1/*
2 * relfilenode.c
3 *
4 * relfilenode functions
5 *
6 * Copyright (c) 2010-2019, PostgreSQL Global Development Group
7 * src/bin/pg_upgrade/relfilenode.c
8 */
9
10#include "postgres_fe.h"
11
12#include "pg_upgrade.h"
13
14#include <sys/stat.h>
15#include "catalog/pg_class_d.h"
16#include "access/transam.h"
17
18
19static void transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace);
20static void transfer_relfile(FileNameMap *map, const char *suffix, bool vm_must_add_frozenbit);
21
22
23/*
24 * transfer_all_new_tablespaces()
25 *
26 * Responsible for upgrading all database. invokes routines to generate mappings and then
27 * physically link the databases.
28 */
29void
30transfer_all_new_tablespaces(DbInfoArr *old_db_arr, DbInfoArr *new_db_arr,
31 char *old_pgdata, char *new_pgdata)
32{
33 switch (user_opts.transfer_mode)
34 {
35 case TRANSFER_MODE_CLONE:
36 pg_log(PG_REPORT, "Cloning user relation files\n");
37 break;
38 case TRANSFER_MODE_COPY:
39 pg_log(PG_REPORT, "Copying user relation files\n");
40 break;
41 case TRANSFER_MODE_LINK:
42 pg_log(PG_REPORT, "Linking user relation files\n");
43 break;
44 }
45
46 /*
47 * Transferring files by tablespace is tricky because a single database
48 * can use multiple tablespaces. For non-parallel mode, we just pass a
49 * NULL tablespace path, which matches all tablespaces. In parallel mode,
50 * we pass the default tablespace and all user-created tablespaces and let
51 * those operations happen in parallel.
52 */
53 if (user_opts.jobs <= 1)
54 parallel_transfer_all_new_dbs(old_db_arr, new_db_arr, old_pgdata,
55 new_pgdata, NULL);
56 else
57 {
58 int tblnum;
59
60 /* transfer default tablespace */
61 parallel_transfer_all_new_dbs(old_db_arr, new_db_arr, old_pgdata,
62 new_pgdata, old_pgdata);
63
64 for (tblnum = 0; tblnum < os_info.num_old_tablespaces; tblnum++)
65 parallel_transfer_all_new_dbs(old_db_arr,
66 new_db_arr,
67 old_pgdata,
68 new_pgdata,
69 os_info.old_tablespaces[tblnum]);
70 /* reap all children */
71 while (reap_child(true) == true)
72 ;
73 }
74
75 end_progress_output();
76 check_ok();
77
78 return;
79}
80
81
82/*
83 * transfer_all_new_dbs()
84 *
85 * Responsible for upgrading all database. invokes routines to generate mappings and then
86 * physically link the databases.
87 */
88void
89transfer_all_new_dbs(DbInfoArr *old_db_arr, DbInfoArr *new_db_arr,
90 char *old_pgdata, char *new_pgdata, char *old_tablespace)
91{
92 int old_dbnum,
93 new_dbnum;
94
95 /* Scan the old cluster databases and transfer their files */
96 for (old_dbnum = new_dbnum = 0;
97 old_dbnum < old_db_arr->ndbs;
98 old_dbnum++, new_dbnum++)
99 {
100 DbInfo *old_db = &old_db_arr->dbs[old_dbnum],
101 *new_db = NULL;
102 FileNameMap *mappings;
103 int n_maps;
104
105 /*
106 * Advance past any databases that exist in the new cluster but not in
107 * the old, e.g. "postgres". (The user might have removed the
108 * 'postgres' database from the old cluster.)
109 */
110 for (; new_dbnum < new_db_arr->ndbs; new_dbnum++)
111 {
112 new_db = &new_db_arr->dbs[new_dbnum];
113 if (strcmp(old_db->db_name, new_db->db_name) == 0)
114 break;
115 }
116
117 if (new_dbnum >= new_db_arr->ndbs)
118 pg_fatal("old database \"%s\" not found in the new cluster\n",
119 old_db->db_name);
120
121 mappings = gen_db_file_maps(old_db, new_db, &n_maps, old_pgdata,
122 new_pgdata);
123 if (n_maps)
124 {
125 print_maps(mappings, n_maps, new_db->db_name);
126
127 transfer_single_new_db(mappings, n_maps, old_tablespace);
128 }
129 /* We allocate something even for n_maps == 0 */
130 pg_free(mappings);
131 }
132
133 return;
134}
135
136/*
137 * transfer_single_new_db()
138 *
139 * create links for mappings stored in "maps" array.
140 */
141static void
142transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace)
143{
144 int mapnum;
145 bool vm_crashsafe_match = true;
146 bool vm_must_add_frozenbit = false;
147
148 /*
149 * Do the old and new cluster disagree on the crash-safetiness of the vm
150 * files? If so, do not copy them.
151 */
152 if (old_cluster.controldata.cat_ver < VISIBILITY_MAP_CRASHSAFE_CAT_VER &&
153 new_cluster.controldata.cat_ver >= VISIBILITY_MAP_CRASHSAFE_CAT_VER)
154 vm_crashsafe_match = false;
155
156 /*
157 * Do we need to rewrite visibilitymap?
158 */
159 if (old_cluster.controldata.cat_ver < VISIBILITY_MAP_FROZEN_BIT_CAT_VER &&
160 new_cluster.controldata.cat_ver >= VISIBILITY_MAP_FROZEN_BIT_CAT_VER)
161 vm_must_add_frozenbit = true;
162
163 for (mapnum = 0; mapnum < size; mapnum++)
164 {
165 if (old_tablespace == NULL ||
166 strcmp(maps[mapnum].old_tablespace, old_tablespace) == 0)
167 {
168 /* transfer primary file */
169 transfer_relfile(&maps[mapnum], "", vm_must_add_frozenbit);
170
171 /* fsm/vm files added in PG 8.4 */
172 if (GET_MAJOR_VERSION(old_cluster.major_version) >= 804)
173 {
174 /*
175 * Copy/link any fsm and vm files, if they exist
176 */
177 transfer_relfile(&maps[mapnum], "_fsm", vm_must_add_frozenbit);
178 if (vm_crashsafe_match)
179 transfer_relfile(&maps[mapnum], "_vm", vm_must_add_frozenbit);
180 }
181 }
182 }
183}
184
185
186/*
187 * transfer_relfile()
188 *
189 * Copy or link file from old cluster to new one. If vm_must_add_frozenbit
190 * is true, visibility map forks are converted and rewritten, even in link
191 * mode.
192 */
193static void
194transfer_relfile(FileNameMap *map, const char *type_suffix, bool vm_must_add_frozenbit)
195{
196 char old_file[MAXPGPATH];
197 char new_file[MAXPGPATH];
198 int segno;
199 char extent_suffix[65];
200 struct stat statbuf;
201
202 /*
203 * Now copy/link any related segments as well. Remember, PG breaks large
204 * files into 1GB segments, the first segment has no extension, subsequent
205 * segments are named relfilenode.1, relfilenode.2, relfilenode.3.
206 */
207 for (segno = 0;; segno++)
208 {
209 if (segno == 0)
210 extent_suffix[0] = '\0';
211 else
212 snprintf(extent_suffix, sizeof(extent_suffix), ".%d", segno);
213
214 snprintf(old_file, sizeof(old_file), "%s%s/%u/%u%s%s",
215 map->old_tablespace,
216 map->old_tablespace_suffix,
217 map->old_db_oid,
218 map->old_relfilenode,
219 type_suffix,
220 extent_suffix);
221 snprintf(new_file, sizeof(new_file), "%s%s/%u/%u%s%s",
222 map->new_tablespace,
223 map->new_tablespace_suffix,
224 map->new_db_oid,
225 map->new_relfilenode,
226 type_suffix,
227 extent_suffix);
228
229 /* Is it an extent, fsm, or vm file? */
230 if (type_suffix[0] != '\0' || segno != 0)
231 {
232 /* Did file open fail? */
233 if (stat(old_file, &statbuf) != 0)
234 {
235 /* File does not exist? That's OK, just return */
236 if (errno == ENOENT)
237 return;
238 else
239 pg_fatal("error while checking for file existence \"%s.%s\" (\"%s\" to \"%s\"): %s\n",
240 map->nspname, map->relname, old_file, new_file,
241 strerror(errno));
242 }
243
244 /* If file is empty, just return */
245 if (statbuf.st_size == 0)
246 return;
247 }
248
249 unlink(new_file);
250
251 /* Copying files might take some time, so give feedback. */
252 pg_log(PG_STATUS, "%s", old_file);
253
254 if (vm_must_add_frozenbit && strcmp(type_suffix, "_vm") == 0)
255 {
256 /* Need to rewrite visibility map format */
257 pg_log(PG_VERBOSE, "rewriting \"%s\" to \"%s\"\n",
258 old_file, new_file);
259 rewriteVisibilityMap(old_file, new_file, map->nspname, map->relname);
260 }
261 else
262 switch (user_opts.transfer_mode)
263 {
264 case TRANSFER_MODE_CLONE:
265 pg_log(PG_VERBOSE, "cloning \"%s\" to \"%s\"\n",
266 old_file, new_file);
267 cloneFile(old_file, new_file, map->nspname, map->relname);
268 break;
269 case TRANSFER_MODE_COPY:
270 pg_log(PG_VERBOSE, "copying \"%s\" to \"%s\"\n",
271 old_file, new_file);
272 copyFile(old_file, new_file, map->nspname, map->relname);
273 break;
274 case TRANSFER_MODE_LINK:
275 pg_log(PG_VERBOSE, "linking \"%s\" to \"%s\"\n",
276 old_file, new_file);
277 linkFile(old_file, new_file, map->nspname, map->relname);
278 }
279 }
280}
281