Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * filemap.c
4 : * A data structure for keeping track of files that have changed.
5 : *
6 : * This source file contains the logic to decide what to do with different
7 : * kinds of files, and the data structure to support it. Before modifying
8 : * anything, pg_rewind collects information about all the files and their
9 : * attributes in the target and source data directories. It also scans the
10 : * WAL log in the target, and collects information about data blocks that
11 : * were changed. All this information is stored in a hash table, using the
12 : * file path relative to the root of the data directory as the key.
13 : *
14 : * After collecting all the information required, the decide_file_actions()
15 : * function scans the hash table and decides what action needs to be taken
16 : * for each file. Finally, it sorts the array to the final order that the
17 : * actions should be executed in.
18 : *
19 : * Copyright (c) 2013-2025, PostgreSQL Global Development Group
20 : *
21 : *-------------------------------------------------------------------------
22 : */
23 :
24 : #include "postgres_fe.h"
25 :
26 : #include <sys/stat.h>
27 : #include <unistd.h>
28 :
29 : #include "access/xlog_internal.h"
30 : #include "catalog/pg_tablespace_d.h"
31 : #include "common/file_utils.h"
32 : #include "common/hashfn_unstable.h"
33 : #include "common/string.h"
34 : #include "datapagemap.h"
35 : #include "filemap.h"
36 : #include "pg_rewind.h"
37 :
38 : /*
39 : * Define a hash table which we can use to store information about the files
40 : * appearing in source and target systems.
41 : */
42 : #define SH_PREFIX filehash
43 : #define SH_ELEMENT_TYPE file_entry_t
44 : #define SH_KEY_TYPE const char *
45 : #define SH_KEY path
46 : #define SH_HASH_KEY(tb, key) hash_string(key)
47 : #define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0)
48 : #define SH_SCOPE static inline
49 : #define SH_RAW_ALLOCATOR pg_malloc0
50 : #define SH_DECLARE
51 : #define SH_DEFINE
52 : #include "lib/simplehash.h"
53 :
54 : #define FILEHASH_INITIAL_SIZE 1000
55 :
56 : static filehash_hash *filehash;
57 :
58 : static bool isRelDataFile(const char *path);
59 : static char *datasegpath(RelFileLocator rlocator, ForkNumber forknum,
60 : BlockNumber segno);
61 :
62 : static file_entry_t *insert_filehash_entry(const char *path);
63 : static file_entry_t *lookup_filehash_entry(const char *path);
64 :
65 : /*
66 : * A separate hash table which tracks WAL files that must not be deleted.
67 : */
68 : typedef struct keepwal_entry
69 : {
70 : const char *path;
71 : uint32 status;
72 : } keepwal_entry;
73 :
74 : #define SH_PREFIX keepwal
75 : #define SH_ELEMENT_TYPE keepwal_entry
76 : #define SH_KEY_TYPE const char *
77 : #define SH_KEY path
78 : #define SH_HASH_KEY(tb, key) hash_string(key)
79 : #define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0)
80 : #define SH_SCOPE static inline
81 : #define SH_RAW_ALLOCATOR pg_malloc0
82 : #define SH_DECLARE
83 : #define SH_DEFINE
84 : #include "lib/simplehash.h"
85 :
86 : #define KEEPWAL_INITIAL_SIZE 1000
87 :
88 :
89 : static keepwal_hash *keepwal = NULL;
90 : static bool keepwal_entry_exists(const char *path);
91 :
92 : static int final_filemap_cmp(const void *a, const void *b);
93 :
94 : static bool check_file_excluded(const char *path, bool is_source);
95 :
96 : /*
97 : * Definition of one element part of an exclusion list, used to exclude
98 : * contents when rewinding. "name" is the name of the file or path to
99 : * check for exclusion. If "match_prefix" is true, any items matching
100 : * the name as prefix are excluded.
101 : */
102 : struct exclude_list_item
103 : {
104 : const char *name;
105 : bool match_prefix;
106 : };
107 :
108 : /*
109 : * The contents of these directories are removed or recreated during server
110 : * start so they are not included in data processed by pg_rewind.
111 : *
112 : * Note: those lists should be kept in sync with what basebackup.c provides.
113 : * Some of the values, contrary to what basebackup.c uses, are hardcoded as
114 : * they are defined in backend-only headers. So this list is maintained
115 : * with a best effort in mind.
116 : */
117 : static const char *const excludeDirContents[] =
118 : {
119 : /*
120 : * Skip temporary statistics files. PG_STAT_TMP_DIR must be skipped
121 : * because extensions like pg_stat_statements store data there.
122 : */
123 : "pg_stat_tmp", /* defined as PG_STAT_TMP_DIR */
124 :
125 : /*
126 : * It is generally not useful to backup the contents of this directory
127 : * even if the intention is to restore to another primary. See backup.sgml
128 : * for a more detailed description.
129 : */
130 : "pg_replslot", /* defined as PG_REPLSLOT_DIR */
131 :
132 : /* Contents removed on startup, see dsm_cleanup_for_mmap(). */
133 : "pg_dynshmem", /* defined as PG_DYNSHMEM_DIR */
134 :
135 : /* Contents removed on startup, see AsyncShmemInit(). */
136 : "pg_notify",
137 :
138 : /*
139 : * Old contents are loaded for possible debugging but are not required for
140 : * normal operation, see SerialInit().
141 : */
142 : "pg_serial",
143 :
144 : /* Contents removed on startup, see DeleteAllExportedSnapshotFiles(). */
145 : "pg_snapshots",
146 :
147 : /* Contents zeroed on startup, see StartupSUBTRANS(). */
148 : "pg_subtrans",
149 :
150 : /* end of list */
151 : NULL
152 : };
153 :
154 : /*
155 : * List of files excluded from filemap processing. Files are excluded
156 : * if their prefix match.
157 : */
158 : static const struct exclude_list_item excludeFiles[] =
159 : {
160 : /* Skip auto conf temporary file. */
161 : {"postgresql.auto.conf.tmp", false}, /* defined as PG_AUTOCONF_FILENAME */
162 :
163 : /* Skip current log file temporary file */
164 : {"current_logfiles.tmp", false}, /* defined as
165 : * LOG_METAINFO_DATAFILE_TMP */
166 :
167 : /* Skip relation cache because it is rebuilt on startup */
168 : {"pg_internal.init", true}, /* defined as RELCACHE_INIT_FILENAME */
169 :
170 : /*
171 : * If there is a backup_label or tablespace_map file, it indicates that a
172 : * recovery failed and this cluster probably can't be rewound, but exclude
173 : * them anyway if they are found.
174 : */
175 : {"backup_label", false}, /* defined as BACKUP_LABEL_FILE */
176 : {"tablespace_map", false}, /* defined as TABLESPACE_MAP */
177 :
178 : /*
179 : * If there's a backup_manifest, it belongs to a backup that was used to
180 : * start this server. It is *not* correct for this backup. Our
181 : * backup_manifest is injected into the backup separately if users want
182 : * it.
183 : */
184 : {"backup_manifest", false},
185 :
186 : {"postmaster.pid", false},
187 : {"postmaster.opts", false},
188 :
189 : /* end of list */
190 : {NULL, false}
191 : };
192 :
193 : /*
194 : * Initialize the hash table for the file map.
195 : */
196 : void
197 28 : filehash_init(void)
198 : {
199 28 : filehash = filehash_create(FILEHASH_INITIAL_SIZE, NULL);
200 28 : }
201 :
202 : /* Look up entry for 'path', creating a new one if it doesn't exist */
203 : static file_entry_t *
204 64032 : insert_filehash_entry(const char *path)
205 : {
206 : file_entry_t *entry;
207 : bool found;
208 :
209 64032 : entry = filehash_insert(filehash, path, &found);
210 64032 : if (!found)
211 : {
212 33404 : entry->path = pg_strdup(path);
213 33404 : entry->isrelfile = isRelDataFile(path);
214 :
215 33404 : entry->target_exists = false;
216 33404 : entry->target_type = FILE_TYPE_UNDEFINED;
217 33404 : entry->target_size = 0;
218 33404 : entry->target_link_target = NULL;
219 33404 : entry->target_pages_to_overwrite.bitmap = NULL;
220 33404 : entry->target_pages_to_overwrite.bitmapsize = 0;
221 :
222 33404 : entry->source_exists = false;
223 33404 : entry->source_type = FILE_TYPE_UNDEFINED;
224 33404 : entry->source_size = 0;
225 33404 : entry->source_link_target = NULL;
226 :
227 33404 : entry->action = FILE_ACTION_UNDECIDED;
228 : }
229 :
230 64032 : return entry;
231 : }
232 :
233 : static file_entry_t *
234 169058 : lookup_filehash_entry(const char *path)
235 : {
236 169058 : return filehash_lookup(filehash, path);
237 : }
238 :
239 : /*
240 : * Initialize a hash table to store WAL file names that must be kept.
241 : */
242 : void
243 28 : keepwal_init(void)
244 : {
245 : /* An initial hash size out of thin air */
246 28 : keepwal = keepwal_create(KEEPWAL_INITIAL_SIZE, NULL);
247 28 : }
248 :
249 : /* Mark the given file to prevent its removal */
250 : void
251 38 : keepwal_add_entry(const char *path)
252 : {
253 : keepwal_entry *entry;
254 : bool found;
255 :
256 : /* Should only be called with keepwal initialized */
257 : Assert(keepwal != NULL);
258 :
259 38 : entry = keepwal_insert(keepwal, path, &found);
260 :
261 38 : if (!found)
262 38 : entry->path = pg_strdup(path);
263 38 : }
264 :
265 : /* Return true if file is marked as not to be removed, false otherwise */
266 : static bool
267 1308 : keepwal_entry_exists(const char *path)
268 : {
269 1308 : return keepwal_lookup(keepwal, path) != NULL;
270 : }
271 :
272 : /*
273 : * Callback for processing source file list.
274 : *
275 : * This is called once for every file in the source server. We record the
276 : * type and size of the file, so that decide_file_action() can later decide what
277 : * to do with it.
278 : */
279 : void
280 32048 : process_source_file(const char *path, file_type_t type, size_t size,
281 : const char *link_target)
282 : {
283 : file_entry_t *entry;
284 :
285 : /*
286 : * Pretend that pg_wal is a directory, even if it's really a symlink. We
287 : * don't want to mess with the symlink itself, nor complain if it's a
288 : * symlink in source but not in target or vice versa.
289 : */
290 32048 : if (strcmp(path, "pg_wal") == 0 && type == FILE_TYPE_SYMLINK)
291 0 : type = FILE_TYPE_DIRECTORY;
292 :
293 : /*
294 : * sanity check: a filename that looks like a data file better be a
295 : * regular file
296 : */
297 32048 : if (type != FILE_TYPE_REGULAR && isRelDataFile(path))
298 0 : pg_fatal("data file \"%s\" in source is not a regular file", path);
299 :
300 : /* Remember this source file */
301 32048 : entry = insert_filehash_entry(path);
302 32048 : if (entry->source_exists)
303 0 : pg_fatal("duplicate source file \"%s\"", path);
304 32048 : entry->source_exists = true;
305 32048 : entry->source_type = type;
306 32048 : entry->source_size = size;
307 32048 : entry->source_link_target = link_target ? pg_strdup(link_target) : NULL;
308 32048 : }
309 :
310 : /*
311 : * Callback for processing target file list.
312 : *
313 : * Record the type and size of the file, like process_source_file() does.
314 : */
315 : void
316 31984 : process_target_file(const char *path, file_type_t type, size_t size,
317 : const char *link_target)
318 : {
319 : file_entry_t *entry;
320 :
321 : /*
322 : * Do not apply any exclusion filters here. This has advantage to remove
323 : * from the target data folder all paths which have been filtered out from
324 : * the source data folder when processing the source files.
325 : */
326 :
327 : /*
328 : * Like in process_source_file, pretend that pg_wal is always a directory.
329 : */
330 31984 : if (strcmp(path, "pg_wal") == 0 && type == FILE_TYPE_SYMLINK)
331 4 : type = FILE_TYPE_DIRECTORY;
332 :
333 : /* Remember this target file */
334 31984 : entry = insert_filehash_entry(path);
335 31984 : if (entry->target_exists)
336 0 : pg_fatal("duplicate source file \"%s\"", path);
337 31984 : entry->target_exists = true;
338 31984 : entry->target_type = type;
339 31984 : entry->target_size = size;
340 31984 : entry->target_link_target = link_target ? pg_strdup(link_target) : NULL;
341 31984 : }
342 :
343 : /*
344 : * This callback gets called while we read the WAL in the target, for every
345 : * block that has changed in the target system. It decides if the given
346 : * 'blkno' in the target relfile needs to be overwritten from the source, and
347 : * if so, records it in 'target_pages_to_overwrite' bitmap.
348 : *
349 : * NOTE: All the files on both systems must have already been added to the
350 : * hash table!
351 : */
352 : void
353 169058 : process_target_wal_block_change(ForkNumber forknum, RelFileLocator rlocator,
354 : BlockNumber blkno)
355 : {
356 : char *path;
357 : file_entry_t *entry;
358 : BlockNumber blkno_inseg;
359 : int segno;
360 :
361 169058 : segno = blkno / RELSEG_SIZE;
362 169058 : blkno_inseg = blkno % RELSEG_SIZE;
363 :
364 169058 : path = datasegpath(rlocator, forknum, segno);
365 169058 : entry = lookup_filehash_entry(path);
366 169058 : pfree(path);
367 :
368 : /*
369 : * If the block still exists in both systems, remember it. Otherwise we
370 : * can safely ignore it.
371 : *
372 : * If the block is beyond the EOF in the source system, or the file
373 : * doesn't exist in the source at all, we're going to truncate/remove it
374 : * away from the target anyway. Likewise, if it doesn't exist in the
375 : * target anymore, we will copy it over with the "tail" from the source
376 : * system, anyway.
377 : *
378 : * It is possible to find WAL for a file that doesn't exist on either
379 : * system anymore. It means that the relation was dropped later in the
380 : * target system, and independently on the source system too, or that it
381 : * was created and dropped in the target system and it never existed in
382 : * the source. Either way, we can safely ignore it.
383 : */
384 169058 : if (entry)
385 : {
386 : Assert(entry->isrelfile);
387 :
388 169058 : if (entry->target_exists)
389 : {
390 169050 : if (entry->target_type != FILE_TYPE_REGULAR)
391 0 : pg_fatal("unexpected page modification for non-regular file \"%s\"",
392 : entry->path);
393 :
394 169050 : if (entry->source_exists)
395 : {
396 : off_t end_offset;
397 :
398 165750 : end_offset = (blkno_inseg + 1) * BLCKSZ;
399 165750 : if (end_offset <= entry->source_size && end_offset <= entry->target_size)
400 5750 : datapagemap_add(&entry->target_pages_to_overwrite, blkno_inseg);
401 : }
402 : }
403 : }
404 169058 : }
405 :
406 : /*
407 : * Is this the path of file that pg_rewind can skip copying?
408 : */
409 : static bool
410 33372 : check_file_excluded(const char *path, bool is_source)
411 : {
412 : char localpath[MAXPGPATH];
413 : int excludeIdx;
414 : const char *filename;
415 :
416 : /*
417 : * Skip all temporary files, .../pgsql_tmp/... and .../pgsql_tmp.*
418 : */
419 33372 : if (strstr(path, "/" PG_TEMP_FILE_PREFIX) != NULL ||
420 33344 : strstr(path, "/" PG_TEMP_FILES_DIR "/") != NULL)
421 : {
422 28 : return true;
423 : }
424 :
425 : /* check individual files... */
426 299528 : for (excludeIdx = 0; excludeFiles[excludeIdx].name != NULL; excludeIdx++)
427 : {
428 266324 : int cmplen = strlen(excludeFiles[excludeIdx].name);
429 :
430 266324 : filename = last_dir_separator(path);
431 266324 : if (filename == NULL)
432 5804 : filename = path;
433 : else
434 260520 : filename++;
435 :
436 266324 : if (!excludeFiles[excludeIdx].match_prefix)
437 232980 : cmplen++;
438 266324 : if (strncmp(filename, excludeFiles[excludeIdx].name, cmplen) == 0)
439 : {
440 140 : if (is_source)
441 140 : pg_log_debug("entry \"%s\" excluded from source file list",
442 : path);
443 : else
444 0 : pg_log_debug("entry \"%s\" excluded from target file list",
445 : path);
446 140 : return true;
447 : }
448 : }
449 :
450 : /*
451 : * ... And check some directories. Note that this includes any contents
452 : * within the directories themselves.
453 : */
454 265604 : for (excludeIdx = 0; excludeDirContents[excludeIdx] != NULL; excludeIdx++)
455 : {
456 232428 : snprintf(localpath, sizeof(localpath), "%s/",
457 : excludeDirContents[excludeIdx]);
458 232428 : if (strstr(path, localpath) == path)
459 : {
460 28 : if (is_source)
461 28 : pg_log_debug("entry \"%s\" excluded from source file list",
462 : path);
463 : else
464 0 : pg_log_debug("entry \"%s\" excluded from target file list",
465 : path);
466 28 : return true;
467 : }
468 : }
469 :
470 33176 : return false;
471 : }
472 :
473 : static const char *
474 11180 : action_to_str(file_action_t action)
475 : {
476 11180 : switch (action)
477 : {
478 830 : case FILE_ACTION_NONE:
479 830 : return "NONE";
480 8872 : case FILE_ACTION_COPY:
481 8872 : return "COPY";
482 8 : case FILE_ACTION_TRUNCATE:
483 8 : return "TRUNCATE";
484 10 : case FILE_ACTION_COPY_TAIL:
485 10 : return "COPY_TAIL";
486 18 : case FILE_ACTION_CREATE:
487 18 : return "CREATE";
488 1442 : case FILE_ACTION_REMOVE:
489 1442 : return "REMOVE";
490 :
491 0 : default:
492 0 : return "unknown";
493 : }
494 : }
495 :
496 : /*
497 : * Calculate the totals needed for progress reports.
498 : */
499 : void
500 0 : calculate_totals(filemap_t *filemap)
501 : {
502 : file_entry_t *entry;
503 : int i;
504 :
505 0 : filemap->total_size = 0;
506 0 : filemap->fetch_size = 0;
507 :
508 0 : for (i = 0; i < filemap->nentries; i++)
509 : {
510 0 : entry = filemap->entries[i];
511 :
512 0 : if (entry->source_type != FILE_TYPE_REGULAR)
513 0 : continue;
514 :
515 0 : filemap->total_size += entry->source_size;
516 :
517 0 : if (entry->action == FILE_ACTION_COPY)
518 : {
519 0 : filemap->fetch_size += entry->source_size;
520 0 : continue;
521 : }
522 :
523 0 : if (entry->action == FILE_ACTION_COPY_TAIL)
524 0 : filemap->fetch_size += (entry->source_size - entry->target_size);
525 :
526 0 : if (entry->target_pages_to_overwrite.bitmapsize > 0)
527 : {
528 : datapagemap_iterator_t *iter;
529 : BlockNumber blk;
530 :
531 0 : iter = datapagemap_iterate(&entry->target_pages_to_overwrite);
532 0 : while (datapagemap_next(iter, &blk))
533 0 : filemap->fetch_size += BLCKSZ;
534 :
535 0 : pg_free(iter);
536 : }
537 : }
538 0 : }
539 :
540 : void
541 28 : print_filemap(filemap_t *filemap)
542 : {
543 : file_entry_t *entry;
544 : int i;
545 :
546 33432 : for (i = 0; i < filemap->nentries; i++)
547 : {
548 33404 : entry = filemap->entries[i];
549 33404 : if (entry->action != FILE_ACTION_NONE ||
550 23054 : entry->target_pages_to_overwrite.bitmapsize > 0)
551 : {
552 11180 : pg_log_debug("%s (%s)", entry->path,
553 : action_to_str(entry->action));
554 :
555 11180 : if (entry->target_pages_to_overwrite.bitmapsize > 0)
556 846 : datapagemap_print(&entry->target_pages_to_overwrite);
557 : }
558 : }
559 28 : fflush(stdout);
560 28 : }
561 :
562 : /*
563 : * Does it look like a relation data file?
564 : *
565 : * For our purposes, only files belonging to the main fork are considered
566 : * relation files. Other forks are always copied in toto, because we cannot
567 : * reliably track changes to them, because WAL only contains block references
568 : * for the main fork.
569 : */
570 : static bool
571 34190 : isRelDataFile(const char *path)
572 : {
573 : RelFileLocator rlocator;
574 : unsigned int segNo;
575 : int nmatch;
576 : bool matched;
577 :
578 : /*----
579 : * Relation data files can be in one of the following directories:
580 : *
581 : * global/
582 : * shared relations
583 : *
584 : * base/<db oid>/
585 : * regular relations, default tablespace
586 : *
587 : * pg_tblspc/<tblspc oid>/<tblspc version>/
588 : * within a non-default tablespace (the name of the directory
589 : * depends on version)
590 : *
591 : * And the relation data files themselves have a filename like:
592 : *
593 : * <oid>.<segment number>
594 : *
595 : *----
596 : */
597 34190 : rlocator.spcOid = InvalidOid;
598 34190 : rlocator.dbOid = InvalidOid;
599 34190 : rlocator.relNumber = InvalidRelFileNumber;
600 34190 : segNo = 0;
601 34190 : matched = false;
602 :
603 34190 : nmatch = sscanf(path, "global/%u.%u", &rlocator.relNumber, &segNo);
604 34190 : if (nmatch == 1 || nmatch == 2)
605 : {
606 1624 : rlocator.spcOid = GLOBALTABLESPACE_OID;
607 1624 : rlocator.dbOid = 0;
608 1624 : matched = true;
609 : }
610 : else
611 : {
612 32566 : nmatch = sscanf(path, "base/%u/%u.%u",
613 : &rlocator.dbOid, &rlocator.relNumber, &segNo);
614 32566 : if (nmatch == 2 || nmatch == 3)
615 : {
616 29974 : rlocator.spcOid = DEFAULTTABLESPACE_OID;
617 29974 : matched = true;
618 : }
619 : else
620 : {
621 2592 : nmatch = sscanf(path, "pg_tblspc/%u/" TABLESPACE_VERSION_DIRECTORY "/%u/%u.%u",
622 : &rlocator.spcOid, &rlocator.dbOid, &rlocator.relNumber,
623 : &segNo);
624 2592 : if (nmatch == 3 || nmatch == 4)
625 24 : matched = true;
626 : }
627 : }
628 :
629 : /*
630 : * The sscanf tests above can match files that have extra characters at
631 : * the end. To eliminate such cases, cross-check that GetRelationPath
632 : * creates the exact same filename, when passed the RelFileLocator
633 : * information we extracted from the filename.
634 : */
635 34190 : if (matched)
636 : {
637 31622 : char *check_path = datasegpath(rlocator, MAIN_FORKNUM, segNo);
638 :
639 31622 : if (strcmp(check_path, path) != 0)
640 7704 : matched = false;
641 :
642 31622 : pfree(check_path);
643 : }
644 :
645 34190 : return matched;
646 : }
647 :
648 : /*
649 : * A helper function to create the path of a relation file and segment.
650 : *
651 : * The returned path is palloc'd
652 : */
653 : static char *
654 200680 : datasegpath(RelFileLocator rlocator, ForkNumber forknum, BlockNumber segno)
655 : {
656 : RelPathStr path;
657 : char *segpath;
658 :
659 200680 : path = relpathperm(rlocator, forknum);
660 200680 : if (segno > 0)
661 : {
662 0 : segpath = psprintf("%s.%u", path.str, segno);
663 0 : return segpath;
664 : }
665 : else
666 200680 : return pstrdup(path.str);
667 : }
668 :
669 : /*
670 : * In the final stage, the filemap is sorted so that removals come last.
671 : * From disk space usage point of view, it would be better to do removals
672 : * first, but for now, safety first. If a whole directory is deleted, all
673 : * files and subdirectories inside it need to removed first. On creation,
674 : * parent directory needs to be created before files and directories inside
675 : * it. To achieve that, the file_action_t enum is ordered so that we can
676 : * just sort on that first. Furthermore, sort REMOVE entries in reverse
677 : * path order, so that "foo/bar" subdirectory is removed before "foo".
678 : */
679 : static int
680 354820 : final_filemap_cmp(const void *a, const void *b)
681 : {
682 354820 : file_entry_t *fa = *((file_entry_t **) a);
683 354820 : file_entry_t *fb = *((file_entry_t **) b);
684 :
685 354820 : if (fa->action > fb->action)
686 12328 : return 1;
687 342492 : if (fa->action < fb->action)
688 14932 : return -1;
689 :
690 327560 : if (fa->action == FILE_ACTION_REMOVE)
691 11514 : return strcmp(fb->path, fa->path);
692 : else
693 316046 : return strcmp(fa->path, fb->path);
694 : }
695 :
696 : /*
697 : * Decide what action to perform to a file.
698 : */
699 : static file_action_t
700 33404 : decide_file_action(file_entry_t *entry)
701 : {
702 33404 : const char *path = entry->path;
703 :
704 : /*
705 : * Don't touch the control file. It is handled specially, after copying
706 : * all the other files.
707 : */
708 33404 : if (strcmp(path, XLOG_CONTROL_FILE) == 0)
709 28 : return FILE_ACTION_NONE;
710 :
711 : /* Skip macOS system files */
712 33376 : if (strstr(path, ".DS_Store") != NULL)
713 4 : return FILE_ACTION_NONE;
714 :
715 : /*
716 : * Remove all files matching the exclusion filters in the target.
717 : */
718 33372 : if (check_file_excluded(path, true))
719 : {
720 196 : if (entry->target_exists)
721 140 : return FILE_ACTION_REMOVE;
722 : else
723 56 : return FILE_ACTION_NONE;
724 : }
725 :
726 : /*
727 : * Handle cases where the file is missing from one of the systems.
728 : */
729 33176 : if (!entry->target_exists && entry->source_exists)
730 : {
731 : /*
732 : * File exists in source, but not in target. Copy it in toto. (If it's
733 : * a relation data file, WAL replay after rewinding should re-create
734 : * it anyway. But there's no harm in copying it now.)
735 : */
736 1360 : switch (entry->source_type)
737 : {
738 18 : case FILE_TYPE_DIRECTORY:
739 : case FILE_TYPE_SYMLINK:
740 18 : return FILE_ACTION_CREATE;
741 1342 : case FILE_TYPE_REGULAR:
742 1342 : return FILE_ACTION_COPY;
743 0 : case FILE_TYPE_UNDEFINED:
744 0 : pg_fatal("unknown file type for \"%s\"", entry->path);
745 : break;
746 : }
747 : }
748 31816 : else if (entry->target_exists && !entry->source_exists)
749 : {
750 : /*
751 : * For files that exist in target but not in source, we check the
752 : * keepwal hash table; any files listed therein must not be removed.
753 : */
754 1308 : if (keepwal_entry_exists(path))
755 : {
756 6 : pg_log_debug("Not removing file \"%s\" because it is required for recovery", path);
757 6 : return FILE_ACTION_NONE;
758 : }
759 1302 : return FILE_ACTION_REMOVE;
760 : }
761 30508 : else if (!entry->target_exists && !entry->source_exists)
762 : {
763 : /*
764 : * Doesn't exist in either server. Why does it have an entry in the
765 : * first place??
766 : */
767 : Assert(false);
768 0 : return FILE_ACTION_NONE;
769 : }
770 :
771 : /*
772 : * Otherwise, the file exists on both systems
773 : */
774 : Assert(entry->target_exists && entry->source_exists);
775 :
776 30508 : if (entry->source_type != entry->target_type)
777 : {
778 : /* But it's a different kind of object. Strange.. */
779 0 : pg_fatal("file \"%s\" is of different type in source and target", entry->path);
780 : }
781 :
782 : /*
783 : * PG_VERSION files should be identical on both systems, but avoid
784 : * overwriting them for paranoia.
785 : */
786 30508 : if (pg_str_endswith(entry->path, "PG_VERSION"))
787 120 : return FILE_ACTION_NONE;
788 :
789 30388 : switch (entry->source_type)
790 : {
791 764 : case FILE_TYPE_DIRECTORY:
792 764 : return FILE_ACTION_NONE;
793 :
794 0 : case FILE_TYPE_SYMLINK:
795 :
796 : /*
797 : * XXX: Should we check if it points to the same target?
798 : */
799 0 : return FILE_ACTION_NONE;
800 :
801 29624 : case FILE_TYPE_REGULAR:
802 29624 : if (!entry->isrelfile)
803 : {
804 : /*
805 : * It's a non-data file that we have no special processing
806 : * for. Copy it in toto.
807 : */
808 7530 : return FILE_ACTION_COPY;
809 : }
810 : else
811 : {
812 : /*
813 : * It's a data file that exists in both systems.
814 : *
815 : * If it's larger in target, we can truncate it. There will
816 : * also be a WAL record of the truncation in the source
817 : * system, so WAL replay would eventually truncate the target
818 : * too, but we might as well do it now.
819 : *
820 : * If it's smaller in the target, it means that it has been
821 : * truncated in the target, or enlarged in the source, or
822 : * both. If it was truncated in the target, we need to copy
823 : * the missing tail from the source system. If it was enlarged
824 : * in the source system, there will be WAL records in the
825 : * source system for the new blocks, so we wouldn't need to
826 : * copy them here. But we don't know which scenario we're
827 : * dealing with, and there's no harm in copying the missing
828 : * blocks now, so do it now.
829 : *
830 : * If it's the same size, do nothing here. Any blocks modified
831 : * in the target will be copied based on parsing the target
832 : * system's WAL, and any blocks modified in the source will be
833 : * updated after rewinding, when the source system's WAL is
834 : * replayed.
835 : */
836 22094 : if (entry->target_size < entry->source_size)
837 10 : return FILE_ACTION_COPY_TAIL;
838 22084 : else if (entry->target_size > entry->source_size)
839 8 : return FILE_ACTION_TRUNCATE;
840 : else
841 22076 : return FILE_ACTION_NONE;
842 : }
843 : break;
844 :
845 0 : case FILE_TYPE_UNDEFINED:
846 0 : pg_fatal("unknown file type for \"%s\"", path);
847 : break;
848 : }
849 :
850 : /* unreachable */
851 0 : pg_fatal("could not decide what to do with file \"%s\"", path);
852 : }
853 :
854 : /*
855 : * Decide what to do with each file.
856 : *
857 : * Returns a 'filemap' with the entries in the order that their actions
858 : * should be executed.
859 : */
860 : filemap_t *
861 28 : decide_file_actions(void)
862 : {
863 : int i;
864 : filehash_iterator it;
865 : file_entry_t *entry;
866 : filemap_t *filemap;
867 :
868 28 : filehash_start_iterate(filehash, &it);
869 33432 : while ((entry = filehash_iterate(filehash, &it)) != NULL)
870 : {
871 33404 : entry->action = decide_file_action(entry);
872 : }
873 :
874 : /*
875 : * Turn the hash table into an array, and sort in the order that the
876 : * actions should be performed.
877 : */
878 28 : filemap = pg_malloc(offsetof(filemap_t, entries) +
879 28 : filehash->members * sizeof(file_entry_t *));
880 28 : filemap->nentries = filehash->members;
881 28 : filehash_start_iterate(filehash, &it);
882 28 : i = 0;
883 33432 : while ((entry = filehash_iterate(filehash, &it)) != NULL)
884 : {
885 33404 : filemap->entries[i++] = entry;
886 : }
887 :
888 28 : qsort(&filemap->entries, filemap->nentries, sizeof(file_entry_t *),
889 : final_filemap_cmp);
890 :
891 28 : return filemap;
892 : }
|
