Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pg_rewind.c
4 : * Synchronizes a PostgreSQL data directory to a new timeline
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : *
8 : *-------------------------------------------------------------------------
9 : */
10 : #include "postgres_fe.h"
11 :
12 : #include <sys/stat.h>
13 : #include <fcntl.h>
14 : #include <time.h>
15 : #include <unistd.h>
16 :
17 : #include "access/timeline.h"
18 : #include "access/xlog_internal.h"
19 : #include "catalog/catversion.h"
20 : #include "catalog/pg_control.h"
21 : #include "common/controldata_utils.h"
22 : #include "common/file_perm.h"
23 : #include "common/restricted_token.h"
24 : #include "common/string.h"
25 : #include "fe_utils/option_utils.h"
26 : #include "fe_utils/recovery_gen.h"
27 : #include "fe_utils/string_utils.h"
28 : #include "file_ops.h"
29 : #include "filemap.h"
30 : #include "getopt_long.h"
31 : #include "pg_rewind.h"
32 : #include "rewind_source.h"
33 : #include "storage/bufpage.h"
34 :
35 : static void usage(const char *progname);
36 :
37 : static void perform_rewind(filemap_t *filemap, rewind_source *source,
38 : XLogRecPtr chkptrec,
39 : TimeLineID chkpttli,
40 : XLogRecPtr chkptredo);
41 :
42 : static void createBackupLabel(XLogRecPtr startpoint, TimeLineID starttli,
43 : XLogRecPtr checkpointloc);
44 :
45 : static void digestControlFile(ControlFileData *ControlFile,
46 : const char *content, size_t size);
47 : static void getRestoreCommand(const char *argv0);
48 : static void sanityChecks(void);
49 : static TimeLineHistoryEntry *getTimelineHistory(TimeLineID tli, bool is_source,
50 : int *nentries);
51 : static void findCommonAncestorTimeline(TimeLineHistoryEntry *a_history,
52 : int a_nentries,
53 : TimeLineHistoryEntry *b_history,
54 : int b_nentries,
55 : XLogRecPtr *recptr, int *tliIndex);
56 : static void ensureCleanShutdown(const char *argv0);
57 : static void disconnect_atexit(void);
58 :
59 : static ControlFileData ControlFile_target;
60 : static ControlFileData ControlFile_source;
61 : static ControlFileData ControlFile_source_after;
62 :
63 : static const char *progname;
64 : int WalSegSz;
65 :
66 : /* Configuration options */
67 : char *datadir_target = NULL;
68 : static char *datadir_source = NULL;
69 : static char *connstr_source = NULL;
70 : static char *restore_command = NULL;
71 : static char *config_file = NULL;
72 :
73 : static bool debug = false;
74 : bool showprogress = false;
75 : bool dry_run = false;
76 : bool do_sync = true;
77 : static bool restore_wal = false;
78 : DataDirSyncMethod sync_method = DATA_DIR_SYNC_METHOD_FSYNC;
79 :
80 : /* Target history */
81 : TimeLineHistoryEntry *targetHistory;
82 : int targetNentries;
83 :
84 : /* Progress counters */
85 : uint64 fetch_size;
86 : uint64 fetch_done;
87 :
88 : static PGconn *conn;
89 : static rewind_source *source;
90 :
91 : static void
92 2 : usage(const char *progname)
93 : {
94 2 : printf(_("%s resynchronizes a PostgreSQL cluster with another copy of the cluster.\n\n"), progname);
95 2 : printf(_("Usage:\n %s [OPTION]...\n\n"), progname);
96 2 : printf(_("Options:\n"));
97 2 : printf(_(" -c, --restore-target-wal use \"restore_command\" in target configuration to\n"
98 : " retrieve WAL files from archives\n"));
99 2 : printf(_(" -D, --target-pgdata=DIRECTORY existing data directory to modify\n"));
100 2 : printf(_(" --source-pgdata=DIRECTORY source data directory to synchronize with\n"));
101 2 : printf(_(" --source-server=CONNSTR source server to synchronize with\n"));
102 2 : printf(_(" -n, --dry-run stop before modifying anything\n"));
103 2 : printf(_(" -N, --no-sync do not wait for changes to be written\n"
104 : " safely to disk\n"));
105 2 : printf(_(" -P, --progress write progress messages\n"));
106 2 : printf(_(" -R, --write-recovery-conf write configuration for replication\n"
107 : " (requires --source-server)\n"));
108 2 : printf(_(" --config-file=FILENAME use specified main server configuration\n"
109 : " file when running target cluster\n"));
110 2 : printf(_(" --debug write a lot of debug messages\n"));
111 2 : printf(_(" --no-ensure-shutdown do not automatically fix unclean shutdown\n"));
112 2 : printf(_(" --sync-method=METHOD set method for syncing files to disk\n"));
113 2 : printf(_(" -V, --version output version information, then exit\n"));
114 2 : printf(_(" -?, --help show this help, then exit\n"));
115 2 : printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT);
116 2 : printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL);
117 2 : }
118 :
119 :
120 : int
121 52 : main(int argc, char **argv)
122 : {
123 : static struct option long_options[] = {
124 : {"help", no_argument, NULL, '?'},
125 : {"target-pgdata", required_argument, NULL, 'D'},
126 : {"write-recovery-conf", no_argument, NULL, 'R'},
127 : {"source-pgdata", required_argument, NULL, 1},
128 : {"source-server", required_argument, NULL, 2},
129 : {"no-ensure-shutdown", no_argument, NULL, 4},
130 : {"config-file", required_argument, NULL, 5},
131 : {"version", no_argument, NULL, 'V'},
132 : {"restore-target-wal", no_argument, NULL, 'c'},
133 : {"dry-run", no_argument, NULL, 'n'},
134 : {"no-sync", no_argument, NULL, 'N'},
135 : {"progress", no_argument, NULL, 'P'},
136 : {"debug", no_argument, NULL, 3},
137 : {"sync-method", required_argument, NULL, 6},
138 : {NULL, 0, NULL, 0}
139 : };
140 : int option_index;
141 : int c;
142 : XLogRecPtr divergerec;
143 : int lastcommontliIndex;
144 : XLogRecPtr chkptrec;
145 : TimeLineID chkpttli;
146 : XLogRecPtr chkptredo;
147 : TimeLineID source_tli;
148 : TimeLineID target_tli;
149 : XLogRecPtr target_wal_endrec;
150 : XLogSegNo last_common_segno;
151 : size_t size;
152 : char *buffer;
153 52 : bool no_ensure_shutdown = false;
154 : bool rewind_needed;
155 52 : bool writerecoveryconf = false;
156 : filemap_t *filemap;
157 :
158 52 : pg_logging_init(argv[0]);
159 52 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_rewind"));
160 52 : progname = get_progname(argv[0]);
161 :
162 : /* Process command-line arguments */
163 52 : if (argc > 1)
164 : {
165 52 : if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
166 : {
167 2 : usage(progname);
168 2 : exit(0);
169 : }
170 50 : if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
171 : {
172 2 : puts("pg_rewind (PostgreSQL) " PG_VERSION);
173 2 : exit(0);
174 : }
175 : }
176 :
177 262 : while ((c = getopt_long(argc, argv, "cD:nNPR", long_options, &option_index)) != -1)
178 : {
179 216 : switch (c)
180 : {
181 2 : case 'c':
182 2 : restore_wal = true;
183 2 : break;
184 :
185 0 : case 'P':
186 0 : showprogress = true;
187 0 : break;
188 :
189 2 : case 'n':
190 2 : dry_run = true;
191 2 : break;
192 :
193 36 : case 'N':
194 36 : do_sync = false;
195 36 : break;
196 :
197 12 : case 'R':
198 12 : writerecoveryconf = true;
199 12 : break;
200 :
201 44 : case 3:
202 44 : debug = true;
203 44 : pg_logging_increase_verbosity();
204 44 : break;
205 :
206 46 : case 'D': /* -D or --target-pgdata */
207 46 : datadir_target = pg_strdup(optarg);
208 46 : break;
209 :
210 32 : case 1: /* --source-pgdata */
211 32 : datadir_source = pg_strdup(optarg);
212 32 : break;
213 :
214 14 : case 2: /* --source-server */
215 14 : connstr_source = pg_strdup(optarg);
216 14 : break;
217 :
218 6 : case 4:
219 6 : no_ensure_shutdown = true;
220 6 : break;
221 :
222 20 : case 5:
223 20 : config_file = pg_strdup(optarg);
224 20 : break;
225 :
226 0 : case 6:
227 0 : if (!parse_sync_method(optarg, &sync_method))
228 0 : exit(1);
229 0 : break;
230 :
231 2 : default:
232 : /* getopt_long already emitted a complaint */
233 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
234 2 : exit(1);
235 : }
236 : }
237 :
238 46 : if (datadir_source == NULL && connstr_source == NULL)
239 : {
240 2 : pg_log_error("no source specified (--source-pgdata or --source-server)");
241 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
242 2 : exit(1);
243 : }
244 :
245 44 : if (datadir_source != NULL && connstr_source != NULL)
246 : {
247 2 : pg_log_error("only one of --source-pgdata or --source-server can be specified");
248 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
249 2 : exit(1);
250 : }
251 :
252 42 : if (datadir_target == NULL)
253 : {
254 0 : pg_log_error("no target data directory specified (--target-pgdata)");
255 0 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
256 0 : exit(1);
257 : }
258 :
259 42 : if (writerecoveryconf && connstr_source == NULL)
260 : {
261 2 : pg_log_error("no source server information (--source-server) specified for --write-recovery-conf");
262 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
263 2 : exit(1);
264 : }
265 :
266 40 : if (optind < argc)
267 : {
268 2 : pg_log_error("too many command-line arguments (first is \"%s\")",
269 : argv[optind]);
270 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
271 2 : exit(1);
272 : }
273 :
274 : /*
275 : * Don't allow pg_rewind to be run as root, to avoid overwriting the
276 : * ownership of files in the data directory. We need only check for root
277 : * -- any other user won't have sufficient permissions to modify files in
278 : * the data directory.
279 : */
280 : #ifndef WIN32
281 38 : if (geteuid() == 0)
282 : {
283 0 : pg_log_error("cannot be executed by \"root\"");
284 0 : pg_log_error_hint("You must run %s as the PostgreSQL superuser.",
285 : progname);
286 0 : exit(1);
287 : }
288 : #endif
289 :
290 38 : get_restricted_token();
291 :
292 : /* Set mask based on PGDATA permissions */
293 38 : if (!GetDataDirectoryCreatePerm(datadir_target))
294 0 : pg_fatal("could not read permissions of directory \"%s\": %m",
295 : datadir_target);
296 :
297 38 : umask(pg_mode_mask);
298 :
299 38 : getRestoreCommand(argv[0]);
300 :
301 38 : atexit(disconnect_atexit);
302 :
303 : /*
304 : * Ok, we have all the options and we're ready to start. First, connect to
305 : * remote server.
306 : */
307 38 : if (connstr_source)
308 : {
309 12 : conn = PQconnectdb(connstr_source);
310 :
311 12 : if (PQstatus(conn) == CONNECTION_BAD)
312 0 : pg_fatal("%s", PQerrorMessage(conn));
313 :
314 12 : if (showprogress)
315 0 : pg_log_info("connected to server");
316 :
317 12 : source = init_libpq_source(conn);
318 : }
319 : else
320 26 : source = init_local_source(datadir_source);
321 :
322 : /*
323 : * Check the status of the target instance.
324 : *
325 : * If the target instance was not cleanly shut down, start and stop the
326 : * target cluster once in single-user mode to enforce recovery to finish,
327 : * ensuring that the cluster can be used by pg_rewind. Note that if
328 : * no_ensure_shutdown is specified, pg_rewind ignores this step, and users
329 : * need to make sure by themselves that the target cluster is in a clean
330 : * state.
331 : */
332 38 : buffer = slurpFile(datadir_target, XLOG_CONTROL_FILE, &size);
333 38 : digestControlFile(&ControlFile_target, buffer, size);
334 38 : pg_free(buffer);
335 :
336 38 : if (!no_ensure_shutdown &&
337 32 : ControlFile_target.state != DB_SHUTDOWNED &&
338 22 : ControlFile_target.state != DB_SHUTDOWNED_IN_RECOVERY)
339 : {
340 20 : ensureCleanShutdown(argv[0]);
341 :
342 18 : buffer = slurpFile(datadir_target, XLOG_CONTROL_FILE, &size);
343 18 : digestControlFile(&ControlFile_target, buffer, size);
344 18 : pg_free(buffer);
345 : }
346 :
347 36 : buffer = source->fetch_file(source, XLOG_CONTROL_FILE, &size);
348 36 : digestControlFile(&ControlFile_source, buffer, size);
349 36 : pg_free(buffer);
350 :
351 36 : sanityChecks();
352 :
353 : /*
354 : * Usually, the TLI can be found in the latest checkpoint record. But if
355 : * the source server is just being promoted (or it's a standby that's
356 : * following a primary that's just being promoted), and the checkpoint
357 : * requested by the promotion hasn't completed yet, the latest timeline is
358 : * in minRecoveryPoint. So we check which is later, the TLI of the
359 : * minRecoveryPoint or the latest checkpoint.
360 : */
361 32 : source_tli = Max(ControlFile_source.minRecoveryPointTLI,
362 : ControlFile_source.checkPointCopy.ThisTimeLineID);
363 :
364 : /* Similarly for the target. */
365 32 : target_tli = Max(ControlFile_target.minRecoveryPointTLI,
366 : ControlFile_target.checkPointCopy.ThisTimeLineID);
367 :
368 : /*
369 : * Find the common ancestor timeline between the clusters.
370 : *
371 : * If both clusters are already on the same timeline, there's nothing to
372 : * do.
373 : */
374 32 : if (target_tli == source_tli)
375 : {
376 2 : pg_log_info("source and target cluster are on the same timeline");
377 2 : rewind_needed = false;
378 2 : target_wal_endrec = 0;
379 : }
380 : else
381 : {
382 : XLogRecPtr chkptendrec;
383 : TimeLineHistoryEntry *sourceHistory;
384 : int sourceNentries;
385 :
386 : /*
387 : * Retrieve timelines for both source and target, and find the point
388 : * where they diverged.
389 : */
390 30 : sourceHistory = getTimelineHistory(source_tli, true, &sourceNentries);
391 30 : targetHistory = getTimelineHistory(target_tli, false, &targetNentries);
392 :
393 30 : findCommonAncestorTimeline(sourceHistory, sourceNentries,
394 : targetHistory, targetNentries,
395 : &divergerec, &lastcommontliIndex);
396 :
397 30 : pg_log_info("servers diverged at WAL location %X/%08X on timeline %u",
398 : LSN_FORMAT_ARGS(divergerec),
399 : targetHistory[lastcommontliIndex].tli);
400 :
401 : /*
402 : * Convert the divergence LSN to a segment number, that will be used
403 : * to decide how WAL segments should be processed.
404 : */
405 30 : XLByteToSeg(divergerec, last_common_segno, ControlFile_target.xlog_seg_size);
406 :
407 : /*
408 : * Don't need the source history anymore. The target history is still
409 : * needed by the routines in parsexlog.c, when we read the target WAL.
410 : */
411 30 : pfree(sourceHistory);
412 :
413 :
414 : /*
415 : * Determine the end-of-WAL on the target.
416 : *
417 : * The WAL ends at the last shutdown checkpoint, or at
418 : * minRecoveryPoint if it was a standby. (If we supported rewinding a
419 : * server that was not shut down cleanly, we would need to replay
420 : * until we reach the first invalid record, like crash recovery does.)
421 : */
422 :
423 : /* read the checkpoint record on the target to see where it ends. */
424 30 : chkptendrec = readOneRecord(datadir_target,
425 : ControlFile_target.checkPoint,
426 : targetNentries - 1,
427 : restore_command);
428 :
429 30 : if (ControlFile_target.minRecoveryPoint > chkptendrec)
430 : {
431 2 : target_wal_endrec = ControlFile_target.minRecoveryPoint;
432 : }
433 : else
434 : {
435 28 : target_wal_endrec = chkptendrec;
436 : }
437 :
438 : /*
439 : * Check for the possibility that the target is in fact a direct
440 : * ancestor of the source. In that case, there is no divergent history
441 : * in the target that needs rewinding.
442 : */
443 30 : if (target_wal_endrec > divergerec)
444 : {
445 30 : rewind_needed = true;
446 : }
447 : else
448 : {
449 : /* the last common checkpoint record must be part of target WAL */
450 : Assert(target_wal_endrec == divergerec);
451 :
452 0 : rewind_needed = false;
453 : }
454 : }
455 :
456 32 : if (!rewind_needed)
457 : {
458 2 : pg_log_info("no rewind required");
459 2 : if (writerecoveryconf && !dry_run)
460 0 : WriteRecoveryConfig(conn, datadir_target,
461 : GenerateRecoveryConfig(conn, NULL,
462 : GetDbnameFromConnectionOptions(connstr_source)));
463 2 : exit(0);
464 : }
465 :
466 : /* Initialize hashtable that tracks WAL files protected from removal */
467 30 : keepwal_init();
468 :
469 30 : findLastCheckpoint(datadir_target, divergerec, lastcommontliIndex,
470 : &chkptrec, &chkpttli, &chkptredo, restore_command);
471 30 : pg_log_info("rewinding from last common checkpoint at %X/%08X on timeline %u",
472 : LSN_FORMAT_ARGS(chkptrec), chkpttli);
473 :
474 : /* Initialize the hash table to track the status of each file */
475 30 : filehash_init();
476 :
477 : /*
478 : * Collect information about all files in the both data directories.
479 : */
480 30 : if (showprogress)
481 0 : pg_log_info("reading source file list");
482 30 : source->traverse_files(source, &process_source_file);
483 :
484 30 : if (showprogress)
485 0 : pg_log_info("reading target file list");
486 30 : traverse_datadir(datadir_target, &process_target_file);
487 :
488 : /*
489 : * Read the target WAL from last checkpoint before the point of fork, to
490 : * extract all the pages that were modified on the target cluster after
491 : * the fork.
492 : */
493 30 : if (showprogress)
494 0 : pg_log_info("reading WAL in target");
495 30 : extractPageMap(datadir_target, chkptrec, lastcommontliIndex,
496 : target_wal_endrec, restore_command);
497 :
498 : /*
499 : * We have collected all information we need from both systems. Decide
500 : * what to do with each file.
501 : */
502 30 : filemap = decide_file_actions(last_common_segno);
503 30 : if (showprogress)
504 0 : calculate_totals(filemap);
505 :
506 : /* this is too verbose even for verbose mode */
507 30 : if (debug)
508 30 : print_filemap(filemap);
509 :
510 : /*
511 : * Ok, we're ready to start copying things over.
512 : */
513 30 : if (showprogress)
514 : {
515 0 : pg_log_info("need to copy %lu MB (total source directory size is %lu MB)",
516 : (unsigned long) (filemap->fetch_size / (1024 * 1024)),
517 : (unsigned long) (filemap->total_size / (1024 * 1024)));
518 :
519 0 : fetch_size = filemap->fetch_size;
520 0 : fetch_done = 0;
521 : }
522 :
523 : /*
524 : * We have now collected all the information we need from both systems,
525 : * and we are ready to start modifying the target directory.
526 : *
527 : * This is the point of no return. Once we start copying things, there is
528 : * no turning back!
529 : */
530 30 : perform_rewind(filemap, source, chkptrec, chkpttli, chkptredo);
531 :
532 28 : if (showprogress)
533 0 : pg_log_info("syncing target data directory");
534 28 : sync_target_dir();
535 :
536 : /* Also update the standby configuration, if requested. */
537 28 : if (writerecoveryconf && !dry_run)
538 10 : WriteRecoveryConfig(conn, datadir_target,
539 : GenerateRecoveryConfig(conn, NULL,
540 : GetDbnameFromConnectionOptions(connstr_source)));
541 :
542 : /* don't need the source connection anymore */
543 28 : source->destroy(source);
544 28 : if (conn)
545 : {
546 12 : PQfinish(conn);
547 12 : conn = NULL;
548 : }
549 :
550 28 : pg_log_info("Done!");
551 :
552 28 : return 0;
553 : }
554 :
555 : /*
556 : * Perform the rewind.
557 : *
558 : * We have already collected all the information we need from the
559 : * target and the source.
560 : */
561 : static void
562 30 : perform_rewind(filemap_t *filemap, rewind_source *source,
563 : XLogRecPtr chkptrec,
564 : TimeLineID chkpttli,
565 : XLogRecPtr chkptredo)
566 : {
567 : XLogRecPtr endrec;
568 : TimeLineID endtli;
569 : ControlFileData ControlFile_new;
570 : size_t size;
571 : char *buffer;
572 :
573 : /*
574 : * Execute the actions in the file map, fetching data from the source
575 : * system as needed.
576 : */
577 33882 : for (int i = 0; i < filemap->nentries; i++)
578 : {
579 33854 : file_entry_t *entry = filemap->entries[i];
580 :
581 : /*
582 : * If this is a relation file, copy the modified blocks.
583 : *
584 : * This is in addition to any other changes.
585 : */
586 33854 : if (entry->target_pages_to_overwrite.bitmapsize > 0)
587 : {
588 : datapagemap_iterator_t *iter;
589 : BlockNumber blkno;
590 : off_t offset;
591 :
592 842 : iter = datapagemap_iterate(&entry->target_pages_to_overwrite);
593 4210 : while (datapagemap_next(iter, &blkno))
594 : {
595 3368 : offset = blkno * BLCKSZ;
596 3368 : source->queue_fetch_range(source, entry->path, offset, BLCKSZ);
597 : }
598 842 : pg_free(iter);
599 : }
600 :
601 33854 : switch (entry->action)
602 : {
603 23026 : case FILE_ACTION_NONE:
604 : /* nothing else to do */
605 23026 : break;
606 :
607 9350 : case FILE_ACTION_COPY:
608 9350 : source->queue_fetch_file(source, entry->path, entry->source_size);
609 9348 : break;
610 :
611 8 : case FILE_ACTION_TRUNCATE:
612 8 : truncate_target_file(entry->path, entry->source_size);
613 8 : break;
614 :
615 10 : case FILE_ACTION_COPY_TAIL:
616 10 : source->queue_fetch_range(source, entry->path,
617 10 : entry->target_size,
618 10 : entry->source_size - entry->target_size);
619 10 : break;
620 :
621 1442 : case FILE_ACTION_REMOVE:
622 1442 : remove_target(entry);
623 1442 : break;
624 :
625 18 : case FILE_ACTION_CREATE:
626 18 : create_target(entry);
627 18 : break;
628 :
629 0 : case FILE_ACTION_UNDECIDED:
630 0 : pg_fatal("no action decided for file \"%s\"", entry->path);
631 : break;
632 : }
633 : }
634 :
635 : /* Complete any remaining range-fetches that we queued up above. */
636 28 : source->finish_fetch(source);
637 :
638 28 : close_target_file();
639 :
640 28 : progress_report(true);
641 :
642 : /*
643 : * Fetch the control file from the source last. This ensures that the
644 : * minRecoveryPoint is up-to-date.
645 : */
646 28 : buffer = source->fetch_file(source, XLOG_CONTROL_FILE, &size);
647 28 : digestControlFile(&ControlFile_source_after, buffer, size);
648 28 : pg_free(buffer);
649 :
650 : /*
651 : * Sanity check: If the source is a local system, the control file should
652 : * not have changed since we started.
653 : *
654 : * XXX: We assume it hasn't been modified, but actually, what could go
655 : * wrong? The logic handles a libpq source that's modified concurrently,
656 : * why not a local datadir?
657 : */
658 28 : if (datadir_source &&
659 16 : memcmp(&ControlFile_source, &ControlFile_source_after,
660 : sizeof(ControlFileData)) != 0)
661 : {
662 0 : pg_fatal("source system was modified while pg_rewind was running");
663 : }
664 :
665 28 : if (showprogress)
666 0 : pg_log_info("creating backup label and updating control file");
667 :
668 : /*
669 : * Create a backup label file, to tell the target where to begin the WAL
670 : * replay. Normally, from the last common checkpoint between the source
671 : * and the target. But if the source is a standby server, it's possible
672 : * that the last common checkpoint is *after* the standby's restartpoint.
673 : * That implies that the source server has applied the checkpoint record,
674 : * but hasn't performed a corresponding restartpoint yet. Make sure we
675 : * start at the restartpoint's redo point in that case.
676 : *
677 : * Use the old version of the source's control file for this. The server
678 : * might have finished the restartpoint after we started copying files,
679 : * but we must begin from the redo point at the time that started copying.
680 : */
681 28 : if (ControlFile_source.checkPointCopy.redo < chkptredo)
682 : {
683 4 : chkptredo = ControlFile_source.checkPointCopy.redo;
684 4 : chkpttli = ControlFile_source.checkPointCopy.ThisTimeLineID;
685 4 : chkptrec = ControlFile_source.checkPoint;
686 : }
687 28 : createBackupLabel(chkptredo, chkpttli, chkptrec);
688 :
689 : /*
690 : * Update control file of target, to tell the target how far it must
691 : * replay the WAL (minRecoveryPoint).
692 : */
693 28 : if (connstr_source)
694 : {
695 : /*
696 : * The source is a live server. Like in an online backup, it's
697 : * important that we recover all the WAL that was generated while we
698 : * were copying files.
699 : */
700 12 : if (ControlFile_source_after.state == DB_IN_ARCHIVE_RECOVERY)
701 : {
702 : /*
703 : * Source is a standby server. We must replay to its
704 : * minRecoveryPoint.
705 : */
706 2 : endrec = ControlFile_source_after.minRecoveryPoint;
707 2 : endtli = ControlFile_source_after.minRecoveryPointTLI;
708 : }
709 : else
710 : {
711 : /*
712 : * Source is a production, non-standby, server. We must replay to
713 : * the last WAL insert location.
714 : */
715 10 : if (ControlFile_source_after.state != DB_IN_PRODUCTION)
716 0 : pg_fatal("source system was in unexpected state at end of rewind");
717 :
718 10 : endrec = source->get_current_wal_insert_lsn(source);
719 10 : endtli = Max(ControlFile_source_after.checkPointCopy.ThisTimeLineID,
720 : ControlFile_source_after.minRecoveryPointTLI);
721 : }
722 : }
723 : else
724 : {
725 : /*
726 : * Source is a local data directory. It should've shut down cleanly,
727 : * and we must replay to the latest shutdown checkpoint.
728 : */
729 16 : endrec = ControlFile_source_after.checkPoint;
730 16 : endtli = ControlFile_source_after.checkPointCopy.ThisTimeLineID;
731 : }
732 :
733 28 : memcpy(&ControlFile_new, &ControlFile_source_after, sizeof(ControlFileData));
734 28 : ControlFile_new.minRecoveryPoint = endrec;
735 28 : ControlFile_new.minRecoveryPointTLI = endtli;
736 28 : ControlFile_new.state = DB_IN_ARCHIVE_RECOVERY;
737 28 : if (!dry_run)
738 26 : update_controlfile(datadir_target, &ControlFile_new, do_sync);
739 28 : }
740 :
741 : static void
742 36 : sanityChecks(void)
743 : {
744 : /* TODO Check that there's no backup_label in either cluster */
745 :
746 : /* Check system_identifier match */
747 36 : if (ControlFile_target.system_identifier != ControlFile_source.system_identifier)
748 0 : pg_fatal("source and target clusters are from different systems");
749 :
750 : /* check version */
751 36 : if (ControlFile_target.pg_control_version != PG_CONTROL_VERSION ||
752 36 : ControlFile_source.pg_control_version != PG_CONTROL_VERSION ||
753 36 : ControlFile_target.catalog_version_no != CATALOG_VERSION_NO ||
754 36 : ControlFile_source.catalog_version_no != CATALOG_VERSION_NO)
755 : {
756 0 : pg_fatal("clusters are not compatible with this version of pg_rewind");
757 : }
758 :
759 : /*
760 : * Target cluster need to use checksums or hint bit wal-logging, this to
761 : * prevent from data corruption that could occur because of hint bits.
762 : */
763 36 : if (ControlFile_target.data_checksum_version != PG_DATA_CHECKSUM_VERSION &&
764 0 : !ControlFile_target.wal_log_hints)
765 : {
766 0 : pg_fatal("target server needs to use either data checksums or \"wal_log_hints = on\"");
767 : }
768 :
769 : /*
770 : * Target cluster better not be running. This doesn't guard against
771 : * someone starting the cluster concurrently. Also, this is probably more
772 : * strict than necessary; it's OK if the target node was not shut down
773 : * cleanly, as long as it isn't running at the moment.
774 : */
775 36 : if (ControlFile_target.state != DB_SHUTDOWNED &&
776 4 : ControlFile_target.state != DB_SHUTDOWNED_IN_RECOVERY)
777 2 : pg_fatal("target server must be shut down cleanly");
778 :
779 : /*
780 : * When the source is a data directory, also require that the source
781 : * server is shut down. There isn't any very strong reason for this
782 : * limitation, but better safe than sorry.
783 : */
784 34 : if (datadir_source &&
785 22 : ControlFile_source.state != DB_SHUTDOWNED &&
786 4 : ControlFile_source.state != DB_SHUTDOWNED_IN_RECOVERY)
787 2 : pg_fatal("source data directory must be shut down cleanly");
788 32 : }
789 :
790 : /*
791 : * Print a progress report based on the fetch_size and fetch_done variables.
792 : *
793 : * Progress report is written at maximum once per second, except that the
794 : * last progress report is always printed.
795 : *
796 : * If finished is set to true, this is the last progress report. The cursor
797 : * is moved to the next line.
798 : */
799 : void
800 100216 : progress_report(bool finished)
801 : {
802 : static pg_time_t last_progress_report = 0;
803 : int percent;
804 : char fetch_done_str[32];
805 : char fetch_size_str[32];
806 : pg_time_t now;
807 :
808 100216 : if (!showprogress)
809 100216 : return;
810 :
811 0 : now = time(NULL);
812 0 : if (now == last_progress_report && !finished)
813 0 : return; /* Max once per second */
814 :
815 0 : last_progress_report = now;
816 0 : percent = fetch_size ? (int) ((fetch_done) * 100 / fetch_size) : 0;
817 :
818 : /*
819 : * Avoid overflowing past 100% or the full size. This may make the total
820 : * size number change as we approach the end of the backup (the estimate
821 : * will always be wrong if WAL is included), but that's better than having
822 : * the done column be bigger than the total.
823 : */
824 0 : if (percent > 100)
825 0 : percent = 100;
826 0 : if (fetch_done > fetch_size)
827 0 : fetch_size = fetch_done;
828 :
829 0 : snprintf(fetch_done_str, sizeof(fetch_done_str), UINT64_FORMAT,
830 : fetch_done / 1024);
831 0 : snprintf(fetch_size_str, sizeof(fetch_size_str), UINT64_FORMAT,
832 : fetch_size / 1024);
833 :
834 0 : fprintf(stderr, _("%*s/%s kB (%d%%) copied"),
835 0 : (int) strlen(fetch_size_str), fetch_done_str, fetch_size_str,
836 : percent);
837 :
838 : /*
839 : * Stay on the same line if reporting to a terminal and we're not done
840 : * yet.
841 : */
842 0 : fputc((!finished && isatty(fileno(stderr))) ? '\r' : '\n', stderr);
843 : }
844 :
845 : /*
846 : * Find minimum from two WAL locations assuming InvalidXLogRecPtr means
847 : * infinity as src/include/access/timeline.h states. This routine should
848 : * be used only when comparing WAL locations related to history files.
849 : */
850 : static XLogRecPtr
851 30 : MinXLogRecPtr(XLogRecPtr a, XLogRecPtr b)
852 : {
853 30 : if (!XLogRecPtrIsValid(a))
854 2 : return b;
855 28 : else if (!XLogRecPtrIsValid(b))
856 28 : return a;
857 : else
858 0 : return Min(a, b);
859 : }
860 :
861 : /*
862 : * Retrieve timeline history for the source or target system.
863 : */
864 : static TimeLineHistoryEntry *
865 60 : getTimelineHistory(TimeLineID tli, bool is_source, int *nentries)
866 : {
867 : TimeLineHistoryEntry *history;
868 :
869 : /*
870 : * Timeline 1 does not have a history file, so there is no need to check
871 : * and fake an entry with infinite start and end positions.
872 : */
873 60 : if (tli == 1)
874 : {
875 28 : history = (TimeLineHistoryEntry *) pg_malloc(sizeof(TimeLineHistoryEntry));
876 28 : history->tli = tli;
877 28 : history->begin = history->end = InvalidXLogRecPtr;
878 28 : *nentries = 1;
879 : }
880 : else
881 : {
882 : char path[MAXPGPATH];
883 : char *histfile;
884 :
885 32 : TLHistoryFilePath(path, tli);
886 :
887 : /* Get history file from appropriate source */
888 32 : if (is_source)
889 28 : histfile = source->fetch_file(source, path, NULL);
890 : else
891 4 : histfile = slurpFile(datadir_target, path, NULL);
892 :
893 32 : history = rewind_parseTimeLineHistory(histfile, tli, nentries);
894 32 : pg_free(histfile);
895 : }
896 :
897 : /* In debugging mode, print what we read */
898 60 : if (debug)
899 : {
900 : int i;
901 :
902 60 : if (is_source)
903 30 : pg_log_debug("Source timeline history:");
904 : else
905 30 : pg_log_debug("Target timeline history:");
906 :
907 154 : for (i = 0; i < *nentries; i++)
908 : {
909 : TimeLineHistoryEntry *entry;
910 :
911 94 : entry = &history[i];
912 94 : pg_log_debug("%u: %X/%08X - %X/%08X", entry->tli,
913 : LSN_FORMAT_ARGS(entry->begin),
914 : LSN_FORMAT_ARGS(entry->end));
915 : }
916 : }
917 :
918 60 : return history;
919 : }
920 :
921 : /*
922 : * Determine the TLI of the last common timeline in the timeline history of
923 : * two clusters. *tliIndex is set to the index of last common timeline in
924 : * the arrays, and *recptr is set to the position where the timeline history
925 : * diverged (ie. the first WAL record that's not the same in both clusters).
926 : */
927 : static void
928 30 : findCommonAncestorTimeline(TimeLineHistoryEntry *a_history, int a_nentries,
929 : TimeLineHistoryEntry *b_history, int b_nentries,
930 : XLogRecPtr *recptr, int *tliIndex)
931 : {
932 : int i,
933 : n;
934 :
935 : /*
936 : * Trace the history forward, until we hit the timeline diverge. It may
937 : * still be possible that the source and target nodes used the same
938 : * timeline number in their history but with different start position
939 : * depending on the history files that each node has fetched in previous
940 : * recovery processes. Hence check the start position of the new timeline
941 : * as well and move down by one extra timeline entry if they do not match.
942 : */
943 30 : n = Min(a_nentries, b_nentries);
944 62 : for (i = 0; i < n; i++)
945 : {
946 32 : if (a_history[i].tli != b_history[i].tli ||
947 32 : a_history[i].begin != b_history[i].begin)
948 : break;
949 : }
950 :
951 30 : if (i > 0)
952 : {
953 30 : i--;
954 30 : *recptr = MinXLogRecPtr(a_history[i].end, b_history[i].end);
955 30 : *tliIndex = i;
956 30 : return;
957 : }
958 : else
959 : {
960 0 : pg_fatal("could not find common ancestor of the source and target cluster's timelines");
961 : }
962 : }
963 :
964 :
965 : /*
966 : * Create a backup_label file that forces recovery to begin at the last common
967 : * checkpoint.
968 : */
969 : static void
970 28 : createBackupLabel(XLogRecPtr startpoint, TimeLineID starttli, XLogRecPtr checkpointloc)
971 : {
972 : XLogSegNo startsegno;
973 : time_t stamp_time;
974 : char strfbuf[128];
975 : char xlogfilename[MAXFNAMELEN];
976 : struct tm *tmp;
977 : char buf[1000];
978 : int len;
979 :
980 28 : XLByteToSeg(startpoint, startsegno, WalSegSz);
981 28 : XLogFileName(xlogfilename, starttli, startsegno, WalSegSz);
982 :
983 : /*
984 : * Construct backup label file
985 : */
986 28 : stamp_time = time(NULL);
987 28 : tmp = localtime(&stamp_time);
988 28 : strftime(strfbuf, sizeof(strfbuf), "%Y-%m-%d %H:%M:%S %Z", tmp);
989 :
990 28 : len = snprintf(buf, sizeof(buf),
991 : "START WAL LOCATION: %X/%08X (file %s)\n"
992 : "CHECKPOINT LOCATION: %X/%08X\n"
993 : "BACKUP METHOD: pg_rewind\n"
994 : "BACKUP FROM: standby\n"
995 : "START TIME: %s\n",
996 : /* omit LABEL: line */
997 28 : LSN_FORMAT_ARGS(startpoint), xlogfilename,
998 28 : LSN_FORMAT_ARGS(checkpointloc),
999 : strfbuf);
1000 28 : if (len >= sizeof(buf))
1001 0 : pg_fatal("backup label buffer too small"); /* shouldn't happen */
1002 :
1003 : /* TODO: move old file out of the way, if any. */
1004 28 : open_target_file("backup_label", true); /* BACKUP_LABEL_FILE */
1005 28 : write_target_range(buf, 0, len);
1006 28 : close_target_file();
1007 28 : }
1008 :
1009 : /*
1010 : * Check CRC of control file
1011 : */
1012 : static void
1013 120 : checkControlFile(ControlFileData *ControlFile)
1014 : {
1015 : pg_crc32c crc;
1016 :
1017 : /* Calculate CRC */
1018 120 : INIT_CRC32C(crc);
1019 120 : COMP_CRC32C(crc, ControlFile, offsetof(ControlFileData, crc));
1020 120 : FIN_CRC32C(crc);
1021 :
1022 : /* And simply compare it */
1023 120 : if (!EQ_CRC32C(crc, ControlFile->crc))
1024 0 : pg_fatal("unexpected control file CRC");
1025 120 : }
1026 :
1027 : /*
1028 : * Verify control file contents in the buffer 'content', and copy it to
1029 : * *ControlFile.
1030 : */
1031 : static void
1032 120 : digestControlFile(ControlFileData *ControlFile, const char *content,
1033 : size_t size)
1034 : {
1035 120 : if (size != PG_CONTROL_FILE_SIZE)
1036 0 : pg_fatal("unexpected control file size %d, expected %d",
1037 : (int) size, PG_CONTROL_FILE_SIZE);
1038 :
1039 120 : memcpy(ControlFile, content, sizeof(ControlFileData));
1040 :
1041 : /* set and validate WalSegSz */
1042 120 : WalSegSz = ControlFile->xlog_seg_size;
1043 :
1044 120 : if (!IsValidWalSegSize(WalSegSz))
1045 : {
1046 0 : pg_log_error(ngettext("invalid WAL segment size in control file (%d byte)",
1047 : "invalid WAL segment size in control file (%d bytes)",
1048 : WalSegSz),
1049 : WalSegSz);
1050 0 : pg_log_error_detail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
1051 0 : exit(1);
1052 : }
1053 :
1054 : /* Additional checks on control file */
1055 120 : checkControlFile(ControlFile);
1056 120 : }
1057 :
1058 : /*
1059 : * Get value of GUC parameter restore_command from the target cluster.
1060 : *
1061 : * This uses a logic based on "postgres -C" to get the value from the
1062 : * cluster.
1063 : */
1064 : static void
1065 38 : getRestoreCommand(const char *argv0)
1066 : {
1067 : int rc;
1068 : char postgres_exec_path[MAXPGPATH];
1069 : PQExpBuffer postgres_cmd;
1070 :
1071 38 : if (!restore_wal)
1072 36 : return;
1073 :
1074 : /* find postgres executable */
1075 2 : rc = find_other_exec(argv0, "postgres",
1076 : PG_BACKEND_VERSIONSTR,
1077 : postgres_exec_path);
1078 :
1079 2 : if (rc < 0)
1080 : {
1081 : char full_path[MAXPGPATH];
1082 :
1083 0 : if (find_my_exec(argv0, full_path) < 0)
1084 0 : strlcpy(full_path, progname, sizeof(full_path));
1085 :
1086 0 : if (rc == -1)
1087 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
1088 : "postgres", progname, full_path);
1089 : else
1090 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
1091 : "postgres", full_path, progname);
1092 : }
1093 :
1094 : /*
1095 : * Build a command able to retrieve the value of GUC parameter
1096 : * restore_command, if set.
1097 : */
1098 2 : postgres_cmd = createPQExpBuffer();
1099 :
1100 : /* path to postgres, properly quoted */
1101 2 : appendShellString(postgres_cmd, postgres_exec_path);
1102 :
1103 : /* add -D switch, with properly quoted data directory */
1104 2 : appendPQExpBufferStr(postgres_cmd, " -D ");
1105 2 : appendShellString(postgres_cmd, datadir_target);
1106 :
1107 : /* add custom configuration file only if requested */
1108 2 : if (config_file != NULL)
1109 : {
1110 2 : appendPQExpBufferStr(postgres_cmd, " -c config_file=");
1111 2 : appendShellString(postgres_cmd, config_file);
1112 : }
1113 :
1114 : /* add -C switch, for restore_command */
1115 2 : appendPQExpBufferStr(postgres_cmd, " -C restore_command");
1116 :
1117 2 : restore_command = pipe_read_line(postgres_cmd->data);
1118 2 : if (restore_command == NULL)
1119 0 : pg_fatal("could not read \"restore_command\" from target cluster");
1120 :
1121 2 : (void) pg_strip_crlf(restore_command);
1122 :
1123 2 : if (strcmp(restore_command, "") == 0)
1124 0 : pg_fatal("\"restore_command\" is not set in the target cluster");
1125 :
1126 2 : pg_log_debug("using for rewind \"restore_command = \'%s\'\"",
1127 : restore_command);
1128 :
1129 2 : destroyPQExpBuffer(postgres_cmd);
1130 : }
1131 :
1132 :
1133 : /*
1134 : * Ensure clean shutdown of target instance by launching single-user mode
1135 : * postgres to do crash recovery.
1136 : */
1137 : static void
1138 20 : ensureCleanShutdown(const char *argv0)
1139 : {
1140 : int ret;
1141 : char exec_path[MAXPGPATH];
1142 : PQExpBuffer postgres_cmd;
1143 :
1144 : /* locate postgres binary */
1145 20 : if ((ret = find_other_exec(argv0, "postgres",
1146 : PG_BACKEND_VERSIONSTR,
1147 : exec_path)) < 0)
1148 : {
1149 : char full_path[MAXPGPATH];
1150 :
1151 0 : if (find_my_exec(argv0, full_path) < 0)
1152 0 : strlcpy(full_path, progname, sizeof(full_path));
1153 :
1154 0 : if (ret == -1)
1155 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
1156 : "postgres", progname, full_path);
1157 : else
1158 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
1159 : "postgres", full_path, progname);
1160 : }
1161 :
1162 20 : pg_log_info("executing \"%s\" for target server to complete crash recovery",
1163 : exec_path);
1164 :
1165 : /*
1166 : * Skip processing if requested, but only after ensuring presence of
1167 : * postgres.
1168 : */
1169 20 : if (dry_run)
1170 0 : return;
1171 :
1172 : /*
1173 : * Finally run postgres in single-user mode. There is no need to use
1174 : * fsync here. This makes the recovery faster, and the target data folder
1175 : * is synced at the end anyway.
1176 : */
1177 20 : postgres_cmd = createPQExpBuffer();
1178 :
1179 : /* path to postgres, properly quoted */
1180 20 : appendShellString(postgres_cmd, exec_path);
1181 :
1182 : /* add set of options with properly quoted data directory */
1183 20 : appendPQExpBufferStr(postgres_cmd, " --single -F -D ");
1184 20 : appendShellString(postgres_cmd, datadir_target);
1185 :
1186 : /* add custom configuration file only if requested */
1187 20 : if (config_file != NULL)
1188 : {
1189 18 : appendPQExpBufferStr(postgres_cmd, " -c config_file=");
1190 18 : appendShellString(postgres_cmd, config_file);
1191 : }
1192 :
1193 : /* finish with the database name, and a properly quoted redirection */
1194 20 : appendPQExpBufferStr(postgres_cmd, " template1 < ");
1195 20 : appendShellString(postgres_cmd, DEVNULL);
1196 :
1197 20 : fflush(NULL);
1198 20 : if (system(postgres_cmd->data) != 0)
1199 : {
1200 2 : pg_log_error("postgres single-user mode in target cluster failed");
1201 2 : pg_log_error_detail("Command was: %s", postgres_cmd->data);
1202 2 : exit(1);
1203 : }
1204 :
1205 18 : destroyPQExpBuffer(postgres_cmd);
1206 : }
1207 :
1208 : static void
1209 38 : disconnect_atexit(void)
1210 : {
1211 38 : if (conn != NULL)
1212 0 : PQfinish(conn);
1213 38 : }
|
