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