Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * pg_combinebackup.c
4 : : * Combine incremental backups with prior backups.
5 : : *
6 : : * Copyright (c) 2017-2025, PostgreSQL Global Development Group
7 : : *
8 : : * IDENTIFICATION
9 : : * src/bin/pg_combinebackup/pg_combinebackup.c
10 : : *
11 : : *-------------------------------------------------------------------------
12 : : */
13 : : #include "postgres_fe.h"
14 : :
15 : : #include <dirent.h>
16 : : #include <fcntl.h>
17 : : #include <limits.h>
18 : :
19 : : #ifdef HAVE_COPYFILE_H
20 : : #include <copyfile.h>
21 : : #endif
22 : : #ifdef __linux__
23 : : #include <sys/ioctl.h>
24 : : #include <linux/fs.h>
25 : : #endif
26 : :
27 : : #include "access/xlog_internal.h"
28 : : #include "backup_label.h"
29 : : #include "common/checksum_helper.h"
30 : : #include "common/controldata_utils.h"
31 : : #include "common/file_perm.h"
32 : : #include "common/file_utils.h"
33 : : #include "common/logging.h"
34 : : #include "common/relpath.h"
35 : : #include "copy_file.h"
36 : : #include "fe_utils/option_utils.h"
37 : : #include "getopt_long.h"
38 : : #include "lib/stringinfo.h"
39 : : #include "load_manifest.h"
40 : : #include "reconstruct.h"
41 : : #include "write_manifest.h"
42 : :
43 : : /* Incremental file naming convention. */
44 : : #define INCREMENTAL_PREFIX "INCREMENTAL."
45 : : #define INCREMENTAL_PREFIX_LENGTH (sizeof(INCREMENTAL_PREFIX) - 1)
46 : :
47 : : /*
48 : : * Tracking for directories that need to be removed, or have their contents
49 : : * removed, if the operation fails.
50 : : */
51 : : typedef struct cb_cleanup_dir
52 : : {
53 : : char *target_path;
54 : : bool rmtopdir;
55 : : struct cb_cleanup_dir *next;
56 : : } cb_cleanup_dir;
57 : :
58 : : /*
59 : : * Stores a tablespace mapping provided using -T, --tablespace-mapping.
60 : : */
61 : : typedef struct cb_tablespace_mapping
62 : : {
63 : : char old_dir[MAXPGPATH];
64 : : char new_dir[MAXPGPATH];
65 : : struct cb_tablespace_mapping *next;
66 : : } cb_tablespace_mapping;
67 : :
68 : : /*
69 : : * Stores data parsed from all command-line options.
70 : : */
71 : : typedef struct cb_options
72 : : {
73 : : bool debug;
74 : : char *output;
75 : : bool dry_run;
76 : : bool no_sync;
77 : : cb_tablespace_mapping *tsmappings;
78 : : pg_checksum_type manifest_checksums;
79 : : bool no_manifest;
80 : : DataDirSyncMethod sync_method;
81 : : CopyMethod copy_method;
82 : : } cb_options;
83 : :
84 : : /*
85 : : * Data about a tablespace.
86 : : *
87 : : * Every normal tablespace needs a tablespace mapping, but in-place tablespaces
88 : : * don't, so the list of tablespaces can contain more entries than the list of
89 : : * tablespace mappings.
90 : : */
91 : : typedef struct cb_tablespace
92 : : {
93 : : Oid oid;
94 : : bool in_place;
95 : : char old_dir[MAXPGPATH];
96 : : char new_dir[MAXPGPATH];
97 : : struct cb_tablespace *next;
98 : : } cb_tablespace;
99 : :
100 : : /* Directories to be removed if we exit uncleanly. */
101 : : static cb_cleanup_dir *cleanup_dir_list = NULL;
102 : :
103 : : static void add_tablespace_mapping(cb_options *opt, char *arg);
104 : : static StringInfo check_backup_label_files(int n_backups, char **backup_dirs);
105 : : static uint64 check_control_files(int n_backups, char **backup_dirs);
106 : : static void check_input_dir_permissions(char *dir);
107 : : static void cleanup_directories_atexit(void);
108 : : static void create_output_directory(char *dirname, cb_options *opt);
109 : : static void help(const char *progname);
110 : : static bool parse_oid(char *s, Oid *result);
111 : : static void process_directory_recursively(Oid tsoid,
112 : : char *input_directory,
113 : : char *output_directory,
114 : : char *relative_path,
115 : : int n_prior_backups,
116 : : char **prior_backup_dirs,
117 : : manifest_data **manifests,
118 : : manifest_writer *mwriter,
119 : : cb_options *opt);
120 : : static int read_pg_version_file(char *directory);
121 : : static void remember_to_cleanup_directory(char *target_path, bool rmtopdir);
122 : : static void reset_directory_cleanup_list(void);
123 : : static cb_tablespace *scan_for_existing_tablespaces(char *pathname,
124 : : cb_options *opt);
125 : : static void slurp_file(int fd, char *filename, StringInfo buf, int maxlen);
126 : :
127 : : /*
128 : : * Main program.
129 : : */
130 : : int
626 rhaas@postgresql.org 131 :CBC 25 : main(int argc, char *argv[])
132 : : {
133 : : static struct option long_options[] = {
134 : : {"debug", no_argument, NULL, 'd'},
135 : : {"dry-run", no_argument, NULL, 'n'},
136 : : {"no-sync", no_argument, NULL, 'N'},
137 : : {"output", required_argument, NULL, 'o'},
138 : : {"tablespace-mapping", required_argument, NULL, 'T'},
139 : : {"link", no_argument, NULL, 'k'},
140 : : {"manifest-checksums", required_argument, NULL, 1},
141 : : {"no-manifest", no_argument, NULL, 2},
142 : : {"sync-method", required_argument, NULL, 3},
143 : : {"clone", no_argument, NULL, 4},
144 : : {"copy", no_argument, NULL, 5},
145 : : {"copy-file-range", no_argument, NULL, 6},
146 : : {NULL, 0, NULL, 0}
147 : : };
148 : :
149 : : const char *progname;
150 : : char *last_input_dir;
151 : : int i;
152 : : int optindex;
153 : : int c;
154 : : int n_backups;
155 : : int n_prior_backups;
156 : : int version;
157 : : uint64 system_identifier;
158 : : char **prior_backup_dirs;
159 : : cb_options opt;
160 : : cb_tablespace *tablespaces;
161 : : cb_tablespace *ts;
162 : : StringInfo last_backup_label;
163 : : manifest_data **manifests;
164 : : manifest_writer *mwriter;
165 : :
166 : 25 : pg_logging_init(argv[0]);
167 : 25 : progname = get_progname(argv[0]);
515 michael@paquier.xyz 168 : 25 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_combinebackup"));
626 rhaas@postgresql.org 169 : 25 : handle_help_version_opts(argc, argv, progname, help);
170 : :
171 : 23 : memset(&opt, 0, sizeof(opt));
172 : 23 : opt.manifest_checksums = CHECKSUM_TYPE_CRC32C;
173 : 23 : opt.sync_method = DATA_DIR_SYNC_METHOD_FSYNC;
519 tomas.vondra@postgre 174 : 23 : opt.copy_method = COPY_METHOD_COPY;
175 : :
176 : : /* process command-line options */
173 rhaas@postgresql.org 177 : 72 : while ((c = getopt_long(argc, argv, "dknNo:T:",
626 178 [ + + ]: 72 : long_options, &optindex)) != -1)
179 : : {
180 [ + + - + : 50 : switch (c)
+ + + + -
- + - + ]
181 : : {
182 : 6 : case 'd':
183 : 6 : opt.debug = true;
184 : 6 : pg_logging_increase_verbosity();
185 : 6 : break;
173 186 : 1 : case 'k':
187 : 1 : opt.copy_method = COPY_METHOD_LINK;
188 : 1 : break;
626 rhaas@postgresql.org 189 :UBC 0 : case 'n':
190 : 0 : opt.dry_run = true;
191 : 0 : break;
626 rhaas@postgresql.org 192 :CBC 3 : case 'N':
193 : 3 : opt.no_sync = true;
194 : 3 : break;
195 : 20 : case 'o':
196 : 20 : opt.output = optarg;
197 : 20 : break;
198 : 1 : case 'T':
199 : 1 : add_tablespace_mapping(&opt, optarg);
200 : 1 : break;
201 : 2 : case 1:
202 [ - + ]: 2 : if (!pg_checksum_parse_type(optarg,
203 : : &opt.manifest_checksums))
626 rhaas@postgresql.org 204 :UBC 0 : pg_fatal("unrecognized checksum algorithm: \"%s\"",
205 : : optarg);
626 rhaas@postgresql.org 206 :CBC 2 : break;
207 : 1 : case 2:
208 : 1 : opt.no_manifest = true;
209 : 1 : break;
626 rhaas@postgresql.org 210 :UBC 0 : case 3:
211 [ # # ]: 0 : if (!parse_sync_method(optarg, &opt.sync_method))
212 : 0 : exit(1);
213 : 0 : break;
519 tomas.vondra@postgre 214 : 0 : case 4:
215 : 0 : opt.copy_method = COPY_METHOD_CLONE;
216 : 0 : break;
519 tomas.vondra@postgre 217 :CBC 15 : case 5:
433 218 : 15 : opt.copy_method = COPY_METHOD_COPY;
219 : 15 : break;
433 tomas.vondra@postgre 220 :UBC 0 : case 6:
519 221 : 0 : opt.copy_method = COPY_METHOD_COPY_FILE_RANGE;
222 : 0 : break;
626 rhaas@postgresql.org 223 :CBC 1 : default:
224 : : /* getopt_long already emitted a complaint */
225 : 1 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
226 : 1 : exit(1);
227 : : }
228 : : }
229 : :
230 [ + + ]: 22 : if (optind >= argc)
231 : : {
443 peter@eisentraut.org 232 : 1 : pg_log_error("no input directories specified");
626 rhaas@postgresql.org 233 : 1 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
234 : 1 : exit(1);
235 : : }
236 : :
237 [ + + ]: 21 : if (opt.output == NULL)
238 : 1 : pg_fatal("no output directory specified");
239 : :
240 : : /* If no manifest is needed, no checksums are needed, either. */
241 [ + + ]: 20 : if (opt.no_manifest)
242 : 1 : opt.manifest_checksums = CHECKSUM_TYPE_NONE;
243 : :
244 : : /* Check that the platform supports the requested copy method. */
519 tomas.vondra@postgre 245 [ - + ]: 20 : if (opt.copy_method == COPY_METHOD_CLONE)
246 : : {
247 : : #if (defined(HAVE_COPYFILE) && defined(COPYFILE_CLONE_FORCE)) || \
248 : : (defined(__linux__) && defined(FICLONE))
249 : :
519 tomas.vondra@postgre 250 [ # # ]:UBC 0 : if (opt.dry_run)
251 [ # # ]: 0 : pg_log_debug("would use cloning to copy files");
252 : : else
253 [ # # ]: 0 : pg_log_debug("will use cloning to copy files");
254 : :
255 : : #else
256 : : pg_fatal("file cloning not supported on this platform");
257 : : #endif
258 : : }
519 tomas.vondra@postgre 259 [ - + ]:CBC 20 : else if (opt.copy_method == COPY_METHOD_COPY_FILE_RANGE)
260 : : {
261 : : #if defined(HAVE_COPY_FILE_RANGE)
262 : :
519 tomas.vondra@postgre 263 [ # # ]:UBC 0 : if (opt.dry_run)
264 [ # # ]: 0 : pg_log_debug("would use copy_file_range to copy blocks");
265 : : else
266 [ # # ]: 0 : pg_log_debug("will use copy_file_range to copy blocks");
267 : :
268 : : #else
269 : : pg_fatal("copy_file_range not supported on this platform");
270 : : #endif
271 : : }
272 : :
273 : : /* Read the server version from the final backup. */
626 rhaas@postgresql.org 274 :CBC 20 : version = read_pg_version_file(argv[argc - 1]);
275 : :
276 : : /* Sanity-check control files. */
277 : 20 : n_backups = argc - optind;
542 278 : 20 : system_identifier = check_control_files(n_backups, argv + optind);
279 : :
280 : : /* Sanity-check backup_label files, and get the contents of the last one. */
626 281 : 19 : last_backup_label = check_backup_label_files(n_backups, argv + optind);
282 : :
283 : : /*
284 : : * We'll need the pathnames to the prior backups. By "prior" we mean all
285 : : * but the last one listed on the command line.
286 : : */
287 : 14 : n_prior_backups = argc - optind - 1;
288 : 14 : prior_backup_dirs = argv + optind;
289 : :
290 : : /* Load backup manifests. */
291 : 14 : manifests = load_backup_manifests(n_backups, prior_backup_dirs);
292 : :
293 : : /*
294 : : * Validate the manifest system identifier against the backup system
295 : : * identifier.
296 : : */
542 297 [ + + ]: 40 : for (i = 0; i < n_backups; i++)
298 : : {
299 [ + - ]: 27 : if (manifests[i] &&
300 [ + + ]: 27 : manifests[i]->system_identifier != system_identifier)
301 : : {
302 : : char *controlpath;
303 : :
152 fujii@postgresql.org 304 : 1 : controlpath = psprintf("%s/%s", prior_backup_dirs[i], XLOG_CONTROL_FILE);
305 : :
161 peter@eisentraut.org 306 : 1 : pg_fatal("%s: manifest system identifier is %" PRIu64 ", but control file has %" PRIu64,
307 : : controlpath,
308 : : manifests[i]->system_identifier,
309 : : system_identifier);
310 : : }
311 : : }
312 : :
313 : : /* Figure out which tablespaces are going to be included in the output. */
626 rhaas@postgresql.org 314 : 13 : last_input_dir = argv[argc - 1];
315 : 13 : check_input_dir_permissions(last_input_dir);
316 : 13 : tablespaces = scan_for_existing_tablespaces(last_input_dir, &opt);
317 : :
318 : : /*
319 : : * Create output directories.
320 : : *
321 : : * We create one output directory for the main data directory plus one for
322 : : * each non-in-place tablespace. create_output_directory() will arrange
323 : : * for those directories to be cleaned up on failure. In-place tablespaces
324 : : * aren't handled at this stage because they're located beneath the main
325 : : * output directory, and thus the cleanup of that directory will get rid
326 : : * of them. Plus, the pg_tblspc directory that needs to contain them
327 : : * doesn't exist yet.
328 : : */
329 : 13 : atexit(cleanup_directories_atexit);
330 : 13 : create_output_directory(opt.output, &opt);
331 [ + + ]: 14 : for (ts = tablespaces; ts != NULL; ts = ts->next)
332 [ + - ]: 1 : if (!ts->in_place)
333 : 1 : create_output_directory(ts->new_dir, &opt);
334 : :
335 : : /* If we need to write a backup_manifest, prepare to do so. */
336 [ + - + + ]: 13 : if (!opt.dry_run && !opt.no_manifest)
337 : : {
542 338 : 12 : mwriter = create_manifest_writer(opt.output, system_identifier);
339 : :
340 : : /*
341 : : * Verify that we have a backup manifest for the final backup; else we
342 : : * won't have the WAL ranges for the resulting manifest.
343 : : */
626 344 [ - + ]: 12 : if (manifests[n_prior_backups] == NULL)
375 peter@eisentraut.org 345 :UBC 0 : pg_fatal("cannot generate a manifest because no manifest is available for the final input backup");
346 : : }
347 : : else
626 rhaas@postgresql.org 348 :CBC 1 : mwriter = NULL;
349 : :
350 : : /* Write backup label into output directory. */
351 [ - + ]: 13 : if (opt.dry_run)
626 rhaas@postgresql.org 352 [ # # ]:UBC 0 : pg_log_debug("would generate \"%s/backup_label\"", opt.output);
353 : : else
354 : : {
626 rhaas@postgresql.org 355 [ + + ]:CBC 13 : pg_log_debug("generating \"%s/backup_label\"", opt.output);
356 : 13 : last_backup_label->cursor = 0;
357 : 13 : write_backup_label(opt.output, last_backup_label,
358 : : opt.manifest_checksums, mwriter);
359 : : }
360 : :
361 : : /* Process everything that's not part of a user-defined tablespace. */
362 [ + + ]: 13 : pg_log_debug("processing backup directory \"%s\"", last_input_dir);
363 : 13 : process_directory_recursively(InvalidOid, last_input_dir, opt.output,
364 : : NULL, n_prior_backups, prior_backup_dirs,
365 : : manifests, mwriter, &opt);
366 : :
367 : : /* Process user-defined tablespaces. */
368 [ + + ]: 13 : for (ts = tablespaces; ts != NULL; ts = ts->next)
369 : : {
370 [ + - ]: 1 : pg_log_debug("processing tablespace directory \"%s\"", ts->old_dir);
371 : :
372 : : /*
373 : : * If it's a normal tablespace, we need to set up a symbolic link from
374 : : * pg_tblspc/${OID} to the target directory; if it's an in-place
375 : : * tablespace, we need to create a directory at pg_tblspc/${OID}.
376 : : */
377 [ + - ]: 1 : if (!ts->in_place)
378 : : {
379 : : char linkpath[MAXPGPATH];
380 : :
368 michael@paquier.xyz 381 : 1 : snprintf(linkpath, MAXPGPATH, "%s/%s/%u", opt.output, PG_TBLSPC_DIR,
382 : : ts->oid);
383 : :
626 rhaas@postgresql.org 384 [ - + ]: 1 : if (opt.dry_run)
626 rhaas@postgresql.org 385 [ # # ]:UBC 0 : pg_log_debug("would create symbolic link from \"%s\" to \"%s\"",
386 : : linkpath, ts->new_dir);
387 : : else
388 : : {
626 rhaas@postgresql.org 389 [ + - ]:CBC 1 : pg_log_debug("creating symbolic link from \"%s\" to \"%s\"",
390 : : linkpath, ts->new_dir);
391 [ - + ]: 1 : if (symlink(ts->new_dir, linkpath) != 0)
626 rhaas@postgresql.org 392 :UBC 0 : pg_fatal("could not create symbolic link from \"%s\" to \"%s\": %m",
393 : : linkpath, ts->new_dir);
394 : : }
395 : : }
396 : : else
397 : : {
398 [ # # ]: 0 : if (opt.dry_run)
399 [ # # ]: 0 : pg_log_debug("would create directory \"%s\"", ts->new_dir);
400 : : else
401 : : {
402 [ # # ]: 0 : pg_log_debug("creating directory \"%s\"", ts->new_dir);
403 [ # # ]: 0 : if (pg_mkdir_p(ts->new_dir, pg_dir_create_mode) == -1)
404 : 0 : pg_fatal("could not create directory \"%s\": %m",
405 : : ts->new_dir);
406 : : }
407 : : }
408 : :
409 : : /* OK, now handle the directory contents. */
626 rhaas@postgresql.org 410 :CBC 1 : process_directory_recursively(ts->oid, ts->old_dir, ts->new_dir,
411 : : NULL, n_prior_backups, prior_backup_dirs,
412 : : manifests, mwriter, &opt);
413 : : }
414 : :
415 : : /* Finalize the backup_manifest, if we're generating one. */
416 [ + + ]: 12 : if (mwriter != NULL)
417 : 11 : finalize_manifest(mwriter,
418 : 11 : manifests[n_prior_backups]->first_wal_range);
419 : :
420 : : /* fsync that output directory unless we've been told not to do so */
421 [ + + ]: 12 : if (!opt.no_sync)
422 : : {
423 [ - + ]: 9 : if (opt.dry_run)
626 rhaas@postgresql.org 424 [ # # ]:UBC 0 : pg_log_debug("would recursively fsync \"%s\"", opt.output);
425 : : else
426 : : {
626 rhaas@postgresql.org 427 [ + + ]:CBC 9 : pg_log_debug("recursively fsyncing \"%s\"", opt.output);
165 nathan@postgresql.or 428 : 9 : sync_pgdata(opt.output, version * 10000, opt.sync_method, true);
429 : : }
430 : : }
431 : :
432 : : /* Warn about the possibility of compromising the backups, when link mode */
173 rhaas@postgresql.org 433 [ + + ]: 12 : if (opt.copy_method == COPY_METHOD_LINK)
434 : 1 : pg_log_warning("--link mode was used; any modifications to the output "
435 : : "directory might destructively modify input directories");
436 : :
437 : : /* It's a success, so don't remove the output directories. */
626 438 : 12 : reset_directory_cleanup_list();
439 : 12 : exit(0);
440 : : }
441 : :
442 : : /*
443 : : * Process the option argument for the -T, --tablespace-mapping switch.
444 : : */
445 : : static void
446 : 1 : add_tablespace_mapping(cb_options *opt, char *arg)
447 : : {
448 : 1 : cb_tablespace_mapping *tsmap = pg_malloc0(sizeof(cb_tablespace_mapping));
449 : : char *dst;
450 : : char *dst_ptr;
451 : : char *arg_ptr;
452 : :
453 : : /*
454 : : * Basically, we just want to copy everything before the equals sign to
455 : : * tsmap->old_dir and everything afterwards to tsmap->new_dir, but if
456 : : * there's more or less than one equals sign, that's an error, and if
457 : : * there's an equals sign preceded by a backslash, don't treat it as a
458 : : * field separator but instead copy a literal equals sign.
459 : : */
460 : 1 : dst_ptr = dst = tsmap->old_dir;
461 [ + + ]: 50 : for (arg_ptr = arg; *arg_ptr != '\0'; arg_ptr++)
462 : : {
463 [ - + ]: 49 : if (dst_ptr - dst >= MAXPGPATH)
626 rhaas@postgresql.org 464 :UBC 0 : pg_fatal("directory name too long");
465 : :
626 rhaas@postgresql.org 466 [ - + - - ]:CBC 49 : if (*arg_ptr == '\\' && *(arg_ptr + 1) == '=')
467 : : ; /* skip backslash escaping = */
468 [ + + + - : 49 : else if (*arg_ptr == '=' && (arg_ptr == arg || *(arg_ptr - 1) != '\\'))
+ - ]
469 : : {
470 [ - + ]: 1 : if (tsmap->new_dir[0] != '\0')
626 rhaas@postgresql.org 471 :UBC 0 : pg_fatal("multiple \"=\" signs in tablespace mapping");
472 : : else
626 rhaas@postgresql.org 473 :CBC 1 : dst = dst_ptr = tsmap->new_dir;
474 : : }
475 : : else
476 : 48 : *dst_ptr++ = *arg_ptr;
477 : : }
478 [ + - - + ]: 1 : if (!tsmap->old_dir[0] || !tsmap->new_dir[0])
626 rhaas@postgresql.org 479 :UBC 0 : pg_fatal("invalid tablespace mapping format \"%s\", must be \"OLDDIR=NEWDIR\"", arg);
480 : :
481 : : /*
482 : : * All tablespaces are created with absolute directories, so specifying a
483 : : * non-absolute path here would never match, possibly confusing users.
484 : : *
485 : : * In contrast to pg_basebackup, both the old and new directories are on
486 : : * the local machine, so the local machine's definition of an absolute
487 : : * path is the only relevant one.
488 : : */
626 rhaas@postgresql.org 489 [ - + ]:CBC 1 : if (!is_absolute_path(tsmap->old_dir))
626 rhaas@postgresql.org 490 :UBC 0 : pg_fatal("old directory is not an absolute path in tablespace mapping: %s",
491 : : tsmap->old_dir);
492 : :
626 rhaas@postgresql.org 493 [ - + ]:CBC 1 : if (!is_absolute_path(tsmap->new_dir))
626 rhaas@postgresql.org 494 :UBC 0 : pg_fatal("old directory is not an absolute path in tablespace mapping: %s",
495 : : tsmap->new_dir);
496 : :
497 : : /* Canonicalize paths to avoid spurious failures when comparing. */
626 rhaas@postgresql.org 498 :CBC 1 : canonicalize_path(tsmap->old_dir);
499 : 1 : canonicalize_path(tsmap->new_dir);
500 : :
501 : : /* Add it to the list. */
502 : 1 : tsmap->next = opt->tsmappings;
503 : 1 : opt->tsmappings = tsmap;
504 : 1 : }
505 : :
506 : : /*
507 : : * Check that the backup_label files form a coherent backup chain, and return
508 : : * the contents of the backup_label file from the latest backup.
509 : : */
510 : : static StringInfo
511 : 19 : check_backup_label_files(int n_backups, char **backup_dirs)
512 : : {
513 : 19 : StringInfo buf = makeStringInfo();
514 : 19 : StringInfo lastbuf = buf;
515 : : int i;
516 : 19 : TimeLineID check_tli = 0;
517 : 19 : XLogRecPtr check_lsn = InvalidXLogRecPtr;
518 : :
519 : : /* Try to read each backup_label file in turn, last to first. */
520 [ + + ]: 51 : for (i = n_backups - 1; i >= 0; --i)
521 : : {
522 : : char pathbuf[MAXPGPATH];
523 : : int fd;
524 : : TimeLineID start_tli;
525 : : TimeLineID previous_tli;
526 : : XLogRecPtr start_lsn;
527 : : XLogRecPtr previous_lsn;
528 : :
529 : : /* Open the backup_label file. */
530 : 37 : snprintf(pathbuf, MAXPGPATH, "%s/backup_label", backup_dirs[i]);
531 [ + + ]: 37 : pg_log_debug("reading \"%s\"", pathbuf);
532 [ - + ]: 37 : if ((fd = open(pathbuf, O_RDONLY, 0)) < 0)
626 rhaas@postgresql.org 533 :UBC 0 : pg_fatal("could not open file \"%s\": %m", pathbuf);
534 : :
535 : : /*
536 : : * Slurp the whole file into memory.
537 : : *
538 : : * The exact size limit that we impose here doesn't really matter --
539 : : * most of what's supposed to be in the file is fixed size and quite
540 : : * short. However, the length of the backup_label is limited (at least
541 : : * by some parts of the code) to MAXPGPATH, so include that value in
542 : : * the maximum length that we tolerate.
543 : : */
626 rhaas@postgresql.org 544 :CBC 37 : slurp_file(fd, pathbuf, buf, 10000 + MAXPGPATH);
545 : :
546 : : /* Close the file. */
547 [ - + ]: 37 : if (close(fd) != 0)
442 peter@eisentraut.org 548 :UBC 0 : pg_fatal("could not close file \"%s\": %m", pathbuf);
549 : :
550 : : /* Parse the file contents. */
626 rhaas@postgresql.org 551 :CBC 37 : parse_backup_label(pathbuf, buf, &start_tli, &start_lsn,
552 : : &previous_tli, &previous_lsn);
553 : :
554 : : /*
555 : : * Sanity checks.
556 : : *
557 : : * XXX. It's actually not required that start_lsn == check_lsn. It
558 : : * would be OK if start_lsn > check_lsn provided that start_lsn is
559 : : * less than or equal to the relevant switchpoint. But at the moment
560 : : * we don't have that information.
561 : : */
562 [ + + + + ]: 37 : if (i > 0 && previous_tli == 0)
563 : 1 : pg_fatal("backup at \"%s\" is a full backup, but only the first backup should be a full backup",
564 : : backup_dirs[i]);
565 [ + + + + ]: 36 : if (i == 0 && previous_tli != 0)
566 : 1 : pg_fatal("backup at \"%s\" is an incremental backup, but the first backup should be a full backup",
567 : : backup_dirs[i]);
568 [ + + - + ]: 35 : if (i < n_backups - 1 && start_tli != check_tli)
626 rhaas@postgresql.org 569 :UBC 0 : pg_fatal("backup at \"%s\" starts on timeline %u, but expected %u",
570 : : backup_dirs[i], start_tli, check_tli);
626 rhaas@postgresql.org 571 [ + + + + ]:CBC 35 : if (i < n_backups - 1 && start_lsn != check_lsn)
61 alvherre@kurilemu.de 572 :GNC 3 : pg_fatal("backup at \"%s\" starts at LSN %X/%08X, but expected %X/%08X",
573 : : backup_dirs[i],
574 : : LSN_FORMAT_ARGS(start_lsn),
575 : : LSN_FORMAT_ARGS(check_lsn));
626 rhaas@postgresql.org 576 :CBC 32 : check_tli = previous_tli;
577 : 32 : check_lsn = previous_lsn;
578 : :
579 : : /*
580 : : * The last backup label in the chain needs to be saved for later use,
581 : : * while the others are only needed within this loop.
582 : : */
583 [ + + ]: 32 : if (lastbuf == buf)
584 : 18 : buf = makeStringInfo();
585 : : else
586 : 14 : resetStringInfo(buf);
587 : : }
588 : :
589 : : /* Free memory that we don't need any more. */
590 [ + - ]: 14 : if (lastbuf != buf)
539 dgustafsson@postgres 591 : 14 : destroyStringInfo(buf);
592 : :
593 : : /*
594 : : * Return the data from the first backup_info that we read (which is the
595 : : * backup_label from the last directory specified on the command line).
596 : : */
626 rhaas@postgresql.org 597 : 14 : return lastbuf;
598 : : }
599 : :
600 : : /*
601 : : * Sanity check control files and return system_identifier.
602 : : */
603 : : static uint64
604 : 20 : check_control_files(int n_backups, char **backup_dirs)
605 : : {
606 : : int i;
607 : 20 : uint64 system_identifier = 0; /* placate compiler */
499 608 : 20 : uint32 data_checksum_version = 0; /* placate compiler */
609 : 20 : bool data_checksum_mismatch = false;
610 : :
611 : : /* Try to read each control file in turn, last to first. */
626 612 [ + + ]: 60 : for (i = n_backups - 1; i >= 0; --i)
613 : : {
614 : : ControlFileData *control_file;
615 : : bool crc_ok;
616 : : char *controlpath;
617 : :
152 fujii@postgresql.org 618 : 41 : controlpath = psprintf("%s/%s", backup_dirs[i], XLOG_CONTROL_FILE);
626 rhaas@postgresql.org 619 [ + + ]: 41 : pg_log_debug("reading \"%s\"", controlpath);
542 620 : 41 : control_file = get_controlfile_by_exact_path(controlpath, &crc_ok);
621 : :
622 : : /* Control file contents not meaningful if CRC is bad. */
626 623 [ - + ]: 41 : if (!crc_ok)
619 john.naylor@postgres 624 :UBC 0 : pg_fatal("%s: CRC is incorrect", controlpath);
625 : :
626 : : /* Can't interpret control file if not current version. */
626 rhaas@postgresql.org 627 [ - + ]:CBC 41 : if (control_file->pg_control_version != PG_CONTROL_VERSION)
626 rhaas@postgresql.org 628 :UBC 0 : pg_fatal("%s: unexpected control file version",
629 : : controlpath);
630 : :
631 : : /* System identifiers should all match. */
626 rhaas@postgresql.org 632 [ + + ]:CBC 41 : if (i == n_backups - 1)
633 : 20 : system_identifier = control_file->system_identifier;
634 [ + + ]: 21 : else if (system_identifier != control_file->system_identifier)
161 peter@eisentraut.org 635 : 1 : pg_fatal("%s: expected system identifier %" PRIu64 ", but found %" PRIu64,
636 : : controlpath, system_identifier,
637 : : control_file->system_identifier);
638 : :
639 : : /*
640 : : * Detect checksum mismatches, but only if the last backup in the
641 : : * chain has checksums enabled.
642 : : */
499 rhaas@postgresql.org 643 [ + + ]: 40 : if (i == n_backups - 1)
644 : 20 : data_checksum_version = control_file->data_checksum_version;
645 [ + - ]: 20 : else if (data_checksum_version != 0 &&
646 [ - + ]: 20 : data_checksum_version != control_file->data_checksum_version)
499 rhaas@postgresql.org 647 :UBC 0 : data_checksum_mismatch = true;
648 : :
649 : : /* Release memory. */
626 rhaas@postgresql.org 650 :CBC 40 : pfree(control_file);
651 : 40 : pfree(controlpath);
652 : : }
653 : :
654 : : /*
655 : : * If debug output is enabled, make a note of the system identifier that
656 : : * we found in all of the relevant control files.
657 : : */
161 peter@eisentraut.org 658 [ + + ]: 19 : pg_log_debug("system identifier is %" PRIu64, system_identifier);
659 : :
660 : : /*
661 : : * Warn the user if not all backups are in the same state with regards to
662 : : * checksums.
663 : : */
499 rhaas@postgresql.org 664 [ - + ]: 19 : if (data_checksum_mismatch)
665 : : {
499 rhaas@postgresql.org 666 :UBC 0 : pg_log_warning("only some backups have checksums enabled");
375 peter@eisentraut.org 667 : 0 : pg_log_warning_hint("Disable, and optionally reenable, checksums on the output directory to avoid failures.");
668 : : }
669 : :
542 rhaas@postgresql.org 670 :CBC 19 : return system_identifier;
671 : : }
672 : :
673 : : /*
674 : : * Set default permissions for new files and directories based on the
675 : : * permissions of the given directory. The intent here is that the output
676 : : * directory should use the same permissions scheme as the final input
677 : : * directory.
678 : : */
679 : : static void
626 680 : 13 : check_input_dir_permissions(char *dir)
681 : : {
682 : : struct stat st;
683 : :
684 [ - + ]: 13 : if (stat(dir, &st) != 0)
442 peter@eisentraut.org 685 :UBC 0 : pg_fatal("could not stat file \"%s\": %m", dir);
686 : :
626 rhaas@postgresql.org 687 :CBC 13 : SetDataDirectoryCreatePerm(st.st_mode);
688 : 13 : }
689 : :
690 : : /*
691 : : * Clean up output directories before exiting.
692 : : */
693 : : static void
694 : 13 : cleanup_directories_atexit(void)
695 : : {
696 [ + + ]: 14 : while (cleanup_dir_list != NULL)
697 : : {
698 : 1 : cb_cleanup_dir *dir = cleanup_dir_list;
699 : :
700 [ + - ]: 1 : if (dir->rmtopdir)
701 : : {
702 : 1 : pg_log_info("removing output directory \"%s\"", dir->target_path);
703 [ - + ]: 1 : if (!rmtree(dir->target_path, dir->rmtopdir))
626 rhaas@postgresql.org 704 :UBC 0 : pg_log_error("failed to remove output directory");
705 : : }
706 : : else
707 : : {
708 : 0 : pg_log_info("removing contents of output directory \"%s\"",
709 : : dir->target_path);
710 [ # # ]: 0 : if (!rmtree(dir->target_path, dir->rmtopdir))
711 : 0 : pg_log_error("failed to remove contents of output directory");
712 : : }
713 : :
626 rhaas@postgresql.org 714 :CBC 1 : cleanup_dir_list = cleanup_dir_list->next;
715 : 1 : pfree(dir);
716 : : }
717 : 13 : }
718 : :
719 : : /*
720 : : * Create the named output directory, unless it already exists or we're in
721 : : * dry-run mode. If it already exists but is not empty, that's a fatal error.
722 : : *
723 : : * Adds the created directory to the list of directories to be cleaned up
724 : : * at process exit.
725 : : */
726 : : static void
727 : 14 : create_output_directory(char *dirname, cb_options *opt)
728 : : {
729 [ + - - - : 14 : switch (pg_check_dir(dirname))
- ]
730 : : {
731 : 14 : case 0:
732 [ - + ]: 14 : if (opt->dry_run)
733 : : {
626 rhaas@postgresql.org 734 [ # # ]:UBC 0 : pg_log_debug("would create directory \"%s\"", dirname);
735 : 0 : return;
736 : : }
626 rhaas@postgresql.org 737 [ + + ]:CBC 14 : pg_log_debug("creating directory \"%s\"", dirname);
738 [ - + ]: 14 : if (pg_mkdir_p(dirname, pg_dir_create_mode) == -1)
626 rhaas@postgresql.org 739 :UBC 0 : pg_fatal("could not create directory \"%s\": %m", dirname);
626 rhaas@postgresql.org 740 :CBC 14 : remember_to_cleanup_directory(dirname, true);
741 : 14 : break;
742 : :
626 rhaas@postgresql.org 743 :UBC 0 : case 1:
744 [ # # ]: 0 : pg_log_debug("using existing directory \"%s\"", dirname);
745 : 0 : remember_to_cleanup_directory(dirname, false);
746 : 0 : break;
747 : :
748 : 0 : case 2:
749 : : case 3:
750 : : case 4:
751 : 0 : pg_fatal("directory \"%s\" exists but is not empty", dirname);
752 : :
753 : 0 : case -1:
754 : 0 : pg_fatal("could not access directory \"%s\": %m", dirname);
755 : : }
756 : : }
757 : :
758 : : /*
759 : : * help
760 : : *
761 : : * Prints help page for the program
762 : : *
763 : : * progname: the name of the executed program, such as "pg_combinebackup"
764 : : */
765 : : static void
626 rhaas@postgresql.org 766 :CBC 1 : help(const char *progname)
767 : : {
768 : 1 : printf(_("%s reconstructs full backups from incrementals.\n\n"), progname);
769 : 1 : printf(_("Usage:\n"));
770 : 1 : printf(_(" %s [OPTION]... DIRECTORY...\n"), progname);
771 : 1 : printf(_("\nOptions:\n"));
772 : 1 : printf(_(" -d, --debug generate lots of debugging output\n"));
173 773 : 1 : printf(_(" -k, --link link files instead of copying\n"));
498 774 : 1 : printf(_(" -n, --dry-run do not actually do anything\n"));
626 775 : 1 : printf(_(" -N, --no-sync do not wait for changes to be written safely to disk\n"));
443 peter@eisentraut.org 776 : 1 : printf(_(" -o, --output=DIRECTORY output directory\n"));
620 michael@paquier.xyz 777 : 1 : printf(_(" -T, --tablespace-mapping=OLDDIR=NEWDIR\n"
778 : : " relocate tablespace in OLDDIR to NEWDIR\n"));
375 peter@eisentraut.org 779 : 1 : printf(_(" --clone clone (reflink) files instead of copying\n"));
433 tomas.vondra@postgre 780 : 1 : printf(_(" --copy copy files (default)\n"));
375 peter@eisentraut.org 781 : 1 : printf(_(" --copy-file-range copy using copy_file_range() system call\n"));
626 rhaas@postgresql.org 782 : 1 : printf(_(" --manifest-checksums=SHA{224,256,384,512}|CRC32C|NONE\n"
783 : : " use algorithm for manifest checksums\n"));
784 : 1 : printf(_(" --no-manifest suppress generation of backup manifest\n"));
785 : 1 : printf(_(" --sync-method=METHOD set method for syncing files to disk\n"));
500 peter@eisentraut.org 786 : 1 : printf(_(" -V, --version output version information, then exit\n"));
626 rhaas@postgresql.org 787 : 1 : printf(_(" -?, --help show this help, then exit\n"));
788 : :
789 : 1 : printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT);
790 : 1 : printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL);
791 : 1 : }
792 : :
793 : : /*
794 : : * Try to parse a string as a non-zero OID without leading zeroes.
795 : : *
796 : : * If it works, return true and set *result to the answer, else return false.
797 : : */
798 : : static bool
799 : 2 : parse_oid(char *s, Oid *result)
800 : : {
801 : : Oid oid;
802 : : char *ep;
803 : :
804 : 2 : errno = 0;
805 : 2 : oid = strtoul(s, &ep, 10);
806 [ + - + - : 2 : if (errno != 0 || *ep != '\0' || oid < 1 || oid > PG_UINT32_MAX)
- + ]
626 rhaas@postgresql.org 807 :UBC 0 : return false;
808 : :
626 rhaas@postgresql.org 809 :CBC 2 : *result = oid;
810 : 2 : return true;
811 : : }
812 : :
813 : : /*
814 : : * Copy files from the input directory to the output directory, reconstructing
815 : : * full files from incremental files as required.
816 : : *
817 : : * If processing a user-defined tablespace, the tsoid should be the OID
818 : : * of that tablespace and input_directory and output_directory should be the
819 : : * toplevel input and output directories for that tablespace. Otherwise,
820 : : * tsoid should be InvalidOid and input_directory and output_directory should
821 : : * be the main input and output directories.
822 : : *
823 : : * relative_path is the path beneath the given input and output directories
824 : : * that we are currently processing. If NULL, it indicates that we're
825 : : * processing the input and output directories themselves.
826 : : *
827 : : * n_prior_backups is the number of prior backups that we have available.
828 : : * This doesn't count the very last backup, which is referenced by
829 : : * input_directory, just the older ones. prior_backup_dirs is an array of
830 : : * the locations of those previous backups.
831 : : */
832 : : static void
833 : 341 : process_directory_recursively(Oid tsoid,
834 : : char *input_directory,
835 : : char *output_directory,
836 : : char *relative_path,
837 : : int n_prior_backups,
838 : : char **prior_backup_dirs,
839 : : manifest_data **manifests,
840 : : manifest_writer *mwriter,
841 : : cb_options *opt)
842 : : {
843 : : char ifulldir[MAXPGPATH];
844 : : char ofulldir[MAXPGPATH];
845 : : char manifest_prefix[MAXPGPATH];
846 : : DIR *dir;
847 : : struct dirent *de;
506 848 : 341 : bool is_pg_tblspc = false;
849 : 341 : bool is_pg_wal = false;
850 : 341 : bool is_incremental_dir = false;
626 851 : 341 : manifest_data *latest_manifest = manifests[n_prior_backups];
852 : : pg_checksum_type checksum_type;
853 : :
854 : : /*
855 : : * Classify this directory.
856 : : *
857 : : * We set is_pg_tblspc only for the toplevel pg_tblspc directory, because
858 : : * the symlinks in that specific directory require special handling.
859 : : *
860 : : * We set is_pg_wal for the toplevel WAL directory and all of its
861 : : * subdirectories, because those files are not included in the backup
862 : : * manifest and hence need special treatment. (Since incremental backup
863 : : * does not exist in pre-v10 versions, we don't have to worry about the
864 : : * old pg_xlog naming.)
865 : : *
866 : : * We set is_incremental_dir for directories that can contain incremental
867 : : * files requiring reconstruction. If such files occur outside these
868 : : * directories, we want to just copy them straight to the output
869 : : * directory. This is to protect against a user creating a file with a
870 : : * strange name like INCREMENTAL.config and then complaining that
871 : : * incremental backups don't work properly. The test here is a bit tricky:
872 : : * incremental files occur in subdirectories of base, in pg_global itself,
873 : : * and in subdirectories of pg_tblspc only if in-place tablespaces are
874 : : * used.
875 : : */
506 876 [ + + ]: 341 : if (OidIsValid(tsoid))
877 : 3 : is_incremental_dir = true;
878 [ + + ]: 338 : else if (relative_path != NULL)
879 : : {
368 michael@paquier.xyz 880 : 325 : is_pg_tblspc = strcmp(relative_path, PG_TBLSPC_DIR) == 0;
506 rhaas@postgresql.org 881 [ + + ]: 638 : is_pg_wal = (strcmp(relative_path, "pg_wal") == 0 ||
882 [ + + ]: 313 : strncmp(relative_path, "pg_wal/", 7) == 0);
883 : 325 : is_incremental_dir = strncmp(relative_path, "base/", 5) == 0 ||
884 [ + + + + ]: 596 : strcmp(relative_path, "global") == 0 ||
368 michael@paquier.xyz 885 [ - + ]: 271 : strncmp(relative_path, PG_TBLSPC_DIR_SLASH, 10) == 0;
886 : : }
887 : :
888 : : /*
889 : : * If we're under pg_wal, then we don't need checksums, because these
890 : : * files aren't included in the backup manifest. Otherwise use whatever
891 : : * type of checksum is configured.
892 : : */
626 rhaas@postgresql.org 893 [ + + ]: 341 : if (!is_pg_wal)
894 : 305 : checksum_type = opt->manifest_checksums;
895 : : else
896 : 36 : checksum_type = CHECKSUM_TYPE_NONE;
897 : :
898 : : /*
899 : : * Append the relative path to the input and output directories, and
900 : : * figure out the appropriate prefix to add to files in this directory
901 : : * when looking them up in a backup manifest.
902 : : */
903 [ + + ]: 341 : if (relative_path == NULL)
904 : : {
604 905 : 14 : strlcpy(ifulldir, input_directory, MAXPGPATH);
906 : 14 : strlcpy(ofulldir, output_directory, MAXPGPATH);
626 907 [ + + ]: 14 : if (OidIsValid(tsoid))
368 michael@paquier.xyz 908 : 1 : snprintf(manifest_prefix, MAXPGPATH, "%s/%u/", PG_TBLSPC_DIR, tsoid);
909 : : else
626 rhaas@postgresql.org 910 : 13 : manifest_prefix[0] = '\0';
911 : : }
912 : : else
913 : : {
914 : 327 : snprintf(ifulldir, MAXPGPATH, "%s/%s", input_directory,
915 : : relative_path);
916 : 327 : snprintf(ofulldir, MAXPGPATH, "%s/%s", output_directory,
917 : : relative_path);
918 [ + + ]: 327 : if (OidIsValid(tsoid))
368 michael@paquier.xyz 919 : 2 : snprintf(manifest_prefix, MAXPGPATH, "%s/%u/%s/",
920 : : PG_TBLSPC_DIR, tsoid, relative_path);
921 : : else
626 rhaas@postgresql.org 922 : 325 : snprintf(manifest_prefix, MAXPGPATH, "%s/", relative_path);
923 : : }
924 : :
925 : : /*
926 : : * Toplevel output directories have already been created by the time this
927 : : * function is called, but any subdirectories are our responsibility.
928 : : */
929 [ + + ]: 341 : if (relative_path != NULL)
930 : : {
931 [ - + ]: 327 : if (opt->dry_run)
626 rhaas@postgresql.org 932 [ # # ]:UBC 0 : pg_log_debug("would create directory \"%s\"", ofulldir);
933 : : else
934 : : {
626 rhaas@postgresql.org 935 [ + + ]:CBC 327 : pg_log_debug("creating directory \"%s\"", ofulldir);
936 [ - + ]: 327 : if (mkdir(ofulldir, pg_dir_create_mode) == -1)
626 rhaas@postgresql.org 937 :UBC 0 : pg_fatal("could not create directory \"%s\": %m", ofulldir);
938 : : }
939 : : }
940 : :
941 : : /* It's time to scan the directory. */
626 rhaas@postgresql.org 942 [ - + ]:CBC 341 : if ((dir = opendir(ifulldir)) == NULL)
626 rhaas@postgresql.org 943 :UBC 0 : pg_fatal("could not open directory \"%s\": %m", ifulldir);
626 rhaas@postgresql.org 944 [ + + ]:CBC 14303 : while (errno = 0, (de = readdir(dir)) != NULL)
945 : : {
946 : : PGFileType type;
947 : : char ifullpath[MAXPGPATH];
948 : : char ofullpath[MAXPGPATH];
949 : : char manifest_path[MAXPGPATH];
950 : 13965 : Oid oid = InvalidOid;
951 : 13965 : int checksum_length = 0;
952 : 13965 : uint8 *checksum_payload = NULL;
953 : : pg_checksum_context checksum_ctx;
954 : :
955 : : /* Ignore "." and ".." entries. */
956 [ + + ]: 13965 : if (strcmp(de->d_name, ".") == 0 ||
957 [ + + ]: 13626 : strcmp(de->d_name, "..") == 0)
958 : 1032 : continue;
959 : :
960 : : /* Construct input path. */
961 : 13285 : snprintf(ifullpath, MAXPGPATH, "%s/%s", ifulldir, de->d_name);
962 : :
963 : : /* Figure out what kind of directory entry this is. */
964 : 13285 : type = get_dirent_type(ifullpath, de, false, PG_LOG_ERROR);
965 [ - + ]: 13285 : if (type == PGFILETYPE_ERROR)
626 rhaas@postgresql.org 966 :UBC 0 : exit(1);
967 : :
968 : : /*
969 : : * If we're processing pg_tblspc, then check whether the filename
970 : : * looks like it could be a tablespace OID. If so, and if the
971 : : * directory entry is a symbolic link or a directory, skip it.
972 : : *
973 : : * Our goal here is to ignore anything that would have been considered
974 : : * by scan_for_existing_tablespaces to be a tablespace.
975 : : */
626 rhaas@postgresql.org 976 [ + + + - :CBC 13285 : if (is_pg_tblspc && parse_oid(de->d_name, &oid) &&
- + ]
626 rhaas@postgresql.org 977 [ # # ]:UBC 0 : (type == PGFILETYPE_LNK || type == PGFILETYPE_DIR))
626 rhaas@postgresql.org 978 :CBC 1 : continue;
979 : :
980 : : /* If it's a directory, recurse. */
981 [ + + ]: 13284 : if (type == PGFILETYPE_DIR)
982 : 325 : {
983 : : char new_relative_path[MAXPGPATH];
984 : :
985 : : /* Append new pathname component to relative path. */
986 [ + + ]: 327 : if (relative_path == NULL)
604 987 : 213 : strlcpy(new_relative_path, de->d_name, MAXPGPATH);
988 : : else
626 989 : 114 : snprintf(new_relative_path, MAXPGPATH, "%s/%s", relative_path,
990 : 114 : de->d_name);
991 : :
992 : : /* And recurse. */
993 : 327 : process_directory_recursively(tsoid,
994 : : input_directory, output_directory,
995 : : new_relative_path,
996 : : n_prior_backups, prior_backup_dirs,
997 : : manifests, mwriter, opt);
998 : 325 : continue;
999 : : }
1000 : :
1001 : : /* Skip anything that's not a regular file. */
1002 [ - + ]: 12957 : if (type != PGFILETYPE_REG)
1003 : : {
626 rhaas@postgresql.org 1004 [ # # ]:UBC 0 : if (type == PGFILETYPE_LNK)
1005 : 0 : pg_log_warning("skipping symbolic link \"%s\"", ifullpath);
1006 : : else
1007 : 0 : pg_log_warning("skipping special file \"%s\"", ifullpath);
1008 : 0 : continue;
1009 : : }
1010 : :
1011 : : /*
1012 : : * Skip the backup_label and backup_manifest files; they require
1013 : : * special handling and are handled elsewhere.
1014 : : */
626 rhaas@postgresql.org 1015 [ + + ]:CBC 12957 : if (relative_path == NULL &&
1016 [ + + ]: 92 : (strcmp(de->d_name, "backup_label") == 0 ||
1017 [ + + ]: 79 : strcmp(de->d_name, "backup_manifest") == 0))
1018 : 26 : continue;
1019 : :
1020 : : /*
1021 : : * If it's an incremental file, hand it off to the reconstruction
1022 : : * code, which will figure out what to do.
1023 : : */
506 1024 [ + + ]: 12931 : if (is_incremental_dir &&
1025 [ + + ]: 12803 : strncmp(de->d_name, INCREMENTAL_PREFIX,
1026 : : INCREMENTAL_PREFIX_LENGTH) == 0)
1027 : : {
1028 : : /* Output path should not include "INCREMENTAL." prefix. */
626 1029 : 6453 : snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir,
1030 : 6453 : de->d_name + INCREMENTAL_PREFIX_LENGTH);
1031 : :
1032 : :
1033 : : /* Manifest path likewise omits incremental prefix. */
1034 : 6453 : snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix,
1035 : 6453 : de->d_name + INCREMENTAL_PREFIX_LENGTH);
1036 : :
1037 : : /* Reconstruction logic will do the rest. */
1038 : 6453 : reconstruct_from_incremental_file(ifullpath, ofullpath,
1039 : : manifest_prefix,
1040 : 6453 : de->d_name + INCREMENTAL_PREFIX_LENGTH,
1041 : : n_prior_backups,
1042 : : prior_backup_dirs,
1043 : : manifests,
1044 : : manifest_path,
1045 : : checksum_type,
1046 : : &checksum_length,
1047 : : &checksum_payload,
1048 : : opt->copy_method,
1049 : 6453 : opt->debug,
1050 : 6453 : opt->dry_run);
1051 : : }
1052 : : else
1053 : : {
1054 : : /* Construct the path that the backup_manifest will use. */
1055 : 6478 : snprintf(manifest_path, MAXPGPATH, "%s%s", manifest_prefix,
1056 : 6478 : de->d_name);
1057 : :
1058 : : /*
1059 : : * It's not an incremental file, so we need to copy the entire
1060 : : * file to the output directory.
1061 : : *
1062 : : * If a checksum of the required type already exists in the
1063 : : * backup_manifest for the final input directory, we can save some
1064 : : * work by reusing that checksum instead of computing a new one.
1065 : : */
1066 [ + + + - ]: 6478 : if (checksum_type != CHECKSUM_TYPE_NONE &&
1067 : : latest_manifest != NULL)
1068 : : {
1069 : : manifest_file *mfile;
1070 : :
1071 : 4531 : mfile = manifest_files_lookup(latest_manifest->files,
1072 : : manifest_path);
1073 [ - + ]: 4531 : if (mfile == NULL)
1074 : : {
1075 : : char *bmpath;
1076 : :
1077 : : /*
1078 : : * The directory is out of sync with the backup_manifest,
1079 : : * so emit a warning.
1080 : : */
626 rhaas@postgresql.org 1081 :UBC 0 : bmpath = psprintf("%s/%s", input_directory,
1082 : : "backup_manifest");
375 peter@eisentraut.org 1083 : 0 : pg_log_warning("manifest file \"%s\" contains no entry for file \"%s\"",
1084 : : bmpath, manifest_path);
626 rhaas@postgresql.org 1085 : 0 : pfree(bmpath);
1086 : : }
626 rhaas@postgresql.org 1087 [ + + ]:CBC 4531 : else if (mfile->checksum_type == checksum_type)
1088 : : {
1089 : 3564 : checksum_length = mfile->checksum_length;
1090 : 3564 : checksum_payload = mfile->checksum_payload;
1091 : : }
1092 : : }
1093 : :
1094 : : /*
1095 : : * If we're reusing a checksum, then we don't need copy_file() to
1096 : : * compute one for us, but otherwise, it needs to compute whatever
1097 : : * type of checksum we need.
1098 : : */
1099 [ + + ]: 6478 : if (checksum_length != 0)
1100 : 3564 : pg_checksum_init(&checksum_ctx, CHECKSUM_TYPE_NONE);
1101 : : else
1102 : 2914 : pg_checksum_init(&checksum_ctx, checksum_type);
1103 : :
1104 : : /* Actually copy the file. */
1105 : 6478 : snprintf(ofullpath, MAXPGPATH, "%s/%s", ofulldir, de->d_name);
519 tomas.vondra@postgre 1106 : 6478 : copy_file(ifullpath, ofullpath, &checksum_ctx,
1107 : 6478 : opt->copy_method, opt->dry_run);
1108 : :
1109 : : /*
1110 : : * If copy_file() performed a checksum calculation for us, then
1111 : : * save the results (except in dry-run mode, when there's no
1112 : : * point).
1113 : : */
626 rhaas@postgresql.org 1114 [ + + + - ]: 6478 : if (checksum_ctx.type != CHECKSUM_TYPE_NONE && !opt->dry_run)
1115 : : {
1116 : 967 : checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH);
1117 : 967 : checksum_length = pg_checksum_final(&checksum_ctx,
1118 : : checksum_payload);
1119 : : }
1120 : : }
1121 : :
1122 : : /* Generate manifest entry, if needed. */
1123 [ + + ]: 12930 : if (mwriter != NULL)
1124 : : {
1125 : : struct stat sb;
1126 : :
1127 : : /*
1128 : : * In order to generate a manifest entry, we need the file size
1129 : : * and mtime. We have no way to know the correct mtime except to
1130 : : * stat() the file, so just do that and get the size as well.
1131 : : *
1132 : : * If we didn't need the mtime here, we could try to obtain the
1133 : : * file size from the reconstruction or file copy process above,
1134 : : * although that is actually not convenient in all cases. If we
1135 : : * write the file ourselves then clearly we can keep a count of
1136 : : * bytes, but if we use something like CopyFile() then it's
1137 : : * trickier. Since we have to stat() anyway to get the mtime,
1138 : : * there's no point in worrying about it.
1139 : : */
1140 [ - + ]: 11962 : if (stat(ofullpath, &sb) < 0)
626 rhaas@postgresql.org 1141 :UBC 0 : pg_fatal("could not stat file \"%s\": %m", ofullpath);
1142 : :
1143 : : /* OK, now do the work. */
626 rhaas@postgresql.org 1144 :CBC 11962 : add_file_to_manifest(mwriter, manifest_path,
1145 : 11962 : sb.st_size, sb.st_mtime,
1146 : : checksum_type, checksum_length,
1147 : : checksum_payload);
1148 : : }
1149 : :
1150 : : /* Avoid leaking memory. */
1151 [ + + ]: 12930 : if (checksum_payload != NULL)
1152 : 10983 : pfree(checksum_payload);
1153 : : }
1154 : :
1155 : 338 : closedir(dir);
1156 : 338 : }
1157 : :
1158 : : /*
1159 : : * Read the version number from PG_VERSION and convert it to the usual server
1160 : : * version number format. (e.g. If PG_VERSION contains "14\n" this function
1161 : : * will return 140000)
1162 : : */
1163 : : static int
1164 : 20 : read_pg_version_file(char *directory)
1165 : : {
1166 : : char filename[MAXPGPATH];
1167 : : StringInfoData buf;
1168 : : int fd;
1169 : : int version;
1170 : : char *ep;
1171 : :
1172 : : /* Construct pathname. */
1173 : 20 : snprintf(filename, MAXPGPATH, "%s/PG_VERSION", directory);
1174 : :
1175 : : /* Open file. */
1176 [ - + ]: 20 : if ((fd = open(filename, O_RDONLY, 0)) < 0)
626 rhaas@postgresql.org 1177 :UBC 0 : pg_fatal("could not open file \"%s\": %m", filename);
1178 : :
1179 : : /* Read into memory. Length limit of 128 should be more than generous. */
626 rhaas@postgresql.org 1180 :CBC 20 : initStringInfo(&buf);
1181 : 20 : slurp_file(fd, filename, &buf, 128);
1182 : :
1183 : : /* Close the file. */
1184 [ - + ]: 20 : if (close(fd) != 0)
442 peter@eisentraut.org 1185 :UBC 0 : pg_fatal("could not close file \"%s\": %m", filename);
1186 : :
1187 : : /* Convert to integer. */
626 rhaas@postgresql.org 1188 :CBC 20 : errno = 0;
1189 : 20 : version = strtoul(buf.data, &ep, 10);
1190 [ + - - + ]: 20 : if (errno != 0 || *ep != '\n')
1191 : : {
1192 : : /*
1193 : : * Incremental backup is not relevant to very old server versions that
1194 : : * used multi-part version number (e.g. 9.6, or 8.4). So if we see
1195 : : * what looks like the beginning of such a version number, just bail
1196 : : * out.
1197 : : */
626 rhaas@postgresql.org 1198 [ # # # # ]:UBC 0 : if (version < 10 && *ep == '.')
442 peter@eisentraut.org 1199 : 0 : pg_fatal("%s: server version too old", filename);
1200 : 0 : pg_fatal("%s: could not parse version number", filename);
1201 : : }
1202 : :
1203 : : /* Debugging output. */
442 peter@eisentraut.org 1204 [ + + ]:CBC 20 : pg_log_debug("read server version %d from file \"%s\"", version, filename);
1205 : :
1206 : : /* Release memory and return result. */
626 rhaas@postgresql.org 1207 : 20 : pfree(buf.data);
1208 : 20 : return version * 10000;
1209 : : }
1210 : :
1211 : : /*
1212 : : * Add a directory to the list of output directories to clean up.
1213 : : */
1214 : : static void
1215 : 14 : remember_to_cleanup_directory(char *target_path, bool rmtopdir)
1216 : : {
1217 : 14 : cb_cleanup_dir *dir = pg_malloc(sizeof(cb_cleanup_dir));
1218 : :
1219 : 14 : dir->target_path = target_path;
1220 : 14 : dir->rmtopdir = rmtopdir;
1221 : 14 : dir->next = cleanup_dir_list;
1222 : 14 : cleanup_dir_list = dir;
1223 : 14 : }
1224 : :
1225 : : /*
1226 : : * Empty out the list of directories scheduled for cleanup at exit.
1227 : : *
1228 : : * We want to remove the output directories only on a failure, so call this
1229 : : * function when we know that the operation has succeeded.
1230 : : *
1231 : : * Since we only expect this to be called when we're about to exit, we could
1232 : : * just set cleanup_dir_list to NULL and be done with it, but we free the
1233 : : * memory to be tidy.
1234 : : */
1235 : : static void
1236 : 12 : reset_directory_cleanup_list(void)
1237 : : {
1238 [ + + ]: 25 : while (cleanup_dir_list != NULL)
1239 : : {
1240 : 13 : cb_cleanup_dir *dir = cleanup_dir_list;
1241 : :
1242 : 13 : cleanup_dir_list = cleanup_dir_list->next;
1243 : 13 : pfree(dir);
1244 : : }
1245 : 12 : }
1246 : :
1247 : : /*
1248 : : * Scan the pg_tblspc directory of the final input backup to get a canonical
1249 : : * list of what tablespaces are part of the backup.
1250 : : *
1251 : : * 'pathname' should be the path to the toplevel backup directory for the
1252 : : * final backup in the backup chain.
1253 : : */
1254 : : static cb_tablespace *
1255 : 13 : scan_for_existing_tablespaces(char *pathname, cb_options *opt)
1256 : : {
1257 : : char pg_tblspc[MAXPGPATH];
1258 : : DIR *dir;
1259 : : struct dirent *de;
1260 : 13 : cb_tablespace *tslist = NULL;
1261 : :
368 michael@paquier.xyz 1262 : 13 : snprintf(pg_tblspc, MAXPGPATH, "%s/%s", pathname, PG_TBLSPC_DIR);
626 rhaas@postgresql.org 1263 [ + + ]: 13 : pg_log_debug("scanning \"%s\"", pg_tblspc);
1264 : :
1265 [ - + ]: 13 : if ((dir = opendir(pg_tblspc)) == NULL)
604 rhaas@postgresql.org 1266 :UBC 0 : pg_fatal("could not open directory \"%s\": %m", pg_tblspc);
1267 : :
626 rhaas@postgresql.org 1268 [ + + ]:CBC 40 : while (errno = 0, (de = readdir(dir)) != NULL)
1269 : : {
1270 : : Oid oid;
1271 : : char tblspcdir[MAXPGPATH];
1272 : : char link_target[MAXPGPATH];
1273 : : int link_length;
1274 : : cb_tablespace *ts;
1275 : : cb_tablespace *otherts;
1276 : : PGFileType type;
1277 : :
1278 : : /* Silently ignore "." and ".." entries. */
1279 [ + + + + ]: 27 : if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
1280 : 26 : continue;
1281 : :
1282 : : /* Construct full pathname. */
1283 : 1 : snprintf(tblspcdir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name);
1284 : :
1285 : : /* Ignore any file name that doesn't look like a proper OID. */
1286 [ - + ]: 1 : if (!parse_oid(de->d_name, &oid))
1287 : : {
626 rhaas@postgresql.org 1288 [ # # ]:UBC 0 : pg_log_debug("skipping \"%s\" because the filename is not a legal tablespace OID",
1289 : : tblspcdir);
1290 : 0 : continue;
1291 : : }
1292 : :
1293 : : /* Only symbolic links and directories are tablespaces. */
626 rhaas@postgresql.org 1294 :CBC 1 : type = get_dirent_type(tblspcdir, de, false, PG_LOG_ERROR);
1295 [ - + ]: 1 : if (type == PGFILETYPE_ERROR)
626 rhaas@postgresql.org 1296 :UBC 0 : exit(1);
626 rhaas@postgresql.org 1297 [ - + - - ]:CBC 1 : if (type != PGFILETYPE_LNK && type != PGFILETYPE_DIR)
1298 : : {
626 rhaas@postgresql.org 1299 [ # # ]:UBC 0 : pg_log_debug("skipping \"%s\" because it is neither a symbolic link nor a directory",
1300 : : tblspcdir);
1301 : 0 : continue;
1302 : : }
1303 : :
1304 : : /* Create a new tablespace object. */
626 rhaas@postgresql.org 1305 :CBC 1 : ts = pg_malloc0(sizeof(cb_tablespace));
1306 : 1 : ts->oid = oid;
1307 : :
1308 : : /*
1309 : : * If it's a link, it's not an in-place tablespace. Otherwise, it must
1310 : : * be a directory, and thus an in-place tablespace.
1311 : : */
1312 [ + - ]: 1 : if (type == PGFILETYPE_LNK)
1313 : : {
1314 : : cb_tablespace_mapping *tsmap;
1315 : :
1316 : : /* Read the link target. */
1317 : 1 : link_length = readlink(tblspcdir, link_target, sizeof(link_target));
1318 [ - + ]: 1 : if (link_length < 0)
626 rhaas@postgresql.org 1319 :UBC 0 : pg_fatal("could not read symbolic link \"%s\": %m",
1320 : : tblspcdir);
626 rhaas@postgresql.org 1321 [ - + ]:CBC 1 : if (link_length >= sizeof(link_target))
442 peter@eisentraut.org 1322 :UBC 0 : pg_fatal("target of symbolic link \"%s\" is too long", tblspcdir);
626 rhaas@postgresql.org 1323 :CBC 1 : link_target[link_length] = '\0';
1324 [ - + ]: 1 : if (!is_absolute_path(link_target))
442 peter@eisentraut.org 1325 :UBC 0 : pg_fatal("target of symbolic link \"%s\" is relative", tblspcdir);
1326 : :
1327 : : /* Canonicalize the link target. */
626 rhaas@postgresql.org 1328 :CBC 1 : canonicalize_path(link_target);
1329 : :
1330 : : /*
1331 : : * Find the corresponding tablespace mapping and copy the relevant
1332 : : * details into the new tablespace entry.
1333 : : */
1334 [ + - ]: 1 : for (tsmap = opt->tsmappings; tsmap != NULL; tsmap = tsmap->next)
1335 : : {
1336 [ + - ]: 1 : if (strcmp(tsmap->old_dir, link_target) == 0)
1337 : : {
604 1338 : 1 : strlcpy(ts->old_dir, tsmap->old_dir, MAXPGPATH);
1339 : 1 : strlcpy(ts->new_dir, tsmap->new_dir, MAXPGPATH);
626 1340 : 1 : ts->in_place = false;
1341 : 1 : break;
1342 : : }
1343 : : }
1344 : :
1345 : : /* Every non-in-place tablespace must be mapped. */
1346 [ - + ]: 1 : if (tsmap == NULL)
626 rhaas@postgresql.org 1347 :UBC 0 : pg_fatal("tablespace at \"%s\" has no tablespace mapping",
1348 : : link_target);
1349 : : }
1350 : : else
1351 : : {
1352 : : /*
1353 : : * For an in-place tablespace, there's no separate directory, so
1354 : : * we just record the paths within the data directories.
1355 : : */
1356 : 0 : snprintf(ts->old_dir, MAXPGPATH, "%s/%s", pg_tblspc, de->d_name);
368 michael@paquier.xyz 1357 : 0 : snprintf(ts->new_dir, MAXPGPATH, "%s/%s/%s", opt->output,
1358 : 0 : PG_TBLSPC_DIR, de->d_name);
626 rhaas@postgresql.org 1359 : 0 : ts->in_place = true;
1360 : : }
1361 : :
1362 : : /* Tablespaces should not share a directory. */
626 rhaas@postgresql.org 1363 [ - + ]:CBC 1 : for (otherts = tslist; otherts != NULL; otherts = otherts->next)
626 rhaas@postgresql.org 1364 [ # # ]:UBC 0 : if (strcmp(ts->new_dir, otherts->new_dir) == 0)
442 peter@eisentraut.org 1365 : 0 : pg_fatal("tablespaces with OIDs %u and %u both point at directory \"%s\"",
1366 : : otherts->oid, oid, ts->new_dir);
1367 : :
1368 : : /* Add this tablespace to the list. */
626 rhaas@postgresql.org 1369 :CBC 1 : ts->next = tslist;
1370 : 1 : tslist = ts;
1371 : : }
1372 : :
604 1373 [ - + ]: 13 : if (closedir(dir) != 0)
604 rhaas@postgresql.org 1374 :UBC 0 : pg_fatal("could not close directory \"%s\": %m", pg_tblspc);
1375 : :
626 rhaas@postgresql.org 1376 :CBC 13 : return tslist;
1377 : : }
1378 : :
1379 : : /*
1380 : : * Read a file into a StringInfo.
1381 : : *
1382 : : * fd is used for the actual file I/O, filename for error reporting purposes.
1383 : : * A file longer than maxlen is a fatal error.
1384 : : */
1385 : : static void
1386 : 57 : slurp_file(int fd, char *filename, StringInfo buf, int maxlen)
1387 : : {
1388 : : struct stat st;
1389 : : ssize_t rb;
1390 : :
1391 : : /* Check file size, and complain if it's too large. */
1392 [ - + ]: 57 : if (fstat(fd, &st) != 0)
442 peter@eisentraut.org 1393 :UBC 0 : pg_fatal("could not stat file \"%s\": %m", filename);
626 rhaas@postgresql.org 1394 [ - + ]:CBC 57 : if (st.st_size > maxlen)
626 rhaas@postgresql.org 1395 :UBC 0 : pg_fatal("file \"%s\" is too large", filename);
1396 : :
1397 : : /* Make sure we have enough space. */
626 rhaas@postgresql.org 1398 :CBC 57 : enlargeStringInfo(buf, st.st_size);
1399 : :
1400 : : /* Read the data. */
1401 : 57 : rb = read(fd, &buf->data[buf->len], st.st_size);
1402 : :
1403 : : /*
1404 : : * We don't expect any concurrent changes, so we should read exactly the
1405 : : * expected number of bytes.
1406 : : */
1407 [ - + ]: 57 : if (rb != st.st_size)
1408 : : {
626 rhaas@postgresql.org 1409 [ # # ]:UBC 0 : if (rb < 0)
1410 : 0 : pg_fatal("could not read file \"%s\": %m", filename);
1411 : : else
442 peter@eisentraut.org 1412 : 0 : pg_fatal("could not read file \"%s\": read %zd of %lld",
1413 : : filename, rb, (long long int) st.st_size);
1414 : : }
1415 : :
1416 : : /* Adjust buffer length for new data and restore trailing-\0 invariant */
626 rhaas@postgresql.org 1417 :CBC 57 : buf->len += rb;
1418 : 57 : buf->data[buf->len] = '\0';
1419 : 57 : }
|
