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 void aofUpdateCurrentSize(void);
  12
  13 /* Called when the user switches from "appendonly yes" to "appendonly no"
  14  * at runtime using the CONFIG command. */
  15 void stopAppendOnly(void) {
  16     flushAppendOnlyFile();
  17     aof_fsync(server.appendfd);
  18     close(server.appendfd);
  19
  20     server.appendfd = -1;
  21     server.appendseldb = -1;
  22     server.appendonly = 0;
  23     /* rewrite operation in progress? kill it, wait child exit */
  24     if (server.bgrewritechildpid != -1) {
  25         int statloc;
  26
  27         if (kill(server.bgrewritechildpid,SIGKILL) != -1)
  28             wait3(&statloc,0,NULL);
  29         /* reset the buffer accumulating changes while the child saves */
  30         sdsfree(server.bgrewritebuf);
  31         server.bgrewritebuf = sdsempty();
  32         server.bgrewritechildpid = -1;
  33     }
  34 }
  35
  36 /* Called when the user switches from "appendonly no" to "appendonly yes"
  37  * at runtime using the CONFIG command. */
  38 int startAppendOnly(void) {
  39     server.appendonly = 1;
  40     server.lastfsync = time(NULL);
  41     server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
  42     if (server.appendfd == -1) {
  43         redisLog(REDIS_WARNING,"Used tried to switch on AOF via CONFIG, but I can't open the AOF file: %s",strerror(errno));
  44         return REDIS_ERR;
  45     }
  46     if (rewriteAppendOnlyFileBackground() == REDIS_ERR) {
  47         server.appendonly = 0;
  48         close(server.appendfd);
  49         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));
  50         return REDIS_ERR;
  51     }
  52     return REDIS_OK;
  53 }
  54
  55 /* Write the append only file buffer on disk.
  56  *
  57  * Since we are required to write the AOF before replying to the client,
  58  * and the only way the client socket can get a write is entering when the
  59  * the event loop, we accumulate all the AOF writes in a memory
  60  * buffer and write it on disk using this function just before entering
  61  * the event loop again. */
  62 void flushAppendOnlyFile(void) {
  63     ssize_t nwritten;
  64
  65     if (sdslen(server.aofbuf) == 0) return;
  66
  67     /* We want to perform a single write. This should be guaranteed atomic
  68      * at least if the filesystem we are writing is a real physical one.
  69      * While this will save us against the server being killed I don't think
  70      * there is much to do about the whole server stopping for power problems
  71      * or alike */
  72      nwritten = write(server.appendfd,server.aofbuf,sdslen(server.aofbuf));
  73      if (nwritten != (signed)sdslen(server.aofbuf)) {
  74         /* Ooops, we are in troubles. The best thing to do for now is
  75          * aborting instead of giving the illusion that everything is
  76          * working as expected. */
  77          if (nwritten == -1) {
  78             redisLog(REDIS_WARNING,"Exiting on error writing to the append-only file: %s",strerror(errno));
  79          } else {
  80             redisLog(REDIS_WARNING,"Exiting on short write while writing to the append-only file: %s",strerror(errno));
  81          }
  82          exit(1);
  83     }
  84     sdsfree(server.aofbuf);
  85     server.aofbuf = sdsempty();
  86     server.appendonly_current_size += nwritten;
  87
  88     /* Don't fsync if no-appendfsync-on-rewrite is set to yes and there are
  89      * children doing I/O in the background. */
  90     if (server.no_appendfsync_on_rewrite &&
  91         (server.bgrewritechildpid != -1 || server.bgsavechildpid != -1))
  92             return;
  93
  94     /* Perform the fsync if needed. */
  95     if (server.appendfsync == APPENDFSYNC_ALWAYS ||
  96         (server.appendfsync == APPENDFSYNC_EVERYSEC &&
  97          server.unixtime > server.lastfsync))
  98     {
  99         /* aof_fsync is defined as fdatasync() for Linux in order to avoid
 100          * flushing metadata. */
 101         aof_fsync(server.appendfd); /* Let's try to get this data on the disk */
 102         server.lastfsync = server.unixtime;
 103     }
 104 }
 105
 106 sds catAppendOnlyGenericCommand(sds buf, int argc, robj **argv) {
 107     int j;
 108     buf = sdscatprintf(buf,"*%d\r\n",argc);
 109     for (j = 0; j < argc; j++) {
 110         robj *o = getDecodedObject(argv[j]);
 111         buf = sdscatprintf(buf,"$%lu\r\n",(unsigned long)sdslen(o->ptr));
 112         buf = sdscatlen(buf,o->ptr,sdslen(o->ptr));
 113         buf = sdscatlen(buf,"\r\n",2);
 114         decrRefCount(o);
 115     }
 116     return buf;
 117 }
 118
 119 sds catAppendOnlyExpireAtCommand(sds buf, robj *key, robj *seconds) {
 120     int argc = 3;
 121     long when;
 122     robj *argv[3];
 123
 124     /* Make sure we can use strtol */
 125     seconds = getDecodedObject(seconds);
 126     when = time(NULL)+strtol(seconds->ptr,NULL,10);
 127     decrRefCount(seconds);
 128
 129     argv[0] = createStringObject("EXPIREAT",8);
 130     argv[1] = key;
 131     argv[2] = createObject(REDIS_STRING,
 132         sdscatprintf(sdsempty(),"%ld",when));
 133     buf = catAppendOnlyGenericCommand(buf, argc, argv);
 134     decrRefCount(argv[0]);
 135     decrRefCount(argv[2]);
 136     return buf;
 137 }
 138
 139 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc) {
 140     sds buf = sdsempty();
 141     robj *tmpargv[3];
 142
 143     /* The DB this command was targetting is not the same as the last command
 144      * we appendend. To issue a SELECT command is needed. */
 145     if (dictid != server.appendseldb) {
 146         char seldb[64];
 147
 148         snprintf(seldb,sizeof(seldb),"%d",dictid);
 149         buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n",
 150             (unsigned long)strlen(seldb),seldb);
 151         server.appendseldb = dictid;
 152     }
 153
 154     if (cmd->proc == expireCommand) {
 155         /* Translate EXPIRE into EXPIREAT */
 156         buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
 157     } else if (cmd->proc == setexCommand) {
 158         /* Translate SETEX to SET and EXPIREAT */
 159         tmpargv[0] = createStringObject("SET",3);
 160         tmpargv[1] = argv[1];
 161         tmpargv[2] = argv[3];
 162         buf = catAppendOnlyGenericCommand(buf,3,tmpargv);
 163         decrRefCount(tmpargv[0]);
 164         buf = catAppendOnlyExpireAtCommand(buf,argv[1],argv[2]);
 165     } else {
 166         buf = catAppendOnlyGenericCommand(buf,argc,argv);
 167     }
 168
 169     /* Append to the AOF buffer. This will be flushed on disk just before
 170      * of re-entering the event loop, so before the client will get a
 171      * positive reply about the operation performed. */
 172     server.aofbuf = sdscatlen(server.aofbuf,buf,sdslen(buf));
 173
 174     /* If a background append only file rewriting is in progress we want to
 175      * accumulate the differences between the child DB and the current one
 176      * in a buffer, so that when the child process will do its work we
 177      * can append the differences to the new append only file. */
 178     if (server.bgrewritechildpid != -1)
 179         server.bgrewritebuf = sdscatlen(server.bgrewritebuf,buf,sdslen(buf));
 180
 181     sdsfree(buf);
 182 }
 183
 184 /* In Redis commands are always executed in the context of a client, so in
 185  * order to load the append only file we need to create a fake client. */
 186 struct redisClient *createFakeClient(void) {
 187     struct redisClient *c = zmalloc(sizeof(*c));
 188
 189     selectDb(c,0);
 190     c->fd = -1;
 191     c->querybuf = sdsempty();
 192     c->argc = 0;
 193     c->argv = NULL;
 194     c->bufpos = 0;
 195     c->flags = 0;
 196     /* We set the fake client as a slave waiting for the synchronization
 197      * so that Redis will not try to send replies to this client. */
 198     c->replstate = REDIS_REPL_WAIT_BGSAVE_START;
 199     c->reply = listCreate();
 200     c->watched_keys = listCreate();
 201     listSetFreeMethod(c->reply,decrRefCount);
 202     listSetDupMethod(c->reply,dupClientReplyValue);
 203     initClientMultiState(c);
 204     return c;
 205 }
 206
 207 void freeFakeClient(struct redisClient *c) {
 208     sdsfree(c->querybuf);
 209     listRelease(c->reply);
 210     listRelease(c->watched_keys);
 211     freeClientMultiState(c);
 212     zfree(c);
 213 }
 214
 215 /* Replay the append log file. On error REDIS_OK is returned. On non fatal
 216  * error (the append only file is zero-length) REDIS_ERR is returned. On
 217  * fatal error an error message is logged and the program exists. */
 218 int loadAppendOnlyFile(char *filename) {
 219     struct redisClient *fakeClient;
 220     FILE *fp = fopen(filename,"r");
 221     struct redis_stat sb;
 222     int appendonly = server.appendonly;
 223     long loops = 0;
 224
 225     if (fp && redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) {
 226         server.appendonly_current_size = 0;
 227         fclose(fp);
 228         return REDIS_ERR;
 229     }
 230
 231     if (fp == NULL) {
 232         redisLog(REDIS_WARNING,"Fatal error: can't open the append log file for reading: %s",strerror(errno));
 233         exit(1);
 234     }
 235
 236     /* Temporarily disable AOF, to prevent EXEC from feeding a MULTI
 237      * to the same file we're about to read. */
 238     server.appendonly = 0;
 239
 240     fakeClient = createFakeClient();
 241     startLoading(fp);
 242
 243     while(1) {
 244         int argc, j;
 245         unsigned long len;
 246         robj **argv;
 247         char buf[128];
 248         sds argsds;
 249         struct redisCommand *cmd;
 250
 251         /* Serve the clients from time to time */
 252         if (!(loops++ % 1000)) {
 253             loadingProgress(ftello(fp));
 254             aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
 255         }
 256
 257         if (fgets(buf,sizeof(buf),fp) == NULL) {
 258             if (feof(fp))
 259                 break;
 260             else
 261                 goto readerr;
 262         }
 263         if (buf[0] != '*') goto fmterr;
 264         argc = atoi(buf+1);
 265         argv = zmalloc(sizeof(robj*)*argc);
 266         for (j = 0; j < argc; j++) {
 267             if (fgets(buf,sizeof(buf),fp) == NULL) goto readerr;
 268             if (buf[0] != '$') goto fmterr;
 269             len = strtol(buf+1,NULL,10);
 270             argsds = sdsnewlen(NULL,len);
 271             if (len && fread(argsds,len,1,fp) == 0) goto fmterr;
 272             argv[j] = createObject(REDIS_STRING,argsds);
 273             if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */
 274         }
 275
 276         /* Command lookup */
 277         cmd = lookupCommand(argv[0]->ptr);
 278         if (!cmd) {
 279             redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr);
 280             exit(1);
 281         }
 282         /* Run the command in the context of a fake client */
 283         fakeClient->argc = argc;
 284         fakeClient->argv = argv;
 285         cmd->proc(fakeClient);
 286
 287         /* The fake client should not have a reply */
 288         redisAssert(fakeClient->bufpos == 0 && listLength(fakeClient->reply) == 0);
 289         /* The fake client should never get blocked */
 290         redisAssert((fakeClient->flags & REDIS_BLOCKED) == 0);
 291
 292         /* Clean up. Command code may have changed argv/argc so we use the
 293          * argv/argc of the client instead of the local variables. */
 294         for (j = 0; j < fakeClient->argc; j++)
 295             decrRefCount(fakeClient->argv[j]);
 296         zfree(fakeClient->argv);
 297     }
 298
 299     /* This point can only be reached when EOF is reached without errors.
 300      * If the client is in the middle of a MULTI/EXEC, log error and quit. */
 301     if (fakeClient->flags & REDIS_MULTI) goto readerr;
 302
 303     fclose(fp);
 304     freeFakeClient(fakeClient);
 305     server.appendonly = appendonly;
 306     stopLoading();
 307     aofUpdateCurrentSize();
 308     server.auto_aofrewrite_base_size = server.appendonly_current_size;
 309     return REDIS_OK;
 310
 311 readerr:
 312     if (feof(fp)) {
 313         redisLog(REDIS_WARNING,"Unexpected end of file reading the append only file");
 314     } else {
 315         redisLog(REDIS_WARNING,"Unrecoverable error reading the append only file: %s", strerror(errno));
 316     }
 317     exit(1);
 318 fmterr:
 319     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>");
 320     exit(1);
 321 }
 322
 323 /* Write a sequence of commands able to fully rebuild the dataset into
 324  * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */
 325 int rewriteAppendOnlyFile(char *filename) {
 326     dictIterator *di = NULL;
 327     dictEntry *de;
 328     FILE *fp;
 329     char tmpfile[256];
 330     int j;
 331     time_t now = time(NULL);
 332
 333     /* Note that we have to use a different temp name here compared to the
 334      * one used by rewriteAppendOnlyFileBackground() function. */
 335     snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid());
 336     fp = fopen(tmpfile,"w");
 337     if (!fp) {
 338         redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno));
 339         return REDIS_ERR;
 340     }
 341     for (j = 0; j < server.dbnum; j++) {
 342         char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n";
 343         redisDb *db = server.db+j;
 344         dict *d = db->dict;
 345         if (dictSize(d) == 0) continue;
 346         di = dictGetSafeIterator(d);
 347         if (!di) {
 348             fclose(fp);
 349             return REDIS_ERR;
 350         }
 351
 352         /* SELECT the new DB */
 353         if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr;
 354         if (fwriteBulkLongLong(fp,j) == 0) goto werr;
 355
 356         /* Iterate this DB writing every entry */
 357         while((de = dictNext(di)) != NULL) {
 358             sds keystr;
 359             robj key, *o;
 360             time_t expiretime;
 361
 362             keystr = dictGetEntryKey(de);
 363             o = dictGetEntryVal(de);
 364             initStaticStringObject(key,keystr);
 365
 366             expiretime = getExpire(db,&key);
 367
 368             /* Save the key and associated value */
 369             if (o->type == REDIS_STRING) {
 370                 /* Emit a SET command */
 371                 char cmd[]="*3\r\n$3\r\nSET\r\n";
 372                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 373                 /* Key and value */
 374                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 375                 if (fwriteBulkObject(fp,o) == 0) goto werr;
 376             } else if (o->type == REDIS_LIST) {
 377                 /* Emit the RPUSHes needed to rebuild the list */
 378                 char cmd[]="*3\r\n$5\r\nRPUSH\r\n";
 379                 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 380                     unsigned char *zl = o->ptr;
 381                     unsigned char *p = ziplistIndex(zl,0);
 382                     unsigned char *vstr;
 383                     unsigned int vlen;
 384                     long long vlong;
 385
 386                     while(ziplistGet(p,&vstr,&vlen,&vlong)) {
 387                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 388                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 389                         if (vstr) {
 390                             if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
 391                                 goto werr;
 392                         } else {
 393                             if (fwriteBulkLongLong(fp,vlong) == 0)
 394                                 goto werr;
 395                         }
 396                         p = ziplistNext(zl,p);
 397                     }
 398                 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
 399                     list *list = o->ptr;
 400                     listNode *ln;
 401                     listIter li;
 402
 403                     listRewind(list,&li);
 404                     while((ln = listNext(&li))) {
 405                         robj *eleobj = listNodeValue(ln);
 406
 407                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 408                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 409                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 410                     }
 411                 } else {
 412                     redisPanic("Unknown list encoding");
 413                 }
 414             } else if (o->type == REDIS_SET) {
 415                 char cmd[]="*3\r\n$4\r\nSADD\r\n";
 416
 417                 /* Emit the SADDs needed to rebuild the set */
 418                 if (o->encoding == REDIS_ENCODING_INTSET) {
 419                     int ii = 0;
 420                     int64_t llval;
 421                     while(intsetGet(o->ptr,ii++,&llval)) {
 422                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 423                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 424                         if (fwriteBulkLongLong(fp,llval) == 0) goto werr;
 425                     }
 426                 } else if (o->encoding == REDIS_ENCODING_HT) {
 427                     dictIterator *di = dictGetIterator(o->ptr);
 428                     dictEntry *de;
 429                     while((de = dictNext(di)) != NULL) {
 430                         robj *eleobj = dictGetEntryKey(de);
 431                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 432                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 433                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 434                     }
 435                     dictReleaseIterator(di);
 436                 } else {
 437                     redisPanic("Unknown set encoding");
 438                 }
 439             } else if (o->type == REDIS_ZSET) {
 440                 /* Emit the ZADDs needed to rebuild the sorted set */
 441                 char cmd[]="*4\r\n$4\r\nZADD\r\n";
 442
 443                 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
 444                     unsigned char *zl = o->ptr;
 445                     unsigned char *eptr, *sptr;
 446                     unsigned char *vstr;
 447                     unsigned int vlen;
 448                     long long vll;
 449                     double score;
 450
 451                     eptr = ziplistIndex(zl,0);
 452                     redisAssert(eptr != NULL);
 453                     sptr = ziplistNext(zl,eptr);
 454                     redisAssert(sptr != NULL);
 455
 456                     while (eptr != NULL) {
 457                         redisAssert(ziplistGet(eptr,&vstr,&vlen,&vll));
 458                         score = zzlGetScore(sptr);
 459
 460                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 461                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 462                         if (fwriteBulkDouble(fp,score) == 0) goto werr;
 463                         if (vstr != NULL) {
 464                             if (fwriteBulkString(fp,(char*)vstr,vlen) == 0)
 465                                 goto werr;
 466                         } else {
 467                             if (fwriteBulkLongLong(fp,vll) == 0)
 468                                 goto werr;
 469                         }
 470                         zzlNext(zl,&eptr,&sptr);
 471                     }
 472                 } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
 473                     zset *zs = o->ptr;
 474                     dictIterator *di = dictGetIterator(zs->dict);
 475                     dictEntry *de;
 476
 477                     while((de = dictNext(di)) != NULL) {
 478                         robj *eleobj = dictGetEntryKey(de);
 479                         double *score = dictGetEntryVal(de);
 480
 481                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 482                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 483                         if (fwriteBulkDouble(fp,*score) == 0) goto werr;
 484                         if (fwriteBulkObject(fp,eleobj) == 0) goto werr;
 485                     }
 486                     dictReleaseIterator(di);
 487                 } else {
 488                     redisPanic("Unknown sorted set encoding");
 489                 }
 490             } else if (o->type == REDIS_HASH) {
 491                 char cmd[]="*4\r\n$4\r\nHSET\r\n";
 492
 493                 /* Emit the HSETs needed to rebuild the hash */
 494                 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
 495                     unsigned char *p = zipmapRewind(o->ptr);
 496                     unsigned char *field, *val;
 497                     unsigned int flen, vlen;
 498
 499                     while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) {
 500                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 501                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 502                         if (fwriteBulkString(fp,(char*)field,flen) == 0)
 503                             goto werr;
 504                         if (fwriteBulkString(fp,(char*)val,vlen) == 0)
 505                             goto werr;
 506                     }
 507                 } else {
 508                     dictIterator *di = dictGetIterator(o->ptr);
 509                     dictEntry *de;
 510
 511                     while((de = dictNext(di)) != NULL) {
 512                         robj *field = dictGetEntryKey(de);
 513                         robj *val = dictGetEntryVal(de);
 514
 515                         if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 516                         if (fwriteBulkObject(fp,&key) == 0) goto werr;
 517                         if (fwriteBulkObject(fp,field) == 0) goto werr;
 518                         if (fwriteBulkObject(fp,val) == 0) goto werr;
 519                     }
 520                     dictReleaseIterator(di);
 521                 }
 522             } else {
 523                 redisPanic("Unknown object type");
 524             }
 525             /* Save the expire time */
 526             if (expiretime != -1) {
 527                 char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n";
 528                 /* If this key is already expired skip it */
 529                 if (expiretime < now) continue;
 530                 if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr;
 531                 if (fwriteBulkObject(fp,&key) == 0) goto werr;
 532                 if (fwriteBulkLongLong(fp,expiretime) == 0) goto werr;
 533             }
 534         }
 535         dictReleaseIterator(di);
 536     }
 537
 538     /* Make sure data will not remain on the OS's output buffers */
 539     fflush(fp);
 540     aof_fsync(fileno(fp));
 541     fclose(fp);
 542
 543     /* Use RENAME to make sure the DB file is changed atomically only
 544      * if the generate DB file is ok. */
 545     if (rename(tmpfile,filename) == -1) {
 546         redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno));
 547         unlink(tmpfile);
 548         return REDIS_ERR;
 549     }
 550     redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed");
 551     return REDIS_OK;
 552
 553 werr:
 554     fclose(fp);
 555     unlink(tmpfile);
 556     redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno));
 557     if (di) dictReleaseIterator(di);
 558     return REDIS_ERR;
 559 }
 560
 561 /* This is how rewriting of the append only file in background works:
 562  *
 563  * 1) The user calls BGREWRITEAOF
 564  * 2) Redis calls this function, that forks():
 565  *    2a) the child rewrite the append only file in a temp file.
 566  *    2b) the parent accumulates differences in server.bgrewritebuf.
 567  * 3) When the child finished '2a' exists.
 568  * 4) The parent will trap the exit code, if it's OK, will append the
 569  *    data accumulated into server.bgrewritebuf into the temp file, and
 570  *    finally will rename(2) the temp file in the actual file name.
 571  *    The the new file is reopened as the new append only file. Profit!
 572  */
 573 int rewriteAppendOnlyFileBackground(void) {
 574     pid_t childpid;
 575     long long start;
 576
 577     if (server.bgrewritechildpid != -1) return REDIS_ERR;
 578     start = ustime();
 579     if ((childpid = fork()) == 0) {
 580         char tmpfile[256];
 581
 582         /* Child */
 583         if (server.ipfd > 0) close(server.ipfd);
 584         if (server.sofd > 0) close(server.sofd);
 585         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid());
 586         if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) {
 587             _exit(0);
 588         } else {
 589             _exit(1);
 590         }
 591     } else {
 592         /* Parent */
 593         server.stat_fork_time = ustime()-start;
 594         if (childpid == -1) {
 595             redisLog(REDIS_WARNING,
 596                 "Can't rewrite append only file in background: fork: %s",
 597                 strerror(errno));
 598             return REDIS_ERR;
 599         }
 600         redisLog(REDIS_NOTICE,
 601             "Background append only file rewriting started by pid %d",childpid);
 602         server.bgrewritechildpid = childpid;
 603         updateDictResizePolicy();
 604         /* We set appendseldb to -1 in order to force the next call to the
 605          * feedAppendOnlyFile() to issue a SELECT command, so the differences
 606          * accumulated by the parent into server.bgrewritebuf will start
 607          * with a SELECT statement and it will be safe to merge. */
 608         server.appendseldb = -1;
 609         return REDIS_OK;
 610     }
 611     return REDIS_OK; /* unreached */
 612 }
 613
 614 void bgrewriteaofCommand(redisClient *c) {
 615     if (server.bgrewritechildpid != -1) {
 616         addReplyError(c,"Background append only file rewriting already in progress");
 617     } else if (server.bgsavechildpid != -1) {
 618         server.aofrewrite_scheduled = 1;
 619         addReplyStatus(c,"Background append only file rewriting scheduled");
 620     } else if (rewriteAppendOnlyFileBackground() == REDIS_OK) {
 621         addReplyStatus(c,"Background append only file rewriting started");
 622     } else {
 623         addReply(c,shared.err);
 624     }
 625 }
 626
 627 void aofRemoveTempFile(pid_t childpid) {
 628     char tmpfile[256];
 629
 630     snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid);
 631     unlink(tmpfile);
 632 }
 633
 634 /* Update the server.appendonly_current_size filed explicitly using stat(2)
 635  * to check the size of the file. This is useful after a rewrite or after
 636  * a restart, normally the size is updated just adding the write length
 637  * to the current lenght, that is much faster. */
 638 void aofUpdateCurrentSize(void) {
 639     struct redis_stat sb;
 640
 641     if (redis_fstat(server.appendfd,&sb) == -1) {
 642         redisLog(REDIS_WARNING,"Unable to check the AOF length: %s",
 643             strerror(errno));
 644     } else {
 645         server.appendonly_current_size = sb.st_size;
 646     }
 647 }
 648
 649 /* A background append only file rewriting (BGREWRITEAOF) terminated its work.
 650  * Handle this. */
 651 void backgroundRewriteDoneHandler(int exitcode, int bysignal) {
 652     if (!bysignal && exitcode == 0) {
 653         int fd;
 654         char tmpfile[256];
 655
 656         redisLog(REDIS_NOTICE,
 657             "Background append only file rewriting terminated with success");
 658         /* Now it's time to flush the differences accumulated by the parent */
 659         snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) server.bgrewritechildpid);
 660         fd = open(tmpfile,O_WRONLY|O_APPEND);
 661         if (fd == -1) {
 662             redisLog(REDIS_WARNING, "Not able to open the temp append only file produced by the child: %s", strerror(errno));
 663             goto cleanup;
 664         }
 665         /* Flush our data... */
 666         if (write(fd,server.bgrewritebuf,sdslen(server.bgrewritebuf)) !=
 667                 (signed) sdslen(server.bgrewritebuf)) {
 668             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));
 669             close(fd);
 670             goto cleanup;
 671         }
 672         redisLog(REDIS_NOTICE,"Parent diff flushed into the new append log file with success (%lu bytes)",sdslen(server.bgrewritebuf));
 673         /* Now our work is to rename the temp file into the stable file. And
 674          * switch the file descriptor used by the server for append only. */
 675         if (rename(tmpfile,server.appendfilename) == -1) {
 676             redisLog(REDIS_WARNING,"Can't rename the temp append only file into the stable one: %s", strerror(errno));
 677             close(fd);
 678             goto cleanup;
 679         }
 680         /* Mission completed... almost */
 681         redisLog(REDIS_NOTICE,"Append only file successfully rewritten.");
 682         if (server.appendfd != -1) {
 683             /* If append only is actually enabled... */
 684             close(server.appendfd);
 685             server.appendfd = fd;
 686             if (server.appendfsync != APPENDFSYNC_NO) aof_fsync(fd);
 687             server.appendseldb = -1; /* Make sure it will issue SELECT */
 688             redisLog(REDIS_NOTICE,"The new append only file was selected for future appends.");
 689             aofUpdateCurrentSize();
 690             server.auto_aofrewrite_base_size = server.appendonly_current_size;
 691         } else {
 692             /* If append only is disabled we just generate a dump in this
 693              * format. Why not? */
 694             close(fd);
 695         }
 696     } else if (!bysignal && exitcode != 0) {
 697         redisLog(REDIS_WARNING, "Background append only file rewriting error");
 698     } else {
 699         redisLog(REDIS_WARNING,
 700             "Background append only file rewriting terminated by signal %d",
 701             bysignal);
 702     }
 703 cleanup:
 704     sdsfree(server.bgrewritebuf);
 705     server.bgrewritebuf = sdsempty();
 706     aofRemoveTempFile(server.bgrewritechildpid);
 707     server.bgrewritechildpid = -1;
 708 }