c->watched_keys);
 209     freeClientMultiState(c);
 210     zfree(c);
 211 }
 212
 213 /* Replay the append log file. On error REDIS_OK is returned. On non fatal
 214  * error (the append only file is zero-length) REDIS_ERR is returned. On
 215  * fatal error an error message is logged and the program exists. */
 216 int loadAppendOnlyFile(char *filename) {
 217     struct redisClient *fakeClient;
 218     FILE *fp = fopen(filename,"r");
 219     struct redis_stat sb;
 220     int appendonly = server.appendonly;
 221     long loops = 0;
 222
 223     if (redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0)
 224         return REDIS_ERR;
 225
 226     if (fp == NULL) {
 227         redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
 228         exit(1);
 229     }
 230
 231     /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
 232      * to the same file we're about to read. */
 233     server.appendonly = 0;
 234
 235     fakeClient = createFakeClient();
 236     startLoading(fp);
 237
 238     while(1) {
 239         int argc, j;
 240         unsigned long len;
 241         robj **argv;
 242         char buf[128];
 243         sds argsds;
 244         struct redisCommand *cmd;
 245         int force_swapout;
 246
 247         /* Serve the clients from time to time */
 248         if (!(loops++ % 1000)) {
 249             loadingProgress(ftello(fp));
 250             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 251         }
 252
 253         if (fgets(buf,sizeof(buf),fp) == NULL) {
 254             if (feof(fp))
 255                 break;
 256             else
 257                 goto readerr;
 258         }
 259         if (buf[0] != '*') goto fmterr;
 260         argc = atoi(buf+1);
 261         argv = zmalloc(sizeof(robj*)*argc);
 262         for (j = 0; j < argc; j++) {
 263             if (fgets(buf,sizeof(buf),fp) == NULL) goto readerr;
 264             if (buf[0] != '$') goto fmterr;
 265             len = strtol(buf+1,NULL,10);
 266             argsds = sdsnewlen(NULL,len);
 267             if (len && fread(argsds,len,1,fp) == 0) goto fmterr;
 268             argv[j] = createObject(REDIS_STRING,argsds);
 269             if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */
 270         }
 271
 272         /* Command lookup */
 273         cmd = lookupCommand(argv[0]->ptr);
 274         if (!cmd) {
 275             redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr);
 276             exit(1);
 277         }
 278         /* Run the command in the context of a fake client */
 279         fakeClient->argc = argc;
 280         fakeClient->argv = argv;
 281         cmd->proc(fakeClient);
 282
 283         /* The fake client should not have a reply */
 284         redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
 285
 286         /* Clean up, ready for the next command */
 287         for (j = 0; j < argc; j++) decrRefCount(argv[j]);
 288         zfree(argv);
 289
 290         /* Handle swapping while loading big datasets when VM is on */
 291         force_swapout = 0;
 292         if ((zmalloc_used_memory() - server.vm_max_memory) > 1024*1024*32)
 293             force_swapout = 1;
 294
 295         if (server.vm_enabled && force_swapout) {
 296             while (zmalloc_used_memory() > server.vm_max_memory) {
 297                 if (vmSwapOneObjectBlocking() == REDIS_ERR) break;
 298             }
 299         }
 300     }
 301
 302     /* This point can only be reached when EOF is reached without errors.
 303      * If the client is in the middle of a MULTI/EXEC, log error and quit. */
 304     if (fakeClient->flags & REDIS_MULTI) goto readerr;
 305
 306     fclose(fp);
 307     freeFakeClient(fakeClient);
 308     server.appendonly = appendonly;
 309     stopLoading();
 310     return REDIS_OK;
 311
 312 readerr:
 313     if (feof(fp)) {
 314         redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
 315     } else {
 316         redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
 317     }
 318     exit(1);
 319 fmterr:
 320     redisLog(REDIS_WARNING,"Bad file format reading the append only file: make a backup of your AOF file, then use ./redis-check-aof --fix <filename>");
 321     exit(1);
 322 }
 323
 324 /* Write a sequence of commands able to fully rebuild the dataset into
 325  * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */
 326 int rewriteAppendOnlyFile(char *filename) {
 327     dictIterator *di = NULL;
 328     dictEntry *de;
 329     FILE *fp;
 330     char tmpfile[256];
 331     int j;
 332     time_t now = time(NULL);
 333
 334     /* Note that we have to use a different temp name here compared to the
 335      * one used by rewriteAppendOnlyFileBackground() function. */
 336     snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
 337     fp = fopen(tmpfile,"w");
 338     if (!fp) {
 339         redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
 340         return REDIS_ERR;
 341     }
 342     for (j = 0; j < server.dbnum; j++) {
 343         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
 344         redisDb *db = server.db+j;
 345         dict *d = db->dict;
 346         if (dictSize(d) == 0) continue;
 347         di = dictGetIterator(d);
 348         if (!di) {
 349             fclose(fp);
 350             return REDIS_ERR;
 351         }
 352
 353         /* SELECT the new DB */
 354         if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr;
 355         if (fwriteBulkLongLong(fp,j) == 0) goto werr;
 356
 357         /* Iterate this DB writing every entry */
 358         while((de = dictNext(di)) != NULL) {
 359             sds keystr = dictGetEntryKey(de);
 360             robj key, *o;
 361             time_t expiretime;
 362             int swapped;
 363
 364             keystr = dictGetEntryKey(de);
 365             o = dictGetEntryVal(de);
 366             initStaticStringObject(key,keystr);
 367             /* If the value for this key is swapped, load a preview in memory.
 368              * We use a "swapped" flag to remember if we need to free the
 369              * value object instead to just increment the ref count anyway
 370              * in order to avoid copy-on-write of pages if we are forked() */
 371             if (!server.vm_enabled || o->storage == REDIS_VM_MEMORY ||
 372                 o->storage == REDIS_VM_SWAPPING) {
 373                 swapped = 0;
 374             } else {
 375                 o = vmPreviewObject(o);
 376                 swapped = 1;
 377             }
 378             expiretime = getExpire(db,&key);
 379
 380             /* Save the key and associated value */
 381             if (o->type == REDIS_STRING) {
 382                 /* Emit a SET command */
 383                 char cmd[]="*3\r\n$3\r\nSET\r\n";
 384                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 385                 /* Key and value */
 386                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 387                 if (fwriteBulkObject(fp,o) == 0) goto werr;
 388             } else if (o->type == REDIS_LIST) {
 389                 /* Emit the RPUSHes needed to rebuild the list */
 390                 char cmd[]="*3\r\n$5\r\nRPUSH\r\n";
 391                 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 392                     unsigned char *zl = o->ptr;
 393                     unsigned char *p = ziplistIndex(zl,0);
 394                     unsigned char *vstr;
 395                     unsigned int vlen;
 396                     long long vlong;
 397
 398                     while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 399                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 400                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 401                         if (vstr) {
 402                             if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
 403                                 goto werr;
 404                         } else {
 405                             if (fwriteBulkLongLong(fp,vlong) == 0)
 406                                 goto werr;
 407                         }
 408                         p = ziplistNext(zl,p);
 409                     }
 410                 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 411                     list *list = o->ptr;
 412                     listNode *ln;
 413                     listIter li;
 414
 415                     listRewind(list,&li);
 416                     while((ln = listNext(&li))) {
 417                         robj *eleobj = listNodeValue(ln);
 418
 419                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 420                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 421                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 422                     }
 423                 } else {
 424                     redisPanic("Unknown list encoding");
 425                 }
 426             } else if (o->type == REDIS_SET) {
 427                 char cmd[]="*3\r\n$4\r\nSADD\r\n";
 428
 429                 /* Emit the SADDs needed to rebuild the set */
 430                 if (o->encoding == REDIS_ENCODING_INTSET) {
 431                     int ii = 0;
 432                     int64_t llval;
 433                     while(intsetGet(o->ptr,ii++,&llval)) {
 434                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 435                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 436                         if (fwriteBulkLongLong(fp,llval) == 0) goto werr;
 437                     }
 438                 } else if (o->encoding == REDIS_ENCODING_HT) {
 439                     dictIterator *di = dictGetIterator(o->ptr);
 440                     dictEntry *de;
 441                     while((de = dictNext(di)) != NULL) {
 442                         robj *eleobj = dictGetEntryKey(de);
 443                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 444                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 445                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 446                     }
 447                     dictReleaseIterator(di);
 448                 } else {
 449                     redisPanic("Unknown set encoding");
 450                 }
 451             } else if (o->type == REDIS_ZSET) {
 452                 /* Emit the ZADDs needed to rebuild the sorted set */
 453                 zset *zs = o->ptr;
 454                 dictIterator *di = dictGetIterator(zs->dict);
 455                 dictEntry *de;
 456
 457                 while((de = dictNext(di)) != NULL) {
 458                     char cmd[]="*4\r\n$4\r\nZADD\r\n";
 459                     robj *eleobj = dictGetEntryKey(de);
 460                     double *score = dictGetEntryVal(de);
 461
 462                     if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 463                     if (fwriteBulkObject(fp,&key) == 0) goto werr;
 464                     if (fwriteBulkDouble(fp,*score) == 0) goto werr;
 465                     if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 466                 }
 467                 dictReleaseIterator(di);
 468             } else if (o->type == REDIS_HASH) {
 469                 char cmd[]="*4\r\n$4\r\nHSET\r\n";
 470
 471                 /* Emit the HSETs needed to rebuild the hash */
 472                 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 473                     unsigned char *p = zipmapRewind(o->ptr);
 474                     unsigned char *field, *val;
 475                     unsigned int flen, vlen;
 476
 477                     while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
 478                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 479                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 480                         if (fwriteBulkString(fp,(char*)field,flen) == 0)
 481                             goto werr;
 482                         if (fwriteBulkString(fp,(char*)val,vlen) == 0)
 483                             goto werr;
 484                     }
 485                 } else {
 486                     dictIterator *di = dictGetIterator(o->ptr);
 487                     dictEntry *de;
 488
 489                     while((de = dictNext(di)) != NULL) {
 490                         robj *field = dictGetEntryKey(de);
 491                         robj *val = dictGetEntryVal(de);
 492
 493                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 494                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 495                         if (fwriteBulkObject(fp,field) == 0) goto werr;
 496                         if (fwriteBulkObject(fp,val) == 0) goto werr;
 497                     }
 498                     dictReleaseIterator(di);
 499                 }
 500             } else {
 501                 redisPanic("Unknown object type");
 502             }
 503             /* Save the expire time */
 504             if (expiretime != -1) {
 505                 char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n";
 506                 /* If this key is already expired skip it */
 507                 if (expiretime < now) continue;
 508                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 509                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 510                 if (fwriteBulkLongLong(fp,expiretime) == 0) goto werr;
 511             }
 512             if (swapped) decrRefCount(o);
 513         }
 514         dictReleaseIterator(di);
 515     }
 516
 517     /* Make sure data will not remain on the OS's output buffers */
 518     fflush(fp);
 519     aof_fsync(fileno(fp));
 520     fclose(fp);
 521
 522     /* Use RENAME to make sure the DB file is changed atomically only
 523      * if the generate DB file is ok. */
 524     if (rename(tmpfile,filename) == -1) {
 525         redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
 526         unlink(tmpfile);
 527         return REDIS_ERR;
 528     }
 529     redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
 530     return REDIS_OK;
 531
 532 werr:
 533     fclose(fp);
 534     unlink(tmpfile);
 535     redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
 536     if (di) dictReleaseIterator(di);
 537     return REDIS_ERR;
 538 }
 539
 540 /* This is how rewriting of the append only file in background works:
 541  *
 542  * 1) The user calls BGREWRITEAOF
 543  * 2) Redis calls this function, that forks():
 544  *    2a) the child rewrite the append only file in a temp file.
 545  *    2b) the parent accumulates differences in server.bgrewritebuf.
 546  * 3) When the child finished '2a' exists.
 547  * 4) The parent will trap the exit code, if it's OK, will append the
 548  *    data accumulated into server.bgrewritebuf into the temp file, and
 549  *    finally will rename(2) the temp file in the actual file name.
 550  *    The the new file is reopened as the new append only file. Profit!
 551  */
 552 int rewriteAppendOnlyFileBackground(void) {
 553     pid_t childpid;
 554
 555     if (server.bgrewritechildpid != -1) return REDIS_ERR;
 556     if (server.vm_enabled) waitEmptyIOJobsQueue();
 557     if ((childpid = fork()) == 0) {
 558         /* Child */
 559         char tmpfile[256];
 560
 561         if (server.vm_enabled) vmReopenSwapFile();
 562         if (server.ipfd > 0) close(server.ipfd);
 563         if (server.sofd > 0) close(server.sofd);
 564         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
 565         if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
 566             _exit(0);
 567         } else {
 568             _exit(1);
 569         }
 570     } else {
 571         /* Parent */
 572         if (childpid == -1) {
 573             redisLog(REDIS_WARNING,
 574                 "Can't rewrite append only file in background: fork: %s",
 575                 strerror(errno));
 576             return REDIS_ERR;
 577         }
 578         redisLog(REDIS_NOTICE,
 579             "Background append only file rewriting started by pid %d",childpid);
 580         server.bgrewritechildpid = childpid;
 581         updateDictResizePolicy();
 582         /* We set appendseldb to -1 in order to force the next call to the
 583          * feedAppendOnlyFile() to issue a SELECT command, so the differences
 584          * accumulated by the parent into server.bgrewritebuf will start
 585          * with a SELECT statement and it will be safe to merge. */
 586         server.appendseldb = -1;
 587         return REDIS_OK;
 588     }
 589     return REDIS_OK; /* unreached */
 590 }
 591
 592 void bgrewriteaofCommand(redisClient *c) {
 593     if (server.bgrewritechildpid != -1) {
 594         addReplyError(c,"Background append only file rewriting already in progress");
 595         return;
 596     }
 597     if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
 598         addReplyStatus(c,"Background append only file rewriting started");
 599     } else {
 600         addReply(c,shared.err);
 601     }
 602 }
 603
 604 void aofRemoveTempFile(pid_t childpid) {
 605     char tmpfile[256];
 606
 607     snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
 608     unlink(tmpfile);
 609 }
 610
 611 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
 612  * Handle this. */
 613 void backgroundRewriteDoneHandler(int statloc) {
 614     int exitcode = WEXITSTATUS(statloc);
 615     int bysignal = WIFSIGNALED(statloc);
 616
 617     if (!bysignal && exitcode == 0) {
 618         int fd;
 619         char tmpfile[256];
 620
 621         redisLog(REDIS_NOTICE,
 622             "Background append only file rewriting terminated with success");
 623         /* Now it's time to flush the differences accumulated by the parent */
 624         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) server.bgrewritechildpid);
 625         fd = open(tmpfile,O_WRONLY|O_APPEND);
 626         if (fd == -1) {
 627             redisLog(REDIS_WARNING, "Not able to open the temp append only file produced by the child: %s", strerror(errno));
 628             goto cleanup;
 629         }
 630         /* Flush our data... */
 631         if (write(fd,server.bgrewritebuf,sdslen(server.bgrewritebuf)) !=
 632                 (signed) sdslen(server.bgrewritebuf)) {
 633             redisLog(REDIS_WARNING, "Error or short write trying to flush the parent diff of the append log file in the child temp file: %s", strerror(errno));
 634             close(fd);
 635             goto cleanup;
 636         }
 637         redisLog(REDIS_NOTICE,"Parent diff flushed into the new append log file with success (%lu bytes)",sdslen(server.bgrewritebuf));
 638         /* Now our work is to rename the temp file into the stable file. And
 639          * switch the file descriptor used by the server for append only. */
 640         if (rename(tmpfile,server.appendfilename) == -1) {
 641             redisLog(REDIS_WARNING,"Can't rename the temp append only file into the stable one: %s", strerror(errno));
 642             close(fd);
 643             goto cleanup;
 644         }
 645         /* Mission completed... almost */
 646         redisLog(REDIS_NOTICE,"Append only file successfully rewritten.");
 647         if (server.appendfd != -1) {
 648             /* If append only is actually enabled... */
 649             close(server.appendfd);
 650             server.appendfd = fd;
 651             if (server.appendfsync != APPENDFSYNC_NO) aof_fsync(fd);
 652             server.appendseldb = -1; /* Make sure it will issue SELECT */
 653             redisLog(REDIS_NOTICE,"The new append only file was selected for future appends.");
 654         } else {
 655             /* If append only is disabled we just generate a dump in this
 656              * format. Why not? */
 657             close(fd);
 658         }
 659     } else if (!bysignal && exitcode != 0) {
 660         redisLog(REDIS_WARNING, "Background append only file rewriting error");
 661     } else {
 662         redisLog(REDIS_WARNING,
 663             "Background append only file rewriting terminated by signal %d",
 664             WTERMSIG(statloc));
 665     }
 666 cleanup:
 667     sdsfree(server.bgrewritebuf);
 668     server.bgrewritebuf = sdsempty();
 669     aofRemoveTempFile(server.bgrewritechildpid);
 670     server.bgrewritechildpid = -1;
 671 }