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) 1996-2008, PostgreSQL Global Development Group
  24  * Portions Copyright (c) 1994, Regents of the University of California
  25  *
  26  * $PostgreSQL$
  27  *
  28  *-------------------------------------------------------------------------
  29  */
  30 #include "postgres.h"
  31
  32 #include <dirent.h>
  33 #include <fcntl.h>
  34 #include <locale.h>
  35 #include <sys/stat.h>
  36 #include <sys/time.h>
  37 #include <time.h>
  38 #include <unistd.h>
  39 #ifdef HAVE_GETOPT_H
  40 #include <getopt.h>
  41 #endif
  42
  43 #include "access/transam.h"
  44 #include "access/tuptoaster.h"
  45 #include "access/multixact.h"
  46 #include "access/xlog_internal.h"
  47 #include "catalog/catversion.h"
  48 #include "catalog/pg_control.h"
  49
  50 extern int      optind;
  51 extern char *optarg;
  52
  53
  54 static ControlFileData ControlFile;             /* pg_control values */
  55 static uint32 newXlogId,
  56                         newXlogSeg;                     /* ID/Segment of new XLOG segment */
  57 static bool guessed = false;    /* T if we had to guess at any values */
  58 static const char *progname;
  59
  60 static bool ReadControlFile(void);
  61 static void GuessControlValues(void);
  62 static void PrintControlValues(bool guessed);
  63 static void RewriteControlFile(void);
  64 static void FindEndOfXLOG(void);
  65 static void KillExistingXLOG(void);
  66 static void WriteEmptyXLOG(void);
  67 static void usage(void);
  68
  69
  70 int
  71 main(int argc, char *argv[])
  72 {
  73         int                     c;
  74         bool            force = false;
  75         bool            noupdate = false;
  76         uint32          set_xid_epoch = (uint32) -1;
  77         TransactionId set_xid = 0;
  78         Oid                     set_oid = 0;
  79         MultiXactId set_mxid = 0;
  80         MultiXactOffset set_mxoff = (MultiXactOffset) -1;
  81         uint32          minXlogTli = 0,
  82                                 minXlogId = 0,
  83                                 minXlogSeg = 0;
  84         char       *endptr;
  85         char       *endptr2;
  86         char       *endptr3;
  87         char       *DataDir;
  88         int                     fd;
  89         char            path[MAXPGPATH];
  90
  91         set_pglocale_pgservice(argv[0], "pg_resetxlog");
  92
  93         progname = get_progname(argv[0]);
  94
  95         if (argc > 1)
  96         {
  97                 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
  98                 {
  99                         usage();
 100                         exit(0);
 101                 }
 102                 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
 103                 {
 104                         puts("pg_resetxlog (PostgreSQL) " PG_VERSION);
 105                         exit(0);
 106                 }
 107         }
 108
 109
 110         while ((c = getopt(argc, argv, "fl:m:no:O:x:e:")) != -1)
 111         {
 112                 switch (c)
 113                 {
 114                         case 'f':
 115                                 force = true;
 116                                 break;
 117
 118                         case 'n':
 119                                 noupdate = true;
 120                                 break;
 121
 122                         case 'e':
 123                                 set_xid_epoch = strtoul(optarg, &endptr, 0);
 124                                 if (endptr == optarg || *endptr != '\0')
 125                                 {
 126                                         fprintf(stderr, _("%s: invalid argument for option -e\n"), progname);
 127                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 128                                         exit(1);
 129                                 }
 130                                 if (set_xid_epoch == -1)
 131                                 {
 132                                         fprintf(stderr, _("%s: transaction ID epoch (-e) must not be -1\n"), progname);
 133                                         exit(1);
 134                                 }
 135                                 break;
 136
 137                         case 'x':
 138                                 set_xid = strtoul(optarg, &endptr, 0);
 139                                 if (endptr == optarg || *endptr != '\0')
 140                                 {
 141                                         fprintf(stderr, _("%s: invalid argument for option -x\n"), progname);
 142                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 143                                         exit(1);
 144                                 }
 145                                 if (set_xid == 0)
 146                                 {
 147                                         fprintf(stderr, _("%s: transaction ID (-x) must not be 0\n"), progname);
 148                                         exit(1);
 149                                 }
 150                                 break;
 151
 152                         case 'o':
 153                                 set_oid = strtoul(optarg, &endptr, 0);
 154                                 if (endptr == optarg || *endptr != '\0')
 155                                 {
 156                                         fprintf(stderr, _("%s: invalid argument for option -o\n"), progname);
 157                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 158                                         exit(1);
 159                                 }
 160                                 if (set_oid == 0)
 161                                 {
 162                                         fprintf(stderr, _("%s: OID (-o) must not be 0\n"), progname);
 163                                         exit(1);
 164                                 }
 165                                 break;
 166
 167                         case 'm':
 168                                 set_mxid = strtoul(optarg, &endptr, 0);
 169                                 if (endptr == optarg || *endptr != '\0')
 170                                 {
 171                                         fprintf(stderr, _("%s: invalid argument for option -m\n"), progname);
 172                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 173                                         exit(1);
 174                                 }
 175                                 if (set_mxid == 0)
 176                                 {
 177                                         fprintf(stderr, _("%s: multitransaction ID (-m) must not be 0\n"), progname);
 178                                         exit(1);
 179                                 }
 180                                 break;
 181
 182                         case 'O':
 183                                 set_mxoff = strtoul(optarg, &endptr, 0);
 184                                 if (endptr == optarg || *endptr != '\0')
 185                                 {
 186                                         fprintf(stderr, _("%s: invalid argument for option -O\n"), progname);
 187                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 188                                         exit(1);
 189                                 }
 190                                 if (set_mxoff == -1)
 191                                 {
 192                                         fprintf(stderr, _("%s: multitransaction offset (-O) must not be -1\n"), progname);
 193                                         exit(1);
 194                                 }
 195                                 break;
 196
 197                         case 'l':
 198                                 minXlogTli = strtoul(optarg, &endptr, 0);
 199                                 if (endptr == optarg || *endptr != ',')
 200                                 {
 201                                         fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
 202                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 203                                         exit(1);
 204                                 }
 205                                 minXlogId = strtoul(endptr + 1, &endptr2, 0);
 206                                 if (endptr2 == endptr + 1 || *endptr2 != ',')
 207                                 {
 208                                         fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
 209                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 210                                         exit(1);
 211                                 }
 212                                 minXlogSeg = strtoul(endptr2 + 1, &endptr3, 0);
 213                                 if (endptr3 == endptr2 + 1 || *endptr3 != '\0')
 214                                 {
 215                                         fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
 216                                         fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 217                                         exit(1);
 218                                 }
 219                                 break;
 220
 221                         default:
 222                                 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 223                                 exit(1);
 224                 }
 225         }
 226
 227         if (optind == argc)
 228         {
 229                 fprintf(stderr, _("%s: no data directory specified\n"), progname);
 230                 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
 231                 exit(1);
 232         }
 233
 234         /*
 235          * Don't allow pg_resetxlog to be run as root, to avoid overwriting the
 236          * ownership of files in the data directory. We need only check for root
 237          * -- any other user won't have sufficient permissions to modify files in
 238          * the data directory.
 239          */
 240 #ifndef WIN32
 241         if (geteuid() == 0)
 242         {
 243                 fprintf(stderr, _("%s: cannot be executed by \"root\"\n"),
 244                                 progname);
 245                 fprintf(stderr, _("You must run %s as the PostgreSQL superuser.\n"),
 246                                 progname);
 247                 exit(1);
 248         }
 249 #endif
 250
 251         DataDir = argv[optind];
 252
 253         if (chdir(DataDir) < 0)
 254         {
 255                 fprintf(stderr, _("%s: could not change directory to \"%s\": %s\n"),
 256                                 progname, DataDir, strerror(errno));
 257                 exit(1);
 258         }
 259
 260         /*
 261          * Check for a postmaster lock file --- if there is one, refuse to
 262          * proceed, on grounds we might be interfering with a live installation.
 263          */
 264         snprintf(path, MAXPGPATH, "%s/postmaster.pid", DataDir);
 265
 266         if ((fd = open(path, O_RDONLY, 0)) < 0)
 267         {
 268                 if (errno != ENOENT)
 269                 {
 270                         fprintf(stderr, _("%s: could not open file \"%s\" for reading: %s\n"), progname, path, strerror(errno));
 271                         exit(1);
 272                 }
 273         }
 274         else
 275         {
 276                 fprintf(stderr, _("%s: lock file \"%s\" exists\n"
 277                                                   "Is a server running?  If not, delete the lock file and try again.\n"),
 278                                 progname, path);
 279                 exit(1);
 280         }
 281
 282         /*
 283          * Attempt to read the existing pg_control file
 284          */
 285         if (!ReadControlFile())
 286                 GuessControlValues();
 287
 288         /*
 289          * Also look at existing segment files to set up newXlogId/newXlogSeg
 290          */
 291         FindEndOfXLOG();
 292
 293         /*
 294          * Adjust fields if required by switches.  (Do this now so that printout,
 295          * if any, includes these values.)
 296          */
 297         if (set_xid_epoch != -1)
 298                 ControlFile.checkPointCopy.nextXidEpoch = set_xid_epoch;
 299
 300         if (set_xid != 0)
 301                 ControlFile.checkPointCopy.nextXid = set_xid;
 302
 303         if (set_oid != 0)
 304                 ControlFile.checkPointCopy.nextOid = set_oid;
 305
 306         if (set_mxid != 0)
 307                 ControlFile.checkPointCopy.nextMulti = set_mxid;
 308
 309         if (set_mxoff != -1)
 310                 ControlFile.checkPointCopy.nextMultiOffset = set_mxoff;
 311
 312         if (minXlogTli > ControlFile.checkPointCopy.ThisTimeLineID)
 313                 ControlFile.checkPointCopy.ThisTimeLineID = minXlogTli;
 314
 315         if (minXlogId > newXlogId ||
 316                 (minXlogId == newXlogId &&
 317                  minXlogSeg > newXlogSeg))
 318         {
 319                 newXlogId = minXlogId;
 320                 newXlogSeg = minXlogSeg;
 321         }
 322
 323         /*
 324          * If we had to guess anything, and -f was not given, just print the
 325          * guessed values and exit.  Also print if -n is given.
 326          */
 327         if ((guessed && !force) || noupdate)
 328         {
 329                 PrintControlValues(guessed);
 330                 if (!noupdate)
 331                 {
 332                         printf(_("\nIf these values seem acceptable, use -f to force reset.\n"));
 333                         exit(1);
 334                 }
 335                 else
 336                         exit(0);
 337         }
 338
 339         /*
 340          * Don't reset from a dirty pg_control without -f, either.
 341          */
 342         if (ControlFile.state != DB_SHUTDOWNED && !force)
 343         {
 344                 printf(_("The database server was not shut down cleanly.\n"
 345                            "Resetting the transaction log might cause data to be lost.\n"
 346                                  "If you want to proceed anyway, use -f to force reset.\n"));
 347                 exit(1);
 348         }
 349
 350         /*
 351          * Else, do the dirty deed.
 352          */
 353         RewriteControlFile();
 354         KillExistingXLOG();
 355         WriteEmptyXLOG();
 356
 357         printf(_("Transaction log reset\n"));
 358         return 0;
 359 }
 360
 361
 362 /*
 363  * Try to read the existing pg_control file.
 364  *
 365  * This routine is also responsible for updating old pg_control versions
 366  * to the current format.  (Currently we don't do anything of the sort.)
 367  */
 368 static bool
 369 ReadControlFile(void)
 370 {
 371         int                     fd;
 372         int                     len;
 373         char       *buffer;
 374         pg_crc32        crc;
 375
 376         if ((fd = open(XLOG_CONTROL_FILE, O_RDONLY | PG_BINARY, 0)) < 0)
 377         {
 378                 /*
 379                  * If pg_control is not there at all, or we can't read it, the odds
 380                  * are we've been handed a bad DataDir path, so give up. User can do
 381                  * "touch pg_control" to force us to proceed.
 382                  */
 383                 fprintf(stderr, _("%s: could not open file \"%s\" for reading: %s\n"),
 384                                 progname, XLOG_CONTROL_FILE, strerror(errno));
 385                 if (errno == ENOENT)
 386                         fprintf(stderr, _("If you are sure the data directory path is correct, execute\n"
 387                                                           "  touch %s\n"
 388                                                           "and try again.\n"),
 389                                         XLOG_CONTROL_FILE);
 390                 exit(1);
 391         }
 392
 393         /* Use malloc to ensure we have a maxaligned buffer */
 394         buffer = (char *) malloc(PG_CONTROL_SIZE);
 395
 396         len = read(fd, buffer, PG_CONTROL_SIZE);
 397         if (len < 0)
 398         {
 399                 fprintf(stderr, _("%s: could not read file \"%s\": %s\n"),
 400                                 progname, XLOG_CONTROL_FILE, strerror(errno));
 401                 exit(1);
 402         }
 403         close(fd);
 404
 405         if (len >= sizeof(ControlFileData) &&
 406           ((ControlFileData *) buffer)->pg_control_version == PG_CONTROL_VERSION)
 407         {
 408                 /* Check the CRC. */
 409                 INIT_CRC32(crc);
 410                 COMP_CRC32(crc,
 411                                    buffer,
 412                                    offsetof(ControlFileData, crc));
 413                 FIN_CRC32(crc);
 414
 415                 if (EQ_CRC32(crc, ((ControlFileData *) buffer)->crc))
 416                 {
 417                         /* Valid data... */
 418                         memcpy(&ControlFile, buffer, sizeof(ControlFile));
 419                         return true;
 420                 }
 421
 422                 fprintf(stderr, _("%s: pg_control exists but has invalid CRC; proceed with caution\n"),
 423                                 progname);
 424                 /* We will use the data anyway, but treat it as guessed. */
 425                 memcpy(&ControlFile, buffer, sizeof(ControlFile));
 426                 guessed = true;
 427                 return true;
 428         }
 429
 430         /* Looks like it's a mess. */
 431         fprintf(stderr, _("%s: pg_control exists but is broken or unknown version; ignoring it\n"),
 432                         progname);
 433         return false;
 434 }
 435
 436
 437 /*
 438  * Guess at pg_control values when we can't read the old ones.
 439  */
 440 static void
 441 GuessControlValues(void)
 442 {
 443         uint64          sysidentifier;
 444         struct timeval tv;
 445
 446         /*
 447          * Set up a completely default set of pg_control values.
 448          */
 449         guessed = true;
 450         memset(&ControlFile, 0, sizeof(ControlFile));
 451
 452         ControlFile.pg_control_version = PG_CONTROL_VERSION;
 453         ControlFile.catalog_version_no = CATALOG_VERSION_NO;
 454
 455         /*
 456          * Create a new unique installation identifier, since we can no longer use
 457          * any old XLOG records.  See notes in xlog.c about the algorithm.
 458          */
 459         gettimeofday(&tv, NULL);
 460         sysidentifier = ((uint64) tv.tv_sec) << 32;
 461         sysidentifier |= (uint32) (tv.tv_sec | tv.tv_usec);
 462
 463         ControlFile.system_identifier = sysidentifier;
 464
 465         ControlFile.checkPointCopy.redo.xlogid = 0;
 466         ControlFile.checkPointCopy.redo.xrecoff = SizeOfXLogLongPHD;
 467         ControlFile.checkPointCopy.ThisTimeLineID = 1;
 468         ControlFile.checkPointCopy.nextXidEpoch = 0;
 469         ControlFile.checkPointCopy.nextXid = (TransactionId) 514;       /* XXX */
 470         ControlFile.checkPointCopy.nextOid = FirstBootstrapObjectId;
 471         ControlFile.checkPointCopy.nextMulti = FirstMultiXactId;
 472         ControlFile.checkPointCopy.nextMultiOffset = 0;
 473         ControlFile.checkPointCopy.time = (pg_time_t) time(NULL);
 474
 475         ControlFile.state = DB_SHUTDOWNED;
 476         ControlFile.time = (pg_time_t) time(NULL);
 477         ControlFile.checkPoint = ControlFile.checkPointCopy.redo;
 478
 479         ControlFile.maxAlign = MAXIMUM_ALIGNOF;
 480         ControlFile.floatFormat = FLOATFORMAT_VALUE;
 481         ControlFile.blcksz = BLCKSZ;
 482         ControlFile.relseg_size = RELSEG_SIZE;
 483         ControlFile.xlog_blcksz = XLOG_BLCKSZ;
 484         ControlFile.xlog_seg_size = XLOG_SEG_SIZE;
 485         ControlFile.nameDataLen = NAMEDATALEN;
 486         ControlFile.indexMaxKeys = INDEX_MAX_KEYS;
 487         ControlFile.toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
 488 #ifdef HAVE_INT64_TIMESTAMP
 489         ControlFile.enableIntTimes = true;
 490 #else
 491         ControlFile.enableIntTimes = false;
 492 #endif
 493         ControlFile.float4ByVal = FLOAT4PASSBYVAL;
 494         ControlFile.float8ByVal = FLOAT8PASSBYVAL;
 495
 496         /*
 497          * XXX eventually, should try to grovel through old XLOG to develop more
 498          * accurate values for TimeLineID, nextXID, etc.
 499          */
 500 }
 501
 502
 503 /*
 504  * Print the guessed pg_control values when we had to guess.
 505  *
 506  * NB: this display should be just those fields that will not be
 507  * reset by RewriteControlFile().
 508  */
 509 static void
 510 PrintControlValues(bool guessed)
 511 {
 512         char            sysident_str[32];
 513
 514         if (guessed)
 515                 printf(_("Guessed pg_control values:\n\n"));
 516         else
 517                 printf(_("pg_control values:\n\n"));
 518
 519         /*
 520          * Format system_identifier separately to keep platform-dependent format
 521          * code out of the translatable message string.
 522          */
 523         snprintf(sysident_str, sizeof(sysident_str), UINT64_FORMAT,
 524                          ControlFile.system_identifier);
 525
 526         printf(_("First log file ID after reset:        %u\n"),
 527                    newXlogId);
 528         printf(_("First log file segment after reset:   %u\n"),
 529                    newXlogSeg);
 530         printf(_("pg_control version number:            %u\n"),
 531                    ControlFile.pg_control_version);
 532         printf(_("Catalog version number:               %u\n"),
 533                    ControlFile.catalog_version_no);
 534         printf(_("Database system identifier:           %s\n"),
 535                    sysident_str);
 536         printf(_("Latest checkpoint's TimeLineID:       %u\n"),
 537                    ControlFile.checkPointCopy.ThisTimeLineID);
 538         printf(_("Latest checkpoint's NextXID:          %u/%u\n"),
 539                    ControlFile.checkPointCopy.nextXidEpoch,
 540                    ControlFile.checkPointCopy.nextXid);
 541         printf(_("Latest checkpoint's NextOID:          %u\n"),
 542                    ControlFile.checkPointCopy.nextOid);
 543         printf(_("Latest checkpoint's NextMultiXactId:  %u\n"),
 544                    ControlFile.checkPointCopy.nextMulti);
 545         printf(_("Latest checkpoint's NextMultiOffset:  %u\n"),
 546                    ControlFile.checkPointCopy.nextMultiOffset);
 547         printf(_("Maximum data alignment:               %u\n"),
 548                    ControlFile.maxAlign);
 549         /* we don't print floatFormat since can't say much useful about it */
 550         printf(_("Database block size:                  %u\n"),
 551                    ControlFile.blcksz);
 552         printf(_("Blocks per segment of large relation: %u\n"),
 553                    ControlFile.relseg_size);
 554         printf(_("WAL block size:                       %u\n"),
 555                    ControlFile.xlog_blcksz);
 556         printf(_("Bytes per WAL segment:                %u\n"),
 557                    ControlFile.xlog_seg_size);
 558         printf(_("Maximum length of identifiers:        %u\n"),
 559                    ControlFile.nameDataLen);
 560         printf(_("Maximum columns in an index:          %u\n"),
 561                    ControlFile.indexMaxKeys);
 562         printf(_("Maximum size of a TOAST chunk:        %u\n"),
 563                    ControlFile.toast_max_chunk_size);
 564         printf(_("Date/time type storage:               %s\n"),
 565                    (ControlFile.enableIntTimes ? _("64-bit integers") : _("floating-point numbers")));
 566         printf(_("Float4 argument passing:              %s\n"),
 567                    (ControlFile.float4ByVal ? _("by value") : _("by reference")));
 568         printf(_("Float8 argument passing:              %s\n"),
 569                    (ControlFile.float8ByVal ? _("by value") : _("by reference")));
 570 }
 571
 572
 573 /*
 574  * Write out the new pg_control file.
 575  */
 576 static void
 577 RewriteControlFile(void)
 578 {
 579         int                     fd;
 580         char            buffer[PG_CONTROL_SIZE];                /* need not be aligned */
 581
 582         /*
 583          * Adjust fields as needed to force an empty XLOG starting at
 584          * newXlogId/newXlogSeg.
 585          */
 586         ControlFile.checkPointCopy.redo.xlogid = newXlogId;
 587         ControlFile.checkPointCopy.redo.xrecoff =
 588                 newXlogSeg * XLogSegSize + SizeOfXLogLongPHD;
 589         ControlFile.checkPointCopy.time = (pg_time_t) time(NULL);
 590
 591         ControlFile.state = DB_SHUTDOWNED;
 592         ControlFile.time = (pg_time_t) time(NULL);
 593         ControlFile.checkPoint = ControlFile.checkPointCopy.redo;
 594         ControlFile.prevCheckPoint.xlogid = 0;
 595         ControlFile.prevCheckPoint.xrecoff = 0;
 596         ControlFile.minRecoveryPoint.xlogid = 0;
 597         ControlFile.minRecoveryPoint.xrecoff = 0;
 598
 599         /* Now we can force the recorded xlog seg size to the right thing. */
 600         ControlFile.xlog_seg_size = XLogSegSize;
 601
 602         /* Contents are protected with a CRC */
 603         INIT_CRC32(ControlFile.crc);
 604         COMP_CRC32(ControlFile.crc,
 605                            (char *) &ControlFile,
 606                            offsetof(ControlFileData, crc));
 607         FIN_CRC32(ControlFile.crc);
 608
 609         /*
 610          * We write out PG_CONTROL_SIZE bytes into pg_control, zero-padding the
 611          * excess over sizeof(ControlFileData).  This reduces the odds of
 612          * premature-EOF errors when reading pg_control.  We'll still fail when we
 613          * check the contents of the file, but hopefully with a more specific
 614          * error than "couldn't read pg_control".
 615          */
 616         if (sizeof(ControlFileData) > PG_CONTROL_SIZE)
 617         {
 618                 fprintf(stderr,
 619                                 _("%s: internal error -- sizeof(ControlFileData) is too large ... fix PG_CONTROL_SIZE\n"),
 620                                 progname);
 621                 exit(1);
 622         }
 623
 624         memset(buffer, 0, PG_CONTROL_SIZE);
 625         memcpy(buffer, &ControlFile, sizeof(ControlFileData));
 626
 627         unlink(XLOG_CONTROL_FILE);
 628
 629         fd = open(XLOG_CONTROL_FILE,
 630                           O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
 631                           S_IRUSR | S_IWUSR);
 632         if (fd < 0)
 633         {
 634                 fprintf(stderr, _("%s: could not create pg_control file: %s\n"),
 635                                 progname, strerror(errno));
 636                 exit(1);
 637         }
 638
 639         errno = 0;
 640         if (write(fd, buffer, PG_CONTROL_SIZE) != PG_CONTROL_SIZE)
 641         {
 642                 /* if write didn't set errno, assume problem is no disk space */
 643                 if (errno == 0)
 644                         errno = ENOSPC;
 645                 fprintf(stderr, _("%s: could not write pg_control file: %s\n"),
 646                                 progname, strerror(errno));
 647                 exit(1);
 648         }
 649
 650         if (fsync(fd) != 0)
 651         {
 652                 fprintf(stderr, _("%s: fsync error: %s\n"), progname, strerror(errno));
 653                 exit(1);
 654         }
 655
 656         close(fd);
 657 }
 658
 659
 660 /*
 661  * Scan existing XLOG files and determine the highest existing WAL address
 662  *
 663  * On entry, ControlFile.checkPointCopy.redo and ControlFile.xlog_seg_size
 664  * are assumed valid (note that we allow the old xlog seg size to differ
 665  * from what we're using).  On exit, newXlogId and newXlogSeg are set to
 666  * suitable values for the beginning of replacement WAL (in our seg size).
 667  */
 668 static void
 669 FindEndOfXLOG(void)
 670 {
 671         DIR                *xldir;
 672         struct dirent *xlde;
 673
 674         /*
 675          * Initialize the max() computation using the last checkpoint address from
 676          * old pg_control.      Note that for the moment we are working with segment
 677          * numbering according to the old xlog seg size.
 678          */
 679         newXlogId = ControlFile.checkPointCopy.redo.xlogid;
 680         newXlogSeg = ControlFile.checkPointCopy.redo.xrecoff / ControlFile.xlog_seg_size;
 681
 682         /*
 683          * Scan the pg_xlog directory to find existing WAL segment files. We
 684          * assume any present have been used; in most scenarios this should be
 685          * conservative, because of xlog.c's attempts to pre-create files.
 686          */
 687         xldir = opendir(XLOGDIR);
 688         if (xldir == NULL)
 689         {
 690                 fprintf(stderr, _("%s: could not open directory \"%s\": %s\n"),
 691                                 progname, XLOGDIR, strerror(errno));
 692                 exit(1);
 693         }
 694
 695         errno = 0;
 696         while ((xlde = readdir(xldir)) != NULL)
 697         {
 698                 if (strlen(xlde->d_name) == 24 &&
 699                         strspn(xlde->d_name, "0123456789ABCDEF") == 24)
 700                 {
 701                         unsigned int tli,
 702                                                 log,
 703                                                 seg;
 704
 705                         sscanf(xlde->d_name, "%08X%08X%08X", &tli, &log, &seg);
 706
 707                         /*
 708                          * Note: we take the max of all files found, regardless of their
 709                          * timelines.  Another possibility would be to ignore files of
 710                          * timelines other than the target TLI, but this seems safer.
 711                          * Better too large a result than too small...
 712                          */
 713                         if (log > newXlogId ||
 714                                 (log == newXlogId && seg > newXlogSeg))
 715                         {
 716                                 newXlogId = log;
 717                                 newXlogSeg = seg;
 718                         }
 719                 }
 720                 errno = 0;
 721         }
 722 #ifdef WIN32
 723
 724         /*
 725          * This fix is in mingw cvs (runtime/mingwex/dirent.c rev 1.4), but not in
 726          * released version
 727          */
 728         if (GetLastError() == ERROR_NO_MORE_FILES)
 729                 errno = 0;
 730 #endif
 731
 732         if (errno)
 733         {
 734                 fprintf(stderr, _("%s: could not read from directory \"%s\": %s\n"),
 735                                 progname, XLOGDIR, strerror(errno));
 736                 exit(1);
 737         }
 738         closedir(xldir);
 739
 740         /*
 741          * Finally, convert to new xlog seg size, and advance by one to ensure we
 742          * are in virgin territory.
 743          */
 744         newXlogSeg *= ControlFile.xlog_seg_size;
 745         newXlogSeg = (newXlogSeg + XLogSegSize - 1) / XLogSegSize;
 746
 747         /* be sure we wrap around correctly at end of a logfile */
 748         NextLogSeg(newXlogId, newXlogSeg);
 749 }
 750
 751
 752 /*
 753  * Remove existing XLOG files
 754  */
 755 static void
 756 KillExistingXLOG(void)
 757 {
 758         DIR                *xldir;
 759         struct dirent *xlde;
 760         char            path[MAXPGPATH];
 761
 762         xldir = opendir(XLOGDIR);
 763         if (xldir == NULL)
 764         {
 765                 fprintf(stderr, _("%s: could not open directory \"%s\": %s\n"),
 766                                 progname, XLOGDIR, strerror(errno));
 767                 exit(1);
 768         }
 769
 770         errno = 0;
 771         while ((xlde = readdir(xldir)) != NULL)
 772         {
 773                 if (strlen(xlde->d_name) == 24 &&
 774                         strspn(xlde->d_name, "0123456789ABCDEF") == 24)
 775                 {
 776                         snprintf(path, MAXPGPATH, "%s/%s", XLOGDIR, xlde->d_name);
 777                         if (unlink(path) < 0)
 778                         {
 779                                 fprintf(stderr, _("%s: could not delete file \"%s\": %s\n"),
 780                                                 progname, path, strerror(errno));
 781                                 exit(1);
 782                         }
 783                 }
 784                 errno = 0;
 785         }
 786 #ifdef WIN32
 787
 788         /*
 789          * This fix is in mingw cvs (runtime/mingwex/dirent.c rev 1.4), but not in
 790          * released version
 791          */
 792         if (GetLastError() == ERROR_NO_MORE_FILES)
 793                 errno = 0;
 794 #endif
 795
 796         if (errno)
 797         {
 798                 fprintf(stderr, _("%s: could not read from directory \"%s\": %s\n"),
 799                                 progname, XLOGDIR, strerror(errno));
 800                 exit(1);
 801         }
 802         closedir(xldir);
 803 }
 804
 805
 806 /*
 807  * Write an empty XLOG file, containing only the checkpoint record
 808  * already set up in ControlFile.
 809  */
 810 static void
 811 WriteEmptyXLOG(void)
 812 {
 813         char       *buffer;
 814         XLogPageHeader page;
 815         XLogLongPageHeader longpage;
 816         XLogRecord *record;
 817         pg_crc32        crc;
 818         char            path[MAXPGPATH];
 819         int                     fd;
 820         int                     nbytes;
 821
 822         /* Use malloc() to ensure buffer is MAXALIGNED */
 823         buffer = (char *) malloc(XLOG_BLCKSZ);
 824         page = (XLogPageHeader) buffer;
 825         memset(buffer, 0, XLOG_BLCKSZ);
 826
 827         /* Set up the XLOG page header */
 828         page->xlp_magic = XLOG_PAGE_MAGIC;
 829         page->xlp_info = XLP_LONG_HEADER;
 830         page->xlp_tli = ControlFile.checkPointCopy.ThisTimeLineID;
 831         page->xlp_pageaddr.xlogid =
 832                 ControlFile.checkPointCopy.redo.xlogid;
 833         page->xlp_pageaddr.xrecoff =
 834                 ControlFile.checkPointCopy.redo.xrecoff - SizeOfXLogLongPHD;
 835         longpage = (XLogLongPageHeader) page;
 836         longpage->xlp_sysid = ControlFile.system_identifier;
 837         longpage->xlp_seg_size = XLogSegSize;
 838         longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
 839
 840         /* Insert the initial checkpoint record */
 841         record = (XLogRecord *) ((char *) page + SizeOfXLogLongPHD);
 842         record->xl_prev.xlogid = 0;
 843         record->xl_prev.xrecoff = 0;
 844         record->xl_xid = InvalidTransactionId;
 845         record->xl_tot_len = SizeOfXLogRecord + sizeof(CheckPoint);
 846         record->xl_len = sizeof(CheckPoint);
 847         record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
 848         record->xl_rmid = RM_XLOG_ID;
 849         memcpy(XLogRecGetData(record), &ControlFile.checkPointCopy,
 850                    sizeof(CheckPoint));
 851
 852         INIT_CRC32(crc);
 853         COMP_CRC32(crc, &ControlFile.checkPointCopy, sizeof(CheckPoint));
 854         COMP_CRC32(crc, (char *) record + sizeof(pg_crc32),
 855                            SizeOfXLogRecord - sizeof(pg_crc32));
 856         FIN_CRC32(crc);
 857         record->xl_crc = crc;
 858
 859         /* Write the first page */
 860         XLogFilePath(path, ControlFile.checkPointCopy.ThisTimeLineID,
 861                                  newXlogId, newXlogSeg);
 862
 863         unlink(path);
 864
 865         fd = open(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
 866                           S_IRUSR | S_IWUSR);
 867         if (fd < 0)
 868         {
 869                 fprintf(stderr, _("%s: could not open file \"%s\": %s\n"),
 870                                 progname, path, strerror(errno));
 871                 exit(1);
 872         }
 873
 874         errno = 0;
 875         if (write(fd, buffer, XLOG_BLCKSZ) != XLOG_BLCKSZ)
 876         {
 877                 /* if write didn't set errno, assume problem is no disk space */
 878                 if (errno == 0)
 879                         errno = ENOSPC;
 880                 fprintf(stderr, _("%s: could not write file \"%s\": %s\n"),
 881                                 progname, path, strerror(errno));
 882                 exit(1);
 883         }
 884
 885         /* Fill the rest of the file with zeroes */
 886         memset(buffer, 0, XLOG_BLCKSZ);
 887         for (nbytes = XLOG_BLCKSZ; nbytes < XLogSegSize; nbytes += XLOG_BLCKSZ)
 888         {
 889                 errno = 0;
 890                 if (write(fd, buffer, XLOG_BLCKSZ) != XLOG_BLCKSZ)
 891                 {
 892                         if (errno == 0)
 893                                 errno = ENOSPC;
 894                         fprintf(stderr, _("%s: could not write file \"%s\": %s\n"),
 895                                         progname, path, strerror(errno));
 896                         exit(1);
 897                 }
 898         }
 899
 900         if (fsync(fd) != 0)
 901         {
 902                 fprintf(stderr, _("%s: fsync error: %s\n"), progname, strerror(errno));
 903                 exit(1);
 904         }
 905
 906         close(fd);
 907 }
 908
 909
 910 static void
 911 usage(void)
 912 {
 913         printf(_("%s resets the PostgreSQL transaction log.\n\n"), progname);
 914         printf(_("Usage:\n  %s [OPTION]... DATADIR\n\n"), progname);
 915         printf(_("Options:\n"));
 916         printf(_("  -f              force update to be done\n"));
 917         printf(_("  -l TLI,FILE,SEG force minimum WAL starting location for new transaction log\n"));
 918         printf(_("  -m XID          set next multitransaction ID\n"));
 919         printf(_("  -n              no update, just show extracted control values (for testing)\n"));
 920         printf(_("  -o OID          set next OID\n"));
 921         printf(_("  -O OFFSET       set next multitransaction offset\n"));
 922         printf(_("  -x XID          set next transaction ID\n"));
 923         printf(_("  -e XIDEPOCH     set next transaction ID epoch\n"));
 924         printf(_("  --help          show this help, then exit\n"));
 925         printf(_("  --version       output version information, then exit\n"));
 926         printf(_("\nReport bugs to <pgsql-bugs@postgresql.org>.\n"));
 927 }