1 /*-------------------------------------------------------------------------
4 * A utility to "zero out" the xlog when it's corrupt beyond recovery.
5 * Can also rebuild pg_control if needed.
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
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
23 * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
24 * Portions Copyright (c) 1994, Regents of the University of California
26 * $PostgreSQL: pgsql/src/bin/pg_resetxlog/pg_resetxlog.c,v 1.20 2004/06/03 00:07:37 momjian Exp $
28 *-------------------------------------------------------------------------
41 #include "access/xlog.h"
42 #include "catalog/catversion.h"
43 #include "catalog/pg_control.h"
48 #define _(x) gettext((x))
51 /******************** stuff copied from xlog.c ********************/
53 /* Increment an xlogid/segment pair */
54 #define NextLogSeg(logId, logSeg) \
56 if ((logSeg) >= XLogSegsPerFile-1) \
65 #define XLogFileName(path, log, seg) \
66 snprintf(path, MAXPGPATH, "%s/%08X%08X", \
69 /******************** end of stuff copied from xlog.c ********************/
71 static char XLogDir[MAXPGPATH];
72 static char ControlFilePath[MAXPGPATH];
74 static ControlFileData ControlFile; /* pg_control values */
75 static uint32 newXlogId,
76 newXlogSeg; /* ID/Segment of new XLOG segment */
77 static bool guessed = false; /* T if we had to guess at any values */
78 static const char *progname;
80 static bool ReadControlFile(void);
81 static void GuessControlValues(void);
82 static void PrintControlValues(bool guessed);
83 static void RewriteControlFile(void);
84 static void KillExistingXLOG(void);
85 static void WriteEmptyXLOG(void);
86 static void usage(void);
90 main(int argc, char *argv[])
94 bool noupdate = false;
95 TransactionId set_xid = 0;
103 char path[MAXPGPATH];
105 set_pglocale_pgservice(argv[0], "pg_resetxlog");
107 progname = get_progname(argv[0]);
111 if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
116 if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
118 puts("pg_resetxlog (PostgreSQL) " PG_VERSION);
124 while ((c = getopt(argc, argv, "fl:no:x:")) != -1)
137 set_xid = strtoul(optarg, &endptr, 0);
138 if (endptr == optarg || *endptr != '\0')
140 fprintf(stderr, _("%s: invalid argument for option -x\n"), progname);
141 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
146 fprintf(stderr, _("%s: transaction ID (-x) must not be 0\n"), progname);
152 set_oid = strtoul(optarg, &endptr, 0);
153 if (endptr == optarg || *endptr != '\0')
155 fprintf(stderr, _("%s: invalid argument for option -o\n"), progname);
156 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
161 fprintf(stderr, _("%s: OID (-o) must not be 0\n"), progname);
167 minXlogId = strtoul(optarg, &endptr, 0);
168 if (endptr == optarg || *endptr != ',')
170 fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
171 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
174 minXlogSeg = strtoul(endptr + 1, &endptr2, 0);
175 if (endptr2 == endptr + 1 || *endptr2 != '\0')
177 fprintf(stderr, _("%s: invalid argument for option -l\n"), progname);
178 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
184 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
191 fprintf(stderr, _("%s: no data directory specified\n"), progname);
192 fprintf(stderr, _("Try \"%s --help\" for more information.\n"), progname);
196 DataDir = argv[optind];
197 snprintf(XLogDir, MAXPGPATH, "%s/pg_xlog", DataDir);
198 snprintf(ControlFilePath, MAXPGPATH, "%s/global/pg_control", DataDir);
201 * Check for a postmaster lock file --- if there is one, refuse to
202 * proceed, on grounds we might be interfering with a live
205 snprintf(path, MAXPGPATH, "%s/postmaster.pid", DataDir);
207 if ((fd = open(path, O_RDONLY)) < 0)
211 fprintf(stderr, _("%s: could not open file \"%s\" for reading: %s\n"), progname, path, strerror(errno));
217 fprintf(stderr, _("%s: lock file \"%s\" exists\n"
218 "Is a server running? If not, delete the lock file and try again.\n"),
224 * Attempt to read the existing pg_control file
226 if (!ReadControlFile())
227 GuessControlValues();
230 * Adjust fields if required by switches. (Do this now so that
231 * printout, if any, includes these values.)
234 ControlFile.checkPointCopy.nextXid = set_xid;
237 ControlFile.checkPointCopy.nextOid = set_oid;
239 if (minXlogId > ControlFile.logId ||
240 (minXlogId == ControlFile.logId &&
241 minXlogSeg > ControlFile.logSeg))
243 ControlFile.logId = minXlogId;
244 ControlFile.logSeg = minXlogSeg;
248 * If we had to guess anything, and -f was not given, just print the
249 * guessed values and exit. Also print if -n is given.
251 if ((guessed && !force) || noupdate)
253 PrintControlValues(guessed);
256 printf(_("\nIf these values seem acceptable, use -f to force reset.\n"));
264 * Don't reset from a dirty pg_control without -f, either.
266 if (ControlFile.state != DB_SHUTDOWNED && !force)
268 printf(_("The database server was not shut down cleanly.\n"
269 "Resetting the transaction log may cause data to be lost.\n"
270 "If you want to proceed anyway, use -f to force reset.\n"));
275 * Else, do the dirty deed.
277 RewriteControlFile();
281 printf(_("Transaction log reset\n"));
287 * Try to read the existing pg_control file.
289 * This routine is also responsible for updating old pg_control versions
290 * to the current format. (Currently we don't do anything of the sort.)
293 ReadControlFile(void)
300 if ((fd = open(ControlFilePath, O_RDONLY)) < 0)
303 * If pg_control is not there at all, or we can't read it, the
304 * odds are we've been handed a bad DataDir path, so give up. User
305 * can do "touch pg_control" to force us to proceed.
307 fprintf(stderr, _("%s: could not open file \"%s\" for reading: %s\n"),
308 progname, ControlFilePath, strerror(errno));
310 fprintf(stderr, _("If you are sure the data directory path is correct, execute\n"
317 /* Use malloc to ensure we have a maxaligned buffer */
318 buffer = (char *) malloc(BLCKSZ);
320 len = read(fd, buffer, BLCKSZ);
323 fprintf(stderr, _("%s: could not read file \"%s\": %s\n"),
324 progname, ControlFilePath, strerror(errno));
329 if (len >= sizeof(ControlFileData) &&
330 ((ControlFileData *) buffer)->pg_control_version == PG_CONTROL_VERSION)
335 buffer + sizeof(crc64),
336 sizeof(ControlFileData) - sizeof(crc64));
339 if (EQ_CRC64(crc, ((ControlFileData *) buffer)->crc))
342 memcpy(&ControlFile, buffer, sizeof(ControlFile));
346 fprintf(stderr, _("%s: pg_control exists but has invalid CRC; proceed with caution\n"),
348 /* We will use the data anyway, but treat it as guessed. */
349 memcpy(&ControlFile, buffer, sizeof(ControlFile));
354 /* Looks like it's a mess. */
355 fprintf(stderr, _("%s: pg_control exists but is broken or unknown version; ignoring it\n"),
362 * Guess at pg_control values when we can't read the old ones.
365 GuessControlValues(void)
367 uint64 sysidentifier;
372 * Set up a completely default set of pg_control values.
375 memset(&ControlFile, 0, sizeof(ControlFile));
377 ControlFile.pg_control_version = PG_CONTROL_VERSION;
378 ControlFile.catalog_version_no = CATALOG_VERSION_NO;
381 * Create a new unique installation identifier, since we can no longer
382 * use any old XLOG records. See notes in xlog.c about the algorithm.
384 gettimeofday(&tv, NULL);
385 sysidentifier = ((uint64) tv.tv_sec) << 32;
386 sysidentifier |= (uint32) (tv.tv_sec | tv.tv_usec);
388 ControlFile.system_identifier = sysidentifier;
390 ControlFile.checkPointCopy.redo.xlogid = 0;
391 ControlFile.checkPointCopy.redo.xrecoff = SizeOfXLogPHD + SizeOfXLogRecord + SizeOfXLogFHD;
392 ControlFile.checkPointCopy.undo = ControlFile.checkPointCopy.redo;
393 ControlFile.checkPointCopy.ThisStartUpID = 0;
394 ControlFile.checkPointCopy.nextXid = (TransactionId) 514; /* XXX */
395 ControlFile.checkPointCopy.nextOid = BootstrapObjectIdData;
396 ControlFile.checkPointCopy.time = time(NULL);
398 ControlFile.state = DB_SHUTDOWNED;
399 ControlFile.time = time(NULL);
400 ControlFile.logId = 0;
401 ControlFile.logSeg = 1;
402 ControlFile.checkPoint = ControlFile.checkPointCopy.redo;
404 ControlFile.blcksz = BLCKSZ;
405 ControlFile.relseg_size = RELSEG_SIZE;
406 ControlFile.xlog_seg_size = XLOG_SEG_SIZE;
407 ControlFile.nameDataLen = NAMEDATALEN;
408 ControlFile.funcMaxArgs = FUNC_MAX_ARGS;
409 #ifdef HAVE_INT64_TIMESTAMP
410 ControlFile.enableIntTimes = TRUE;
412 ControlFile.enableIntTimes = FALSE;
414 ControlFile.localeBuflen = LOCALE_NAME_BUFLEN;
416 localeptr = setlocale(LC_COLLATE, "");
419 fprintf(stderr, _("%s: invalid LC_COLLATE setting\n"), progname);
422 StrNCpy(ControlFile.lc_collate, localeptr, LOCALE_NAME_BUFLEN);
423 localeptr = setlocale(LC_CTYPE, "");
426 fprintf(stderr, _("%s: invalid LC_CTYPE setting\n"), progname);
429 StrNCpy(ControlFile.lc_ctype, localeptr, LOCALE_NAME_BUFLEN);
432 * XXX eventually, should try to grovel through old XLOG to develop
433 * more accurate values for startupid, nextXID, and nextOID.
439 * Print the guessed pg_control values when we had to guess.
441 * NB: this display should be just those fields that will not be
442 * reset by RewriteControlFile().
445 PrintControlValues(bool guessed)
447 char sysident_str[32];
450 printf(_("Guessed pg_control values:\n\n"));
452 printf(_("pg_control values:\n\n"));
455 * Format system_identifier separately to keep platform-dependent format
456 * code out of the translatable message string.
458 snprintf(sysident_str, sizeof(sysident_str), UINT64_FORMAT,
459 ControlFile.system_identifier);
461 printf(_("pg_control version number: %u\n"), ControlFile.pg_control_version);
462 printf(_("Catalog version number: %u\n"), ControlFile.catalog_version_no);
463 printf(_("Database system identifier: %s\n"), sysident_str);
464 printf(_("Current log file ID: %u\n"), ControlFile.logId);
465 printf(_("Next log file segment: %u\n"), ControlFile.logSeg);
466 printf(_("Latest checkpoint's StartUpID: %u\n"), ControlFile.checkPointCopy.ThisStartUpID);
467 printf(_("Latest checkpoint's NextXID: %u\n"), ControlFile.checkPointCopy.nextXid);
468 printf(_("Latest checkpoint's NextOID: %u\n"), ControlFile.checkPointCopy.nextOid);
469 printf(_("Database block size: %u\n"), ControlFile.blcksz);
470 printf(_("Blocks per segment of large relation: %u\n"), ControlFile.relseg_size);
471 printf(_("Maximum length of identifiers: %u\n"), ControlFile.nameDataLen);
472 printf(_("Maximum number of function arguments: %u\n"), ControlFile.funcMaxArgs);
473 printf(_("Date/time type storage: %s\n"),
474 (ControlFile.enableIntTimes ? _("64-bit integers") : _("floating-point numbers")));
475 printf(_("Maximum length of locale name: %u\n"), ControlFile.localeBuflen);
476 printf(_("LC_COLLATE: %s\n"), ControlFile.lc_collate);
477 printf(_("LC_CTYPE: %s\n"), ControlFile.lc_ctype);
482 * Write out the new pg_control file.
485 RewriteControlFile(void)
488 char buffer[BLCKSZ]; /* need not be aligned */
491 * Adjust fields as needed to force an empty XLOG starting at the next
494 newXlogId = ControlFile.logId;
495 newXlogSeg = ControlFile.logSeg;
497 /* adjust in case we are changing segment size */
498 newXlogSeg *= ControlFile.xlog_seg_size;
499 newXlogSeg = (newXlogSeg + XLogSegSize-1) / XLogSegSize;
501 /* be sure we wrap around correctly at end of a logfile */
502 NextLogSeg(newXlogId, newXlogSeg);
504 /* Now we can force the recorded xlog seg size to the right thing. */
505 ControlFile.xlog_seg_size = XLogSegSize;
507 ControlFile.checkPointCopy.redo.xlogid = newXlogId;
508 ControlFile.checkPointCopy.redo.xrecoff =
509 newXlogSeg * XLogSegSize + SizeOfXLogPHD + SizeOfXLogRecord + SizeOfXLogFHD;
510 ControlFile.checkPointCopy.undo = ControlFile.checkPointCopy.redo;
511 ControlFile.checkPointCopy.time = time(NULL);
513 ControlFile.state = DB_SHUTDOWNED;
514 ControlFile.time = time(NULL);
515 ControlFile.logId = newXlogId;
516 ControlFile.logSeg = newXlogSeg + 1;
517 ControlFile.checkPoint = ControlFile.checkPointCopy.redo;
518 ControlFile.prevCheckPoint.xlogid = 0;
519 ControlFile.prevCheckPoint.xrecoff = 0;
521 /* Contents are protected with a CRC */
522 INIT_CRC64(ControlFile.crc);
523 COMP_CRC64(ControlFile.crc,
524 (char *) &ControlFile + sizeof(crc64),
525 sizeof(ControlFileData) - sizeof(crc64));
526 FIN_CRC64(ControlFile.crc);
529 * We write out BLCKSZ bytes into pg_control, zero-padding the excess
530 * over sizeof(ControlFileData). This reduces the odds of
531 * premature-EOF errors when reading pg_control. We'll still fail
532 * when we check the contents of the file, but hopefully with a more
533 * specific error than "couldn't read pg_control".
535 if (sizeof(ControlFileData) > BLCKSZ)
538 _("%s: internal error -- sizeof(ControlFileData) is too large ... fix xlog.c\n"),
543 memset(buffer, 0, BLCKSZ);
544 memcpy(buffer, &ControlFile, sizeof(ControlFileData));
546 unlink(ControlFilePath);
548 fd = open(ControlFilePath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY, S_IRUSR | S_IWUSR);
551 fprintf(stderr, _("%s: could not create pg_control file: %s\n"),
552 progname, strerror(errno));
557 if (write(fd, buffer, BLCKSZ) != BLCKSZ)
559 /* if write didn't set errno, assume problem is no disk space */
562 fprintf(stderr, _("%s: could not write pg_control file: %s\n"),
563 progname, strerror(errno));
569 fprintf(stderr, _("%s: fsync error: %s\n"), progname, strerror(errno));
578 * Remove existing XLOG files
581 KillExistingXLOG(void)
585 char path[MAXPGPATH];
587 xldir = opendir(XLogDir);
590 fprintf(stderr, _("%s: could not open directory \"%s\": %s\n"),
591 progname, XLogDir, strerror(errno));
596 while ((xlde = readdir(xldir)) != NULL)
598 if (strlen(xlde->d_name) == 16 &&
599 strspn(xlde->d_name, "0123456789ABCDEF") == 16)
601 snprintf(path, MAXPGPATH, "%s/%s", XLogDir, xlde->d_name);
602 if (unlink(path) < 0)
604 fprintf(stderr, _("%s: could not delete file \"%s\": %s\n"),
605 progname, path, strerror(errno));
612 /* This fix is in mingw cvs (runtime/mingwex/dirent.c rev 1.4), but
613 not in released version */
614 if (GetLastError() == ERROR_NO_MORE_FILES)
620 fprintf(stderr, _("%s: could not read from directory \"%s\": %s\n"),
621 progname, XLogDir, strerror(errno));
629 * Write an empty XLOG file, containing only the checkpoint record
630 * already set up in ControlFile.
638 XLogFileHeaderData *fhdr;
640 char path[MAXPGPATH];
644 /* Use malloc() to ensure buffer is MAXALIGNED */
645 buffer = (char *) malloc(BLCKSZ);
646 page = (XLogPageHeader) buffer;
647 memset(buffer, 0, BLCKSZ);
649 /* Set up the XLOG page header */
650 page->xlp_magic = XLOG_PAGE_MAGIC;
652 page->xlp_sui = ControlFile.checkPointCopy.ThisStartUpID;
653 page->xlp_pageaddr.xlogid =
654 ControlFile.checkPointCopy.redo.xlogid;
655 page->xlp_pageaddr.xrecoff =
656 ControlFile.checkPointCopy.redo.xrecoff -
657 (SizeOfXLogPHD + SizeOfXLogRecord + SizeOfXLogFHD);
659 /* Insert the file header record */
660 record = (XLogRecord *) ((char *) page + SizeOfXLogPHD);
661 record->xl_prev.xlogid = 0;
662 record->xl_prev.xrecoff = 0;
663 record->xl_xact_prev.xlogid = 0;
664 record->xl_xact_prev.xrecoff = 0;
665 record->xl_xid = InvalidTransactionId;
666 record->xl_len = SizeOfXLogFHD;
667 record->xl_info = XLOG_FILE_HEADER;
668 record->xl_rmid = RM_XLOG_ID;
669 fhdr = (XLogFileHeaderData *) XLogRecGetData(record);
670 fhdr->xlfhd_sysid = ControlFile.system_identifier;
671 fhdr->xlfhd_xlogid = page->xlp_pageaddr.xlogid;
672 fhdr->xlfhd_segno = page->xlp_pageaddr.xrecoff / XLogSegSize;
673 fhdr->xlfhd_seg_size = XLogSegSize;
676 COMP_CRC64(crc, fhdr, SizeOfXLogFHD);
677 COMP_CRC64(crc, (char *) record + sizeof(crc64),
678 SizeOfXLogRecord - sizeof(crc64));
680 record->xl_crc = crc;
682 /* Insert the initial checkpoint record */
683 record = (XLogRecord *) ((char *) page + SizeOfXLogPHD + SizeOfXLogRecord + SizeOfXLogFHD);
684 record->xl_prev.xlogid = page->xlp_pageaddr.xlogid;
685 record->xl_prev.xrecoff = page->xlp_pageaddr.xrecoff + SizeOfXLogPHD;
686 record->xl_xact_prev.xlogid = 0;
687 record->xl_xact_prev.xrecoff = 0;
688 record->xl_xid = InvalidTransactionId;
689 record->xl_len = sizeof(CheckPoint);
690 record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
691 record->xl_rmid = RM_XLOG_ID;
692 memcpy(XLogRecGetData(record), &ControlFile.checkPointCopy,
696 COMP_CRC64(crc, &ControlFile.checkPointCopy, sizeof(CheckPoint));
697 COMP_CRC64(crc, (char *) record + sizeof(crc64),
698 SizeOfXLogRecord - sizeof(crc64));
700 record->xl_crc = crc;
702 /* Write the first page */
703 XLogFileName(path, newXlogId, newXlogSeg);
707 fd = open(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY,
711 fprintf(stderr, _("%s: could not open file \"%s\": %s\n"),
712 progname, path, strerror(errno));
717 if (write(fd, buffer, BLCKSZ) != BLCKSZ)
719 /* if write didn't set errno, assume problem is no disk space */
722 fprintf(stderr, _("%s: could not write file \"%s\": %s\n"),
723 progname, path, strerror(errno));
727 /* Fill the rest of the file with zeroes */
728 memset(buffer, 0, BLCKSZ);
729 for (nbytes = BLCKSZ; nbytes < XLogSegSize; nbytes += BLCKSZ)
732 if (write(fd, buffer, BLCKSZ) != BLCKSZ)
736 fprintf(stderr, _("%s: could not write file \"%s\": %s\n"),
737 progname, path, strerror(errno));
744 fprintf(stderr, _("%s: fsync error: %s\n"), progname, strerror(errno));
755 printf(_("%s resets the PostgreSQL transaction log.\n\n"), progname);
756 printf(_("Usage:\n %s [OPTION]... DATADIR\n\n"), progname);
757 printf(_("Options:\n"));
758 printf(_(" -f force update to be done\n"));
759 printf(_(" -l FILEID,SEG force minimum WAL starting location for new transaction log\n"));
760 printf(_(" -n no update, just show extracted control values (for testing)\n"));
761 printf(_(" -o OID set next OID\n"));
762 printf(_(" -x XID set next transaction ID\n"));
763 printf(_(" --help show this help, then exit\n"));
764 printf(_(" --version output version information, then exit\n"));
765 printf(_("\nReport bugs to <pgsql-bugs@postgresql.org>.\n"));