Line data Source code
1 : /*
2 : * pg_upgrade.c
3 : *
4 : * main source file
5 : *
6 : * Copyright (c) 2010-2026, PostgreSQL Global Development Group
7 : * src/bin/pg_upgrade/pg_upgrade.c
8 : */
9 :
10 : /*
11 : * To simplify the upgrade process, we force certain system values to be
12 : * identical between old and new clusters:
13 : *
14 : * We control all assignments of pg_class.oid (and relfilenode) so toast
15 : * oids are the same between old and new clusters. This is important
16 : * because toast oids are stored as toast pointers in user tables.
17 : *
18 : * While pg_class.oid and pg_class.relfilenode are initially the same in a
19 : * cluster, they can diverge due to CLUSTER, REINDEX, or VACUUM FULL. We
20 : * control assignments of pg_class.relfilenode because we want the filenames
21 : * to match between the old and new cluster.
22 : *
23 : * We control assignment of pg_tablespace.oid because we want the oid to match
24 : * between the old and new cluster.
25 : *
26 : * We control all assignments of pg_type.oid because these oids are stored
27 : * in user composite type values.
28 : *
29 : * We control all assignments of pg_enum.oid because these oids are stored
30 : * in user tables as enum values.
31 : *
32 : * We control all assignments of pg_authid.oid because the oids are stored in
33 : * pg_largeobject_metadata, which is copied via file transfer for upgrades
34 : * from v16 and newer.
35 : *
36 : * We control all assignments of pg_database.oid because we want the directory
37 : * names to match between the old and new cluster.
38 : */
39 :
40 :
41 :
42 : #include "postgres_fe.h"
43 :
44 : #include <time.h>
45 :
46 : #include "access/multixact.h"
47 : #include "catalog/pg_class_d.h"
48 : #include "common/file_perm.h"
49 : #include "common/logging.h"
50 : #include "common/restricted_token.h"
51 : #include "fe_utils/string_utils.h"
52 : #include "pg_upgrade.h"
53 :
54 : /*
55 : * Maximum number of pg_restore actions (TOC entries) to process within one
56 : * transaction. At some point we might want to make this user-controllable,
57 : * but for now a hard-wired setting will suffice.
58 : */
59 : #define RESTORE_TRANSACTION_SIZE 1000
60 :
61 : static void set_new_cluster_char_signedness(void);
62 : static void set_locale_and_encoding(void);
63 : static void prepare_new_cluster(void);
64 : static void prepare_new_globals(void);
65 : static void create_new_objects(void);
66 : static void copy_xact_xlog_xid(void);
67 : static void set_frozenxids(bool minmxid_only);
68 : static void make_outputdirs(char *pgdata);
69 : static void setup(char *argv0);
70 : static void create_logical_replication_slots(void);
71 : static void create_conflict_detection_slot(void);
72 :
73 : ClusterInfo old_cluster,
74 : new_cluster;
75 : OSInfo os_info;
76 :
77 : char *output_files[] = {
78 : SERVER_LOG_FILE,
79 : #ifdef WIN32
80 : /* unique file for pg_ctl start */
81 : SERVER_START_LOG_FILE,
82 : #endif
83 : UTILITY_LOG_FILE,
84 : INTERNAL_LOG_FILE,
85 : NULL
86 : };
87 :
88 :
89 : int
90 0 : main(int argc, char **argv)
91 : {
92 0 : char *deletion_script_file_name = NULL;
93 0 : bool migrate_logical_slots;
94 :
95 : /*
96 : * pg_upgrade doesn't currently use common/logging.c, but initialize it
97 : * anyway because we might call common code that does.
98 : */
99 0 : pg_logging_init(argv[0]);
100 0 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_upgrade"));
101 :
102 : /* Set default restrictive mask until new cluster permissions are read */
103 0 : umask(PG_MODE_MASK_OWNER);
104 :
105 0 : parseCommandLine(argc, argv);
106 :
107 0 : get_restricted_token();
108 :
109 0 : adjust_data_dir(&old_cluster);
110 0 : adjust_data_dir(&new_cluster);
111 :
112 : /*
113 : * Set mask based on PGDATA permissions, needed for the creation of the
114 : * output directories with correct permissions.
115 : */
116 0 : if (!GetDataDirectoryCreatePerm(new_cluster.pgdata))
117 0 : pg_fatal("could not read permissions of directory \"%s\": %m",
118 0 : new_cluster.pgdata);
119 :
120 0 : umask(pg_mode_mask);
121 :
122 : /*
123 : * This needs to happen after adjusting the data directory of the new
124 : * cluster in adjust_data_dir().
125 : */
126 0 : make_outputdirs(new_cluster.pgdata);
127 :
128 0 : setup(argv[0]);
129 :
130 0 : output_check_banner();
131 :
132 0 : check_cluster_versions();
133 :
134 0 : get_sock_dir(&old_cluster);
135 0 : get_sock_dir(&new_cluster);
136 :
137 0 : check_cluster_compatibility();
138 :
139 0 : check_and_dump_old_cluster();
140 :
141 :
142 : /* -- NEW -- */
143 0 : start_postmaster(&new_cluster, true);
144 :
145 0 : check_new_cluster();
146 0 : report_clusters_compatible();
147 :
148 0 : pg_log(PG_REPORT,
149 : "\n"
150 : "Performing Upgrade\n"
151 : "------------------");
152 :
153 0 : set_locale_and_encoding();
154 :
155 0 : prepare_new_cluster();
156 :
157 0 : stop_postmaster(false);
158 :
159 : /*
160 : * Destructive Changes to New Cluster
161 : */
162 :
163 0 : copy_xact_xlog_xid();
164 0 : set_new_cluster_char_signedness();
165 :
166 : /* New now using xids of the old system */
167 :
168 : /* -- NEW -- */
169 0 : start_postmaster(&new_cluster, true);
170 :
171 0 : prepare_new_globals();
172 :
173 0 : create_new_objects();
174 :
175 0 : stop_postmaster(false);
176 :
177 : /*
178 : * Most failures happen in create_new_objects(), which has completed at
179 : * this point. We do this here because it is just before file transfer,
180 : * which for --link will make it unsafe to start the old cluster once the
181 : * new cluster is started, and for --swap will make it unsafe to start the
182 : * old cluster at all.
183 : */
184 0 : if (user_opts.transfer_mode == TRANSFER_MODE_LINK ||
185 0 : user_opts.transfer_mode == TRANSFER_MODE_SWAP)
186 0 : disable_old_cluster(user_opts.transfer_mode);
187 :
188 0 : transfer_all_new_tablespaces(&old_cluster.dbarr, &new_cluster.dbarr,
189 0 : old_cluster.pgdata, new_cluster.pgdata);
190 :
191 : /*
192 : * Assuming OIDs are only used in system tables, there is no need to
193 : * restore the OID counter because we have not transferred any OIDs from
194 : * the old system, but we do it anyway just in case. We do it late here
195 : * because there is no need to have the schema load use new oids.
196 : */
197 0 : prep_status("Setting next OID for new cluster");
198 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
199 : "\"%s/pg_resetwal\" -o %u \"%s\"",
200 0 : new_cluster.bindir, old_cluster.controldata.chkpnt_nxtoid,
201 0 : new_cluster.pgdata);
202 0 : check_ok();
203 :
204 0 : migrate_logical_slots = count_old_cluster_logical_slots();
205 :
206 : /*
207 : * Migrate replication slots to the new cluster.
208 : *
209 : * Note that we must migrate logical slots after resetting WAL because
210 : * otherwise the required WAL would be removed and slots would become
211 : * unusable. There is a possibility that background processes might
212 : * generate some WAL before we could create the slots in the new cluster
213 : * but we can ignore that WAL as that won't be required downstream.
214 : *
215 : * The conflict detection slot is not affected by concerns related to WALs
216 : * as it only retains the dead tuples. It is created here for consistency.
217 : * Note that the new conflict detection slot uses the latest transaction
218 : * ID as xmin, so it cannot protect dead tuples that existed before the
219 : * upgrade. Additionally, commit timestamps and origin data are not
220 : * preserved during the upgrade. So, even after creating the slot, the
221 : * upgraded subscriber may be unable to detect conflicts or log relevant
222 : * commit timestamps and origins when applying changes from the publisher
223 : * occurred before the upgrade especially if those changes were not
224 : * replicated. It can only protect tuples that might be deleted after the
225 : * new cluster starts.
226 : */
227 0 : if (migrate_logical_slots || old_cluster.sub_retain_dead_tuples)
228 : {
229 0 : start_postmaster(&new_cluster, true);
230 :
231 0 : if (migrate_logical_slots)
232 0 : create_logical_replication_slots();
233 :
234 0 : if (old_cluster.sub_retain_dead_tuples)
235 0 : create_conflict_detection_slot();
236 :
237 0 : stop_postmaster(false);
238 0 : }
239 :
240 0 : if (user_opts.do_sync)
241 : {
242 0 : prep_status("Sync data directory to disk");
243 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
244 : "\"%s/initdb\" --sync-only %s \"%s\" --sync-method %s",
245 0 : new_cluster.bindir,
246 0 : (user_opts.transfer_mode == TRANSFER_MODE_SWAP) ?
247 : "--no-sync-data-files" : "",
248 0 : new_cluster.pgdata,
249 0 : user_opts.sync_method);
250 0 : check_ok();
251 0 : }
252 :
253 0 : create_script_for_old_cluster_deletion(&deletion_script_file_name);
254 :
255 0 : issue_warnings_and_set_wal_level();
256 :
257 0 : pg_log(PG_REPORT,
258 : "\n"
259 : "Upgrade Complete\n"
260 : "----------------");
261 :
262 0 : output_completion_banner(deletion_script_file_name);
263 :
264 0 : pg_free(deletion_script_file_name);
265 :
266 0 : cleanup_output_dirs();
267 :
268 0 : return 0;
269 0 : }
270 :
271 : /*
272 : * Create and assign proper permissions to the set of output directories
273 : * used to store any data generated internally, filling in log_opts in
274 : * the process.
275 : */
276 : static void
277 0 : make_outputdirs(char *pgdata)
278 : {
279 0 : FILE *fp;
280 0 : char **filename;
281 0 : time_t run_time = time(NULL);
282 0 : char filename_path[MAXPGPATH];
283 0 : char timebuf[128];
284 0 : struct timeval time;
285 0 : time_t tt;
286 0 : int len;
287 :
288 0 : log_opts.rootdir = (char *) pg_malloc0(MAXPGPATH);
289 0 : len = snprintf(log_opts.rootdir, MAXPGPATH, "%s/%s", pgdata, BASE_OUTPUTDIR);
290 0 : if (len >= MAXPGPATH)
291 0 : pg_fatal("directory path for new cluster is too long");
292 :
293 : /* BASE_OUTPUTDIR/$timestamp/ */
294 0 : gettimeofday(&time, NULL);
295 0 : tt = (time_t) time.tv_sec;
296 0 : strftime(timebuf, sizeof(timebuf), "%Y%m%dT%H%M%S", localtime(&tt));
297 : /* append milliseconds */
298 0 : snprintf(timebuf + strlen(timebuf), sizeof(timebuf) - strlen(timebuf),
299 0 : ".%03d", (int) (time.tv_usec / 1000));
300 0 : log_opts.basedir = (char *) pg_malloc0(MAXPGPATH);
301 0 : len = snprintf(log_opts.basedir, MAXPGPATH, "%s/%s", log_opts.rootdir,
302 0 : timebuf);
303 0 : if (len >= MAXPGPATH)
304 0 : pg_fatal("directory path for new cluster is too long");
305 :
306 : /* BASE_OUTPUTDIR/$timestamp/dump/ */
307 0 : log_opts.dumpdir = (char *) pg_malloc0(MAXPGPATH);
308 0 : len = snprintf(log_opts.dumpdir, MAXPGPATH, "%s/%s/%s", log_opts.rootdir,
309 0 : timebuf, DUMP_OUTPUTDIR);
310 0 : if (len >= MAXPGPATH)
311 0 : pg_fatal("directory path for new cluster is too long");
312 :
313 : /* BASE_OUTPUTDIR/$timestamp/log/ */
314 0 : log_opts.logdir = (char *) pg_malloc0(MAXPGPATH);
315 0 : len = snprintf(log_opts.logdir, MAXPGPATH, "%s/%s/%s", log_opts.rootdir,
316 0 : timebuf, LOG_OUTPUTDIR);
317 0 : if (len >= MAXPGPATH)
318 0 : pg_fatal("directory path for new cluster is too long");
319 :
320 : /*
321 : * Ignore the error case where the root path exists, as it is kept the
322 : * same across runs.
323 : */
324 0 : if (mkdir(log_opts.rootdir, pg_dir_create_mode) < 0 && errno != EEXIST)
325 0 : pg_fatal("could not create directory \"%s\": %m", log_opts.rootdir);
326 0 : if (mkdir(log_opts.basedir, pg_dir_create_mode) < 0)
327 0 : pg_fatal("could not create directory \"%s\": %m", log_opts.basedir);
328 0 : if (mkdir(log_opts.dumpdir, pg_dir_create_mode) < 0)
329 0 : pg_fatal("could not create directory \"%s\": %m", log_opts.dumpdir);
330 0 : if (mkdir(log_opts.logdir, pg_dir_create_mode) < 0)
331 0 : pg_fatal("could not create directory \"%s\": %m", log_opts.logdir);
332 :
333 0 : len = snprintf(filename_path, sizeof(filename_path), "%s/%s",
334 0 : log_opts.logdir, INTERNAL_LOG_FILE);
335 0 : if (len >= sizeof(filename_path))
336 0 : pg_fatal("directory path for new cluster is too long");
337 :
338 0 : if ((log_opts.internal = fopen_priv(filename_path, "a")) == NULL)
339 0 : pg_fatal("could not open log file \"%s\": %m", filename_path);
340 :
341 : /* label start of upgrade in logfiles */
342 0 : for (filename = output_files; *filename != NULL; filename++)
343 : {
344 0 : len = snprintf(filename_path, sizeof(filename_path), "%s/%s",
345 0 : log_opts.logdir, *filename);
346 0 : if (len >= sizeof(filename_path))
347 0 : pg_fatal("directory path for new cluster is too long");
348 0 : if ((fp = fopen_priv(filename_path, "a")) == NULL)
349 0 : pg_fatal("could not write to log file \"%s\": %m", filename_path);
350 :
351 0 : fprintf(fp,
352 : "-----------------------------------------------------------------\n"
353 : " pg_upgrade run on %s"
354 : "-----------------------------------------------------------------\n\n",
355 0 : ctime(&run_time));
356 0 : fclose(fp);
357 0 : }
358 0 : }
359 :
360 :
361 : static void
362 0 : setup(char *argv0)
363 : {
364 : /*
365 : * make sure the user has a clean environment, otherwise, we may confuse
366 : * libpq when we connect to one (or both) of the servers.
367 : */
368 0 : check_pghost_envvar();
369 :
370 : /*
371 : * In case the user hasn't specified the directory for the new binaries
372 : * with -B, default to using the path of the currently executed pg_upgrade
373 : * binary.
374 : */
375 0 : if (!new_cluster.bindir)
376 : {
377 0 : char exec_path[MAXPGPATH];
378 :
379 0 : if (find_my_exec(argv0, exec_path) < 0)
380 0 : pg_fatal("%s: could not find own program executable", argv0);
381 : /* Trim off program name and keep just path */
382 0 : *last_dir_separator(exec_path) = '\0';
383 0 : canonicalize_path(exec_path);
384 0 : new_cluster.bindir = pg_strdup(exec_path);
385 0 : }
386 :
387 0 : verify_directories();
388 :
389 : /* no postmasters should be running, except for a live check */
390 0 : if (pid_lock_file_exists(old_cluster.pgdata))
391 : {
392 : /*
393 : * If we have a postmaster.pid file, try to start the server. If it
394 : * starts, the pid file was stale, so stop the server. If it doesn't
395 : * start, assume the server is running. If the pid file is left over
396 : * from a server crash, this also allows any committed transactions
397 : * stored in the WAL to be replayed so they are not lost, because WAL
398 : * files are not transferred from old to new servers. We later check
399 : * for a clean shutdown.
400 : */
401 0 : if (start_postmaster(&old_cluster, false))
402 0 : stop_postmaster(false);
403 : else
404 : {
405 0 : if (!user_opts.check)
406 0 : pg_fatal("There seems to be a postmaster servicing the old cluster.\n"
407 : "Please shutdown that postmaster and try again.");
408 : else
409 0 : user_opts.live_check = true;
410 : }
411 0 : }
412 :
413 : /* same goes for the new postmaster */
414 0 : if (pid_lock_file_exists(new_cluster.pgdata))
415 : {
416 0 : if (start_postmaster(&new_cluster, false))
417 0 : stop_postmaster(false);
418 : else
419 0 : pg_fatal("There seems to be a postmaster servicing the new cluster.\n"
420 : "Please shutdown that postmaster and try again.");
421 0 : }
422 0 : }
423 :
424 : /*
425 : * Set the new cluster's default char signedness using the old cluster's
426 : * value.
427 : */
428 : static void
429 0 : set_new_cluster_char_signedness(void)
430 : {
431 0 : bool new_char_signedness;
432 :
433 : /*
434 : * Use the specified char signedness if specified. Otherwise we inherit
435 : * the source database's signedness.
436 : */
437 0 : if (user_opts.char_signedness != -1)
438 0 : new_char_signedness = (user_opts.char_signedness == 1);
439 : else
440 0 : new_char_signedness = old_cluster.controldata.default_char_signedness;
441 :
442 : /* Change the char signedness of the new cluster, if necessary */
443 0 : if (new_cluster.controldata.default_char_signedness != new_char_signedness)
444 : {
445 0 : prep_status("Setting the default char signedness for new cluster");
446 :
447 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
448 : "\"%s/pg_resetwal\" --char-signedness %s \"%s\"",
449 0 : new_cluster.bindir,
450 0 : new_char_signedness ? "signed" : "unsigned",
451 0 : new_cluster.pgdata);
452 :
453 0 : check_ok();
454 0 : }
455 0 : }
456 :
457 : /*
458 : * Copy locale and encoding information into the new cluster's template0.
459 : *
460 : * We need to copy the encoding, datlocprovider, datcollate, datctype, and
461 : * datlocale. We don't need datcollversion because that's never set for
462 : * template0.
463 : */
464 : static void
465 0 : set_locale_and_encoding(void)
466 : {
467 0 : PGconn *conn_new_template1;
468 0 : char *datcollate_literal;
469 0 : char *datctype_literal;
470 0 : char *datlocale_literal = NULL;
471 0 : DbLocaleInfo *locale = old_cluster.template0;
472 :
473 0 : prep_status("Setting locale and encoding for new cluster");
474 :
475 : /* escape literals with respect to new cluster */
476 0 : conn_new_template1 = connectToServer(&new_cluster, "template1");
477 :
478 0 : datcollate_literal = PQescapeLiteral(conn_new_template1,
479 0 : locale->db_collate,
480 0 : strlen(locale->db_collate));
481 0 : datctype_literal = PQescapeLiteral(conn_new_template1,
482 0 : locale->db_ctype,
483 0 : strlen(locale->db_ctype));
484 :
485 0 : if (locale->db_locale)
486 0 : datlocale_literal = PQescapeLiteral(conn_new_template1,
487 0 : locale->db_locale,
488 0 : strlen(locale->db_locale));
489 : else
490 0 : datlocale_literal = "NULL";
491 :
492 : /* update template0 in new cluster */
493 0 : if (GET_MAJOR_VERSION(new_cluster.major_version) >= 1700)
494 0 : PQclear(executeQueryOrDie(conn_new_template1,
495 : "UPDATE pg_catalog.pg_database "
496 : " SET encoding = %d, "
497 : " datlocprovider = '%c', "
498 : " datcollate = %s, "
499 : " datctype = %s, "
500 : " datlocale = %s "
501 : " WHERE datname = 'template0' ",
502 0 : locale->db_encoding,
503 0 : locale->db_collprovider,
504 0 : datcollate_literal,
505 0 : datctype_literal,
506 0 : datlocale_literal));
507 0 : else if (GET_MAJOR_VERSION(new_cluster.major_version) >= 1500)
508 0 : PQclear(executeQueryOrDie(conn_new_template1,
509 : "UPDATE pg_catalog.pg_database "
510 : " SET encoding = %d, "
511 : " datlocprovider = '%c', "
512 : " datcollate = %s, "
513 : " datctype = %s, "
514 : " daticulocale = %s "
515 : " WHERE datname = 'template0' ",
516 0 : locale->db_encoding,
517 0 : locale->db_collprovider,
518 0 : datcollate_literal,
519 0 : datctype_literal,
520 0 : datlocale_literal));
521 : else
522 0 : PQclear(executeQueryOrDie(conn_new_template1,
523 : "UPDATE pg_catalog.pg_database "
524 : " SET encoding = %d, "
525 : " datcollate = %s, "
526 : " datctype = %s "
527 : " WHERE datname = 'template0' ",
528 0 : locale->db_encoding,
529 0 : datcollate_literal,
530 0 : datctype_literal));
531 :
532 0 : PQfreemem(datcollate_literal);
533 0 : PQfreemem(datctype_literal);
534 0 : if (locale->db_locale)
535 0 : PQfreemem(datlocale_literal);
536 :
537 0 : PQfinish(conn_new_template1);
538 :
539 0 : check_ok();
540 0 : }
541 :
542 :
543 : static void
544 0 : prepare_new_cluster(void)
545 : {
546 : /*
547 : * It would make more sense to freeze after loading the schema, but that
548 : * would cause us to lose the frozenxids restored by the load. We use
549 : * --analyze so autovacuum doesn't update statistics later
550 : */
551 0 : prep_status("Analyzing all rows in the new cluster");
552 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
553 : "\"%s/vacuumdb\" %s --all --analyze %s",
554 0 : new_cluster.bindir, cluster_conn_opts(&new_cluster),
555 0 : log_opts.verbose ? "--verbose" : "");
556 0 : check_ok();
557 :
558 : /*
559 : * We do freeze after analyze so pg_statistic is also frozen. template0 is
560 : * not frozen here, but data rows were frozen by initdb, and we set its
561 : * datfrozenxid, relfrozenxids, and relminmxid later to match the new xid
562 : * counter later.
563 : */
564 0 : prep_status("Freezing all rows in the new cluster");
565 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
566 : "\"%s/vacuumdb\" %s --all --freeze %s",
567 0 : new_cluster.bindir, cluster_conn_opts(&new_cluster),
568 0 : log_opts.verbose ? "--verbose" : "");
569 0 : check_ok();
570 0 : }
571 :
572 :
573 : static void
574 0 : prepare_new_globals(void)
575 : {
576 : /*
577 : * Before we restore anything, set frozenxids of initdb-created tables.
578 : */
579 0 : set_frozenxids(false);
580 :
581 : /*
582 : * Now restore global objects (roles and tablespaces).
583 : */
584 0 : prep_status("Restoring global objects in the new cluster");
585 :
586 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
587 : "\"%s/psql\" " EXEC_PSQL_ARGS " %s -f \"%s/%s\"",
588 0 : new_cluster.bindir, cluster_conn_opts(&new_cluster),
589 0 : log_opts.dumpdir,
590 : GLOBALS_DUMP_FILE);
591 0 : check_ok();
592 0 : }
593 :
594 :
595 : static void
596 0 : create_new_objects(void)
597 : {
598 0 : int dbnum;
599 0 : PGconn *conn_new_template1;
600 :
601 0 : prep_status_progress("Restoring database schemas in the new cluster");
602 :
603 : /*
604 : * Ensure that any changes to template0 are fully written out to disk
605 : * prior to restoring the databases. This is necessary because we use the
606 : * FILE_COPY strategy to create the databases (which testing has shown to
607 : * be faster), and when the server is in binary upgrade mode, it skips the
608 : * checkpoints this strategy ordinarily performs.
609 : */
610 0 : conn_new_template1 = connectToServer(&new_cluster, "template1");
611 0 : PQclear(executeQueryOrDie(conn_new_template1, "CHECKPOINT"));
612 0 : PQfinish(conn_new_template1);
613 :
614 : /*
615 : * We cannot process the template1 database concurrently with others,
616 : * because when it's transiently dropped, connection attempts would fail.
617 : * So handle it in a separate non-parallelized pass.
618 : */
619 0 : for (dbnum = 0; dbnum < old_cluster.dbarr.ndbs; dbnum++)
620 : {
621 0 : char sql_file_name[MAXPGPATH],
622 : log_file_name[MAXPGPATH];
623 0 : DbInfo *old_db = &old_cluster.dbarr.dbs[dbnum];
624 0 : const char *create_opts;
625 :
626 : /* Process only template1 in this pass */
627 0 : if (strcmp(old_db->db_name, "template1") != 0)
628 0 : continue;
629 :
630 0 : pg_log(PG_STATUS, "%s", old_db->db_name);
631 0 : snprintf(sql_file_name, sizeof(sql_file_name), DB_DUMP_FILE_MASK, old_db->db_oid);
632 0 : snprintf(log_file_name, sizeof(log_file_name), DB_DUMP_LOG_FILE_MASK, old_db->db_oid);
633 :
634 : /*
635 : * template1 database will already exist in the target installation,
636 : * so tell pg_restore to drop and recreate it; otherwise we would fail
637 : * to propagate its database-level properties.
638 : */
639 0 : create_opts = "--clean --create";
640 :
641 0 : exec_prog(log_file_name,
642 : NULL,
643 : true,
644 : true,
645 : "\"%s/pg_restore\" %s %s --exit-on-error --verbose "
646 : "--transaction-size=%d "
647 : "--dbname postgres \"%s/%s\"",
648 0 : new_cluster.bindir,
649 0 : cluster_conn_opts(&new_cluster),
650 0 : create_opts,
651 : RESTORE_TRANSACTION_SIZE,
652 0 : log_opts.dumpdir,
653 0 : sql_file_name);
654 :
655 0 : break; /* done once we've processed template1 */
656 0 : }
657 :
658 0 : for (dbnum = 0; dbnum < old_cluster.dbarr.ndbs; dbnum++)
659 : {
660 0 : char sql_file_name[MAXPGPATH],
661 : log_file_name[MAXPGPATH];
662 0 : DbInfo *old_db = &old_cluster.dbarr.dbs[dbnum];
663 0 : const char *create_opts;
664 0 : int txn_size;
665 :
666 : /* Skip template1 in this pass */
667 0 : if (strcmp(old_db->db_name, "template1") == 0)
668 0 : continue;
669 :
670 0 : pg_log(PG_STATUS, "%s", old_db->db_name);
671 0 : snprintf(sql_file_name, sizeof(sql_file_name), DB_DUMP_FILE_MASK, old_db->db_oid);
672 0 : snprintf(log_file_name, sizeof(log_file_name), DB_DUMP_LOG_FILE_MASK, old_db->db_oid);
673 :
674 : /*
675 : * postgres database will already exist in the target installation, so
676 : * tell pg_restore to drop and recreate it; otherwise we would fail to
677 : * propagate its database-level properties.
678 : */
679 0 : if (strcmp(old_db->db_name, "postgres") == 0)
680 0 : create_opts = "--clean --create";
681 : else
682 0 : create_opts = "--create";
683 :
684 : /*
685 : * In parallel mode, reduce the --transaction-size of each restore job
686 : * so that the total number of locks that could be held across all the
687 : * jobs stays in bounds.
688 : */
689 0 : txn_size = RESTORE_TRANSACTION_SIZE;
690 0 : if (user_opts.jobs > 1)
691 : {
692 0 : txn_size /= user_opts.jobs;
693 : /* Keep some sanity if -j is huge */
694 0 : txn_size = Max(txn_size, 10);
695 0 : }
696 :
697 0 : parallel_exec_prog(log_file_name,
698 : NULL,
699 : "\"%s/pg_restore\" %s %s --exit-on-error --verbose "
700 : "--transaction-size=%d "
701 : "--dbname template1 \"%s/%s\"",
702 0 : new_cluster.bindir,
703 0 : cluster_conn_opts(&new_cluster),
704 0 : create_opts,
705 0 : txn_size,
706 0 : log_opts.dumpdir,
707 0 : sql_file_name);
708 0 : }
709 :
710 : /* reap all children */
711 0 : while (reap_child(true) == true)
712 : ;
713 :
714 0 : end_progress_output();
715 0 : check_ok();
716 :
717 : /*
718 : * We don't have minmxids for databases or relations in pre-9.3 clusters,
719 : * so set those after we have restored the schema.
720 : */
721 0 : if (GET_MAJOR_VERSION(old_cluster.major_version) <= 902)
722 0 : set_frozenxids(true);
723 :
724 : /* update new_cluster info now that we have objects in the databases */
725 0 : get_db_rel_and_slot_infos(&new_cluster);
726 0 : }
727 :
728 : /*
729 : * Delete the given subdirectory contents from the new cluster
730 : */
731 : static void
732 0 : remove_new_subdir(const char *subdir, bool rmtopdir)
733 : {
734 0 : char new_path[MAXPGPATH];
735 :
736 0 : prep_status("Deleting files from new %s", subdir);
737 :
738 0 : snprintf(new_path, sizeof(new_path), "%s/%s", new_cluster.pgdata, subdir);
739 0 : if (!rmtree(new_path, rmtopdir))
740 0 : pg_fatal("could not delete directory \"%s\"", new_path);
741 :
742 0 : check_ok();
743 0 : }
744 :
745 : /*
746 : * Copy the files from the old cluster into it
747 : */
748 : static void
749 0 : copy_subdir_files(const char *old_subdir, const char *new_subdir)
750 : {
751 0 : char old_path[MAXPGPATH];
752 0 : char new_path[MAXPGPATH];
753 :
754 0 : remove_new_subdir(new_subdir, true);
755 :
756 0 : snprintf(old_path, sizeof(old_path), "%s/%s", old_cluster.pgdata, old_subdir);
757 0 : snprintf(new_path, sizeof(new_path), "%s/%s", new_cluster.pgdata, new_subdir);
758 :
759 0 : prep_status("Copying old %s to new server", old_subdir);
760 :
761 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
762 : #ifndef WIN32
763 : "cp -Rf \"%s\" \"%s\"",
764 : #else
765 : /* flags: everything, no confirm, quiet, overwrite read-only */
766 : "xcopy /e /y /q /r \"%s\" \"%s\\\"",
767 : #endif
768 0 : old_path, new_path);
769 :
770 0 : check_ok();
771 0 : }
772 :
773 : static void
774 0 : copy_xact_xlog_xid(void)
775 : {
776 : /*
777 : * Copy old commit logs to new data dir. pg_clog has been renamed to
778 : * pg_xact in post-10 clusters.
779 : */
780 0 : copy_subdir_files(GET_MAJOR_VERSION(old_cluster.major_version) <= 906 ?
781 : "pg_clog" : "pg_xact",
782 0 : GET_MAJOR_VERSION(new_cluster.major_version) <= 906 ?
783 : "pg_clog" : "pg_xact");
784 :
785 0 : prep_status("Setting oldest XID for new cluster");
786 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
787 : "\"%s/pg_resetwal\" -f -u %u \"%s\"",
788 0 : new_cluster.bindir, old_cluster.controldata.chkpnt_oldstxid,
789 0 : new_cluster.pgdata);
790 0 : check_ok();
791 :
792 : /* set the next transaction id and epoch of the new cluster */
793 0 : prep_status("Setting next transaction ID and epoch for new cluster");
794 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
795 : "\"%s/pg_resetwal\" -f -x %u \"%s\"",
796 0 : new_cluster.bindir, old_cluster.controldata.chkpnt_nxtxid,
797 0 : new_cluster.pgdata);
798 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
799 : "\"%s/pg_resetwal\" -f -e %u \"%s\"",
800 0 : new_cluster.bindir, old_cluster.controldata.chkpnt_nxtepoch,
801 0 : new_cluster.pgdata);
802 : /* must reset commit timestamp limits also */
803 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
804 : "\"%s/pg_resetwal\" -f -c %u,%u \"%s\"",
805 0 : new_cluster.bindir,
806 0 : old_cluster.controldata.chkpnt_nxtxid,
807 0 : old_cluster.controldata.chkpnt_nxtxid,
808 0 : new_cluster.pgdata);
809 0 : check_ok();
810 :
811 : /* Copy or convert pg_multixact files */
812 0 : Assert(new_cluster.controldata.cat_ver >= MULTIXACT_FORMATCHANGE_CAT_VER);
813 0 : Assert(new_cluster.controldata.cat_ver >= MULTIXACTOFFSET_FORMATCHANGE_CAT_VER);
814 0 : if (old_cluster.controldata.cat_ver >= MULTIXACTOFFSET_FORMATCHANGE_CAT_VER)
815 : {
816 : /* No change in multixact format, just copy the files */
817 0 : MultiXactId new_nxtmulti = old_cluster.controldata.chkpnt_nxtmulti;
818 0 : MultiXactOffset new_nxtmxoff = old_cluster.controldata.chkpnt_nxtmxoff;
819 :
820 0 : copy_subdir_files("pg_multixact/offsets", "pg_multixact/offsets");
821 0 : copy_subdir_files("pg_multixact/members", "pg_multixact/members");
822 :
823 0 : prep_status("Setting next multixact ID and offset for new cluster");
824 :
825 : /*
826 : * we preserve all files and contents, so we must preserve both "next"
827 : * counters here and the oldest multi present on system.
828 : */
829 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
830 : "\"%s/pg_resetwal\" -O %" PRIu64 " -m %u,%u \"%s\"",
831 0 : new_cluster.bindir, new_nxtmxoff, new_nxtmulti,
832 0 : old_cluster.controldata.chkpnt_oldstMulti,
833 0 : new_cluster.pgdata);
834 0 : check_ok();
835 0 : }
836 : else
837 : {
838 : /* Conversion is needed */
839 0 : MultiXactId nxtmulti;
840 0 : MultiXactId oldstMulti;
841 0 : MultiXactOffset nxtmxoff;
842 :
843 : /*
844 : * Determine the range of multixacts to convert.
845 : */
846 0 : nxtmulti = old_cluster.controldata.chkpnt_nxtmulti;
847 0 : if (old_cluster.controldata.cat_ver >= MULTIXACT_FORMATCHANGE_CAT_VER)
848 : {
849 : /* Versions 9.3 - 18: convert all multixids */
850 0 : oldstMulti = old_cluster.controldata.chkpnt_oldstMulti;
851 0 : }
852 : else
853 : {
854 : /*
855 : * In PostgreSQL 9.2 and below, multitransactions were only used
856 : * for row locking, and as such don't need to be preserved during
857 : * upgrade. In that case, we utilize rewrite_multixacts() just to
858 : * initialize new, empty files in the new format.
859 : *
860 : * It's important that the oldest multi is set to the latest value
861 : * used by the old system, so that multixact.c returns the empty
862 : * set for multis that might be present on disk.
863 : */
864 0 : oldstMulti = nxtmulti;
865 : }
866 : /* handle wraparound */
867 0 : if (nxtmulti < FirstMultiXactId)
868 0 : nxtmulti = FirstMultiXactId;
869 0 : if (oldstMulti < FirstMultiXactId)
870 0 : oldstMulti = FirstMultiXactId;
871 :
872 : /*
873 : * Remove the files created by initdb in the new cluster.
874 : * rewrite_multixacts() will create new ones.
875 : */
876 0 : remove_new_subdir("pg_multixact/members", false);
877 0 : remove_new_subdir("pg_multixact/offsets", false);
878 :
879 : /*
880 : * Create new pg_multixact files, converting old ones if needed.
881 : */
882 0 : prep_status("Converting pg_multixact files");
883 0 : nxtmxoff = rewrite_multixacts(oldstMulti, nxtmulti);
884 0 : check_ok();
885 :
886 0 : prep_status("Setting next multixact ID and offset for new cluster");
887 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
888 : "\"%s/pg_resetwal\" -O %" PRIu64 " -m %u,%u \"%s\"",
889 0 : new_cluster.bindir,
890 0 : nxtmxoff, nxtmulti, oldstMulti,
891 0 : new_cluster.pgdata);
892 0 : check_ok();
893 0 : }
894 :
895 : /* now reset the wal archives in the new cluster */
896 0 : prep_status("Resetting WAL archives");
897 0 : exec_prog(UTILITY_LOG_FILE, NULL, true, true,
898 : /* use timeline 1 to match controldata and no WAL history file */
899 0 : "\"%s/pg_resetwal\" -l 00000001%s \"%s\"", new_cluster.bindir,
900 : old_cluster.controldata.nextxlogfile + 8,
901 0 : new_cluster.pgdata);
902 0 : check_ok();
903 0 : }
904 :
905 :
906 : /*
907 : * set_frozenxids()
908 : *
909 : * This is called on the new cluster before we restore anything, with
910 : * minmxid_only = false. Its purpose is to ensure that all initdb-created
911 : * vacuumable tables have relfrozenxid/relminmxid matching the old cluster's
912 : * xid/mxid counters. We also initialize the datfrozenxid/datminmxid of the
913 : * built-in databases to match.
914 : *
915 : * As we create user tables later, their relfrozenxid/relminmxid fields will
916 : * be restored properly by the binary-upgrade restore script. Likewise for
917 : * user-database datfrozenxid/datminmxid. However, if we're upgrading from a
918 : * pre-9.3 database, which does not store per-table or per-DB minmxid, then
919 : * the relminmxid/datminmxid values filled in by the restore script will just
920 : * be zeroes.
921 : *
922 : * Hence, with a pre-9.3 source database, a second call occurs after
923 : * everything is restored, with minmxid_only = true. This pass will
924 : * initialize all tables and databases, both those made by initdb and user
925 : * objects, with the desired minmxid value. frozenxid values are left alone.
926 : */
927 : static void
928 0 : set_frozenxids(bool minmxid_only)
929 : {
930 0 : int dbnum;
931 0 : PGconn *conn,
932 : *conn_template1;
933 0 : PGresult *dbres;
934 0 : int ntups;
935 0 : int i_datname;
936 0 : int i_datallowconn;
937 :
938 0 : if (!minmxid_only)
939 0 : prep_status("Setting frozenxid and minmxid counters in new cluster");
940 : else
941 0 : prep_status("Setting minmxid counter in new cluster");
942 :
943 0 : conn_template1 = connectToServer(&new_cluster, "template1");
944 :
945 0 : if (!minmxid_only)
946 : /* set pg_database.datfrozenxid */
947 0 : PQclear(executeQueryOrDie(conn_template1,
948 : "UPDATE pg_catalog.pg_database "
949 : "SET datfrozenxid = '%u'",
950 0 : old_cluster.controldata.chkpnt_nxtxid));
951 :
952 : /* set pg_database.datminmxid */
953 0 : PQclear(executeQueryOrDie(conn_template1,
954 : "UPDATE pg_catalog.pg_database "
955 : "SET datminmxid = '%u'",
956 0 : old_cluster.controldata.chkpnt_nxtmulti));
957 :
958 : /* get database names */
959 0 : dbres = executeQueryOrDie(conn_template1,
960 : "SELECT datname, datallowconn "
961 : "FROM pg_catalog.pg_database");
962 :
963 0 : i_datname = PQfnumber(dbres, "datname");
964 0 : i_datallowconn = PQfnumber(dbres, "datallowconn");
965 :
966 0 : ntups = PQntuples(dbres);
967 0 : for (dbnum = 0; dbnum < ntups; dbnum++)
968 : {
969 0 : char *datname = PQgetvalue(dbres, dbnum, i_datname);
970 0 : char *datallowconn = PQgetvalue(dbres, dbnum, i_datallowconn);
971 :
972 : /*
973 : * We must update databases where datallowconn = false, e.g.
974 : * template0, because autovacuum increments their datfrozenxids,
975 : * relfrozenxids, and relminmxid even if autovacuum is turned off, and
976 : * even though all the data rows are already frozen. To enable this,
977 : * we temporarily change datallowconn.
978 : */
979 0 : if (strcmp(datallowconn, "f") == 0)
980 0 : PQclear(executeQueryOrDie(conn_template1,
981 : "ALTER DATABASE %s ALLOW_CONNECTIONS = true",
982 0 : quote_identifier(datname)));
983 :
984 0 : conn = connectToServer(&new_cluster, datname);
985 :
986 0 : if (!minmxid_only)
987 : /* set pg_class.relfrozenxid */
988 0 : PQclear(executeQueryOrDie(conn,
989 : "UPDATE pg_catalog.pg_class "
990 : "SET relfrozenxid = '%u' "
991 : /* only heap, materialized view, and TOAST are vacuumed */
992 : "WHERE relkind IN ("
993 : CppAsString2(RELKIND_RELATION) ", "
994 : CppAsString2(RELKIND_MATVIEW) ", "
995 : CppAsString2(RELKIND_TOASTVALUE) ")",
996 0 : old_cluster.controldata.chkpnt_nxtxid));
997 :
998 : /* set pg_class.relminmxid */
999 0 : PQclear(executeQueryOrDie(conn,
1000 : "UPDATE pg_catalog.pg_class "
1001 : "SET relminmxid = '%u' "
1002 : /* only heap, materialized view, and TOAST are vacuumed */
1003 : "WHERE relkind IN ("
1004 : CppAsString2(RELKIND_RELATION) ", "
1005 : CppAsString2(RELKIND_MATVIEW) ", "
1006 : CppAsString2(RELKIND_TOASTVALUE) ")",
1007 0 : old_cluster.controldata.chkpnt_nxtmulti));
1008 0 : PQfinish(conn);
1009 :
1010 : /* Reset datallowconn flag */
1011 0 : if (strcmp(datallowconn, "f") == 0)
1012 0 : PQclear(executeQueryOrDie(conn_template1,
1013 : "ALTER DATABASE %s ALLOW_CONNECTIONS = false",
1014 0 : quote_identifier(datname)));
1015 0 : }
1016 :
1017 0 : PQclear(dbres);
1018 :
1019 0 : PQfinish(conn_template1);
1020 :
1021 0 : check_ok();
1022 0 : }
1023 :
1024 : /*
1025 : * create_logical_replication_slots()
1026 : *
1027 : * Similar to create_new_objects() but only restores logical replication slots.
1028 : */
1029 : static void
1030 0 : create_logical_replication_slots(void)
1031 : {
1032 0 : prep_status_progress("Restoring logical replication slots in the new cluster");
1033 :
1034 0 : for (int dbnum = 0; dbnum < old_cluster.dbarr.ndbs; dbnum++)
1035 : {
1036 0 : DbInfo *old_db = &old_cluster.dbarr.dbs[dbnum];
1037 0 : LogicalSlotInfoArr *slot_arr = &old_db->slot_arr;
1038 0 : PGconn *conn;
1039 0 : PQExpBuffer query;
1040 :
1041 : /* Skip this database if there are no slots */
1042 0 : if (slot_arr->nslots == 0)
1043 0 : continue;
1044 :
1045 0 : conn = connectToServer(&new_cluster, old_db->db_name);
1046 0 : query = createPQExpBuffer();
1047 :
1048 0 : pg_log(PG_STATUS, "%s", old_db->db_name);
1049 :
1050 0 : for (int slotnum = 0; slotnum < slot_arr->nslots; slotnum++)
1051 : {
1052 0 : LogicalSlotInfo *slot_info = &slot_arr->slots[slotnum];
1053 :
1054 : /* Constructs a query for creating logical replication slots */
1055 0 : appendPQExpBufferStr(query,
1056 : "SELECT * FROM "
1057 : "pg_catalog.pg_create_logical_replication_slot(");
1058 0 : appendStringLiteralConn(query, slot_info->slotname, conn);
1059 0 : appendPQExpBufferStr(query, ", ");
1060 0 : appendStringLiteralConn(query, slot_info->plugin, conn);
1061 :
1062 0 : appendPQExpBuffer(query, ", false, %s, %s);",
1063 0 : slot_info->two_phase ? "true" : "false",
1064 0 : slot_info->failover ? "true" : "false");
1065 :
1066 0 : PQclear(executeQueryOrDie(conn, "%s", query->data));
1067 :
1068 0 : resetPQExpBuffer(query);
1069 0 : }
1070 :
1071 0 : PQfinish(conn);
1072 :
1073 0 : destroyPQExpBuffer(query);
1074 0 : }
1075 :
1076 0 : end_progress_output();
1077 0 : check_ok();
1078 :
1079 0 : return;
1080 : }
1081 :
1082 : /*
1083 : * create_conflict_detection_slot()
1084 : *
1085 : * Create a replication slot to retain information necessary for conflict
1086 : * detection such as dead tuples, commit timestamps, and origins, for migrated
1087 : * subscriptions with retain_dead_tuples enabled.
1088 : */
1089 : static void
1090 0 : create_conflict_detection_slot(void)
1091 : {
1092 0 : PGconn *conn_new_template1;
1093 :
1094 0 : prep_status("Creating the replication conflict detection slot");
1095 :
1096 0 : conn_new_template1 = connectToServer(&new_cluster, "template1");
1097 0 : PQclear(executeQueryOrDie(conn_new_template1, "SELECT pg_catalog.binary_upgrade_create_conflict_detection_slot()"));
1098 0 : PQfinish(conn_new_template1);
1099 :
1100 0 : check_ok();
1101 0 : }
|
