src/bin/pg_resetxlog/pg_resetxlog.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * pg_resetxlog.c
   4  *        A utility to "zero out" the xlog when it's corrupt beyond recovery.
   5  *        Can also rebuild pg_control if needed.
   6  *
   7  * The theory of operation is fairly simple:
   8  *        1. Read the existing pg_control (which will include the last
   9  *               checkpoint record).  If it is an old format then update to
  10  *               current format.
  11  *        2. If pg_control is corrupt, attempt to intuit reasonable values,
  12  *               by scanning the old xlog if necessary.
  13  *        3. Modify pg_control to reflect a "shutdown" state with a checkpoint
  14  *               record at the start of xlog.
  15  *        4. Flush the existing xlog files and write a new segment with
  16  *               just a checkpoint record in it.  The new segment is positioned
  17  *               just past the end of the old xlog, so that existing LSNs in
  18  *               data pages will appear to be "in the past".
  19  * This is all pretty straightforward except for the intuition part of
  20  * step 2 ...
  21  *
  22  *
  23  * Portions Copyright (c) 1996-2009, PostgreSQL Global Development Group
  24  * Portions Copyright (c) 1994, Regents of the University of California
  25  *
  26  * $PostgreSQL$
  27  *
  28  *-------------------------------------------------------------------------
  29  */
  30
  31 /*
  32  * We have to use postgres.h not postgres_fe.h here, because there's so much
  33  * backend-only stuff in the XLOG include files we need.  But we need a
  34  * frontend-ish environment otherwise.  Hence this ugly hack.
  35  */
  36 #define FRONTEND 1
  37
  38 #include "postgres.h"
  39
  40 #include <dirent.h>
  41 #include <fcntl.h>
  42 #include <locale.h>
  43 #include <sys/stat.h>
  44 #include <sys/time.h>
  45 #include <time.h>
  46 #include <unistd.h>
  47 #ifdef HAVE_GETOPT_H
  48 #include <getopt.h>
  49 #endif
  50
  51 #include "access/transam.h"
  52 #include "access/tuptoaster.h"
  53 #include "access/multixact.h"
  54 #include "access/xlog_internal.h"
  55 #include "catalog/catversion.h"
  56 #include "catalog/pg_control.h"
  57
  58 extern int      optind;
  59 extern char *optarg;
  60
  61
  62 static ControlFileData ControlFile;             /* pg_control values */
  63 static uint32 newXlogId,
  64                         newXlogSeg;                     /* ID/Segment of new XLOG segment */
  65 static bool guessed = false;    /* T if we had to guess at any values */
  66 static const char *progname;
  67
  68 static bool ReadControlFile(void);
  69 static void GuessControlValues(void);
  70 static void PrintControlValues(bool guessed);
  71 static void RewriteControlFile(void);
  72 static void FindEndOfXLOG(void);
  73 static void KillExistingXLOG(void);
  74 static void KillExistingArchiveStatus(void);
  75 static void WriteEmptyXLOG(void);
  76 static void usage(void);
  77
  78
  79 int
  80 main(int argc, char *argv[])
  81 {
  82         int                     c;
  83         bool            force = false;
  84         bool            noupdate = false;
  85         uint32          set_xid_epoch = (uint32) -1;
  86         TransactionId set_xid = 0;
  87         Oid                     set_oid = 0;
  88         MultiXactId set_mxid = 0;
  89         MultiXactOffset set_mxoff = (MultiXactOffset) -1;
  90         uint32          minXlogTli = 0,
  91                                 minXlogId = 0,
  92                                 minXlogSeg = 0;
  93         char       *endptr;
  94         char       *endptr2;
  95         char       *endptr3;
  96         char       *DataDir;
  97         int                     fd;
  98         char            path[MAXPGPATH];
  99
 100         set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_resetxlog"));
 101
 102         progname = get_progname(argv[0]);
 103
 104         if (argc > 1)
 105         {
 106                 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
 107                 {
 108                         usage();
 109                         exit(0);
 110                 }
 111                 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
 112                 {
 113                         puts("pg_resetxlog (PostgreSQL) " PG_VERSION);
 114                         exit(0);
 115                 }
 116         }
 117
 118
 119         while ((c = getopt(argc, argv, "fl:m:no:O:x:e:")) != -1)
 120         {
 121                 switch (c)
 122                 {
 123                         case 'f':
 124                                 force = true;
 125                                 break;
 126
 127                         case 'n':
 128                                 noupdate = true;
 129                                 break;
 130
 131                         case 'e':
 132                                 set_xid_epoch = strtoul(optarg, &endptr, 0);
 133                                 if (endptr == optarg || *endptr != '\0')
 134                                 {
 135                                         fprintf(stderr, _("%s: invalid argument for option -e\n"), progname);
 136                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 137                                         exit(1);
 138                                 }
 139                                 if (set_xid_epoch == -1)
 140                                 {
 141                                         fprintf(stderr, _("%s: transaction ID epoch (-e) must not be -1\n"), progname);
 142                                         exit(1);
 143                                 }
 144                                 break;
 145
 146                         case 'x':
 147                                 set_xid = strtoul(optarg, &endptr, 0);
 148                                 if (endptr == optarg || *endptr != '\0')
 149                                 {
 150                                         fprintf(stderr, _("%s: invalid argument for option -x\n"), progname);
 151                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 152                                         exit(1);
 153                                 }
 154                                 if (set_xid == 0)
 155                                 {
 156                                         fprintf(stderr, _("%s: transaction ID (-x) must not be 0\n"), progname);
 157                                         exit(1);
 158                                 }
 159                                 break;
 160
 161                         case 'o':
 162                                 set_oid = strtoul(optarg, &endptr, 0);
 163                                 if (endptr == optarg || *endptr != '\0')
 164                                 {
 165                                         fprintf(stderr, _("%s: invalid argument for option -o\n"), progname);
 166                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 167                                         exit(1);
 168                                 }
 169                                 if (set_oid == 0)
 170                                 {
 171                                         fprintf(stderr, _("%s: OID (-o) must not be 0\n"), progname);
 172                                         exit(1);
 173                                 }
 174                                 break;
 175
 176                         case 'm':
 177                                 set_mxid = strtoul(optarg, &endptr, 0);
 178                                 if (endptr == optarg || *endptr != '\0')
 179                                 {
 180                                         fprintf(stderr, _("%s: invalid argument for option -m\n"), progname);
 181                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 182                                         exit(1);
 183                                 }
 184                                 if (set_mxid == 0)
 185                                 {
 186                                         fprintf(stderr, _("%s: multitransaction ID (-m) must not be 0\n"), progname);
 187                                         exit(1);
 188                                 }
 189                                 break;
 190
 191                         case 'O':
 192                                 set_mxoff = strtoul(optarg, &endptr, 0);
 193                                 if (endptr == optarg || *endptr != '\0')
 194                                 {
 195                                         fprintf(stderr, _("%s: invalid argument for option -O\n"), progname);
 196                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 197                                         exit(1);
 198                                 }
 199                                 if (set_mxoff == -1)
 200                                 {
 201                                         fprintf(stderr, _("%s: multitransaction offset (-O) must not be -1\n"), progname);
 202                                         exit(1);
 203                                 }
 204                                 break;
 205
 206                         case 'l':
 207                                 minXlogTli = strtoul(optarg, &endptr, 0);
 208                                 if (endptr == optarg || *endptr != ',')
 209                                 {
 210                                         fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
 211                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 212                                         exit(1);
 213                                 }
 214                                 minXlogId = strtoul(endptr + 1, &endptr2, 0);
 215                                 if (endptr2 == endptr + 1 || *endptr2 != ',')
 216                                 {
 217                                         fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
 218                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 219                                         exit(1);
 220                                 }
 221                                 minXlogSeg = strtoul(endptr2 + 1, &endptr3, 0);
 222                                 if (endptr3 == endptr2 + 1 || *endptr3 != '\0')
 223                                 {
 224                                         fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
 225                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 226                                         exit(1);
 227                                 }
 228                                 break;
 229
 230                         default:
 231                                 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 232                                 exit(1);
 233                 }
 234         }
 235
 236         if (optind == argc)
 237         {
 238                 fprintf(stderr, _("%s: no data directory specified\n"), progname);
 239                 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 240                 exit(1);
 241         }
 242
 243         /*
 244          * Don't allow pg_resetxlog to be run as root, to avoid overwriting the
 245          * ownership of files in the data directory. We need only check for root
 246          * -- any other user won't have sufficient permissions to modify files in
 247          * the data directory.
 248          */
 249 #ifndef WIN32
 250         if (geteuid() == 0)
 251         {
 252                 fprintf(stderr, _("%s: cannot be executed by \"root\"\n"),
 253                                 progname);
 254                 fprintf(stderr, _("You must run %s as the PostgreSQL superuser.\n"),
 255                                 progname);
 256                 exit(1);
 257         }
 258 #endif
 259
 260         DataDir = argv[optind];
 261
 262         if (chdir(DataDir) < 0)
 263         {
 264                 fprintf(stderr, _("%s: could not change directory to \"%s\": %s\n"),
 265                                 progname, DataDir, strerror(errno));
 266                 exit(1);
 267         }
 268
 269         /*
 270          * Check for a postmaster lock file --- if there is one, refuse to
 271          * proceed, on grounds we might be interfering with a live installation.
 272          */
 273         snprintf(path, MAXPGPATH, "%s/postmaster.pid", DataDir);
 274
 275         if ((fd = open(path, O_RDONLY, 0)) < 0)
 276         {
 277                 if (errno != ENOENT)
 278                 {
 279                         fprintf(stderr, _("%s: could not open file \"%s\" for reading: %s\n"), progname, path, strerror(errno));
 280                         exit(1);
 281                 }
 282         }
 283         else
 284         {
 285                 fprintf(stderr, _("%s: lock file \"%s\" exists\n"
 286                                                   "Is a server running?  If not, delete the lock file and try again.\n"),
 287                                 progname, path);
 288                 exit(1);
 289         }
 290
 291         /*
 292          * Attempt to read the existing pg_control file
 293          */
 294         if (!ReadControlFile())
 295                 GuessControlValues();
 296
 297         /*
 298          * Also look at existing segment files to set up newXlogId/newXlogSeg
 299          */
 300         FindEndOfXLOG();
 301
 302         /*
 303          * Adjust fields if required by switches.  (Do this now so that printout,
 304          * if any, includes these values.)
 305          */
 306         if (set_xid_epoch != -1)
 307                 ControlFile.checkPointCopy.nextXidEpoch = set_xid_epoch;
 308
 309         if (set_xid != 0)
 310                 ControlFile.checkPointCopy.nextXid = set_xid;
 311
 312         if (set_oid != 0)
 313                 ControlFile.checkPointCopy.nextOid = set_oid;
 314
 315         if (set_mxid != 0)
 316                 ControlFile.checkPointCopy.nextMulti = set_mxid;
 317
 318         if (set_mxoff != -1)
 319                 ControlFile.checkPointCopy.nextMultiOffset = set_mxoff;
 320
 321         if (minXlogTli > ControlFile.checkPointCopy.ThisTimeLineID)
 322                 ControlFile.checkPointCopy.ThisTimeLineID = minXlogTli;
 323
 324         if (minXlogId > newXlogId ||
 325                 (minXlogId == newXlogId &&
 326                  minXlogSeg > newXlogSeg))
 327         {
 328                 newXlogId = minXlogId;
 329                 newXlogSeg = minXlogSeg;
 330         }
 331
 332         /*
 333          * If we had to guess anything, and -f was not given, just print the
 334          * guessed values and exit.  Also print if -n is given.
 335          */
 336         if ((guessed && !force) || noupdate)
 337         {
 338                 PrintControlValues(guessed);
 339                 if (!noupdate)
 340                 {
 341                         printf(_("\nIf these values seem acceptable, use -f to force reset.\n"));
 342                         exit(1);
 343                 }
 344                 else
 345                         exit(0);
 346         }
 347
 348         /*
 349          * Don't reset from a dirty pg_control without -f, either.
 350          */
 351         if (ControlFile.state != DB_SHUTDOWNED && !force)
 352         {
 353                 printf(_("The database server was not shut down cleanly.\n"
 354                            "Resetting the transaction log might cause data to be lost.\n"
 355                                  "If you want to proceed anyway, use -f to force reset.\n"));
 356                 exit(1);
 357         }
 358
 359         /*
 360          * Else, do the dirty deed.
 361          */
 362         RewriteControlFile();
 363         KillExistingXLOG();
 364         KillExistingArchiveStatus();
 365         WriteEmptyXLOG();
 366
 367         printf(_("Transaction log reset\n"));
 368         return 0;
 369 }
 370
 371
 372 /*
 373  * Try to read the existing pg_control file.
 374  *
 375  * This routine is also responsible for updating old pg_control versions
 376  * to the current format.  (Currently we don't do anything of the sort.)
 377  */
 378 static bool
 379 ReadControlFile(void)
 380 {
 381         int                     fd;
 382         int                     len;
 383         char       *buffer;
 384         pg_crc32        crc;
 385
 386         if ((fd = open(XLOG_CONTROL_FILE, O_RDONLY | PG_BINARY, 0)) < 0)
 387         {
 388                 /*
 389                  * If pg_control is not there at all, or we can't read it, the odds
 390                  * are we've been handed a bad DataDir path, so give up. User can do
 391                  * "touch pg_control" to force us to proceed.
 392                  */
 393                 fprintf(stderr, _("%s: could not open file \"%s\" for reading: %s\n"),
 394                                 progname, XLOG_CONTROL_FILE, strerror(errno));
 395                 if (errno == ENOENT)
 396                         fprintf(stderr, _("If you are sure the data directory path is correct, execute\n"
 397                                                           "  touch %s\n"
 398                                                           "and try again.\n"),
 399                                         XLOG_CONTROL_FILE);
 400                 exit(1);
 401         }
 402
 403         /* Use malloc to ensure we have a maxaligned buffer */
 404         buffer = (char *) malloc(PG_CONTROL_SIZE);
 405
 406         len = read(fd, buffer, PG_CONTROL_SIZE);
 407         if (len < 0)
 408         {
 409                 fprintf(stderr, _("%s: could not read file \"%s\": %s\n"),
 410                                 progname, XLOG_CONTROL_FILE, strerror(errno));
 411                 exit(1);
 412         }
 413         close(fd);
 414
 415         if (len >= sizeof(ControlFileData) &&
 416           ((ControlFileData *) buffer)->pg_control_version == PG_CONTROL_VERSION)
 417         {
 418                 /* Check the CRC. */
 419                 INIT_CRC32(crc);
 420                 COMP_CRC32(crc,
 421                                    buffer,
 422                                    offsetof(ControlFileData, crc));
 423                 FIN_CRC32(crc);
 424
 425                 if (EQ_CRC32(crc, ((ControlFileData *) buffer)->crc))
 426                 {
 427                         /* Valid data... */
 428                         memcpy(&ControlFile, buffer, sizeof(ControlFile));
 429                         return true;
 430                 }
 431
 432                 fprintf(stderr, _("%s: pg_control exists but has invalid CRC; proceed with caution\n"),
 433                                 progname);
 434                 /* We will use the data anyway, but treat it as guessed. */
 435                 memcpy(&ControlFile, buffer, sizeof(ControlFile));
 436                 guessed = true;
 437                 return true;
 438         }
 439
 440         /* Looks like it's a mess. */
 441         fprintf(stderr, _("%s: pg_control exists but is broken or unknown version; ignoring it\n"),
 442                         progname);
 443         return false;
 444 }
 445
 446
 447 /*
 448  * Guess at pg_control values when we can't read the old ones.
 449  */
 450 static void
 451 GuessControlValues(void)
 452 {
 453         uint64          sysidentifier;
 454         struct timeval tv;
 455
 456         /*
 457          * Set up a completely default set of pg_control values.
 458          */
 459         guessed = true;
 460         memset(&ControlFile, 0, sizeof(ControlFile));
 461
 462         ControlFile.pg_control_version = PG_CONTROL_VERSION;
 463         ControlFile.catalog_version_no = CATALOG_VERSION_NO;
 464
 465         /*
 466          * Create a new unique installation identifier, since we can no longer use
 467          * any old XLOG records.  See notes in xlog.c about the algorithm.
 468          */
 469         gettimeofday(&tv, NULL);
 470         sysidentifier = ((uint64) tv.tv_sec) << 32;
 471         sysidentifier |= (uint32) (tv.tv_sec | tv.tv_usec);
 472
 473         ControlFile.system_identifier = sysidentifier;
 474
 475         ControlFile.checkPointCopy.redo.xlogid = 0;
 476         ControlFile.checkPointCopy.redo.xrecoff = SizeOfXLogLongPHD;
 477         ControlFile.checkPointCopy.ThisTimeLineID = 1;
 478         ControlFile.checkPointCopy.nextXidEpoch = 0;
 479         ControlFile.checkPointCopy.nextXid = (TransactionId) 514;       /* XXX */
 480         ControlFile.checkPointCopy.nextOid = FirstBootstrapObjectId;
 481         ControlFile.checkPointCopy.nextMulti = FirstMultiXactId;
 482         ControlFile.checkPointCopy.nextMultiOffset = 0;
 483         ControlFile.checkPointCopy.time = (pg_time_t) time(NULL);
 484
 485         ControlFile.state = DB_SHUTDOWNED;
 486         ControlFile.time = (pg_time_t) time(NULL);
 487         ControlFile.checkPoint = ControlFile.checkPointCopy.redo;
 488
 489         ControlFile.maxAlign = MAXIMUM_ALIGNOF;
 490         ControlFile.floatFormat = FLOATFORMAT_VALUE;
 491         ControlFile.blcksz = BLCKSZ;
 492         ControlFile.relseg_size = RELSEG_SIZE;
 493         ControlFile.xlog_blcksz = XLOG_BLCKSZ;
 494         ControlFile.xlog_seg_size = XLOG_SEG_SIZE;
 495         ControlFile.nameDataLen = NAMEDATALEN;
 496         ControlFile.indexMaxKeys = INDEX_MAX_KEYS;
 497         ControlFile.toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
 498 #ifdef HAVE_INT64_TIMESTAMP
 499         ControlFile.enableIntTimes = true;
 500 #else
 501         ControlFile.enableIntTimes = false;
 502 #endif
 503         ControlFile.float4ByVal = FLOAT4PASSBYVAL;
 504         ControlFile.float8ByVal = FLOAT8PASSBYVAL;
 505
 506         /*
 507          * XXX eventually, should try to grovel through old XLOG to develop more
 508          * accurate values for TimeLineID, nextXID, etc.
 509          */
 510 }
 511
 512
 513 /*
 514  * Print the guessed pg_control values when we had to guess.
 515  *
 516  * NB: this display should be just those fields that will not be
 517  * reset by RewriteControlFile().
 518  */
 519 static void
 520 PrintControlValues(bool guessed)
 521 {
 522         char            sysident_str[32];
 523
 524         if (guessed)
 525                 printf(_("Guessed pg_control values:\n\n"));
 526         else
 527                 printf(_("pg_control values:\n\n"));
 528
 529         /*
 530          * Format system_identifier separately to keep platform-dependent format
 531          * code out of the translatable message string.
 532          */
 533         snprintf(sysident_str, sizeof(sysident_str), UINT64_FORMAT,
 534                          ControlFile.system_identifier);
 535
 536         printf(_("First log file ID after reset:        %u\n"),
 537                    newXlogId);
 538         printf(_("First log file segment after reset:   %u\n"),
 539                    newXlogSeg);
 540         printf(_("pg_control version number:            %u\n"),
 541                    ControlFile.pg_control_version);
 542         printf(_("Catalog version number:               %u\n"),
 543                    ControlFile.catalog_version_no);
 544         printf(_("Database system identifier:           %s\n"),
 545                    sysident_str);
 546         printf(_("Latest checkpoint's TimeLineID:       %u\n"),
 547                    ControlFile.checkPointCopy.ThisTimeLineID);
 548         printf(_("Latest checkpoint's NextXID:          %u/%u\n"),
 549                    ControlFile.checkPointCopy.nextXidEpoch,
 550                    ControlFile.checkPointCopy.nextXid);
 551         printf(_("Latest checkpoint's NextOID:          %u\n"),
 552                    ControlFile.checkPointCopy.nextOid);
 553         printf(_("Latest checkpoint's NextMultiXactId:  %u\n"),
 554                    ControlFile.checkPointCopy.nextMulti);
 555         printf(_("Latest checkpoint's NextMultiOffset:  %u\n"),
 556                    ControlFile.checkPointCopy.nextMultiOffset);
 557         printf(_("Maximum data alignment:               %u\n"),
 558                    ControlFile.maxAlign);
 559         /* we don't print floatFormat since can't say much useful about it */
 560         printf(_("Database block size:                  %u\n"),
 561                    ControlFile.blcksz);
 562         printf(_("Blocks per segment of large relation: %u\n"),
 563                    ControlFile.relseg_size);
 564         printf(_("WAL block size:                       %u\n"),
 565                    ControlFile.xlog_blcksz);
 566         printf(_("Bytes per WAL segment:                %u\n"),
 567                    ControlFile.xlog_seg_size);
 568         printf(_("Maximum length of identifiers:        %u\n"),
 569                    ControlFile.nameDataLen);
 570         printf(_("Maximum columns in an index:          %u\n"),
 571                    ControlFile.indexMaxKeys);
 572         printf(_("Maximum size of a TOAST chunk:        %u\n"),
 573                    ControlFile.toast_max_chunk_size);
 574         printf(_("Date/time type storage:               %s\n"),
 575                    (ControlFile.enableIntTimes ? _("64-bit integers") : _("floating-point numbers")));
 576         printf(_("Float4 argument passing:              %s\n"),
 577                    (ControlFile.float4ByVal ? _("by value") : _("by reference")));
 578         printf(_("Float8 argument passing:              %s\n"),
 579                    (ControlFile.float8ByVal ? _("by value") : _("by reference")));
 580 }
 581
 582
 583 /*
 584  * Write out the new pg_control file.
 585  */
 586 static void
 587 RewriteControlFile(void)
 588 {
 589         int                     fd;
 590         char            buffer[PG_CONTROL_SIZE];                /* need not be aligned */
 591
 592         /*
 593          * Adjust fields as needed to force an empty XLOG starting at
 594          * newXlogId/newXlogSeg.
 595          */
 596         ControlFile.checkPointCopy.redo.xlogid = newXlogId;
 597         ControlFile.checkPointCopy.redo.xrecoff =
 598                 newXlogSeg * XLogSegSize + SizeOfXLogLongPHD;
 599         ControlFile.checkPointCopy.time = (pg_time_t) time(NULL);
 600
 601         ControlFile.state = DB_SHUTDOWNED;
 602         ControlFile.time = (pg_time_t) time(NULL);
 603         ControlFile.checkPoint = ControlFile.checkPointCopy.redo;
 604         ControlFile.prevCheckPoint.xlogid = 0;
 605         ControlFile.prevCheckPoint.xrecoff = 0;
 606         ControlFile.minRecoveryPoint.xlogid = 0;
 607         ControlFile.minRecoveryPoint.xrecoff = 0;
 608
 609         /* Now we can force the recorded xlog seg size to the right thing. */
 610         ControlFile.xlog_seg_size = XLogSegSize;
 611
 612         /* Contents are protected with a CRC */
 613         INIT_CRC32(ControlFile.crc);
 614         COMP_CRC32(ControlFile.crc,
 615                            (char *) &ControlFile,
 616                            offsetof(ControlFileData, crc));
 617         FIN_CRC32(ControlFile.crc);
 618
 619         /*
 620          * We write out PG_CONTROL_SIZE bytes into pg_control, zero-padding the
 621          * excess over sizeof(ControlFileData).  This reduces the odds of
 622          * premature-EOF errors when reading pg_control.  We'll still fail when we
 623          * check the contents of the file, but hopefully with a more specific
 624          * error than "couldn't read pg_control".
 625          */
 626         if (sizeof(ControlFileData) > PG_CONTROL_SIZE)
 627         {
 628                 fprintf(stderr,
 629                                 _("%s: internal error -- sizeof(ControlFileData) is too large ... fix PG_CONTROL_SIZE\n"),
 630                                 progname);
 631                 exit(1);
 632         }
 633
 634         memset(buffer, 0, PG_CONTROL_SIZE);
 635         memcpy(buffer, &ControlFile, sizeof(ControlFileData));
 636
 637         unlink(XLOG_CONTROL_FILE);
 638
 639         fd = open(XLOG_CONTROL_FILE,
 640                           O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
 641                           S_IRUSR | S_IWUSR);
 642         if (fd < 0)
 643         {
 644                 fprintf(stderr, _("%s: could not create pg_control file: %s\n"),
 645                                 progname, strerror(errno));
 646                 exit(1);
 647         }
 648
 649         errno = 0;
 650         if (write(fd, buffer, PG_CONTROL_SIZE) != PG_CONTROL_SIZE)
 651         {
 652                 /* if write didn't set errno, assume problem is no disk space */
 653                 if (errno == 0)
 654                         errno = ENOSPC;
 655                 fprintf(stderr, _("%s: could not write pg_control file: %s\n"),
 656                                 progname, strerror(errno));
 657                 exit(1);
 658         }
 659
 660         if (fsync(fd) != 0)
 661         {
 662                 fprintf(stderr, _("%s: fsync error: %s\n"), progname, strerror(errno));
 663                 exit(1);
 664         }
 665
 666         close(fd);
 667 }
 668
 669
 670 /*
 671  * Scan existing XLOG files and determine the highest existing WAL address
 672  *
 673  * On entry, ControlFile.checkPointCopy.redo and ControlFile.xlog_seg_size
 674  * are assumed valid (note that we allow the old xlog seg size to differ
 675  * from what we're using).  On exit, newXlogId and newXlogSeg are set to
 676  * suitable values for the beginning of replacement WAL (in our seg size).
 677  */
 678 static void
 679 FindEndOfXLOG(void)
 680 {
 681         DIR                *xldir;
 682         struct dirent *xlde;
 683
 684         /*
 685          * Initialize the max() computation using the last checkpoint address from
 686          * old pg_control.      Note that for the moment we are working with segment
 687          * numbering according to the old xlog seg size.
 688          */
 689         newXlogId = ControlFile.checkPointCopy.redo.xlogid;
 690         newXlogSeg = ControlFile.checkPointCopy.redo.xrecoff / ControlFile.xlog_seg_size;
 691
 692         /*
 693          * Scan the pg_xlog directory to find existing WAL segment files. We
 694          * assume any present have been used; in most scenarios this should be
 695          * conservative, because of xlog.c's attempts to pre-create files.
 696          */
 697         xldir = opendir(XLOGDIR);
 698         if (xldir == NULL)
 699         {
 700                 fprintf(stderr, _("%s: could not open directory \"%s\": %s\n"),
 701                                 progname, XLOGDIR, strerror(errno));
 702                 exit(1);
 703         }
 704
 705         errno = 0;
 706         while ((xlde = readdir(xldir)) != NULL)
 707         {
 708                 if (strlen(xlde->d_name) == 24 &&
 709                         strspn(xlde->d_name, "0123456789ABCDEF") == 24)
 710                 {
 711                         unsigned int tli,
 712                                                 log,
 713                                                 seg;
 714
 715                         sscanf(xlde->d_name, "%08X%08X%08X", &tli, &log, &seg);
 716
 717                         /*
 718                          * Note: we take the max of all files found, regardless of their
 719                          * timelines.  Another possibility would be to ignore files of
 720                          * timelines other than the target TLI, but this seems safer.
 721                          * Better too large a result than too small...
 722                          */
 723                         if (log > newXlogId ||
 724                                 (log == newXlogId && seg > newXlogSeg))
 725                         {
 726                                 newXlogId = log;
 727                                 newXlogSeg = seg;
 728                         }
 729                 }
 730                 errno = 0;
 731         }
 732 #ifdef WIN32
 733
 734         /*
 735          * This fix is in mingw cvs (runtime/mingwex/dirent.c rev 1.4), but not in
 736          * released version
 737          */
 738         if (GetLastError() == ERROR_NO_MORE_FILES)
 739                 errno = 0;
 740 #endif
 741
 742         if (errno)
 743         {
 744                 fprintf(stderr, _("%s: could not read from directory \"%s\": %s\n"),
 745                                 progname, XLOGDIR, strerror(errno));
 746                 exit(1);
 747         }
 748         closedir(xldir);
 749
 750         /*
 751          * Finally, convert to new xlog seg size, and advance by one to ensure we
 752          * are in virgin territory.
 753          */
 754         newXlogSeg *= ControlFile.xlog_seg_size;
 755         newXlogSeg = (newXlogSeg + XLogSegSize - 1) / XLogSegSize;
 756
 757         /* be sure we wrap around correctly at end of a logfile */
 758         NextLogSeg(newXlogId, newXlogSeg);
 759 }
 760
 761
 762 /*
 763  * Remove existing XLOG files
 764  */
 765 static void
 766 KillExistingXLOG(void)
 767 {
 768         DIR                *xldir;
 769         struct dirent *xlde;
 770         char            path[MAXPGPATH];
 771
 772         xldir = opendir(XLOGDIR);
 773         if (xldir == NULL)
 774         {
 775                 fprintf(stderr, _("%s: could not open directory \"%s\": %s\n"),
 776                                 progname, XLOGDIR, strerror(errno));
 777                 exit(1);
 778         }
 779
 780         errno = 0;
 781         while ((xlde = readdir(xldir)) != NULL)
 782         {
 783                 if (strlen(xlde->d_name) == 24 &&
 784                         strspn(xlde->d_name, "0123456789ABCDEF") == 24)
 785                 {
 786                         snprintf(path, MAXPGPATH, "%s/%s", XLOGDIR, xlde->d_name);
 787                         if (unlink(path) < 0)
 788                         {
 789                                 fprintf(stderr, _("%s: could not delete file \"%s\": %s\n"),
 790                                                 progname, path, strerror(errno));
 791                                 exit(1);
 792                         }
 793                 }
 794                 errno = 0;
 795         }
 796 #ifdef WIN32
 797
 798         /*
 799          * This fix is in mingw cvs (runtime/mingwex/dirent.c rev 1.4), but not in
 800          * released version
 801          */
 802         if (GetLastError() == ERROR_NO_MORE_FILES)
 803                 errno = 0;
 804 #endif
 805
 806         if (errno)
 807         {
 808                 fprintf(stderr, _("%s: could not read from directory \"%s\": %s\n"),
 809                                 progname, XLOGDIR, strerror(errno));
 810                 exit(1);
 811         }
 812         closedir(xldir);
 813 }
 814
 815
 816 /*
 817  * Remove existing archive status files
 818  */
 819 static void
 820 KillExistingArchiveStatus(void)
 821 {
 822         DIR                *xldir;
 823         struct dirent *xlde;
 824         char            path[MAXPGPATH];
 825
 826 #define ARCHSTATDIR XLOGDIR "/archive_status"
 827
 828         xldir = opendir(ARCHSTATDIR);
 829         if (xldir == NULL)
 830         {
 831                 fprintf(stderr, _("%s: could not open directory \"%s\": %s\n"),
 832                                 progname, ARCHSTATDIR, strerror(errno));
 833                 exit(1);
 834         }
 835
 836         errno = 0;
 837         while ((xlde = readdir(xldir)) != NULL)
 838         {
 839                 if (strspn(xlde->d_name, "0123456789ABCDEF") == 24 &&
 840                         (strcmp(xlde->d_name + 24, ".ready") == 0 ||
 841                          strcmp(xlde->d_name + 24, ".done") == 0))
 842                 {
 843                         snprintf(path, MAXPGPATH, "%s/%s", ARCHSTATDIR, xlde->d_name);
 844                         if (unlink(path) < 0)
 845                         {
 846                                 fprintf(stderr, _("%s: could not delete file \"%s\": %s\n"),
 847                                                 progname, path, strerror(errno));
 848                                 exit(1);
 849                         }
 850                 }
 851                 errno = 0;
 852         }
 853 #ifdef WIN32
 854
 855         /*
 856          * This fix is in mingw cvs (runtime/mingwex/dirent.c rev 1.4), but not in
 857          * released version
 858          */
 859         if (GetLastError() == ERROR_NO_MORE_FILES)
 860                 errno = 0;
 861 #endif
 862
 863         if (errno)
 864         {
 865                 fprintf(stderr, _("%s: could not read from directory \"%s\": %s\n"),
 866                                 progname, ARCHSTATDIR, strerror(errno));
 867                 exit(1);
 868         }
 869         closedir(xldir);
 870 }
 871
 872
 873 /*
 874  * Write an empty XLOG file, containing only the checkpoint record
 875  * already set up in ControlFile.
 876  */
 877 static void
 878 WriteEmptyXLOG(void)
 879 {
 880         char       *buffer;
 881         XLogPageHeader page;
 882         XLogLongPageHeader longpage;
 883         XLogRecord *record;
 884         pg_crc32        crc;
 885         char            path[MAXPGPATH];
 886         int                     fd;
 887         int                     nbytes;
 888
 889         /* Use malloc() to ensure buffer is MAXALIGNED */
 890         buffer = (char *) malloc(XLOG_BLCKSZ);
 891         page = (XLogPageHeader) buffer;
 892         memset(buffer, 0, XLOG_BLCKSZ);
 893
 894         /* Set up the XLOG page header */
 895         page->xlp_magic = XLOG_PAGE_MAGIC;
 896         page->xlp_info = XLP_LONG_HEADER;
 897         page->xlp_tli = ControlFile.checkPointCopy.ThisTimeLineID;
 898         page->xlp_pageaddr.xlogid =
 899                 ControlFile.checkPointCopy.redo.xlogid;
 900         page->xlp_pageaddr.xrecoff =
 901                 ControlFile.checkPointCopy.redo.xrecoff - SizeOfXLogLongPHD;
 902         longpage = (XLogLongPageHeader) page;
 903         longpage->xlp_sysid = ControlFile.system_identifier;
 904         longpage->xlp_seg_size = XLogSegSize;
 905         longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
 906
 907         /* Insert the initial checkpoint record */
 908         record = (XLogRecord *) ((char *) page + SizeOfXLogLongPHD);
 909         record->xl_prev.xlogid = 0;
 910         record->xl_prev.xrecoff = 0;
 911         record->xl_xid = InvalidTransactionId;
 912         record->xl_tot_len = SizeOfXLogRecord + sizeof(CheckPoint);
 913         record->xl_len = sizeof(CheckPoint);
 914         record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
 915         record->xl_rmid = RM_XLOG_ID;
 916         memcpy(XLogRecGetData(record), &ControlFile.checkPointCopy,
 917                    sizeof(CheckPoint));
 918
 919         INIT_CRC32(crc);
 920         COMP_CRC32(crc, &ControlFile.checkPointCopy, sizeof(CheckPoint));
 921         COMP_CRC32(crc, (char *) record + sizeof(pg_crc32),
 922                            SizeOfXLogRecord - sizeof(pg_crc32));
 923         FIN_CRC32(crc);
 924         record->xl_crc = crc;
 925
 926         /* Write the first page */
 927         XLogFilePath(path, ControlFile.checkPointCopy.ThisTimeLineID,
 928                                  newXlogId, newXlogSeg);
 929
 930         unlink(path);
 931
 932         fd = open(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
 933                           S_IRUSR | S_IWUSR);
 934         if (fd < 0)
 935         {
 936                 fprintf(stderr, _("%s: could not open file \"%s\": %s\n"),
 937                                 progname, path, strerror(errno));
 938                 exit(1);
 939         }
 940
 941         errno = 0;
 942         if (write(fd, buffer, XLOG_BLCKSZ) != XLOG_BLCKSZ)
 943         {
 944                 /* if write didn't set errno, assume problem is no disk space */
 945                 if (errno == 0)
 946                         errno = ENOSPC;
 947                 fprintf(stderr, _("%s: could not write file \"%s\": %s\n"),
 948                                 progname, path, strerror(errno));
 949                 exit(1);
 950         }
 951
 952         /* Fill the rest of the file with zeroes */
 953         memset(buffer, 0, XLOG_BLCKSZ);
 954         for (nbytes = XLOG_BLCKSZ; nbytes < XLogSegSize; nbytes += XLOG_BLCKSZ)
 955         {
 956                 errno = 0;
 957                 if (write(fd, buffer, XLOG_BLCKSZ) != XLOG_BLCKSZ)
 958                 {
 959                         if (errno == 0)
 960                                 errno = ENOSPC;
 961                         fprintf(stderr, _("%s: could not write file \"%s\": %s\n"),
 962                                         progname, path, strerror(errno));
 963                         exit(1);
 964                 }
 965         }
 966
 967         if (fsync(fd) != 0)
 968         {
 969                 fprintf(stderr, _("%s: fsync error: %s\n"), progname, strerror(errno));
 970                 exit(1);
 971         }
 972
 973         close(fd);
 974 }
 975
 976
 977 static void
 978 usage(void)
 979 {
 980         printf(_("%s resets the PostgreSQL transaction log.\n\n"), progname);
 981         printf(_("Usage:\n  %s [OPTION]... DATADIR\n\n"), progname);
 982         printf(_("Options:\n"));
 983         printf(_("  -e XIDEPOCH     set next transaction ID epoch\n"));
 984         printf(_("  -f              force update to be done\n"));
 985         printf(_("  -l TLI,FILE,SEG force minimum WAL starting location for new transaction log\n"));
 986         printf(_("  -m XID          set next multitransaction ID\n"));
 987         printf(_("  -n              no update, just show extracted control values (for testing)\n"));
 988         printf(_("  -o OID          set next OID\n"));
 989         printf(_("  -O OFFSET       set next multitransaction offset\n"));
 990         printf(_("  -x XID          set next transaction ID\n"));
 991         printf(_("  --help          show this help, then exit\n"));
 992         printf(_("  --version       output version information, then exit\n"));
 993         printf(_("\nReport bugs to <pgsql-bugs@postgresql.org>.\n"));
 994 }