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) 1996-2024, 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 50 : 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 : size_t size;
151 : char *buffer;
152 50 : bool no_ensure_shutdown = false;
153 : bool rewind_needed;
154 50 : bool writerecoveryconf = false;
155 : filemap_t *filemap;
156 :
157 50 : pg_logging_init(argv[0]);
158 50 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_rewind"));
159 50 : progname = get_progname(argv[0]);
160 :
161 : /* Process command-line arguments */
162 50 : if (argc > 1)
163 : {
164 50 : if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
165 : {
166 2 : usage(progname);
167 2 : exit(0);
168 : }
169 48 : if (strcmp(argv[1], "--version") == 0 || strcmp(argv[1], "-V") == 0)
170 : {
171 2 : puts("pg_rewind (PostgreSQL) " PG_VERSION);
172 2 : exit(0);
173 : }
174 : }
175 :
176 252 : while ((c = getopt_long(argc, argv, "cD:nNPR", long_options, &option_index)) != -1)
177 : {
178 208 : switch (c)
179 : {
180 2 : case 'c':
181 2 : restore_wal = true;
182 2 : break;
183 :
184 0 : case 'P':
185 0 : showprogress = true;
186 0 : break;
187 :
188 2 : case 'n':
189 2 : dry_run = true;
190 2 : break;
191 :
192 34 : case 'N':
193 34 : do_sync = false;
194 34 : break;
195 :
196 12 : case 'R':
197 12 : writerecoveryconf = true;
198 12 : break;
199 :
200 42 : case 3:
201 42 : debug = true;
202 42 : pg_logging_increase_verbosity();
203 42 : break;
204 :
205 44 : case 'D': /* -D or --target-pgdata */
206 44 : datadir_target = pg_strdup(optarg);
207 44 : break;
208 :
209 30 : case 1: /* --source-pgdata */
210 30 : datadir_source = pg_strdup(optarg);
211 30 : break;
212 :
213 14 : case 2: /* --source-server */
214 14 : connstr_source = pg_strdup(optarg);
215 14 : break;
216 :
217 6 : case 4:
218 6 : no_ensure_shutdown = true;
219 6 : break;
220 :
221 20 : case 5:
222 20 : config_file = pg_strdup(optarg);
223 20 : break;
224 :
225 0 : case 6:
226 0 : if (!parse_sync_method(optarg, &sync_method))
227 0 : exit(1);
228 0 : break;
229 :
230 2 : default:
231 : /* getopt_long already emitted a complaint */
232 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
233 2 : exit(1);
234 : }
235 : }
236 :
237 44 : if (datadir_source == NULL && connstr_source == NULL)
238 : {
239 2 : pg_log_error("no source specified (--source-pgdata or --source-server)");
240 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
241 2 : exit(1);
242 : }
243 :
244 42 : if (datadir_source != NULL && connstr_source != NULL)
245 : {
246 2 : pg_log_error("only one of --source-pgdata or --source-server can be specified");
247 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
248 2 : exit(1);
249 : }
250 :
251 40 : if (datadir_target == NULL)
252 : {
253 0 : pg_log_error("no target data directory specified (--target-pgdata)");
254 0 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
255 0 : exit(1);
256 : }
257 :
258 40 : if (writerecoveryconf && connstr_source == NULL)
259 : {
260 2 : pg_log_error("no source server information (--source-server) specified for --write-recovery-conf");
261 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
262 2 : exit(1);
263 : }
264 :
265 38 : if (optind < argc)
266 : {
267 2 : pg_log_error("too many command-line arguments (first is \"%s\")",
268 : argv[optind]);
269 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
270 2 : exit(1);
271 : }
272 :
273 : /*
274 : * Don't allow pg_rewind to be run as root, to avoid overwriting the
275 : * ownership of files in the data directory. We need only check for root
276 : * -- any other user won't have sufficient permissions to modify files in
277 : * the data directory.
278 : */
279 : #ifndef WIN32
280 36 : if (geteuid() == 0)
281 : {
282 0 : pg_log_error("cannot be executed by \"root\"");
283 0 : pg_log_error_hint("You must run %s as the PostgreSQL superuser.",
284 : progname);
285 0 : exit(1);
286 : }
287 : #endif
288 :
289 36 : get_restricted_token();
290 :
291 : /* Set mask based on PGDATA permissions */
292 36 : if (!GetDataDirectoryCreatePerm(datadir_target))
293 0 : pg_fatal("could not read permissions of directory \"%s\": %m",
294 : datadir_target);
295 :
296 36 : umask(pg_mode_mask);
297 :
298 36 : getRestoreCommand(argv[0]);
299 :
300 36 : atexit(disconnect_atexit);
301 :
302 : /*
303 : * Ok, we have all the options and we're ready to start. First, connect to
304 : * remote server.
305 : */
306 36 : if (connstr_source)
307 : {
308 12 : conn = PQconnectdb(connstr_source);
309 :
310 12 : if (PQstatus(conn) == CONNECTION_BAD)
311 0 : pg_fatal("%s", PQerrorMessage(conn));
312 :
313 12 : if (showprogress)
314 0 : pg_log_info("connected to server");
315 :
316 12 : source = init_libpq_source(conn);
317 : }
318 : else
319 24 : source = init_local_source(datadir_source);
320 :
321 : /*
322 : * Check the status of the target instance.
323 : *
324 : * If the target instance was not cleanly shut down, start and stop the
325 : * target cluster once in single-user mode to enforce recovery to finish,
326 : * ensuring that the cluster can be used by pg_rewind. Note that if
327 : * no_ensure_shutdown is specified, pg_rewind ignores this step, and users
328 : * need to make sure by themselves that the target cluster is in a clean
329 : * state.
330 : */
331 36 : buffer = slurpFile(datadir_target, "global/pg_control", &size);
332 36 : digestControlFile(&ControlFile_target, buffer, size);
333 36 : pg_free(buffer);
334 :
335 36 : if (!no_ensure_shutdown &&
336 30 : ControlFile_target.state != DB_SHUTDOWNED &&
337 22 : ControlFile_target.state != DB_SHUTDOWNED_IN_RECOVERY)
338 : {
339 20 : ensureCleanShutdown(argv[0]);
340 :
341 18 : buffer = slurpFile(datadir_target, "global/pg_control", &size);
342 18 : digestControlFile(&ControlFile_target, buffer, size);
343 18 : pg_free(buffer);
344 : }
345 :
346 34 : buffer = source->fetch_file(source, "global/pg_control", &size);
347 34 : digestControlFile(&ControlFile_source, buffer, size);
348 34 : pg_free(buffer);
349 :
350 34 : sanityChecks();
351 :
352 : /*
353 : * Usually, the TLI can be found in the latest checkpoint record. But if
354 : * the source server is just being promoted (or it's a standby that's
355 : * following a primary that's just being promoted), and the checkpoint
356 : * requested by the promotion hasn't completed yet, the latest timeline is
357 : * in minRecoveryPoint. So we check which is later, the TLI of the
358 : * minRecoveryPoint or the latest checkpoint.
359 : */
360 30 : source_tli = Max(ControlFile_source.minRecoveryPointTLI,
361 : ControlFile_source.checkPointCopy.ThisTimeLineID);
362 :
363 : /* Similarly for the target. */
364 30 : target_tli = Max(ControlFile_target.minRecoveryPointTLI,
365 : ControlFile_target.checkPointCopy.ThisTimeLineID);
366 :
367 : /*
368 : * Find the common ancestor timeline between the clusters.
369 : *
370 : * If both clusters are already on the same timeline, there's nothing to
371 : * do.
372 : */
373 30 : if (target_tli == source_tli)
374 : {
375 2 : pg_log_info("source and target cluster are on the same timeline");
376 2 : rewind_needed = false;
377 2 : target_wal_endrec = 0;
378 : }
379 : else
380 : {
381 : XLogRecPtr chkptendrec;
382 : TimeLineHistoryEntry *sourceHistory;
383 : int sourceNentries;
384 :
385 : /*
386 : * Retrieve timelines for both source and target, and find the point
387 : * where they diverged.
388 : */
389 28 : sourceHistory = getTimelineHistory(source_tli, true, &sourceNentries);
390 28 : targetHistory = getTimelineHistory(target_tli, false, &targetNentries);
391 :
392 28 : findCommonAncestorTimeline(sourceHistory, sourceNentries,
393 : targetHistory, targetNentries,
394 : &divergerec, &lastcommontliIndex);
395 :
396 28 : pg_log_info("servers diverged at WAL location %X/%X on timeline %u",
397 : LSN_FORMAT_ARGS(divergerec),
398 : targetHistory[lastcommontliIndex].tli);
399 :
400 : /*
401 : * Don't need the source history anymore. The target history is still
402 : * needed by the routines in parsexlog.c, when we read the target WAL.
403 : */
404 28 : pfree(sourceHistory);
405 :
406 :
407 : /*
408 : * Determine the end-of-WAL on the target.
409 : *
410 : * The WAL ends at the last shutdown checkpoint, or at
411 : * minRecoveryPoint if it was a standby. (If we supported rewinding a
412 : * server that was not shut down cleanly, we would need to replay
413 : * until we reach the first invalid record, like crash recovery does.)
414 : */
415 :
416 : /* read the checkpoint record on the target to see where it ends. */
417 28 : chkptendrec = readOneRecord(datadir_target,
418 : ControlFile_target.checkPoint,
419 : targetNentries - 1,
420 : restore_command);
421 :
422 28 : if (ControlFile_target.minRecoveryPoint > chkptendrec)
423 : {
424 2 : target_wal_endrec = ControlFile_target.minRecoveryPoint;
425 : }
426 : else
427 : {
428 26 : target_wal_endrec = chkptendrec;
429 : }
430 :
431 : /*
432 : * Check for the possibility that the target is in fact a direct
433 : * ancestor of the source. In that case, there is no divergent history
434 : * in the target that needs rewinding.
435 : */
436 28 : if (target_wal_endrec > divergerec)
437 : {
438 28 : rewind_needed = true;
439 : }
440 : else
441 : {
442 : /* the last common checkpoint record must be part of target WAL */
443 : Assert(target_wal_endrec == divergerec);
444 :
445 0 : rewind_needed = false;
446 : }
447 : }
448 :
449 30 : if (!rewind_needed)
450 : {
451 2 : pg_log_info("no rewind required");
452 2 : if (writerecoveryconf && !dry_run)
453 0 : WriteRecoveryConfig(conn, datadir_target,
454 : GenerateRecoveryConfig(conn, NULL, NULL));
455 2 : exit(0);
456 : }
457 :
458 : /* Initialize hashtable that tracks WAL files protected from removal */
459 28 : keepwal_init();
460 :
461 28 : findLastCheckpoint(datadir_target, divergerec, lastcommontliIndex,
462 : &chkptrec, &chkpttli, &chkptredo, restore_command);
463 28 : pg_log_info("rewinding from last common checkpoint at %X/%X on timeline %u",
464 : LSN_FORMAT_ARGS(chkptrec), chkpttli);
465 :
466 : /* Initialize the hash table to track the status of each file */
467 28 : filehash_init();
468 :
469 : /*
470 : * Collect information about all files in the both data directories.
471 : */
472 28 : if (showprogress)
473 0 : pg_log_info("reading source file list");
474 28 : source->traverse_files(source, &process_source_file);
475 :
476 28 : if (showprogress)
477 0 : pg_log_info("reading target file list");
478 28 : traverse_datadir(datadir_target, &process_target_file);
479 :
480 : /*
481 : * Read the target WAL from last checkpoint before the point of fork, to
482 : * extract all the pages that were modified on the target cluster after
483 : * the fork.
484 : */
485 28 : if (showprogress)
486 0 : pg_log_info("reading WAL in target");
487 28 : extractPageMap(datadir_target, chkptrec, lastcommontliIndex,
488 : target_wal_endrec, restore_command);
489 :
490 : /*
491 : * We have collected all information we need from both systems. Decide
492 : * what to do with each file.
493 : */
494 28 : filemap = decide_file_actions();
495 28 : if (showprogress)
496 0 : calculate_totals(filemap);
497 :
498 : /* this is too verbose even for verbose mode */
499 28 : if (debug)
500 28 : print_filemap(filemap);
501 :
502 : /*
503 : * Ok, we're ready to start copying things over.
504 : */
505 28 : if (showprogress)
506 : {
507 0 : pg_log_info("need to copy %lu MB (total source directory size is %lu MB)",
508 : (unsigned long) (filemap->fetch_size / (1024 * 1024)),
509 : (unsigned long) (filemap->total_size / (1024 * 1024)));
510 :
511 0 : fetch_size = filemap->fetch_size;
512 0 : fetch_done = 0;
513 : }
514 :
515 : /*
516 : * We have now collected all the information we need from both systems,
517 : * and we are ready to start modifying the target directory.
518 : *
519 : * This is the point of no return. Once we start copying things, there is
520 : * no turning back!
521 : */
522 28 : perform_rewind(filemap, source, chkptrec, chkpttli, chkptredo);
523 :
524 26 : if (showprogress)
525 0 : pg_log_info("syncing target data directory");
526 26 : sync_target_dir();
527 :
528 : /* Also update the standby configuration, if requested. */
529 26 : if (writerecoveryconf && !dry_run)
530 10 : WriteRecoveryConfig(conn, datadir_target,
531 : GenerateRecoveryConfig(conn, NULL, NULL));
532 :
533 : /* don't need the source connection anymore */
534 26 : source->destroy(source);
535 26 : if (conn)
536 : {
537 12 : PQfinish(conn);
538 12 : conn = NULL;
539 : }
540 :
541 26 : pg_log_info("Done!");
542 :
543 26 : return 0;
544 : }
545 :
546 : /*
547 : * Perform the rewind.
548 : *
549 : * We have already collected all the information we need from the
550 : * target and the source.
551 : */
552 : static void
553 28 : perform_rewind(filemap_t *filemap, rewind_source *source,
554 : XLogRecPtr chkptrec,
555 : TimeLineID chkpttli,
556 : XLogRecPtr chkptredo)
557 : {
558 : XLogRecPtr endrec;
559 : TimeLineID endtli;
560 : ControlFileData ControlFile_new;
561 : size_t size;
562 : char *buffer;
563 :
564 : /*
565 : * Execute the actions in the file map, fetching data from the source
566 : * system as needed.
567 : */
568 31908 : for (int i = 0; i < filemap->nentries; i++)
569 : {
570 31882 : file_entry_t *entry = filemap->entries[i];
571 :
572 : /*
573 : * If this is a relation file, copy the modified blocks.
574 : *
575 : * This is in addition to any other changes.
576 : */
577 31882 : if (entry->target_pages_to_overwrite.bitmapsize > 0)
578 : {
579 : datapagemap_iterator_t *iter;
580 : BlockNumber blkno;
581 : off_t offset;
582 :
583 842 : iter = datapagemap_iterate(&entry->target_pages_to_overwrite);
584 4172 : while (datapagemap_next(iter, &blkno))
585 : {
586 3330 : offset = blkno * BLCKSZ;
587 3330 : source->queue_fetch_range(source, entry->path, offset, BLCKSZ);
588 : }
589 842 : pg_free(iter);
590 : }
591 :
592 31882 : switch (entry->action)
593 : {
594 21546 : case FILE_ACTION_NONE:
595 : /* nothing else to do */
596 21546 : break;
597 :
598 8872 : case FILE_ACTION_COPY:
599 8872 : source->queue_fetch_file(source, entry->path, entry->source_size);
600 8870 : break;
601 :
602 8 : case FILE_ACTION_TRUNCATE:
603 8 : truncate_target_file(entry->path, entry->source_size);
604 8 : break;
605 :
606 10 : case FILE_ACTION_COPY_TAIL:
607 10 : source->queue_fetch_range(source, entry->path,
608 10 : entry->target_size,
609 10 : entry->source_size - entry->target_size);
610 10 : break;
611 :
612 1428 : case FILE_ACTION_REMOVE:
613 1428 : remove_target(entry);
614 1428 : break;
615 :
616 18 : case FILE_ACTION_CREATE:
617 18 : create_target(entry);
618 18 : break;
619 :
620 0 : case FILE_ACTION_UNDECIDED:
621 0 : pg_fatal("no action decided for file \"%s\"", entry->path);
622 : break;
623 : }
624 31880 : }
625 :
626 : /* Complete any remaining range-fetches that we queued up above. */
627 26 : source->finish_fetch(source);
628 :
629 26 : close_target_file();
630 :
631 26 : progress_report(true);
632 :
633 : /*
634 : * Fetch the control file from the source last. This ensures that the
635 : * minRecoveryPoint is up-to-date.
636 : */
637 26 : buffer = source->fetch_file(source, "global/pg_control", &size);
638 26 : digestControlFile(&ControlFile_source_after, buffer, size);
639 26 : pg_free(buffer);
640 :
641 : /*
642 : * Sanity check: If the source is a local system, the control file should
643 : * not have changed since we started.
644 : *
645 : * XXX: We assume it hasn't been modified, but actually, what could go
646 : * wrong? The logic handles a libpq source that's modified concurrently,
647 : * why not a local datadir?
648 : */
649 26 : if (datadir_source &&
650 14 : memcmp(&ControlFile_source, &ControlFile_source_after,
651 : sizeof(ControlFileData)) != 0)
652 : {
653 0 : pg_fatal("source system was modified while pg_rewind was running");
654 : }
655 :
656 26 : if (showprogress)
657 0 : pg_log_info("creating backup label and updating control file");
658 :
659 : /*
660 : * Create a backup label file, to tell the target where to begin the WAL
661 : * replay. Normally, from the last common checkpoint between the source
662 : * and the target. But if the source is a standby server, it's possible
663 : * that the last common checkpoint is *after* the standby's restartpoint.
664 : * That implies that the source server has applied the checkpoint record,
665 : * but hasn't performed a corresponding restartpoint yet. Make sure we
666 : * start at the restartpoint's redo point in that case.
667 : *
668 : * Use the old version of the source's control file for this. The server
669 : * might have finished the restartpoint after we started copying files,
670 : * but we must begin from the redo point at the time that started copying.
671 : */
672 26 : if (ControlFile_source.checkPointCopy.redo < chkptredo)
673 : {
674 4 : chkptredo = ControlFile_source.checkPointCopy.redo;
675 4 : chkpttli = ControlFile_source.checkPointCopy.ThisTimeLineID;
676 4 : chkptrec = ControlFile_source.checkPoint;
677 : }
678 26 : createBackupLabel(chkptredo, chkpttli, chkptrec);
679 :
680 : /*
681 : * Update control file of target, to tell the target how far it must
682 : * replay the WAL (minRecoveryPoint).
683 : */
684 26 : if (connstr_source)
685 : {
686 : /*
687 : * The source is a live server. Like in an online backup, it's
688 : * important that we recover all the WAL that was generated while we
689 : * were copying files.
690 : */
691 12 : if (ControlFile_source_after.state == DB_IN_ARCHIVE_RECOVERY)
692 : {
693 : /*
694 : * Source is a standby server. We must replay to its
695 : * minRecoveryPoint.
696 : */
697 2 : endrec = ControlFile_source_after.minRecoveryPoint;
698 2 : endtli = ControlFile_source_after.minRecoveryPointTLI;
699 : }
700 : else
701 : {
702 : /*
703 : * Source is a production, non-standby, server. We must replay to
704 : * the last WAL insert location.
705 : */
706 10 : if (ControlFile_source_after.state != DB_IN_PRODUCTION)
707 0 : pg_fatal("source system was in unexpected state at end of rewind");
708 :
709 10 : endrec = source->get_current_wal_insert_lsn(source);
710 10 : endtli = Max(ControlFile_source_after.checkPointCopy.ThisTimeLineID,
711 : ControlFile_source_after.minRecoveryPointTLI);
712 : }
713 : }
714 : else
715 : {
716 : /*
717 : * Source is a local data directory. It should've shut down cleanly,
718 : * and we must replay to the latest shutdown checkpoint.
719 : */
720 14 : endrec = ControlFile_source_after.checkPoint;
721 14 : endtli = ControlFile_source_after.checkPointCopy.ThisTimeLineID;
722 : }
723 :
724 26 : memcpy(&ControlFile_new, &ControlFile_source_after, sizeof(ControlFileData));
725 26 : ControlFile_new.minRecoveryPoint = endrec;
726 26 : ControlFile_new.minRecoveryPointTLI = endtli;
727 26 : ControlFile_new.state = DB_IN_ARCHIVE_RECOVERY;
728 26 : if (!dry_run)
729 24 : update_controlfile(datadir_target, &ControlFile_new, do_sync);
730 26 : }
731 :
732 : static void
733 34 : sanityChecks(void)
734 : {
735 : /* TODO Check that there's no backup_label in either cluster */
736 :
737 : /* Check system_identifier match */
738 34 : if (ControlFile_target.system_identifier != ControlFile_source.system_identifier)
739 0 : pg_fatal("source and target clusters are from different systems");
740 :
741 : /* check version */
742 34 : if (ControlFile_target.pg_control_version != PG_CONTROL_VERSION ||
743 34 : ControlFile_source.pg_control_version != PG_CONTROL_VERSION ||
744 34 : ControlFile_target.catalog_version_no != CATALOG_VERSION_NO ||
745 34 : ControlFile_source.catalog_version_no != CATALOG_VERSION_NO)
746 : {
747 0 : pg_fatal("clusters are not compatible with this version of pg_rewind");
748 : }
749 :
750 : /*
751 : * Target cluster need to use checksums or hint bit wal-logging, this to
752 : * prevent from data corruption that could occur because of hint bits.
753 : */
754 34 : if (ControlFile_target.data_checksum_version != PG_DATA_CHECKSUM_VERSION &&
755 0 : !ControlFile_target.wal_log_hints)
756 : {
757 0 : pg_fatal("target server needs to use either data checksums or \"wal_log_hints = on\"");
758 : }
759 :
760 : /*
761 : * Target cluster better not be running. This doesn't guard against
762 : * someone starting the cluster concurrently. Also, this is probably more
763 : * strict than necessary; it's OK if the target node was not shut down
764 : * cleanly, as long as it isn't running at the moment.
765 : */
766 34 : if (ControlFile_target.state != DB_SHUTDOWNED &&
767 4 : ControlFile_target.state != DB_SHUTDOWNED_IN_RECOVERY)
768 2 : pg_fatal("target server must be shut down cleanly");
769 :
770 : /*
771 : * When the source is a data directory, also require that the source
772 : * server is shut down. There isn't any very strong reason for this
773 : * limitation, but better safe than sorry.
774 : */
775 32 : if (datadir_source &&
776 20 : ControlFile_source.state != DB_SHUTDOWNED &&
777 4 : ControlFile_source.state != DB_SHUTDOWNED_IN_RECOVERY)
778 2 : pg_fatal("source data directory must be shut down cleanly");
779 30 : }
780 :
781 : /*
782 : * Print a progress report based on the fetch_size and fetch_done variables.
783 : *
784 : * Progress report is written at maximum once per second, except that the
785 : * last progress report is always printed.
786 : *
787 : * If finished is set to true, this is the last progress report. The cursor
788 : * is moved to the next line.
789 : */
790 : void
791 119770 : progress_report(bool finished)
792 : {
793 : static pg_time_t last_progress_report = 0;
794 : int percent;
795 : char fetch_done_str[32];
796 : char fetch_size_str[32];
797 : pg_time_t now;
798 :
799 119770 : if (!showprogress)
800 119770 : return;
801 :
802 0 : now = time(NULL);
803 0 : if (now == last_progress_report && !finished)
804 0 : return; /* Max once per second */
805 :
806 0 : last_progress_report = now;
807 0 : percent = fetch_size ? (int) ((fetch_done) * 100 / fetch_size) : 0;
808 :
809 : /*
810 : * Avoid overflowing past 100% or the full size. This may make the total
811 : * size number change as we approach the end of the backup (the estimate
812 : * will always be wrong if WAL is included), but that's better than having
813 : * the done column be bigger than the total.
814 : */
815 0 : if (percent > 100)
816 0 : percent = 100;
817 0 : if (fetch_done > fetch_size)
818 0 : fetch_size = fetch_done;
819 :
820 0 : snprintf(fetch_done_str, sizeof(fetch_done_str), UINT64_FORMAT,
821 : fetch_done / 1024);
822 0 : snprintf(fetch_size_str, sizeof(fetch_size_str), UINT64_FORMAT,
823 : fetch_size / 1024);
824 :
825 0 : fprintf(stderr, _("%*s/%s kB (%d%%) copied"),
826 0 : (int) strlen(fetch_size_str), fetch_done_str, fetch_size_str,
827 : percent);
828 :
829 : /*
830 : * Stay on the same line if reporting to a terminal and we're not done
831 : * yet.
832 : */
833 0 : fputc((!finished && isatty(fileno(stderr))) ? '\r' : '\n', stderr);
834 : }
835 :
836 : /*
837 : * Find minimum from two WAL locations assuming InvalidXLogRecPtr means
838 : * infinity as src/include/access/timeline.h states. This routine should
839 : * be used only when comparing WAL locations related to history files.
840 : */
841 : static XLogRecPtr
842 28 : MinXLogRecPtr(XLogRecPtr a, XLogRecPtr b)
843 : {
844 28 : if (XLogRecPtrIsInvalid(a))
845 2 : return b;
846 26 : else if (XLogRecPtrIsInvalid(b))
847 26 : return a;
848 : else
849 0 : return Min(a, b);
850 : }
851 :
852 : /*
853 : * Retrieve timeline history for the source or target system.
854 : */
855 : static TimeLineHistoryEntry *
856 56 : getTimelineHistory(TimeLineID tli, bool is_source, int *nentries)
857 : {
858 : TimeLineHistoryEntry *history;
859 :
860 : /*
861 : * Timeline 1 does not have a history file, so there is no need to check
862 : * and fake an entry with infinite start and end positions.
863 : */
864 56 : if (tli == 1)
865 : {
866 26 : history = (TimeLineHistoryEntry *) pg_malloc(sizeof(TimeLineHistoryEntry));
867 26 : history->tli = tli;
868 26 : history->begin = history->end = InvalidXLogRecPtr;
869 26 : *nentries = 1;
870 : }
871 : else
872 : {
873 : char path[MAXPGPATH];
874 : char *histfile;
875 :
876 30 : TLHistoryFilePath(path, tli);
877 :
878 : /* Get history file from appropriate source */
879 30 : if (is_source)
880 26 : histfile = source->fetch_file(source, path, NULL);
881 : else
882 4 : histfile = slurpFile(datadir_target, path, NULL);
883 :
884 30 : history = rewind_parseTimeLineHistory(histfile, tli, nentries);
885 30 : pg_free(histfile);
886 : }
887 :
888 : /* In debugging mode, print what we read */
889 56 : if (debug)
890 : {
891 : int i;
892 :
893 56 : if (is_source)
894 28 : pg_log_debug("Source timeline history:");
895 : else
896 28 : pg_log_debug("Target timeline history:");
897 :
898 144 : for (i = 0; i < *nentries; i++)
899 : {
900 : TimeLineHistoryEntry *entry;
901 :
902 88 : entry = &history[i];
903 88 : pg_log_debug("%u: %X/%X - %X/%X", entry->tli,
904 : LSN_FORMAT_ARGS(entry->begin),
905 : LSN_FORMAT_ARGS(entry->end));
906 : }
907 : }
908 :
909 56 : return history;
910 : }
911 :
912 : /*
913 : * Determine the TLI of the last common timeline in the timeline history of
914 : * two clusters. *tliIndex is set to the index of last common timeline in
915 : * the arrays, and *recptr is set to the position where the timeline history
916 : * diverged (ie. the first WAL record that's not the same in both clusters).
917 : */
918 : static void
919 28 : findCommonAncestorTimeline(TimeLineHistoryEntry *a_history, int a_nentries,
920 : TimeLineHistoryEntry *b_history, int b_nentries,
921 : XLogRecPtr *recptr, int *tliIndex)
922 : {
923 : int i,
924 : n;
925 :
926 : /*
927 : * Trace the history forward, until we hit the timeline diverge. It may
928 : * still be possible that the source and target nodes used the same
929 : * timeline number in their history but with different start position
930 : * depending on the history files that each node has fetched in previous
931 : * recovery processes. Hence check the start position of the new timeline
932 : * as well and move down by one extra timeline entry if they do not match.
933 : */
934 28 : n = Min(a_nentries, b_nentries);
935 58 : for (i = 0; i < n; i++)
936 : {
937 30 : if (a_history[i].tli != b_history[i].tli ||
938 30 : a_history[i].begin != b_history[i].begin)
939 : break;
940 : }
941 :
942 28 : if (i > 0)
943 : {
944 28 : i--;
945 28 : *recptr = MinXLogRecPtr(a_history[i].end, b_history[i].end);
946 28 : *tliIndex = i;
947 28 : return;
948 : }
949 : else
950 : {
951 0 : pg_fatal("could not find common ancestor of the source and target cluster's timelines");
952 : }
953 : }
954 :
955 :
956 : /*
957 : * Create a backup_label file that forces recovery to begin at the last common
958 : * checkpoint.
959 : */
960 : static void
961 26 : createBackupLabel(XLogRecPtr startpoint, TimeLineID starttli, XLogRecPtr checkpointloc)
962 : {
963 : XLogSegNo startsegno;
964 : time_t stamp_time;
965 : char strfbuf[128];
966 : char xlogfilename[MAXFNAMELEN];
967 : struct tm *tmp;
968 : char buf[1000];
969 : int len;
970 :
971 26 : XLByteToSeg(startpoint, startsegno, WalSegSz);
972 26 : XLogFileName(xlogfilename, starttli, startsegno, WalSegSz);
973 :
974 : /*
975 : * Construct backup label file
976 : */
977 26 : stamp_time = time(NULL);
978 26 : tmp = localtime(&stamp_time);
979 26 : strftime(strfbuf, sizeof(strfbuf), "%Y-%m-%d %H:%M:%S %Z", tmp);
980 :
981 26 : len = snprintf(buf, sizeof(buf),
982 : "START WAL LOCATION: %X/%X (file %s)\n"
983 : "CHECKPOINT LOCATION: %X/%X\n"
984 : "BACKUP METHOD: pg_rewind\n"
985 : "BACKUP FROM: standby\n"
986 : "START TIME: %s\n",
987 : /* omit LABEL: line */
988 26 : LSN_FORMAT_ARGS(startpoint), xlogfilename,
989 26 : LSN_FORMAT_ARGS(checkpointloc),
990 : strfbuf);
991 26 : if (len >= sizeof(buf))
992 0 : pg_fatal("backup label buffer too small"); /* shouldn't happen */
993 :
994 : /* TODO: move old file out of the way, if any. */
995 26 : open_target_file("backup_label", true); /* BACKUP_LABEL_FILE */
996 26 : write_target_range(buf, 0, len);
997 26 : close_target_file();
998 26 : }
999 :
1000 : /*
1001 : * Check CRC of control file
1002 : */
1003 : static void
1004 114 : checkControlFile(ControlFileData *ControlFile)
1005 : {
1006 : pg_crc32c crc;
1007 :
1008 : /* Calculate CRC */
1009 114 : INIT_CRC32C(crc);
1010 114 : COMP_CRC32C(crc, (char *) ControlFile, offsetof(ControlFileData, crc));
1011 114 : FIN_CRC32C(crc);
1012 :
1013 : /* And simply compare it */
1014 114 : if (!EQ_CRC32C(crc, ControlFile->crc))
1015 0 : pg_fatal("unexpected control file CRC");
1016 114 : }
1017 :
1018 : /*
1019 : * Verify control file contents in the buffer 'content', and copy it to
1020 : * *ControlFile.
1021 : */
1022 : static void
1023 114 : digestControlFile(ControlFileData *ControlFile, const char *content,
1024 : size_t size)
1025 : {
1026 114 : if (size != PG_CONTROL_FILE_SIZE)
1027 0 : pg_fatal("unexpected control file size %d, expected %d",
1028 : (int) size, PG_CONTROL_FILE_SIZE);
1029 :
1030 114 : memcpy(ControlFile, content, sizeof(ControlFileData));
1031 :
1032 : /* set and validate WalSegSz */
1033 114 : WalSegSz = ControlFile->xlog_seg_size;
1034 :
1035 114 : if (!IsValidWalSegSize(WalSegSz))
1036 : {
1037 0 : pg_log_error(ngettext("invalid WAL segment size in control file (%d byte)",
1038 : "invalid WAL segment size in control file (%d bytes)",
1039 : WalSegSz),
1040 : WalSegSz);
1041 0 : pg_log_error_detail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
1042 0 : exit(1);
1043 : }
1044 :
1045 : /* Additional checks on control file */
1046 114 : checkControlFile(ControlFile);
1047 114 : }
1048 :
1049 : /*
1050 : * Get value of GUC parameter restore_command from the target cluster.
1051 : *
1052 : * This uses a logic based on "postgres -C" to get the value from the
1053 : * cluster.
1054 : */
1055 : static void
1056 36 : getRestoreCommand(const char *argv0)
1057 : {
1058 : int rc;
1059 : char postgres_exec_path[MAXPGPATH];
1060 : PQExpBuffer postgres_cmd;
1061 :
1062 36 : if (!restore_wal)
1063 34 : return;
1064 :
1065 : /* find postgres executable */
1066 2 : rc = find_other_exec(argv0, "postgres",
1067 : PG_BACKEND_VERSIONSTR,
1068 : postgres_exec_path);
1069 :
1070 2 : if (rc < 0)
1071 : {
1072 : char full_path[MAXPGPATH];
1073 :
1074 0 : if (find_my_exec(argv0, full_path) < 0)
1075 0 : strlcpy(full_path, progname, sizeof(full_path));
1076 :
1077 0 : if (rc == -1)
1078 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
1079 : "postgres", progname, full_path);
1080 : else
1081 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
1082 : "postgres", full_path, progname);
1083 : }
1084 :
1085 : /*
1086 : * Build a command able to retrieve the value of GUC parameter
1087 : * restore_command, if set.
1088 : */
1089 2 : postgres_cmd = createPQExpBuffer();
1090 :
1091 : /* path to postgres, properly quoted */
1092 2 : appendShellString(postgres_cmd, postgres_exec_path);
1093 :
1094 : /* add -D switch, with properly quoted data directory */
1095 2 : appendPQExpBufferStr(postgres_cmd, " -D ");
1096 2 : appendShellString(postgres_cmd, datadir_target);
1097 :
1098 : /* add custom configuration file only if requested */
1099 2 : if (config_file != NULL)
1100 : {
1101 2 : appendPQExpBufferStr(postgres_cmd, " -c config_file=");
1102 2 : appendShellString(postgres_cmd, config_file);
1103 : }
1104 :
1105 : /* add -C switch, for restore_command */
1106 2 : appendPQExpBufferStr(postgres_cmd, " -C restore_command");
1107 :
1108 2 : restore_command = pipe_read_line(postgres_cmd->data);
1109 2 : if (restore_command == NULL)
1110 0 : pg_fatal("could not read \"restore_command\" from target cluster");
1111 :
1112 2 : (void) pg_strip_crlf(restore_command);
1113 :
1114 2 : if (strcmp(restore_command, "") == 0)
1115 0 : pg_fatal("\"restore_command\" is not set in the target cluster");
1116 :
1117 2 : pg_log_debug("using for rewind \"restore_command = \'%s\'\"",
1118 : restore_command);
1119 :
1120 2 : destroyPQExpBuffer(postgres_cmd);
1121 : }
1122 :
1123 :
1124 : /*
1125 : * Ensure clean shutdown of target instance by launching single-user mode
1126 : * postgres to do crash recovery.
1127 : */
1128 : static void
1129 20 : ensureCleanShutdown(const char *argv0)
1130 : {
1131 : int ret;
1132 : char exec_path[MAXPGPATH];
1133 : PQExpBuffer postgres_cmd;
1134 :
1135 : /* locate postgres binary */
1136 20 : if ((ret = find_other_exec(argv0, "postgres",
1137 : PG_BACKEND_VERSIONSTR,
1138 : exec_path)) < 0)
1139 : {
1140 : char full_path[MAXPGPATH];
1141 :
1142 0 : if (find_my_exec(argv0, full_path) < 0)
1143 0 : strlcpy(full_path, progname, sizeof(full_path));
1144 :
1145 0 : if (ret == -1)
1146 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
1147 : "postgres", progname, full_path);
1148 : else
1149 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
1150 : "postgres", full_path, progname);
1151 : }
1152 :
1153 20 : pg_log_info("executing \"%s\" for target server to complete crash recovery",
1154 : exec_path);
1155 :
1156 : /*
1157 : * Skip processing if requested, but only after ensuring presence of
1158 : * postgres.
1159 : */
1160 20 : if (dry_run)
1161 0 : return;
1162 :
1163 : /*
1164 : * Finally run postgres in single-user mode. There is no need to use
1165 : * fsync here. This makes the recovery faster, and the target data folder
1166 : * is synced at the end anyway.
1167 : */
1168 20 : postgres_cmd = createPQExpBuffer();
1169 :
1170 : /* path to postgres, properly quoted */
1171 20 : appendShellString(postgres_cmd, exec_path);
1172 :
1173 : /* add set of options with properly quoted data directory */
1174 20 : appendPQExpBufferStr(postgres_cmd, " --single -F -D ");
1175 20 : appendShellString(postgres_cmd, datadir_target);
1176 :
1177 : /* add custom configuration file only if requested */
1178 20 : if (config_file != NULL)
1179 : {
1180 18 : appendPQExpBufferStr(postgres_cmd, " -c config_file=");
1181 18 : appendShellString(postgres_cmd, config_file);
1182 : }
1183 :
1184 : /* finish with the database name, and a properly quoted redirection */
1185 20 : appendPQExpBufferStr(postgres_cmd, " template1 < ");
1186 20 : appendShellString(postgres_cmd, DEVNULL);
1187 :
1188 20 : fflush(NULL);
1189 20 : if (system(postgres_cmd->data) != 0)
1190 : {
1191 2 : pg_log_error("postgres single-user mode in target cluster failed");
1192 2 : pg_log_error_detail("Command was: %s", postgres_cmd->data);
1193 2 : exit(1);
1194 : }
1195 :
1196 18 : destroyPQExpBuffer(postgres_cmd);
1197 : }
1198 :
1199 : static void
1200 36 : disconnect_atexit(void)
1201 : {
1202 36 : if (conn != NULL)
1203 0 : PQfinish(conn);
1204 36 : }
|
