Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * dbcommands.c
4 : * Database management commands (create/drop database).
5 : *
6 : * Note: database creation/destruction commands use exclusive locks on
7 : * the database objects (as expressed by LockSharedObject()) to avoid
8 : * stepping on each others' toes. Formerly we used table-level locks
9 : * on pg_database, but that's too coarse-grained.
10 : *
11 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
12 : * Portions Copyright (c) 1994, Regents of the University of California
13 : *
14 : *
15 : * IDENTIFICATION
16 : * src/backend/commands/dbcommands.c
17 : *
18 : *-------------------------------------------------------------------------
19 : */
20 : #include "postgres.h"
21 :
22 : #include <fcntl.h>
23 : #include <unistd.h>
24 : #include <sys/stat.h>
25 :
26 : #include "access/genam.h"
27 : #include "access/heapam.h"
28 : #include "access/htup_details.h"
29 : #include "access/multixact.h"
30 : #include "access/tableam.h"
31 : #include "access/xact.h"
32 : #include "access/xloginsert.h"
33 : #include "access/xlogrecovery.h"
34 : #include "access/xlogutils.h"
35 : #include "catalog/catalog.h"
36 : #include "catalog/dependency.h"
37 : #include "catalog/indexing.h"
38 : #include "catalog/objectaccess.h"
39 : #include "catalog/pg_authid.h"
40 : #include "catalog/pg_collation.h"
41 : #include "catalog/pg_database.h"
42 : #include "catalog/pg_db_role_setting.h"
43 : #include "catalog/pg_subscription.h"
44 : #include "catalog/pg_tablespace.h"
45 : #include "commands/comment.h"
46 : #include "commands/dbcommands.h"
47 : #include "commands/dbcommands_xlog.h"
48 : #include "commands/defrem.h"
49 : #include "commands/seclabel.h"
50 : #include "commands/tablespace.h"
51 : #include "common/file_perm.h"
52 : #include "mb/pg_wchar.h"
53 : #include "miscadmin.h"
54 : #include "pgstat.h"
55 : #include "postmaster/bgwriter.h"
56 : #include "replication/slot.h"
57 : #include "storage/copydir.h"
58 : #include "storage/fd.h"
59 : #include "storage/ipc.h"
60 : #include "storage/lmgr.h"
61 : #include "storage/md.h"
62 : #include "storage/procarray.h"
63 : #include "storage/smgr.h"
64 : #include "utils/acl.h"
65 : #include "utils/builtins.h"
66 : #include "utils/fmgroids.h"
67 : #include "utils/pg_locale.h"
68 : #include "utils/relmapper.h"
69 : #include "utils/snapmgr.h"
70 : #include "utils/syscache.h"
71 :
72 : /*
73 : * Create database strategy.
74 : *
75 : * CREATEDB_WAL_LOG will copy the database at the block level and WAL log each
76 : * copied block.
77 : *
78 : * CREATEDB_FILE_COPY will simply perform a file system level copy of the
79 : * database and log a single record for each tablespace copied. To make this
80 : * safe, it also triggers checkpoints before and after the operation.
81 : */
82 : typedef enum CreateDBStrategy
83 : {
84 : CREATEDB_WAL_LOG,
85 : CREATEDB_FILE_COPY,
86 : } CreateDBStrategy;
87 :
88 : typedef struct
89 : {
90 : Oid src_dboid; /* source (template) DB */
91 : Oid dest_dboid; /* DB we are trying to create */
92 : CreateDBStrategy strategy; /* create db strategy */
93 : } createdb_failure_params;
94 :
95 : typedef struct
96 : {
97 : Oid dest_dboid; /* DB we are trying to move */
98 : Oid dest_tsoid; /* tablespace we are trying to move to */
99 : } movedb_failure_params;
100 :
101 : /*
102 : * Information about a relation to be copied when creating a database.
103 : */
104 : typedef struct CreateDBRelInfo
105 : {
106 : RelFileLocator rlocator; /* physical relation identifier */
107 : Oid reloid; /* relation oid */
108 : bool permanent; /* relation is permanent or unlogged */
109 : } CreateDBRelInfo;
110 :
111 :
112 : /* non-export function prototypes */
113 : static void createdb_failure_callback(int code, Datum arg);
114 : static void movedb(const char *dbname, const char *tblspcname);
115 : static void movedb_failure_callback(int code, Datum arg);
116 : static bool get_db_info(const char *name, LOCKMODE lockmode,
117 : Oid *dbIdP, Oid *ownerIdP,
118 : int *encodingP, bool *dbIsTemplateP, bool *dbAllowConnP, bool *dbHasLoginEvtP,
119 : TransactionId *dbFrozenXidP, MultiXactId *dbMinMultiP,
120 : Oid *dbTablespace, char **dbCollate, char **dbCtype, char **dbLocale,
121 : char **dbIcurules,
122 : char *dbLocProvider,
123 : char **dbCollversion);
124 : static void remove_dbtablespaces(Oid db_id);
125 : static bool check_db_file_conflict(Oid db_id);
126 : static int errdetail_busy_db(int notherbackends, int npreparedxacts);
127 : static void CreateDatabaseUsingWalLog(Oid src_dboid, Oid dst_dboid, Oid src_tsid,
128 : Oid dst_tsid);
129 : static List *ScanSourceDatabasePgClass(Oid tbid, Oid dbid, char *srcpath);
130 : static List *ScanSourceDatabasePgClassPage(Page page, Buffer buf, Oid tbid,
131 : Oid dbid, char *srcpath,
132 : List *rlocatorlist, Snapshot snapshot);
133 : static CreateDBRelInfo *ScanSourceDatabasePgClassTuple(HeapTupleData *tuple,
134 : Oid tbid, Oid dbid,
135 : char *srcpath);
136 : static void CreateDirAndVersionFile(char *dbpath, Oid dbid, Oid tsid,
137 : bool isRedo);
138 : static void CreateDatabaseUsingFileCopy(Oid src_dboid, Oid dst_dboid,
139 : Oid src_tsid, Oid dst_tsid);
140 : static void recovery_create_dbdir(char *path, bool only_tblspc);
141 :
142 : /*
143 : * Create a new database using the WAL_LOG strategy.
144 : *
145 : * Each copied block is separately written to the write-ahead log.
146 : */
147 : static void
148 432 : CreateDatabaseUsingWalLog(Oid src_dboid, Oid dst_dboid,
149 : Oid src_tsid, Oid dst_tsid)
150 : {
151 : char *srcpath;
152 : char *dstpath;
153 432 : List *rlocatorlist = NULL;
154 : ListCell *cell;
155 : LockRelId srcrelid;
156 : LockRelId dstrelid;
157 : RelFileLocator srcrlocator;
158 : RelFileLocator dstrlocator;
159 : CreateDBRelInfo *relinfo;
160 :
161 : /* Get source and destination database paths. */
162 432 : srcpath = GetDatabasePath(src_dboid, src_tsid);
163 432 : dstpath = GetDatabasePath(dst_dboid, dst_tsid);
164 :
165 : /* Create database directory and write PG_VERSION file. */
166 432 : CreateDirAndVersionFile(dstpath, dst_dboid, dst_tsid, false);
167 :
168 : /* Copy relmap file from source database to the destination database. */
169 432 : RelationMapCopy(dst_dboid, dst_tsid, srcpath, dstpath);
170 :
171 : /* Get list of relfilelocators to copy from the source database. */
172 432 : rlocatorlist = ScanSourceDatabasePgClass(src_tsid, src_dboid, srcpath);
173 : Assert(rlocatorlist != NIL);
174 :
175 : /*
176 : * Database IDs will be the same for all relations so set them before
177 : * entering the loop.
178 : */
179 432 : srcrelid.dbId = src_dboid;
180 432 : dstrelid.dbId = dst_dboid;
181 :
182 : /* Loop over our list of relfilelocators and copy each one. */
183 96366 : foreach(cell, rlocatorlist)
184 : {
185 95934 : relinfo = lfirst(cell);
186 95934 : srcrlocator = relinfo->rlocator;
187 :
188 : /*
189 : * If the relation is from the source db's default tablespace then we
190 : * need to create it in the destination db's default tablespace.
191 : * Otherwise, we need to create in the same tablespace as it is in the
192 : * source database.
193 : */
194 95934 : if (srcrlocator.spcOid == src_tsid)
195 95934 : dstrlocator.spcOid = dst_tsid;
196 : else
197 0 : dstrlocator.spcOid = srcrlocator.spcOid;
198 :
199 95934 : dstrlocator.dbOid = dst_dboid;
200 95934 : dstrlocator.relNumber = srcrlocator.relNumber;
201 :
202 : /*
203 : * Acquire locks on source and target relations before copying.
204 : *
205 : * We typically do not read relation data into shared_buffers without
206 : * holding a relation lock. It's unclear what could go wrong if we
207 : * skipped it in this case, because nobody can be modifying either the
208 : * source or destination database at this point, and we have locks on
209 : * both databases, too, but let's take the conservative route.
210 : */
211 95934 : dstrelid.relId = srcrelid.relId = relinfo->reloid;
212 95934 : LockRelationId(&srcrelid, AccessShareLock);
213 95934 : LockRelationId(&dstrelid, AccessShareLock);
214 :
215 : /* Copy relation storage from source to the destination. */
216 95934 : CreateAndCopyRelationData(srcrlocator, dstrlocator, relinfo->permanent);
217 :
218 : /* Release the relation locks. */
219 95934 : UnlockRelationId(&srcrelid, AccessShareLock);
220 95934 : UnlockRelationId(&dstrelid, AccessShareLock);
221 : }
222 :
223 432 : pfree(srcpath);
224 432 : pfree(dstpath);
225 432 : list_free_deep(rlocatorlist);
226 432 : }
227 :
228 : /*
229 : * Scan the pg_class table in the source database to identify the relations
230 : * that need to be copied to the destination database.
231 : *
232 : * This is an exception to the usual rule that cross-database access is
233 : * not possible. We can make it work here because we know that there are no
234 : * connections to the source database and (since there can't be prepared
235 : * transactions touching that database) no in-doubt tuples either. This
236 : * means that we don't need to worry about pruning removing anything from
237 : * under us, and we don't need to be too picky about our snapshot either.
238 : * As long as it sees all previously-committed XIDs as committed and all
239 : * aborted XIDs as aborted, we should be fine: nothing else is possible
240 : * here.
241 : *
242 : * We can't rely on the relcache for anything here, because that only knows
243 : * about the database to which we are connected, and can't handle access to
244 : * other databases. That also means we can't rely on the heap scan
245 : * infrastructure, which would be a bad idea anyway since it might try
246 : * to do things like HOT pruning which we definitely can't do safely in
247 : * a database to which we're not even connected.
248 : */
249 : static List *
250 432 : ScanSourceDatabasePgClass(Oid tbid, Oid dbid, char *srcpath)
251 : {
252 : RelFileLocator rlocator;
253 : BlockNumber nblocks;
254 : BlockNumber blkno;
255 : Buffer buf;
256 : RelFileNumber relfilenumber;
257 : Page page;
258 432 : List *rlocatorlist = NIL;
259 : LockRelId relid;
260 : Snapshot snapshot;
261 : SMgrRelation smgr;
262 : BufferAccessStrategy bstrategy;
263 :
264 : /* Get pg_class relfilenumber. */
265 432 : relfilenumber = RelationMapOidToFilenumberForDatabase(srcpath,
266 : RelationRelationId);
267 :
268 : /* Don't read data into shared_buffers without holding a relation lock. */
269 432 : relid.dbId = dbid;
270 432 : relid.relId = RelationRelationId;
271 432 : LockRelationId(&relid, AccessShareLock);
272 :
273 : /* Prepare a RelFileLocator for the pg_class relation. */
274 432 : rlocator.spcOid = tbid;
275 432 : rlocator.dbOid = dbid;
276 432 : rlocator.relNumber = relfilenumber;
277 :
278 432 : smgr = smgropen(rlocator, INVALID_PROC_NUMBER);
279 432 : nblocks = smgrnblocks(smgr, MAIN_FORKNUM);
280 432 : smgrclose(smgr);
281 :
282 : /* Use a buffer access strategy since this is a bulk read operation. */
283 432 : bstrategy = GetAccessStrategy(BAS_BULKREAD);
284 :
285 : /*
286 : * As explained in the function header comments, we need a snapshot that
287 : * will see all committed transactions as committed, and our transaction
288 : * snapshot - or the active snapshot - might not be new enough for that,
289 : * but the return value of GetLatestSnapshot() should work fine.
290 : */
291 432 : snapshot = GetLatestSnapshot();
292 :
293 : /* Process the relation block by block. */
294 6520 : for (blkno = 0; blkno < nblocks; blkno++)
295 : {
296 6088 : CHECK_FOR_INTERRUPTS();
297 :
298 6088 : buf = ReadBufferWithoutRelcache(rlocator, MAIN_FORKNUM, blkno,
299 : RBM_NORMAL, bstrategy, true);
300 :
301 6088 : LockBuffer(buf, BUFFER_LOCK_SHARE);
302 6088 : page = BufferGetPage(buf);
303 6088 : if (PageIsNew(page) || PageIsEmpty(page))
304 : {
305 0 : UnlockReleaseBuffer(buf);
306 0 : continue;
307 : }
308 :
309 : /* Append relevant pg_class tuples for current page to rlocatorlist. */
310 6088 : rlocatorlist = ScanSourceDatabasePgClassPage(page, buf, tbid, dbid,
311 : srcpath, rlocatorlist,
312 : snapshot);
313 :
314 6088 : UnlockReleaseBuffer(buf);
315 : }
316 :
317 : /* Release relation lock. */
318 432 : UnlockRelationId(&relid, AccessShareLock);
319 :
320 432 : return rlocatorlist;
321 : }
322 :
323 : /*
324 : * Scan one page of the source database's pg_class relation and add relevant
325 : * entries to rlocatorlist. The return value is the updated list.
326 : */
327 : static List *
328 6088 : ScanSourceDatabasePgClassPage(Page page, Buffer buf, Oid tbid, Oid dbid,
329 : char *srcpath, List *rlocatorlist,
330 : Snapshot snapshot)
331 : {
332 6088 : BlockNumber blkno = BufferGetBlockNumber(buf);
333 : OffsetNumber offnum;
334 : OffsetNumber maxoff;
335 : HeapTupleData tuple;
336 :
337 6088 : maxoff = PageGetMaxOffsetNumber(page);
338 :
339 : /* Loop over offsets. */
340 311882 : for (offnum = FirstOffsetNumber;
341 : offnum <= maxoff;
342 305794 : offnum = OffsetNumberNext(offnum))
343 : {
344 : ItemId itemid;
345 :
346 305794 : itemid = PageGetItemId(page, offnum);
347 :
348 : /* Nothing to do if slot is empty or already dead. */
349 305794 : if (!ItemIdIsUsed(itemid) || ItemIdIsDead(itemid) ||
350 218638 : ItemIdIsRedirected(itemid))
351 124296 : continue;
352 :
353 : Assert(ItemIdIsNormal(itemid));
354 181498 : ItemPointerSet(&(tuple.t_self), blkno, offnum);
355 :
356 : /* Initialize a HeapTupleData structure. */
357 181498 : tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid);
358 181498 : tuple.t_len = ItemIdGetLength(itemid);
359 181498 : tuple.t_tableOid = RelationRelationId;
360 :
361 : /* Skip tuples that are not visible to this snapshot. */
362 181498 : if (HeapTupleSatisfiesVisibility(&tuple, snapshot, buf))
363 : {
364 : CreateDBRelInfo *relinfo;
365 :
366 : /*
367 : * ScanSourceDatabasePgClassTuple is in charge of constructing a
368 : * CreateDBRelInfo object for this tuple, but can also decide that
369 : * this tuple isn't something we need to copy. If we do need to
370 : * copy the relation, add it to the list.
371 : */
372 179310 : relinfo = ScanSourceDatabasePgClassTuple(&tuple, tbid, dbid,
373 : srcpath);
374 179310 : if (relinfo != NULL)
375 95934 : rlocatorlist = lappend(rlocatorlist, relinfo);
376 : }
377 : }
378 :
379 6088 : return rlocatorlist;
380 : }
381 :
382 : /*
383 : * Decide whether a certain pg_class tuple represents something that
384 : * needs to be copied from the source database to the destination database,
385 : * and if so, construct a CreateDBRelInfo for it.
386 : *
387 : * Visibility checks are handled by the caller, so our job here is just
388 : * to assess the data stored in the tuple.
389 : */
390 : CreateDBRelInfo *
391 179310 : ScanSourceDatabasePgClassTuple(HeapTupleData *tuple, Oid tbid, Oid dbid,
392 : char *srcpath)
393 : {
394 : CreateDBRelInfo *relinfo;
395 : Form_pg_class classForm;
396 179310 : RelFileNumber relfilenumber = InvalidRelFileNumber;
397 :
398 179310 : classForm = (Form_pg_class) GETSTRUCT(tuple);
399 :
400 : /*
401 : * Return NULL if this object does not need to be copied.
402 : *
403 : * Shared objects don't need to be copied, because they are shared.
404 : * Objects without storage can't be copied, because there's nothing to
405 : * copy. Temporary relations don't need to be copied either, because they
406 : * are inaccessible outside of the session that created them, which must
407 : * be gone already, and couldn't connect to a different database if it
408 : * still existed. autovacuum will eventually remove the pg_class entries
409 : * as well.
410 : */
411 179310 : if (classForm->reltablespace == GLOBALTABLESPACE_OID ||
412 157710 : !RELKIND_HAS_STORAGE(classForm->relkind) ||
413 95934 : classForm->relpersistence == RELPERSISTENCE_TEMP)
414 83376 : return NULL;
415 :
416 : /*
417 : * If relfilenumber is valid then directly use it. Otherwise, consult the
418 : * relmap.
419 : */
420 95934 : if (RelFileNumberIsValid(classForm->relfilenode))
421 88590 : relfilenumber = classForm->relfilenode;
422 : else
423 7344 : relfilenumber = RelationMapOidToFilenumberForDatabase(srcpath,
424 : classForm->oid);
425 :
426 : /* We must have a valid relfilenumber. */
427 95934 : if (!RelFileNumberIsValid(relfilenumber))
428 0 : elog(ERROR, "relation with OID %u does not have a valid relfilenumber",
429 : classForm->oid);
430 :
431 : /* Prepare a rel info element and add it to the list. */
432 95934 : relinfo = (CreateDBRelInfo *) palloc(sizeof(CreateDBRelInfo));
433 95934 : if (OidIsValid(classForm->reltablespace))
434 0 : relinfo->rlocator.spcOid = classForm->reltablespace;
435 : else
436 95934 : relinfo->rlocator.spcOid = tbid;
437 :
438 95934 : relinfo->rlocator.dbOid = dbid;
439 95934 : relinfo->rlocator.relNumber = relfilenumber;
440 95934 : relinfo->reloid = classForm->oid;
441 :
442 : /* Temporary relations were rejected above. */
443 : Assert(classForm->relpersistence != RELPERSISTENCE_TEMP);
444 95934 : relinfo->permanent =
445 95934 : (classForm->relpersistence == RELPERSISTENCE_PERMANENT) ? true : false;
446 :
447 95934 : return relinfo;
448 : }
449 :
450 : /*
451 : * Create database directory and write out the PG_VERSION file in the database
452 : * path. If isRedo is true, it's okay for the database directory to exist
453 : * already.
454 : */
455 : static void
456 472 : CreateDirAndVersionFile(char *dbpath, Oid dbid, Oid tsid, bool isRedo)
457 : {
458 : int fd;
459 : int nbytes;
460 : char versionfile[MAXPGPATH];
461 : char buf[16];
462 :
463 : /*
464 : * Note that we don't have to copy version data from the source database;
465 : * there's only one legal value.
466 : */
467 472 : sprintf(buf, "%s\n", PG_MAJORVERSION);
468 472 : nbytes = strlen(PG_MAJORVERSION) + 1;
469 :
470 : /* Create database directory. */
471 472 : if (MakePGDirectory(dbpath) < 0)
472 : {
473 : /* Failure other than already exists or not in WAL replay? */
474 16 : if (errno != EEXIST || !isRedo)
475 0 : ereport(ERROR,
476 : (errcode_for_file_access(),
477 : errmsg("could not create directory \"%s\": %m", dbpath)));
478 : }
479 :
480 : /*
481 : * Create PG_VERSION file in the database path. If the file already
482 : * exists and we are in WAL replay then try again to open it in write
483 : * mode.
484 : */
485 472 : snprintf(versionfile, sizeof(versionfile), "%s/%s", dbpath, "PG_VERSION");
486 :
487 472 : fd = OpenTransientFile(versionfile, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY);
488 472 : if (fd < 0 && errno == EEXIST && isRedo)
489 16 : fd = OpenTransientFile(versionfile, O_WRONLY | O_TRUNC | PG_BINARY);
490 :
491 472 : if (fd < 0)
492 0 : ereport(ERROR,
493 : (errcode_for_file_access(),
494 : errmsg("could not create file \"%s\": %m", versionfile)));
495 :
496 : /* Write PG_MAJORVERSION in the PG_VERSION file. */
497 472 : pgstat_report_wait_start(WAIT_EVENT_VERSION_FILE_WRITE);
498 472 : errno = 0;
499 472 : if ((int) write(fd, buf, nbytes) != nbytes)
500 : {
501 : /* If write didn't set errno, assume problem is no disk space. */
502 0 : if (errno == 0)
503 0 : errno = ENOSPC;
504 0 : ereport(ERROR,
505 : (errcode_for_file_access(),
506 : errmsg("could not write to file \"%s\": %m", versionfile)));
507 : }
508 472 : pgstat_report_wait_end();
509 :
510 472 : pgstat_report_wait_start(WAIT_EVENT_VERSION_FILE_SYNC);
511 472 : if (pg_fsync(fd) != 0)
512 0 : ereport(data_sync_elevel(ERROR),
513 : (errcode_for_file_access(),
514 : errmsg("could not fsync file \"%s\": %m", versionfile)));
515 472 : fsync_fname(dbpath, true);
516 472 : pgstat_report_wait_end();
517 :
518 : /* Close the version file. */
519 472 : CloseTransientFile(fd);
520 :
521 : /* If we are not in WAL replay then write the WAL. */
522 472 : if (!isRedo)
523 : {
524 : xl_dbase_create_wal_log_rec xlrec;
525 :
526 432 : START_CRIT_SECTION();
527 :
528 432 : xlrec.db_id = dbid;
529 432 : xlrec.tablespace_id = tsid;
530 :
531 432 : XLogBeginInsert();
532 432 : XLogRegisterData((char *) (&xlrec),
533 : sizeof(xl_dbase_create_wal_log_rec));
534 :
535 432 : (void) XLogInsert(RM_DBASE_ID, XLOG_DBASE_CREATE_WAL_LOG);
536 :
537 432 : END_CRIT_SECTION();
538 : }
539 472 : }
540 :
541 : /*
542 : * Create a new database using the FILE_COPY strategy.
543 : *
544 : * Copy each tablespace at the filesystem level, and log a single WAL record
545 : * for each tablespace copied. This requires a checkpoint before and after the
546 : * copy, which may be expensive, but it does greatly reduce WAL generation
547 : * if the copied database is large.
548 : */
549 : static void
550 170 : CreateDatabaseUsingFileCopy(Oid src_dboid, Oid dst_dboid, Oid src_tsid,
551 : Oid dst_tsid)
552 : {
553 : TableScanDesc scan;
554 : Relation rel;
555 : HeapTuple tuple;
556 :
557 : /*
558 : * Force a checkpoint before starting the copy. This will force all dirty
559 : * buffers, including those of unlogged tables, out to disk, to ensure
560 : * source database is up-to-date on disk for the copy.
561 : * FlushDatabaseBuffers() would suffice for that, but we also want to
562 : * process any pending unlink requests. Otherwise, if a checkpoint
563 : * happened while we're copying files, a file might be deleted just when
564 : * we're about to copy it, causing the lstat() call in copydir() to fail
565 : * with ENOENT.
566 : */
567 170 : RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE |
568 : CHECKPOINT_WAIT | CHECKPOINT_FLUSH_ALL);
569 :
570 : /*
571 : * Iterate through all tablespaces of the template database, and copy each
572 : * one to the new database.
573 : */
574 170 : rel = table_open(TableSpaceRelationId, AccessShareLock);
575 170 : scan = table_beginscan_catalog(rel, 0, NULL);
576 546 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
577 : {
578 376 : Form_pg_tablespace spaceform = (Form_pg_tablespace) GETSTRUCT(tuple);
579 376 : Oid srctablespace = spaceform->oid;
580 : Oid dsttablespace;
581 : char *srcpath;
582 : char *dstpath;
583 : struct stat st;
584 :
585 : /* No need to copy global tablespace */
586 376 : if (srctablespace == GLOBALTABLESPACE_OID)
587 206 : continue;
588 :
589 206 : srcpath = GetDatabasePath(src_dboid, srctablespace);
590 :
591 376 : if (stat(srcpath, &st) < 0 || !S_ISDIR(st.st_mode) ||
592 170 : directory_is_empty(srcpath))
593 : {
594 : /* Assume we can ignore it */
595 36 : pfree(srcpath);
596 36 : continue;
597 : }
598 :
599 170 : if (srctablespace == src_tsid)
600 170 : dsttablespace = dst_tsid;
601 : else
602 0 : dsttablespace = srctablespace;
603 :
604 170 : dstpath = GetDatabasePath(dst_dboid, dsttablespace);
605 :
606 : /*
607 : * Copy this subdirectory to the new location
608 : *
609 : * We don't need to copy subdirectories
610 : */
611 170 : copydir(srcpath, dstpath, false);
612 :
613 : /* Record the filesystem change in XLOG */
614 : {
615 : xl_dbase_create_file_copy_rec xlrec;
616 :
617 170 : xlrec.db_id = dst_dboid;
618 170 : xlrec.tablespace_id = dsttablespace;
619 170 : xlrec.src_db_id = src_dboid;
620 170 : xlrec.src_tablespace_id = srctablespace;
621 :
622 170 : XLogBeginInsert();
623 170 : XLogRegisterData((char *) &xlrec,
624 : sizeof(xl_dbase_create_file_copy_rec));
625 :
626 170 : (void) XLogInsert(RM_DBASE_ID,
627 : XLOG_DBASE_CREATE_FILE_COPY | XLR_SPECIAL_REL_UPDATE);
628 : }
629 170 : pfree(srcpath);
630 170 : pfree(dstpath);
631 : }
632 170 : table_endscan(scan);
633 170 : table_close(rel, AccessShareLock);
634 :
635 : /*
636 : * We force a checkpoint before committing. This effectively means that
637 : * committed XLOG_DBASE_CREATE_FILE_COPY operations will never need to be
638 : * replayed (at least not in ordinary crash recovery; we still have to
639 : * make the XLOG entry for the benefit of PITR operations). This avoids
640 : * two nasty scenarios:
641 : *
642 : * #1: When PITR is off, we don't XLOG the contents of newly created
643 : * indexes; therefore the drop-and-recreate-whole-directory behavior of
644 : * DBASE_CREATE replay would lose such indexes.
645 : *
646 : * #2: Since we have to recopy the source database during DBASE_CREATE
647 : * replay, we run the risk of copying changes in it that were committed
648 : * after the original CREATE DATABASE command but before the system crash
649 : * that led to the replay. This is at least unexpected and at worst could
650 : * lead to inconsistencies, eg duplicate table names.
651 : *
652 : * (Both of these were real bugs in releases 8.0 through 8.0.3.)
653 : *
654 : * In PITR replay, the first of these isn't an issue, and the second is
655 : * only a risk if the CREATE DATABASE and subsequent template database
656 : * change both occur while a base backup is being taken. There doesn't
657 : * seem to be much we can do about that except document it as a
658 : * limitation.
659 : *
660 : * See CreateDatabaseUsingWalLog() for a less cheesy CREATE DATABASE
661 : * strategy that avoids these problems.
662 : */
663 170 : RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT);
664 170 : }
665 :
666 : /*
667 : * CREATE DATABASE
668 : */
669 : Oid
670 636 : createdb(ParseState *pstate, const CreatedbStmt *stmt)
671 : {
672 : Oid src_dboid;
673 : Oid src_owner;
674 636 : int src_encoding = -1;
675 636 : char *src_collate = NULL;
676 636 : char *src_ctype = NULL;
677 636 : char *src_locale = NULL;
678 636 : char *src_icurules = NULL;
679 636 : char src_locprovider = '\0';
680 636 : char *src_collversion = NULL;
681 : bool src_istemplate;
682 636 : bool src_hasloginevt = false;
683 : bool src_allowconn;
684 636 : TransactionId src_frozenxid = InvalidTransactionId;
685 636 : MultiXactId src_minmxid = InvalidMultiXactId;
686 : Oid src_deftablespace;
687 : volatile Oid dst_deftablespace;
688 : Relation pg_database_rel;
689 : HeapTuple tuple;
690 636 : Datum new_record[Natts_pg_database] = {0};
691 636 : bool new_record_nulls[Natts_pg_database] = {0};
692 636 : Oid dboid = InvalidOid;
693 : Oid datdba;
694 : ListCell *option;
695 636 : DefElem *dtablespacename = NULL;
696 636 : DefElem *downer = NULL;
697 636 : DefElem *dtemplate = NULL;
698 636 : DefElem *dencoding = NULL;
699 636 : DefElem *dlocale = NULL;
700 636 : DefElem *dbuiltinlocale = NULL;
701 636 : DefElem *dcollate = NULL;
702 636 : DefElem *dctype = NULL;
703 636 : DefElem *diculocale = NULL;
704 636 : DefElem *dicurules = NULL;
705 636 : DefElem *dlocprovider = NULL;
706 636 : DefElem *distemplate = NULL;
707 636 : DefElem *dallowconnections = NULL;
708 636 : DefElem *dconnlimit = NULL;
709 636 : DefElem *dcollversion = NULL;
710 636 : DefElem *dstrategy = NULL;
711 636 : char *dbname = stmt->dbname;
712 636 : char *dbowner = NULL;
713 636 : const char *dbtemplate = NULL;
714 636 : char *dbcollate = NULL;
715 636 : char *dbctype = NULL;
716 636 : const char *dblocale = NULL;
717 636 : char *dbicurules = NULL;
718 636 : char dblocprovider = '\0';
719 : char *canonname;
720 636 : int encoding = -1;
721 636 : bool dbistemplate = false;
722 636 : bool dballowconnections = true;
723 636 : int dbconnlimit = DATCONNLIMIT_UNLIMITED;
724 636 : char *dbcollversion = NULL;
725 : int notherbackends;
726 : int npreparedxacts;
727 636 : CreateDBStrategy dbstrategy = CREATEDB_WAL_LOG;
728 : createdb_failure_params fparms;
729 :
730 : /* Extract options from the statement node tree */
731 1732 : foreach(option, stmt->options)
732 : {
733 1096 : DefElem *defel = (DefElem *) lfirst(option);
734 :
735 1096 : if (strcmp(defel->defname, "tablespace") == 0)
736 : {
737 16 : if (dtablespacename)
738 0 : errorConflictingDefElem(defel, pstate);
739 16 : dtablespacename = defel;
740 : }
741 1080 : else if (strcmp(defel->defname, "owner") == 0)
742 : {
743 2 : if (downer)
744 0 : errorConflictingDefElem(defel, pstate);
745 2 : downer = defel;
746 : }
747 1078 : else if (strcmp(defel->defname, "template") == 0)
748 : {
749 276 : if (dtemplate)
750 0 : errorConflictingDefElem(defel, pstate);
751 276 : dtemplate = defel;
752 : }
753 802 : else if (strcmp(defel->defname, "encoding") == 0)
754 : {
755 60 : if (dencoding)
756 0 : errorConflictingDefElem(defel, pstate);
757 60 : dencoding = defel;
758 : }
759 742 : else if (strcmp(defel->defname, "locale") == 0)
760 : {
761 70 : if (dlocale)
762 0 : errorConflictingDefElem(defel, pstate);
763 70 : dlocale = defel;
764 : }
765 672 : else if (strcmp(defel->defname, "builtin_locale") == 0)
766 : {
767 16 : if (dbuiltinlocale)
768 0 : errorConflictingDefElem(defel, pstate);
769 16 : dbuiltinlocale = defel;
770 : }
771 656 : else if (strcmp(defel->defname, "lc_collate") == 0)
772 : {
773 12 : if (dcollate)
774 0 : errorConflictingDefElem(defel, pstate);
775 12 : dcollate = defel;
776 : }
777 644 : else if (strcmp(defel->defname, "lc_ctype") == 0)
778 : {
779 12 : if (dctype)
780 0 : errorConflictingDefElem(defel, pstate);
781 12 : dctype = defel;
782 : }
783 632 : else if (strcmp(defel->defname, "icu_locale") == 0)
784 : {
785 10 : if (diculocale)
786 0 : errorConflictingDefElem(defel, pstate);
787 10 : diculocale = defel;
788 : }
789 622 : else if (strcmp(defel->defname, "icu_rules") == 0)
790 : {
791 2 : if (dicurules)
792 0 : errorConflictingDefElem(defel, pstate);
793 2 : dicurules = defel;
794 : }
795 620 : else if (strcmp(defel->defname, "locale_provider") == 0)
796 : {
797 76 : if (dlocprovider)
798 0 : errorConflictingDefElem(defel, pstate);
799 76 : dlocprovider = defel;
800 : }
801 544 : else if (strcmp(defel->defname, "is_template") == 0)
802 : {
803 80 : if (distemplate)
804 0 : errorConflictingDefElem(defel, pstate);
805 80 : distemplate = defel;
806 : }
807 464 : else if (strcmp(defel->defname, "allow_connections") == 0)
808 : {
809 78 : if (dallowconnections)
810 0 : errorConflictingDefElem(defel, pstate);
811 78 : dallowconnections = defel;
812 : }
813 386 : else if (strcmp(defel->defname, "connection_limit") == 0)
814 : {
815 0 : if (dconnlimit)
816 0 : errorConflictingDefElem(defel, pstate);
817 0 : dconnlimit = defel;
818 : }
819 386 : else if (strcmp(defel->defname, "collation_version") == 0)
820 : {
821 20 : if (dcollversion)
822 0 : errorConflictingDefElem(defel, pstate);
823 20 : dcollversion = defel;
824 : }
825 366 : else if (strcmp(defel->defname, "location") == 0)
826 : {
827 0 : ereport(WARNING,
828 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
829 : errmsg("LOCATION is not supported anymore"),
830 : errhint("Consider using tablespaces instead."),
831 : parser_errposition(pstate, defel->location)));
832 : }
833 366 : else if (strcmp(defel->defname, "oid") == 0)
834 : {
835 182 : dboid = defGetObjectId(defel);
836 :
837 : /*
838 : * We don't normally permit new databases to be created with
839 : * system-assigned OIDs. pg_upgrade tries to preserve database
840 : * OIDs, so we can't allow any database to be created with an OID
841 : * that might be in use in a freshly-initialized cluster created
842 : * by some future version. We assume all such OIDs will be from
843 : * the system-managed OID range.
844 : *
845 : * As an exception, however, we permit any OID to be assigned when
846 : * allow_system_table_mods=on (so that initdb can assign system
847 : * OIDs to template0 and postgres) or when performing a binary
848 : * upgrade (so that pg_upgrade can preserve whatever OIDs it finds
849 : * in the source cluster).
850 : */
851 182 : if (dboid < FirstNormalObjectId &&
852 164 : !allowSystemTableMods && !IsBinaryUpgrade)
853 0 : ereport(ERROR,
854 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE)),
855 : errmsg("OIDs less than %u are reserved for system objects", FirstNormalObjectId));
856 : }
857 184 : else if (strcmp(defel->defname, "strategy") == 0)
858 : {
859 184 : if (dstrategy)
860 0 : errorConflictingDefElem(defel, pstate);
861 184 : dstrategy = defel;
862 : }
863 : else
864 0 : ereport(ERROR,
865 : (errcode(ERRCODE_SYNTAX_ERROR),
866 : errmsg("option \"%s\" not recognized", defel->defname),
867 : parser_errposition(pstate, defel->location)));
868 : }
869 :
870 636 : if (downer && downer->arg)
871 2 : dbowner = defGetString(downer);
872 636 : if (dtemplate && dtemplate->arg)
873 276 : dbtemplate = defGetString(dtemplate);
874 636 : if (dencoding && dencoding->arg)
875 : {
876 : const char *encoding_name;
877 :
878 60 : if (IsA(dencoding->arg, Integer))
879 : {
880 0 : encoding = defGetInt32(dencoding);
881 0 : encoding_name = pg_encoding_to_char(encoding);
882 0 : if (strcmp(encoding_name, "") == 0 ||
883 0 : pg_valid_server_encoding(encoding_name) < 0)
884 0 : ereport(ERROR,
885 : (errcode(ERRCODE_UNDEFINED_OBJECT),
886 : errmsg("%d is not a valid encoding code",
887 : encoding),
888 : parser_errposition(pstate, dencoding->location)));
889 : }
890 : else
891 : {
892 60 : encoding_name = defGetString(dencoding);
893 60 : encoding = pg_valid_server_encoding(encoding_name);
894 60 : if (encoding < 0)
895 0 : ereport(ERROR,
896 : (errcode(ERRCODE_UNDEFINED_OBJECT),
897 : errmsg("%s is not a valid encoding name",
898 : encoding_name),
899 : parser_errposition(pstate, dencoding->location)));
900 : }
901 : }
902 636 : if (dlocale && dlocale->arg)
903 : {
904 70 : dbcollate = defGetString(dlocale);
905 70 : dbctype = defGetString(dlocale);
906 70 : dblocale = defGetString(dlocale);
907 : }
908 636 : if (dbuiltinlocale && dbuiltinlocale->arg)
909 16 : dblocale = defGetString(dbuiltinlocale);
910 636 : if (dcollate && dcollate->arg)
911 12 : dbcollate = defGetString(dcollate);
912 636 : if (dctype && dctype->arg)
913 12 : dbctype = defGetString(dctype);
914 636 : if (diculocale && diculocale->arg)
915 10 : dblocale = defGetString(diculocale);
916 636 : if (dicurules && dicurules->arg)
917 2 : dbicurules = defGetString(dicurules);
918 636 : if (dlocprovider && dlocprovider->arg)
919 : {
920 76 : char *locproviderstr = defGetString(dlocprovider);
921 :
922 76 : if (pg_strcasecmp(locproviderstr, "builtin") == 0)
923 30 : dblocprovider = COLLPROVIDER_BUILTIN;
924 46 : else if (pg_strcasecmp(locproviderstr, "icu") == 0)
925 16 : dblocprovider = COLLPROVIDER_ICU;
926 30 : else if (pg_strcasecmp(locproviderstr, "libc") == 0)
927 28 : dblocprovider = COLLPROVIDER_LIBC;
928 : else
929 2 : ereport(ERROR,
930 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
931 : errmsg("unrecognized locale provider: %s",
932 : locproviderstr)));
933 : }
934 634 : if (distemplate && distemplate->arg)
935 80 : dbistemplate = defGetBoolean(distemplate);
936 634 : if (dallowconnections && dallowconnections->arg)
937 78 : dballowconnections = defGetBoolean(dallowconnections);
938 634 : if (dconnlimit && dconnlimit->arg)
939 : {
940 0 : dbconnlimit = defGetInt32(dconnlimit);
941 0 : if (dbconnlimit < DATCONNLIMIT_UNLIMITED)
942 0 : ereport(ERROR,
943 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
944 : errmsg("invalid connection limit: %d", dbconnlimit)));
945 : }
946 634 : if (dcollversion)
947 20 : dbcollversion = defGetString(dcollversion);
948 :
949 : /* obtain OID of proposed owner */
950 634 : if (dbowner)
951 2 : datdba = get_role_oid(dbowner, false);
952 : else
953 632 : datdba = GetUserId();
954 :
955 : /*
956 : * To create a database, must have createdb privilege and must be able to
957 : * become the target role (this does not imply that the target role itself
958 : * must have createdb privilege). The latter provision guards against
959 : * "giveaway" attacks. Note that a superuser will always have both of
960 : * these privileges a fortiori.
961 : */
962 634 : if (!have_createdb_privilege())
963 6 : ereport(ERROR,
964 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
965 : errmsg("permission denied to create database")));
966 :
967 628 : check_can_set_role(GetUserId(), datdba);
968 :
969 : /*
970 : * Lookup database (template) to be cloned, and obtain share lock on it.
971 : * ShareLock allows two CREATE DATABASEs to work from the same template
972 : * concurrently, while ensuring no one is busy dropping it in parallel
973 : * (which would be Very Bad since we'd likely get an incomplete copy
974 : * without knowing it). This also prevents any new connections from being
975 : * made to the source until we finish copying it, so we can be sure it
976 : * won't change underneath us.
977 : */
978 628 : if (!dbtemplate)
979 354 : dbtemplate = "template1"; /* Default template database name */
980 :
981 628 : if (!get_db_info(dbtemplate, ShareLock,
982 : &src_dboid, &src_owner, &src_encoding,
983 : &src_istemplate, &src_allowconn, &src_hasloginevt,
984 : &src_frozenxid, &src_minmxid, &src_deftablespace,
985 : &src_collate, &src_ctype, &src_locale, &src_icurules, &src_locprovider,
986 : &src_collversion))
987 0 : ereport(ERROR,
988 : (errcode(ERRCODE_UNDEFINED_DATABASE),
989 : errmsg("template database \"%s\" does not exist",
990 : dbtemplate)));
991 :
992 : /*
993 : * If the source database was in the process of being dropped, we can't
994 : * use it as a template.
995 : */
996 628 : if (database_is_invalid_oid(src_dboid))
997 2 : ereport(ERROR,
998 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
999 : errmsg("cannot use invalid database \"%s\" as template", dbtemplate),
1000 : errhint("Use DROP DATABASE to drop invalid databases."));
1001 :
1002 : /*
1003 : * Permission check: to copy a DB that's not marked datistemplate, you
1004 : * must be superuser or the owner thereof.
1005 : */
1006 626 : if (!src_istemplate)
1007 : {
1008 12 : if (!object_ownercheck(DatabaseRelationId, src_dboid, GetUserId()))
1009 0 : ereport(ERROR,
1010 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1011 : errmsg("permission denied to copy database \"%s\"",
1012 : dbtemplate)));
1013 : }
1014 :
1015 : /* Validate the database creation strategy. */
1016 626 : if (dstrategy && dstrategy->arg)
1017 : {
1018 : char *strategy;
1019 :
1020 184 : strategy = defGetString(dstrategy);
1021 184 : if (strcmp(strategy, "wal_log") == 0)
1022 12 : dbstrategy = CREATEDB_WAL_LOG;
1023 172 : else if (strcmp(strategy, "file_copy") == 0)
1024 170 : dbstrategy = CREATEDB_FILE_COPY;
1025 : else
1026 2 : ereport(ERROR,
1027 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1028 : errmsg("invalid create database strategy \"%s\"", strategy),
1029 : errhint("Valid strategies are \"wal_log\", and \"file_copy\".")));
1030 : }
1031 :
1032 : /* If encoding or locales are defaulted, use source's setting */
1033 624 : if (encoding < 0)
1034 564 : encoding = src_encoding;
1035 624 : if (dbcollate == NULL)
1036 548 : dbcollate = src_collate;
1037 624 : if (dbctype == NULL)
1038 548 : dbctype = src_ctype;
1039 624 : if (dblocprovider == '\0')
1040 550 : dblocprovider = src_locprovider;
1041 624 : if (dblocale == NULL)
1042 546 : dblocale = src_locale;
1043 624 : if (dbicurules == NULL)
1044 622 : dbicurules = src_icurules;
1045 :
1046 : /* Some encodings are client only */
1047 624 : if (!PG_VALID_BE_ENCODING(encoding))
1048 0 : ereport(ERROR,
1049 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1050 : errmsg("invalid server encoding %d", encoding)));
1051 :
1052 : /* Check that the chosen locales are valid, and get canonical spellings */
1053 624 : if (!check_locale(LC_COLLATE, dbcollate, &canonname))
1054 2 : ereport(ERROR,
1055 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1056 : errmsg("invalid LC_COLLATE locale name: \"%s\"", dbcollate),
1057 : errhint("If the locale name is specific to ICU, use ICU_LOCALE.")));
1058 622 : dbcollate = canonname;
1059 622 : if (!check_locale(LC_CTYPE, dbctype, &canonname))
1060 2 : ereport(ERROR,
1061 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1062 : errmsg("invalid LC_CTYPE locale name: \"%s\"", dbctype),
1063 : errhint("If the locale name is specific to ICU, use ICU_LOCALE.")));
1064 620 : dbctype = canonname;
1065 :
1066 620 : check_encoding_locale_matches(encoding, dbcollate, dbctype);
1067 :
1068 : /* validate provider-specific parameters */
1069 620 : if (dblocprovider != COLLPROVIDER_BUILTIN)
1070 : {
1071 564 : if (dbuiltinlocale)
1072 0 : ereport(ERROR,
1073 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1074 : errmsg("BUILTIN_LOCALE cannot be specified unless locale provider is builtin")));
1075 : }
1076 56 : else if (dblocprovider != COLLPROVIDER_ICU)
1077 : {
1078 56 : if (diculocale)
1079 2 : ereport(ERROR,
1080 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1081 : errmsg("ICU locale cannot be specified unless locale provider is ICU")));
1082 :
1083 54 : if (dbicurules)
1084 2 : ereport(ERROR,
1085 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
1086 : errmsg("ICU rules cannot be specified unless locale provider is ICU")));
1087 : }
1088 :
1089 : /* validate and canonicalize locale for the provider */
1090 616 : if (dblocprovider == COLLPROVIDER_BUILTIN)
1091 : {
1092 : /*
1093 : * This would happen if template0 uses the libc provider but the new
1094 : * database uses builtin.
1095 : */
1096 52 : if (!dblocale)
1097 2 : ereport(ERROR,
1098 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1099 : errmsg("LOCALE or BUILTIN_LOCALE must be specified")));
1100 :
1101 50 : dblocale = builtin_validate_locale(encoding, dblocale);
1102 : }
1103 564 : else if (dblocprovider == COLLPROVIDER_ICU)
1104 : {
1105 30 : if (!(is_encoding_supported_by_icu(encoding)))
1106 2 : ereport(ERROR,
1107 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1108 : errmsg("encoding \"%s\" is not supported with ICU provider",
1109 : pg_encoding_to_char(encoding))));
1110 :
1111 : /*
1112 : * This would happen if template0 uses the libc provider but the new
1113 : * database uses icu.
1114 : */
1115 28 : if (!dblocale)
1116 2 : ereport(ERROR,
1117 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1118 : errmsg("LOCALE or ICU_LOCALE must be specified")));
1119 :
1120 : /*
1121 : * During binary upgrade, or when the locale came from the template
1122 : * database, preserve locale string. Otherwise, canonicalize to a
1123 : * language tag.
1124 : */
1125 26 : if (!IsBinaryUpgrade && dblocale != src_locale)
1126 : {
1127 14 : char *langtag = icu_language_tag(dblocale,
1128 : icu_validation_level);
1129 :
1130 14 : if (langtag && strcmp(dblocale, langtag) != 0)
1131 : {
1132 6 : ereport(NOTICE,
1133 : (errmsg("using standard form \"%s\" for ICU locale \"%s\"",
1134 : langtag, dblocale)));
1135 :
1136 6 : dblocale = langtag;
1137 : }
1138 : }
1139 :
1140 26 : icu_validate_locale(dblocale);
1141 : }
1142 :
1143 : /* for libc, locale comes from datcollate and datctype */
1144 606 : if (dblocprovider == COLLPROVIDER_LIBC)
1145 534 : dblocale = NULL;
1146 :
1147 : /*
1148 : * Check that the new encoding and locale settings match the source
1149 : * database. We insist on this because we simply copy the source data ---
1150 : * any non-ASCII data would be wrongly encoded, and any indexes sorted
1151 : * according to the source locale would be wrong.
1152 : *
1153 : * However, we assume that template0 doesn't contain any non-ASCII data
1154 : * nor any indexes that depend on collation or ctype, so template0 can be
1155 : * used as template for creating a database with any encoding or locale.
1156 : */
1157 606 : if (strcmp(dbtemplate, "template0") != 0)
1158 : {
1159 368 : if (encoding != src_encoding)
1160 0 : ereport(ERROR,
1161 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1162 : errmsg("new encoding (%s) is incompatible with the encoding of the template database (%s)",
1163 : pg_encoding_to_char(encoding),
1164 : pg_encoding_to_char(src_encoding)),
1165 : errhint("Use the same encoding as in the template database, or use template0 as template.")));
1166 :
1167 368 : if (strcmp(dbcollate, src_collate) != 0)
1168 2 : ereport(ERROR,
1169 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1170 : errmsg("new collation (%s) is incompatible with the collation of the template database (%s)",
1171 : dbcollate, src_collate),
1172 : errhint("Use the same collation as in the template database, or use template0 as template.")));
1173 :
1174 366 : if (strcmp(dbctype, src_ctype) != 0)
1175 0 : ereport(ERROR,
1176 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1177 : errmsg("new LC_CTYPE (%s) is incompatible with the LC_CTYPE of the template database (%s)",
1178 : dbctype, src_ctype),
1179 : errhint("Use the same LC_CTYPE as in the template database, or use template0 as template.")));
1180 :
1181 366 : if (dblocprovider != src_locprovider)
1182 0 : ereport(ERROR,
1183 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1184 : errmsg("new locale provider (%s) does not match locale provider of the template database (%s)",
1185 : collprovider_name(dblocprovider), collprovider_name(src_locprovider)),
1186 : errhint("Use the same locale provider as in the template database, or use template0 as template.")));
1187 :
1188 366 : if (dblocprovider == COLLPROVIDER_ICU)
1189 : {
1190 : char *val1;
1191 : char *val2;
1192 :
1193 : Assert(dblocale);
1194 : Assert(src_locale);
1195 12 : if (strcmp(dblocale, src_locale) != 0)
1196 0 : ereport(ERROR,
1197 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1198 : errmsg("new ICU locale (%s) is incompatible with the ICU locale of the template database (%s)",
1199 : dblocale, src_locale),
1200 : errhint("Use the same ICU locale as in the template database, or use template0 as template.")));
1201 :
1202 12 : val1 = dbicurules;
1203 12 : if (!val1)
1204 12 : val1 = "";
1205 12 : val2 = src_icurules;
1206 12 : if (!val2)
1207 12 : val2 = "";
1208 12 : if (strcmp(val1, val2) != 0)
1209 0 : ereport(ERROR,
1210 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1211 : errmsg("new ICU collation rules (%s) are incompatible with the ICU collation rules of the template database (%s)",
1212 : val1, val2),
1213 : errhint("Use the same ICU collation rules as in the template database, or use template0 as template.")));
1214 : }
1215 : }
1216 :
1217 : /*
1218 : * If we got a collation version for the template database, check that it
1219 : * matches the actual OS collation version. Otherwise error; the user
1220 : * needs to fix the template database first. Don't complain if a
1221 : * collation version was specified explicitly as a statement option; that
1222 : * is used by pg_upgrade to reproduce the old state exactly.
1223 : *
1224 : * (If the template database has no collation version, then either the
1225 : * platform/provider does not support collation versioning, or it's
1226 : * template0, for which we stipulate that it does not contain
1227 : * collation-using objects.)
1228 : */
1229 604 : if (src_collversion && !dcollversion)
1230 : {
1231 : char *actual_versionstr;
1232 : const char *locale;
1233 :
1234 232 : if (dblocprovider == COLLPROVIDER_LIBC)
1235 210 : locale = dbcollate;
1236 : else
1237 22 : locale = dblocale;
1238 :
1239 232 : actual_versionstr = get_collation_actual_version(dblocprovider, locale);
1240 232 : if (!actual_versionstr)
1241 0 : ereport(ERROR,
1242 : (errmsg("template database \"%s\" has a collation version, but no actual collation version could be determined",
1243 : dbtemplate)));
1244 :
1245 232 : if (strcmp(actual_versionstr, src_collversion) != 0)
1246 0 : ereport(ERROR,
1247 : (errmsg("template database \"%s\" has a collation version mismatch",
1248 : dbtemplate),
1249 : errdetail("The template database was created using collation version %s, "
1250 : "but the operating system provides version %s.",
1251 : src_collversion, actual_versionstr),
1252 : errhint("Rebuild all objects in the template database that use the default collation and run "
1253 : "ALTER DATABASE %s REFRESH COLLATION VERSION, "
1254 : "or build PostgreSQL with the right library version.",
1255 : quote_identifier(dbtemplate))));
1256 : }
1257 :
1258 604 : if (dbcollversion == NULL)
1259 584 : dbcollversion = src_collversion;
1260 :
1261 : /*
1262 : * Normally, we copy the collation version from the template database.
1263 : * This last resort only applies if the template database does not have a
1264 : * collation version, which is normally only the case for template0.
1265 : */
1266 604 : if (dbcollversion == NULL)
1267 : {
1268 : const char *locale;
1269 :
1270 352 : if (dblocprovider == COLLPROVIDER_LIBC)
1271 316 : locale = dbcollate;
1272 : else
1273 36 : locale = dblocale;
1274 :
1275 352 : dbcollversion = get_collation_actual_version(dblocprovider, locale);
1276 : }
1277 :
1278 : /* Resolve default tablespace for new database */
1279 604 : if (dtablespacename && dtablespacename->arg)
1280 16 : {
1281 : char *tablespacename;
1282 : AclResult aclresult;
1283 :
1284 16 : tablespacename = defGetString(dtablespacename);
1285 16 : dst_deftablespace = get_tablespace_oid(tablespacename, false);
1286 : /* check permissions */
1287 16 : aclresult = object_aclcheck(TableSpaceRelationId, dst_deftablespace, GetUserId(),
1288 : ACL_CREATE);
1289 16 : if (aclresult != ACLCHECK_OK)
1290 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
1291 : tablespacename);
1292 :
1293 : /* pg_global must never be the default tablespace */
1294 16 : if (dst_deftablespace == GLOBALTABLESPACE_OID)
1295 0 : ereport(ERROR,
1296 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1297 : errmsg("pg_global cannot be used as default tablespace")));
1298 :
1299 : /*
1300 : * If we are trying to change the default tablespace of the template,
1301 : * we require that the template not have any files in the new default
1302 : * tablespace. This is necessary because otherwise the copied
1303 : * database would contain pg_class rows that refer to its default
1304 : * tablespace both explicitly (by OID) and implicitly (as zero), which
1305 : * would cause problems. For example another CREATE DATABASE using
1306 : * the copied database as template, and trying to change its default
1307 : * tablespace again, would yield outright incorrect results (it would
1308 : * improperly move tables to the new default tablespace that should
1309 : * stay in the same tablespace).
1310 : */
1311 16 : if (dst_deftablespace != src_deftablespace)
1312 : {
1313 : char *srcpath;
1314 : struct stat st;
1315 :
1316 16 : srcpath = GetDatabasePath(src_dboid, dst_deftablespace);
1317 :
1318 16 : if (stat(srcpath, &st) == 0 &&
1319 0 : S_ISDIR(st.st_mode) &&
1320 0 : !directory_is_empty(srcpath))
1321 0 : ereport(ERROR,
1322 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1323 : errmsg("cannot assign new default tablespace \"%s\"",
1324 : tablespacename),
1325 : errdetail("There is a conflict because database \"%s\" already has some tables in this tablespace.",
1326 : dbtemplate)));
1327 16 : pfree(srcpath);
1328 : }
1329 : }
1330 : else
1331 : {
1332 : /* Use template database's default tablespace */
1333 588 : dst_deftablespace = src_deftablespace;
1334 : /* Note there is no additional permission check in this path */
1335 : }
1336 :
1337 : /*
1338 : * If built with appropriate switch, whine when regression-testing
1339 : * conventions for database names are violated. But don't complain during
1340 : * initdb.
1341 : */
1342 : #ifdef ENFORCE_REGRESSION_TEST_NAME_RESTRICTIONS
1343 : if (IsUnderPostmaster && strstr(dbname, "regression") == NULL)
1344 : elog(WARNING, "databases created by regression test cases should have names including \"regression\"");
1345 : #endif
1346 :
1347 : /*
1348 : * Check for db name conflict. This is just to give a more friendly error
1349 : * message than "unique index violation". There's a race condition but
1350 : * we're willing to accept the less friendly message in that case.
1351 : */
1352 604 : if (OidIsValid(get_database_oid(dbname, true)))
1353 2 : ereport(ERROR,
1354 : (errcode(ERRCODE_DUPLICATE_DATABASE),
1355 : errmsg("database \"%s\" already exists", dbname)));
1356 :
1357 : /*
1358 : * The source DB can't have any active backends, except this one
1359 : * (exception is to allow CREATE DB while connected to template1).
1360 : * Otherwise we might copy inconsistent data.
1361 : *
1362 : * This should be last among the basic error checks, because it involves
1363 : * potential waiting; we may as well throw an error first if we're gonna
1364 : * throw one.
1365 : */
1366 602 : if (CountOtherDBBackends(src_dboid, ¬herbackends, &npreparedxacts))
1367 0 : ereport(ERROR,
1368 : (errcode(ERRCODE_OBJECT_IN_USE),
1369 : errmsg("source database \"%s\" is being accessed by other users",
1370 : dbtemplate),
1371 : errdetail_busy_db(notherbackends, npreparedxacts)));
1372 :
1373 : /*
1374 : * Select an OID for the new database, checking that it doesn't have a
1375 : * filename conflict with anything already existing in the tablespace
1376 : * directories.
1377 : */
1378 602 : pg_database_rel = table_open(DatabaseRelationId, RowExclusiveLock);
1379 :
1380 : /*
1381 : * If database OID is configured, check if the OID is already in use or
1382 : * data directory already exists.
1383 : */
1384 602 : if (OidIsValid(dboid))
1385 : {
1386 182 : char *existing_dbname = get_database_name(dboid);
1387 :
1388 182 : if (existing_dbname != NULL)
1389 0 : ereport(ERROR,
1390 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE)),
1391 : errmsg("database OID %u is already in use by database \"%s\"",
1392 : dboid, existing_dbname));
1393 :
1394 182 : if (check_db_file_conflict(dboid))
1395 0 : ereport(ERROR,
1396 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE)),
1397 : errmsg("data directory with the specified OID %u already exists", dboid));
1398 : }
1399 : else
1400 : {
1401 : /* Select an OID for the new database if is not explicitly configured. */
1402 : do
1403 : {
1404 420 : dboid = GetNewOidWithIndex(pg_database_rel, DatabaseOidIndexId,
1405 : Anum_pg_database_oid);
1406 420 : } while (check_db_file_conflict(dboid));
1407 : }
1408 :
1409 : /*
1410 : * Insert a new tuple into pg_database. This establishes our ownership of
1411 : * the new database name (anyone else trying to insert the same name will
1412 : * block on the unique index, and fail after we commit).
1413 : */
1414 :
1415 : Assert((dblocprovider != COLLPROVIDER_LIBC && dblocale) ||
1416 : (dblocprovider == COLLPROVIDER_LIBC && !dblocale));
1417 :
1418 : /* Form tuple */
1419 602 : new_record[Anum_pg_database_oid - 1] = ObjectIdGetDatum(dboid);
1420 602 : new_record[Anum_pg_database_datname - 1] =
1421 602 : DirectFunctionCall1(namein, CStringGetDatum(dbname));
1422 602 : new_record[Anum_pg_database_datdba - 1] = ObjectIdGetDatum(datdba);
1423 602 : new_record[Anum_pg_database_encoding - 1] = Int32GetDatum(encoding);
1424 602 : new_record[Anum_pg_database_datlocprovider - 1] = CharGetDatum(dblocprovider);
1425 602 : new_record[Anum_pg_database_datistemplate - 1] = BoolGetDatum(dbistemplate);
1426 602 : new_record[Anum_pg_database_datallowconn - 1] = BoolGetDatum(dballowconnections);
1427 602 : new_record[Anum_pg_database_dathasloginevt - 1] = BoolGetDatum(src_hasloginevt);
1428 602 : new_record[Anum_pg_database_datconnlimit - 1] = Int32GetDatum(dbconnlimit);
1429 602 : new_record[Anum_pg_database_datfrozenxid - 1] = TransactionIdGetDatum(src_frozenxid);
1430 602 : new_record[Anum_pg_database_datminmxid - 1] = TransactionIdGetDatum(src_minmxid);
1431 602 : new_record[Anum_pg_database_dattablespace - 1] = ObjectIdGetDatum(dst_deftablespace);
1432 602 : new_record[Anum_pg_database_datcollate - 1] = CStringGetTextDatum(dbcollate);
1433 602 : new_record[Anum_pg_database_datctype - 1] = CStringGetTextDatum(dbctype);
1434 602 : if (dblocale)
1435 70 : new_record[Anum_pg_database_datlocale - 1] = CStringGetTextDatum(dblocale);
1436 : else
1437 532 : new_record_nulls[Anum_pg_database_datlocale - 1] = true;
1438 602 : if (dbicurules)
1439 0 : new_record[Anum_pg_database_daticurules - 1] = CStringGetTextDatum(dbicurules);
1440 : else
1441 602 : new_record_nulls[Anum_pg_database_daticurules - 1] = true;
1442 602 : if (dbcollversion)
1443 490 : new_record[Anum_pg_database_datcollversion - 1] = CStringGetTextDatum(dbcollversion);
1444 : else
1445 112 : new_record_nulls[Anum_pg_database_datcollversion - 1] = true;
1446 :
1447 : /*
1448 : * We deliberately set datacl to default (NULL), rather than copying it
1449 : * from the template database. Copying it would be a bad idea when the
1450 : * owner is not the same as the template's owner.
1451 : */
1452 602 : new_record_nulls[Anum_pg_database_datacl - 1] = true;
1453 :
1454 602 : tuple = heap_form_tuple(RelationGetDescr(pg_database_rel),
1455 : new_record, new_record_nulls);
1456 :
1457 602 : CatalogTupleInsert(pg_database_rel, tuple);
1458 :
1459 : /*
1460 : * Now generate additional catalog entries associated with the new DB
1461 : */
1462 :
1463 : /* Register owner dependency */
1464 602 : recordDependencyOnOwner(DatabaseRelationId, dboid, datdba);
1465 :
1466 : /* Create pg_shdepend entries for objects within database */
1467 602 : copyTemplateDependencies(src_dboid, dboid);
1468 :
1469 : /* Post creation hook for new database */
1470 602 : InvokeObjectPostCreateHook(DatabaseRelationId, dboid, 0);
1471 :
1472 : /*
1473 : * If we're going to be reading data for the to-be-created database into
1474 : * shared_buffers, take a lock on it. Nobody should know that this
1475 : * database exists yet, but it's good to maintain the invariant that an
1476 : * AccessExclusiveLock on the database is sufficient to drop all of its
1477 : * buffers without worrying about more being read later.
1478 : *
1479 : * Note that we need to do this before entering the
1480 : * PG_ENSURE_ERROR_CLEANUP block below, because createdb_failure_callback
1481 : * expects this lock to be held already.
1482 : */
1483 602 : if (dbstrategy == CREATEDB_WAL_LOG)
1484 432 : LockSharedObject(DatabaseRelationId, dboid, 0, AccessShareLock);
1485 :
1486 : /*
1487 : * Once we start copying subdirectories, we need to be able to clean 'em
1488 : * up if we fail. Use an ENSURE block to make sure this happens. (This
1489 : * is not a 100% solution, because of the possibility of failure during
1490 : * transaction commit after we leave this routine, but it should handle
1491 : * most scenarios.)
1492 : */
1493 602 : fparms.src_dboid = src_dboid;
1494 602 : fparms.dest_dboid = dboid;
1495 602 : fparms.strategy = dbstrategy;
1496 :
1497 602 : PG_ENSURE_ERROR_CLEANUP(createdb_failure_callback,
1498 : PointerGetDatum(&fparms));
1499 : {
1500 : /*
1501 : * If the user has asked to create a database with WAL_LOG strategy
1502 : * then call CreateDatabaseUsingWalLog, which will copy the database
1503 : * at the block level and it will WAL log each copied block.
1504 : * Otherwise, call CreateDatabaseUsingFileCopy that will copy the
1505 : * database file by file.
1506 : */
1507 602 : if (dbstrategy == CREATEDB_WAL_LOG)
1508 432 : CreateDatabaseUsingWalLog(src_dboid, dboid, src_deftablespace,
1509 : dst_deftablespace);
1510 : else
1511 170 : CreateDatabaseUsingFileCopy(src_dboid, dboid, src_deftablespace,
1512 : dst_deftablespace);
1513 :
1514 : /*
1515 : * Close pg_database, but keep lock till commit.
1516 : */
1517 602 : table_close(pg_database_rel, NoLock);
1518 :
1519 : /*
1520 : * Force synchronous commit, thus minimizing the window between
1521 : * creation of the database files and committal of the transaction. If
1522 : * we crash before committing, we'll have a DB that's taking up disk
1523 : * space but is not in pg_database, which is not good.
1524 : */
1525 602 : ForceSyncCommit();
1526 : }
1527 602 : PG_END_ENSURE_ERROR_CLEANUP(createdb_failure_callback,
1528 : PointerGetDatum(&fparms));
1529 :
1530 602 : return dboid;
1531 : }
1532 :
1533 : /*
1534 : * Check whether chosen encoding matches chosen locale settings. This
1535 : * restriction is necessary because libc's locale-specific code usually
1536 : * fails when presented with data in an encoding it's not expecting. We
1537 : * allow mismatch in four cases:
1538 : *
1539 : * 1. locale encoding = SQL_ASCII, which means that the locale is C/POSIX
1540 : * which works with any encoding.
1541 : *
1542 : * 2. locale encoding = -1, which means that we couldn't determine the
1543 : * locale's encoding and have to trust the user to get it right.
1544 : *
1545 : * 3. selected encoding is UTF8 and platform is win32. This is because
1546 : * UTF8 is a pseudo codepage that is supported in all locales since it's
1547 : * converted to UTF16 before being used.
1548 : *
1549 : * 4. selected encoding is SQL_ASCII, but only if you're a superuser. This
1550 : * is risky but we have historically allowed it --- notably, the
1551 : * regression tests require it.
1552 : *
1553 : * Note: if you change this policy, fix initdb to match.
1554 : */
1555 : void
1556 648 : check_encoding_locale_matches(int encoding, const char *collate, const char *ctype)
1557 : {
1558 648 : int ctype_encoding = pg_get_encoding_from_locale(ctype, true);
1559 648 : int collate_encoding = pg_get_encoding_from_locale(collate, true);
1560 :
1561 654 : if (!(ctype_encoding == encoding ||
1562 6 : ctype_encoding == PG_SQL_ASCII ||
1563 : ctype_encoding == -1 ||
1564 : #ifdef WIN32
1565 : encoding == PG_UTF8 ||
1566 : #endif
1567 6 : (encoding == PG_SQL_ASCII && superuser())))
1568 0 : ereport(ERROR,
1569 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1570 : errmsg("encoding \"%s\" does not match locale \"%s\"",
1571 : pg_encoding_to_char(encoding),
1572 : ctype),
1573 : errdetail("The chosen LC_CTYPE setting requires encoding \"%s\".",
1574 : pg_encoding_to_char(ctype_encoding))));
1575 :
1576 654 : if (!(collate_encoding == encoding ||
1577 6 : collate_encoding == PG_SQL_ASCII ||
1578 : collate_encoding == -1 ||
1579 : #ifdef WIN32
1580 : encoding == PG_UTF8 ||
1581 : #endif
1582 6 : (encoding == PG_SQL_ASCII && superuser())))
1583 0 : ereport(ERROR,
1584 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1585 : errmsg("encoding \"%s\" does not match locale \"%s\"",
1586 : pg_encoding_to_char(encoding),
1587 : collate),
1588 : errdetail("The chosen LC_COLLATE setting requires encoding \"%s\".",
1589 : pg_encoding_to_char(collate_encoding))));
1590 648 : }
1591 :
1592 : /* Error cleanup callback for createdb */
1593 : static void
1594 0 : createdb_failure_callback(int code, Datum arg)
1595 : {
1596 0 : createdb_failure_params *fparms = (createdb_failure_params *) DatumGetPointer(arg);
1597 :
1598 : /*
1599 : * If we were copying database at block levels then drop pages for the
1600 : * destination database that are in the shared buffer cache. And tell
1601 : * checkpointer to forget any pending fsync and unlink requests for files
1602 : * in the database. The reasoning behind doing this is same as explained
1603 : * in dropdb function. But unlike dropdb we don't need to call
1604 : * pgstat_drop_database because this database is still not created so
1605 : * there should not be any stat for this.
1606 : */
1607 0 : if (fparms->strategy == CREATEDB_WAL_LOG)
1608 : {
1609 0 : DropDatabaseBuffers(fparms->dest_dboid);
1610 0 : ForgetDatabaseSyncRequests(fparms->dest_dboid);
1611 :
1612 : /* Release lock on the target database. */
1613 0 : UnlockSharedObject(DatabaseRelationId, fparms->dest_dboid, 0,
1614 : AccessShareLock);
1615 : }
1616 :
1617 : /*
1618 : * Release lock on source database before doing recursive remove. This is
1619 : * not essential but it seems desirable to release the lock as soon as
1620 : * possible.
1621 : */
1622 0 : UnlockSharedObject(DatabaseRelationId, fparms->src_dboid, 0, ShareLock);
1623 :
1624 : /* Throw away any successfully copied subdirectories */
1625 0 : remove_dbtablespaces(fparms->dest_dboid);
1626 0 : }
1627 :
1628 :
1629 : /*
1630 : * DROP DATABASE
1631 : */
1632 : void
1633 94 : dropdb(const char *dbname, bool missing_ok, bool force)
1634 : {
1635 : Oid db_id;
1636 : bool db_istemplate;
1637 : Relation pgdbrel;
1638 : HeapTuple tup;
1639 : Form_pg_database datform;
1640 : int notherbackends;
1641 : int npreparedxacts;
1642 : int nslots,
1643 : nslots_active;
1644 : int nsubscriptions;
1645 :
1646 : /*
1647 : * Look up the target database's OID, and get exclusive lock on it. We
1648 : * need this to ensure that no new backend starts up in the target
1649 : * database while we are deleting it (see postinit.c), and that no one is
1650 : * using it as a CREATE DATABASE template or trying to delete it for
1651 : * themselves.
1652 : */
1653 94 : pgdbrel = table_open(DatabaseRelationId, RowExclusiveLock);
1654 :
1655 94 : if (!get_db_info(dbname, AccessExclusiveLock, &db_id, NULL, NULL,
1656 : &db_istemplate, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL))
1657 : {
1658 32 : if (!missing_ok)
1659 : {
1660 16 : ereport(ERROR,
1661 : (errcode(ERRCODE_UNDEFINED_DATABASE),
1662 : errmsg("database \"%s\" does not exist", dbname)));
1663 : }
1664 : else
1665 : {
1666 : /* Close pg_database, release the lock, since we changed nothing */
1667 16 : table_close(pgdbrel, RowExclusiveLock);
1668 16 : ereport(NOTICE,
1669 : (errmsg("database \"%s\" does not exist, skipping",
1670 : dbname)));
1671 16 : return;
1672 : }
1673 : }
1674 :
1675 : /*
1676 : * Permission checks
1677 : */
1678 62 : if (!object_ownercheck(DatabaseRelationId, db_id, GetUserId()))
1679 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
1680 : dbname);
1681 :
1682 : /* DROP hook for the database being removed */
1683 62 : InvokeObjectDropHook(DatabaseRelationId, db_id, 0);
1684 :
1685 : /*
1686 : * Disallow dropping a DB that is marked istemplate. This is just to
1687 : * prevent people from accidentally dropping template0 or template1; they
1688 : * can do so if they're really determined ...
1689 : */
1690 62 : if (db_istemplate)
1691 0 : ereport(ERROR,
1692 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1693 : errmsg("cannot drop a template database")));
1694 :
1695 : /* Obviously can't drop my own database */
1696 62 : if (db_id == MyDatabaseId)
1697 0 : ereport(ERROR,
1698 : (errcode(ERRCODE_OBJECT_IN_USE),
1699 : errmsg("cannot drop the currently open database")));
1700 :
1701 : /*
1702 : * Check whether there are active logical slots that refer to the
1703 : * to-be-dropped database. The database lock we are holding prevents the
1704 : * creation of new slots using the database or existing slots becoming
1705 : * active.
1706 : */
1707 62 : (void) ReplicationSlotsCountDBSlots(db_id, &nslots, &nslots_active);
1708 62 : if (nslots_active)
1709 : {
1710 2 : ereport(ERROR,
1711 : (errcode(ERRCODE_OBJECT_IN_USE),
1712 : errmsg("database \"%s\" is used by an active logical replication slot",
1713 : dbname),
1714 : errdetail_plural("There is %d active slot.",
1715 : "There are %d active slots.",
1716 : nslots_active, nslots_active)));
1717 : }
1718 :
1719 : /*
1720 : * Check if there are subscriptions defined in the target database.
1721 : *
1722 : * We can't drop them automatically because they might be holding
1723 : * resources in other databases/instances.
1724 : */
1725 60 : if ((nsubscriptions = CountDBSubscriptions(db_id)) > 0)
1726 0 : ereport(ERROR,
1727 : (errcode(ERRCODE_OBJECT_IN_USE),
1728 : errmsg("database \"%s\" is being used by logical replication subscription",
1729 : dbname),
1730 : errdetail_plural("There is %d subscription.",
1731 : "There are %d subscriptions.",
1732 : nsubscriptions, nsubscriptions)));
1733 :
1734 :
1735 : /*
1736 : * Attempt to terminate all existing connections to the target database if
1737 : * the user has requested to do so.
1738 : */
1739 60 : if (force)
1740 2 : TerminateOtherDBBackends(db_id);
1741 :
1742 : /*
1743 : * Check for other backends in the target database. (Because we hold the
1744 : * database lock, no new ones can start after this.)
1745 : *
1746 : * As in CREATE DATABASE, check this after other error conditions.
1747 : */
1748 60 : if (CountOtherDBBackends(db_id, ¬herbackends, &npreparedxacts))
1749 0 : ereport(ERROR,
1750 : (errcode(ERRCODE_OBJECT_IN_USE),
1751 : errmsg("database \"%s\" is being accessed by other users",
1752 : dbname),
1753 : errdetail_busy_db(notherbackends, npreparedxacts)));
1754 :
1755 : /*
1756 : * Delete any comments or security labels associated with the database.
1757 : */
1758 60 : DeleteSharedComments(db_id, DatabaseRelationId);
1759 60 : DeleteSharedSecurityLabel(db_id, DatabaseRelationId);
1760 :
1761 : /*
1762 : * Remove settings associated with this database
1763 : */
1764 60 : DropSetting(db_id, InvalidOid);
1765 :
1766 : /*
1767 : * Remove shared dependency references for the database.
1768 : */
1769 60 : dropDatabaseDependencies(db_id);
1770 :
1771 : /*
1772 : * Tell the cumulative stats system to forget it immediately, too.
1773 : */
1774 60 : pgstat_drop_database(db_id);
1775 :
1776 60 : tup = SearchSysCacheCopy1(DATABASEOID, ObjectIdGetDatum(db_id));
1777 60 : if (!HeapTupleIsValid(tup))
1778 0 : elog(ERROR, "cache lookup failed for database %u", db_id);
1779 60 : datform = (Form_pg_database) GETSTRUCT(tup);
1780 :
1781 : /*
1782 : * Except for the deletion of the catalog row, subsequent actions are not
1783 : * transactional (consider DropDatabaseBuffers() discarding modified
1784 : * buffers). But we might crash or get interrupted below. To prevent
1785 : * accesses to a database with invalid contents, mark the database as
1786 : * invalid using an in-place update.
1787 : *
1788 : * We need to flush the WAL before continuing, to guarantee the
1789 : * modification is durable before performing irreversible filesystem
1790 : * operations.
1791 : */
1792 60 : datform->datconnlimit = DATCONNLIMIT_INVALID_DB;
1793 60 : heap_inplace_update(pgdbrel, tup);
1794 60 : XLogFlush(XactLastRecEnd);
1795 :
1796 : /*
1797 : * Also delete the tuple - transactionally. If this transaction commits,
1798 : * the row will be gone, but if we fail, dropdb() can be invoked again.
1799 : */
1800 60 : CatalogTupleDelete(pgdbrel, &tup->t_self);
1801 :
1802 : /*
1803 : * Drop db-specific replication slots.
1804 : */
1805 60 : ReplicationSlotsDropDBSlots(db_id);
1806 :
1807 : /*
1808 : * Drop pages for this database that are in the shared buffer cache. This
1809 : * is important to ensure that no remaining backend tries to write out a
1810 : * dirty buffer to the dead database later...
1811 : */
1812 60 : DropDatabaseBuffers(db_id);
1813 :
1814 : /*
1815 : * Tell checkpointer to forget any pending fsync and unlink requests for
1816 : * files in the database; else the fsyncs will fail at next checkpoint, or
1817 : * worse, it will delete files that belong to a newly created database
1818 : * with the same OID.
1819 : */
1820 60 : ForgetDatabaseSyncRequests(db_id);
1821 :
1822 : /*
1823 : * Force a checkpoint to make sure the checkpointer has received the
1824 : * message sent by ForgetDatabaseSyncRequests.
1825 : */
1826 60 : RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT);
1827 :
1828 : /* Close all smgr fds in all backends. */
1829 60 : WaitForProcSignalBarrier(EmitProcSignalBarrier(PROCSIGNAL_BARRIER_SMGRRELEASE));
1830 :
1831 : /*
1832 : * Remove all tablespace subdirs belonging to the database.
1833 : */
1834 60 : remove_dbtablespaces(db_id);
1835 :
1836 : /*
1837 : * Close pg_database, but keep lock till commit.
1838 : */
1839 60 : table_close(pgdbrel, NoLock);
1840 :
1841 : /*
1842 : * Force synchronous commit, thus minimizing the window between removal of
1843 : * the database files and committal of the transaction. If we crash before
1844 : * committing, we'll have a DB that's gone on disk but still there
1845 : * according to pg_database, which is not good.
1846 : */
1847 60 : ForceSyncCommit();
1848 : }
1849 :
1850 :
1851 : /*
1852 : * Rename database
1853 : */
1854 : ObjectAddress
1855 0 : RenameDatabase(const char *oldname, const char *newname)
1856 : {
1857 : Oid db_id;
1858 : HeapTuple newtup;
1859 : Relation rel;
1860 : int notherbackends;
1861 : int npreparedxacts;
1862 : ObjectAddress address;
1863 :
1864 : /*
1865 : * Look up the target database's OID, and get exclusive lock on it. We
1866 : * need this for the same reasons as DROP DATABASE.
1867 : */
1868 0 : rel = table_open(DatabaseRelationId, RowExclusiveLock);
1869 :
1870 0 : if (!get_db_info(oldname, AccessExclusiveLock, &db_id, NULL, NULL, NULL,
1871 : NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL))
1872 0 : ereport(ERROR,
1873 : (errcode(ERRCODE_UNDEFINED_DATABASE),
1874 : errmsg("database \"%s\" does not exist", oldname)));
1875 :
1876 : /* must be owner */
1877 0 : if (!object_ownercheck(DatabaseRelationId, db_id, GetUserId()))
1878 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
1879 : oldname);
1880 :
1881 : /* must have createdb rights */
1882 0 : if (!have_createdb_privilege())
1883 0 : ereport(ERROR,
1884 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1885 : errmsg("permission denied to rename database")));
1886 :
1887 : /*
1888 : * If built with appropriate switch, whine when regression-testing
1889 : * conventions for database names are violated.
1890 : */
1891 : #ifdef ENFORCE_REGRESSION_TEST_NAME_RESTRICTIONS
1892 : if (strstr(newname, "regression") == NULL)
1893 : elog(WARNING, "databases created by regression test cases should have names including \"regression\"");
1894 : #endif
1895 :
1896 : /*
1897 : * Make sure the new name doesn't exist. See notes for same error in
1898 : * CREATE DATABASE.
1899 : */
1900 0 : if (OidIsValid(get_database_oid(newname, true)))
1901 0 : ereport(ERROR,
1902 : (errcode(ERRCODE_DUPLICATE_DATABASE),
1903 : errmsg("database \"%s\" already exists", newname)));
1904 :
1905 : /*
1906 : * XXX Client applications probably store the current database somewhere,
1907 : * so renaming it could cause confusion. On the other hand, there may not
1908 : * be an actual problem besides a little confusion, so think about this
1909 : * and decide.
1910 : */
1911 0 : if (db_id == MyDatabaseId)
1912 0 : ereport(ERROR,
1913 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1914 : errmsg("current database cannot be renamed")));
1915 :
1916 : /*
1917 : * Make sure the database does not have active sessions. This is the same
1918 : * concern as above, but applied to other sessions.
1919 : *
1920 : * As in CREATE DATABASE, check this after other error conditions.
1921 : */
1922 0 : if (CountOtherDBBackends(db_id, ¬herbackends, &npreparedxacts))
1923 0 : ereport(ERROR,
1924 : (errcode(ERRCODE_OBJECT_IN_USE),
1925 : errmsg("database \"%s\" is being accessed by other users",
1926 : oldname),
1927 : errdetail_busy_db(notherbackends, npreparedxacts)));
1928 :
1929 : /* rename */
1930 0 : newtup = SearchSysCacheCopy1(DATABASEOID, ObjectIdGetDatum(db_id));
1931 0 : if (!HeapTupleIsValid(newtup))
1932 0 : elog(ERROR, "cache lookup failed for database %u", db_id);
1933 0 : namestrcpy(&(((Form_pg_database) GETSTRUCT(newtup))->datname), newname);
1934 0 : CatalogTupleUpdate(rel, &newtup->t_self, newtup);
1935 :
1936 0 : InvokeObjectPostAlterHook(DatabaseRelationId, db_id, 0);
1937 :
1938 0 : ObjectAddressSet(address, DatabaseRelationId, db_id);
1939 :
1940 : /*
1941 : * Close pg_database, but keep lock till commit.
1942 : */
1943 0 : table_close(rel, NoLock);
1944 :
1945 0 : return address;
1946 : }
1947 :
1948 :
1949 : /*
1950 : * ALTER DATABASE SET TABLESPACE
1951 : */
1952 : static void
1953 10 : movedb(const char *dbname, const char *tblspcname)
1954 : {
1955 : Oid db_id;
1956 : Relation pgdbrel;
1957 : int notherbackends;
1958 : int npreparedxacts;
1959 : HeapTuple oldtuple,
1960 : newtuple;
1961 : Oid src_tblspcoid,
1962 : dst_tblspcoid;
1963 : ScanKeyData scankey;
1964 : SysScanDesc sysscan;
1965 : AclResult aclresult;
1966 : char *src_dbpath;
1967 : char *dst_dbpath;
1968 : DIR *dstdir;
1969 : struct dirent *xlde;
1970 : movedb_failure_params fparms;
1971 :
1972 : /*
1973 : * Look up the target database's OID, and get exclusive lock on it. We
1974 : * need this to ensure that no new backend starts up in the database while
1975 : * we are moving it, and that no one is using it as a CREATE DATABASE
1976 : * template or trying to delete it.
1977 : */
1978 10 : pgdbrel = table_open(DatabaseRelationId, RowExclusiveLock);
1979 :
1980 10 : if (!get_db_info(dbname, AccessExclusiveLock, &db_id, NULL, NULL, NULL,
1981 : NULL, NULL, NULL, NULL, &src_tblspcoid, NULL, NULL, NULL, NULL, NULL, NULL))
1982 0 : ereport(ERROR,
1983 : (errcode(ERRCODE_UNDEFINED_DATABASE),
1984 : errmsg("database \"%s\" does not exist", dbname)));
1985 :
1986 : /*
1987 : * We actually need a session lock, so that the lock will persist across
1988 : * the commit/restart below. (We could almost get away with letting the
1989 : * lock be released at commit, except that someone could try to move
1990 : * relations of the DB back into the old directory while we rmtree() it.)
1991 : */
1992 10 : LockSharedObjectForSession(DatabaseRelationId, db_id, 0,
1993 : AccessExclusiveLock);
1994 :
1995 : /*
1996 : * Permission checks
1997 : */
1998 10 : if (!object_ownercheck(DatabaseRelationId, db_id, GetUserId()))
1999 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
2000 : dbname);
2001 :
2002 : /*
2003 : * Obviously can't move the tables of my own database
2004 : */
2005 10 : if (db_id == MyDatabaseId)
2006 0 : ereport(ERROR,
2007 : (errcode(ERRCODE_OBJECT_IN_USE),
2008 : errmsg("cannot change the tablespace of the currently open database")));
2009 :
2010 : /*
2011 : * Get tablespace's oid
2012 : */
2013 10 : dst_tblspcoid = get_tablespace_oid(tblspcname, false);
2014 :
2015 : /*
2016 : * Permission checks
2017 : */
2018 10 : aclresult = object_aclcheck(TableSpaceRelationId, dst_tblspcoid, GetUserId(),
2019 : ACL_CREATE);
2020 10 : if (aclresult != ACLCHECK_OK)
2021 0 : aclcheck_error(aclresult, OBJECT_TABLESPACE,
2022 : tblspcname);
2023 :
2024 : /*
2025 : * pg_global must never be the default tablespace
2026 : */
2027 10 : if (dst_tblspcoid == GLOBALTABLESPACE_OID)
2028 0 : ereport(ERROR,
2029 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2030 : errmsg("pg_global cannot be used as default tablespace")));
2031 :
2032 : /*
2033 : * No-op if same tablespace
2034 : */
2035 10 : if (src_tblspcoid == dst_tblspcoid)
2036 : {
2037 0 : table_close(pgdbrel, NoLock);
2038 0 : UnlockSharedObjectForSession(DatabaseRelationId, db_id, 0,
2039 : AccessExclusiveLock);
2040 0 : return;
2041 : }
2042 :
2043 : /*
2044 : * Check for other backends in the target database. (Because we hold the
2045 : * database lock, no new ones can start after this.)
2046 : *
2047 : * As in CREATE DATABASE, check this after other error conditions.
2048 : */
2049 10 : if (CountOtherDBBackends(db_id, ¬herbackends, &npreparedxacts))
2050 0 : ereport(ERROR,
2051 : (errcode(ERRCODE_OBJECT_IN_USE),
2052 : errmsg("database \"%s\" is being accessed by other users",
2053 : dbname),
2054 : errdetail_busy_db(notherbackends, npreparedxacts)));
2055 :
2056 : /*
2057 : * Get old and new database paths
2058 : */
2059 10 : src_dbpath = GetDatabasePath(db_id, src_tblspcoid);
2060 10 : dst_dbpath = GetDatabasePath(db_id, dst_tblspcoid);
2061 :
2062 : /*
2063 : * Force a checkpoint before proceeding. This will force all dirty
2064 : * buffers, including those of unlogged tables, out to disk, to ensure
2065 : * source database is up-to-date on disk for the copy.
2066 : * FlushDatabaseBuffers() would suffice for that, but we also want to
2067 : * process any pending unlink requests. Otherwise, the check for existing
2068 : * files in the target directory might fail unnecessarily, not to mention
2069 : * that the copy might fail due to source files getting deleted under it.
2070 : * On Windows, this also ensures that background procs don't hold any open
2071 : * files, which would cause rmdir() to fail.
2072 : */
2073 10 : RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT
2074 : | CHECKPOINT_FLUSH_ALL);
2075 :
2076 : /* Close all smgr fds in all backends. */
2077 10 : WaitForProcSignalBarrier(EmitProcSignalBarrier(PROCSIGNAL_BARRIER_SMGRRELEASE));
2078 :
2079 : /*
2080 : * Now drop all buffers holding data of the target database; they should
2081 : * no longer be dirty so DropDatabaseBuffers is safe.
2082 : *
2083 : * It might seem that we could just let these buffers age out of shared
2084 : * buffers naturally, since they should not get referenced anymore. The
2085 : * problem with that is that if the user later moves the database back to
2086 : * its original tablespace, any still-surviving buffers would appear to
2087 : * contain valid data again --- but they'd be missing any changes made in
2088 : * the database while it was in the new tablespace. In any case, freeing
2089 : * buffers that should never be used again seems worth the cycles.
2090 : *
2091 : * Note: it'd be sufficient to get rid of buffers matching db_id and
2092 : * src_tblspcoid, but bufmgr.c presently provides no API for that.
2093 : */
2094 10 : DropDatabaseBuffers(db_id);
2095 :
2096 : /*
2097 : * Check for existence of files in the target directory, i.e., objects of
2098 : * this database that are already in the target tablespace. We can't
2099 : * allow the move in such a case, because we would need to change those
2100 : * relations' pg_class.reltablespace entries to zero, and we don't have
2101 : * access to the DB's pg_class to do so.
2102 : */
2103 10 : dstdir = AllocateDir(dst_dbpath);
2104 10 : if (dstdir != NULL)
2105 : {
2106 0 : while ((xlde = ReadDir(dstdir, dst_dbpath)) != NULL)
2107 : {
2108 0 : if (strcmp(xlde->d_name, ".") == 0 ||
2109 0 : strcmp(xlde->d_name, "..") == 0)
2110 0 : continue;
2111 :
2112 0 : ereport(ERROR,
2113 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2114 : errmsg("some relations of database \"%s\" are already in tablespace \"%s\"",
2115 : dbname, tblspcname),
2116 : errhint("You must move them back to the database's default tablespace before using this command.")));
2117 : }
2118 :
2119 0 : FreeDir(dstdir);
2120 :
2121 : /*
2122 : * The directory exists but is empty. We must remove it before using
2123 : * the copydir function.
2124 : */
2125 0 : if (rmdir(dst_dbpath) != 0)
2126 0 : elog(ERROR, "could not remove directory \"%s\": %m",
2127 : dst_dbpath);
2128 : }
2129 :
2130 : /*
2131 : * Use an ENSURE block to make sure we remove the debris if the copy fails
2132 : * (eg, due to out-of-disk-space). This is not a 100% solution, because
2133 : * of the possibility of failure during transaction commit, but it should
2134 : * handle most scenarios.
2135 : */
2136 10 : fparms.dest_dboid = db_id;
2137 10 : fparms.dest_tsoid = dst_tblspcoid;
2138 10 : PG_ENSURE_ERROR_CLEANUP(movedb_failure_callback,
2139 : PointerGetDatum(&fparms));
2140 : {
2141 10 : Datum new_record[Natts_pg_database] = {0};
2142 10 : bool new_record_nulls[Natts_pg_database] = {0};
2143 10 : bool new_record_repl[Natts_pg_database] = {0};
2144 :
2145 : /*
2146 : * Copy files from the old tablespace to the new one
2147 : */
2148 10 : copydir(src_dbpath, dst_dbpath, false);
2149 :
2150 : /*
2151 : * Record the filesystem change in XLOG
2152 : */
2153 : {
2154 : xl_dbase_create_file_copy_rec xlrec;
2155 :
2156 10 : xlrec.db_id = db_id;
2157 10 : xlrec.tablespace_id = dst_tblspcoid;
2158 10 : xlrec.src_db_id = db_id;
2159 10 : xlrec.src_tablespace_id = src_tblspcoid;
2160 :
2161 10 : XLogBeginInsert();
2162 10 : XLogRegisterData((char *) &xlrec,
2163 : sizeof(xl_dbase_create_file_copy_rec));
2164 :
2165 10 : (void) XLogInsert(RM_DBASE_ID,
2166 : XLOG_DBASE_CREATE_FILE_COPY | XLR_SPECIAL_REL_UPDATE);
2167 : }
2168 :
2169 : /*
2170 : * Update the database's pg_database tuple
2171 : */
2172 10 : ScanKeyInit(&scankey,
2173 : Anum_pg_database_datname,
2174 : BTEqualStrategyNumber, F_NAMEEQ,
2175 : CStringGetDatum(dbname));
2176 10 : sysscan = systable_beginscan(pgdbrel, DatabaseNameIndexId, true,
2177 : NULL, 1, &scankey);
2178 10 : oldtuple = systable_getnext(sysscan);
2179 10 : if (!HeapTupleIsValid(oldtuple)) /* shouldn't happen... */
2180 0 : ereport(ERROR,
2181 : (errcode(ERRCODE_UNDEFINED_DATABASE),
2182 : errmsg("database \"%s\" does not exist", dbname)));
2183 :
2184 10 : new_record[Anum_pg_database_dattablespace - 1] = ObjectIdGetDatum(dst_tblspcoid);
2185 10 : new_record_repl[Anum_pg_database_dattablespace - 1] = true;
2186 :
2187 10 : newtuple = heap_modify_tuple(oldtuple, RelationGetDescr(pgdbrel),
2188 : new_record,
2189 : new_record_nulls, new_record_repl);
2190 10 : CatalogTupleUpdate(pgdbrel, &oldtuple->t_self, newtuple);
2191 :
2192 10 : InvokeObjectPostAlterHook(DatabaseRelationId, db_id, 0);
2193 :
2194 10 : systable_endscan(sysscan);
2195 :
2196 : /*
2197 : * Force another checkpoint here. As in CREATE DATABASE, this is to
2198 : * ensure that we don't have to replay a committed
2199 : * XLOG_DBASE_CREATE_FILE_COPY operation, which would cause us to lose
2200 : * any unlogged operations done in the new DB tablespace before the
2201 : * next checkpoint.
2202 : */
2203 10 : RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT);
2204 :
2205 : /*
2206 : * Force synchronous commit, thus minimizing the window between
2207 : * copying the database files and committal of the transaction. If we
2208 : * crash before committing, we'll leave an orphaned set of files on
2209 : * disk, which is not fatal but not good either.
2210 : */
2211 10 : ForceSyncCommit();
2212 :
2213 : /*
2214 : * Close pg_database, but keep lock till commit.
2215 : */
2216 10 : table_close(pgdbrel, NoLock);
2217 : }
2218 10 : PG_END_ENSURE_ERROR_CLEANUP(movedb_failure_callback,
2219 : PointerGetDatum(&fparms));
2220 :
2221 : /*
2222 : * Commit the transaction so that the pg_database update is committed. If
2223 : * we crash while removing files, the database won't be corrupt, we'll
2224 : * just leave some orphaned files in the old directory.
2225 : *
2226 : * (This is OK because we know we aren't inside a transaction block.)
2227 : *
2228 : * XXX would it be safe/better to do this inside the ensure block? Not
2229 : * convinced it's a good idea; consider elog just after the transaction
2230 : * really commits.
2231 : */
2232 10 : PopActiveSnapshot();
2233 10 : CommitTransactionCommand();
2234 :
2235 : /* Start new transaction for the remaining work; don't need a snapshot */
2236 10 : StartTransactionCommand();
2237 :
2238 : /*
2239 : * Remove files from the old tablespace
2240 : */
2241 10 : if (!rmtree(src_dbpath, true))
2242 0 : ereport(WARNING,
2243 : (errmsg("some useless files may be left behind in old database directory \"%s\"",
2244 : src_dbpath)));
2245 :
2246 : /*
2247 : * Record the filesystem change in XLOG
2248 : */
2249 : {
2250 : xl_dbase_drop_rec xlrec;
2251 :
2252 10 : xlrec.db_id = db_id;
2253 10 : xlrec.ntablespaces = 1;
2254 :
2255 10 : XLogBeginInsert();
2256 10 : XLogRegisterData((char *) &xlrec, sizeof(xl_dbase_drop_rec));
2257 10 : XLogRegisterData((char *) &src_tblspcoid, sizeof(Oid));
2258 :
2259 10 : (void) XLogInsert(RM_DBASE_ID,
2260 : XLOG_DBASE_DROP | XLR_SPECIAL_REL_UPDATE);
2261 : }
2262 :
2263 : /* Now it's safe to release the database lock */
2264 10 : UnlockSharedObjectForSession(DatabaseRelationId, db_id, 0,
2265 : AccessExclusiveLock);
2266 :
2267 10 : pfree(src_dbpath);
2268 10 : pfree(dst_dbpath);
2269 : }
2270 :
2271 : /* Error cleanup callback for movedb */
2272 : static void
2273 0 : movedb_failure_callback(int code, Datum arg)
2274 : {
2275 0 : movedb_failure_params *fparms = (movedb_failure_params *) DatumGetPointer(arg);
2276 : char *dstpath;
2277 :
2278 : /* Get rid of anything we managed to copy to the target directory */
2279 0 : dstpath = GetDatabasePath(fparms->dest_dboid, fparms->dest_tsoid);
2280 :
2281 0 : (void) rmtree(dstpath, true);
2282 :
2283 0 : pfree(dstpath);
2284 0 : }
2285 :
2286 : /*
2287 : * Process options and call dropdb function.
2288 : */
2289 : void
2290 94 : DropDatabase(ParseState *pstate, DropdbStmt *stmt)
2291 : {
2292 94 : bool force = false;
2293 : ListCell *lc;
2294 :
2295 120 : foreach(lc, stmt->options)
2296 : {
2297 26 : DefElem *opt = (DefElem *) lfirst(lc);
2298 :
2299 26 : if (strcmp(opt->defname, "force") == 0)
2300 26 : force = true;
2301 : else
2302 0 : ereport(ERROR,
2303 : (errcode(ERRCODE_SYNTAX_ERROR),
2304 : errmsg("unrecognized DROP DATABASE option \"%s\"", opt->defname),
2305 : parser_errposition(pstate, opt->location)));
2306 : }
2307 :
2308 94 : dropdb(stmt->dbname, stmt->missing_ok, force);
2309 76 : }
2310 :
2311 : /*
2312 : * ALTER DATABASE name ...
2313 : */
2314 : Oid
2315 34 : AlterDatabase(ParseState *pstate, AlterDatabaseStmt *stmt, bool isTopLevel)
2316 : {
2317 : Relation rel;
2318 : Oid dboid;
2319 : HeapTuple tuple,
2320 : newtuple;
2321 : Form_pg_database datform;
2322 : ScanKeyData scankey;
2323 : SysScanDesc scan;
2324 : ListCell *option;
2325 34 : bool dbistemplate = false;
2326 34 : bool dballowconnections = true;
2327 34 : int dbconnlimit = DATCONNLIMIT_UNLIMITED;
2328 34 : DefElem *distemplate = NULL;
2329 34 : DefElem *dallowconnections = NULL;
2330 34 : DefElem *dconnlimit = NULL;
2331 34 : DefElem *dtablespace = NULL;
2332 34 : Datum new_record[Natts_pg_database] = {0};
2333 34 : bool new_record_nulls[Natts_pg_database] = {0};
2334 34 : bool new_record_repl[Natts_pg_database] = {0};
2335 :
2336 : /* Extract options from the statement node tree */
2337 68 : foreach(option, stmt->options)
2338 : {
2339 34 : DefElem *defel = (DefElem *) lfirst(option);
2340 :
2341 34 : if (strcmp(defel->defname, "is_template") == 0)
2342 : {
2343 10 : if (distemplate)
2344 0 : errorConflictingDefElem(defel, pstate);
2345 10 : distemplate = defel;
2346 : }
2347 24 : else if (strcmp(defel->defname, "allow_connections") == 0)
2348 : {
2349 12 : if (dallowconnections)
2350 0 : errorConflictingDefElem(defel, pstate);
2351 12 : dallowconnections = defel;
2352 : }
2353 12 : else if (strcmp(defel->defname, "connection_limit") == 0)
2354 : {
2355 2 : if (dconnlimit)
2356 0 : errorConflictingDefElem(defel, pstate);
2357 2 : dconnlimit = defel;
2358 : }
2359 10 : else if (strcmp(defel->defname, "tablespace") == 0)
2360 : {
2361 10 : if (dtablespace)
2362 0 : errorConflictingDefElem(defel, pstate);
2363 10 : dtablespace = defel;
2364 : }
2365 : else
2366 0 : ereport(ERROR,
2367 : (errcode(ERRCODE_SYNTAX_ERROR),
2368 : errmsg("option \"%s\" not recognized", defel->defname),
2369 : parser_errposition(pstate, defel->location)));
2370 : }
2371 :
2372 34 : if (dtablespace)
2373 : {
2374 : /*
2375 : * While the SET TABLESPACE syntax doesn't allow any other options,
2376 : * somebody could write "WITH TABLESPACE ...". Forbid any other
2377 : * options from being specified in that case.
2378 : */
2379 10 : if (list_length(stmt->options) != 1)
2380 0 : ereport(ERROR,
2381 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2382 : errmsg("option \"%s\" cannot be specified with other options",
2383 : dtablespace->defname),
2384 : parser_errposition(pstate, dtablespace->location)));
2385 : /* this case isn't allowed within a transaction block */
2386 10 : PreventInTransactionBlock(isTopLevel, "ALTER DATABASE SET TABLESPACE");
2387 10 : movedb(stmt->dbname, defGetString(dtablespace));
2388 10 : return InvalidOid;
2389 : }
2390 :
2391 24 : if (distemplate && distemplate->arg)
2392 10 : dbistemplate = defGetBoolean(distemplate);
2393 24 : if (dallowconnections && dallowconnections->arg)
2394 12 : dballowconnections = defGetBoolean(dallowconnections);
2395 24 : if (dconnlimit && dconnlimit->arg)
2396 : {
2397 2 : dbconnlimit = defGetInt32(dconnlimit);
2398 2 : if (dbconnlimit < DATCONNLIMIT_UNLIMITED)
2399 0 : ereport(ERROR,
2400 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2401 : errmsg("invalid connection limit: %d", dbconnlimit)));
2402 : }
2403 :
2404 : /*
2405 : * Get the old tuple. We don't need a lock on the database per se,
2406 : * because we're not going to do anything that would mess up incoming
2407 : * connections.
2408 : */
2409 24 : rel = table_open(DatabaseRelationId, RowExclusiveLock);
2410 24 : ScanKeyInit(&scankey,
2411 : Anum_pg_database_datname,
2412 : BTEqualStrategyNumber, F_NAMEEQ,
2413 24 : CStringGetDatum(stmt->dbname));
2414 24 : scan = systable_beginscan(rel, DatabaseNameIndexId, true,
2415 : NULL, 1, &scankey);
2416 24 : tuple = systable_getnext(scan);
2417 24 : if (!HeapTupleIsValid(tuple))
2418 0 : ereport(ERROR,
2419 : (errcode(ERRCODE_UNDEFINED_DATABASE),
2420 : errmsg("database \"%s\" does not exist", stmt->dbname)));
2421 :
2422 24 : datform = (Form_pg_database) GETSTRUCT(tuple);
2423 24 : dboid = datform->oid;
2424 :
2425 24 : if (database_is_invalid_form(datform))
2426 : {
2427 2 : ereport(FATAL,
2428 : errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2429 : errmsg("cannot alter invalid database \"%s\"", stmt->dbname),
2430 : errhint("Use DROP DATABASE to drop invalid databases."));
2431 : }
2432 :
2433 22 : if (!object_ownercheck(DatabaseRelationId, dboid, GetUserId()))
2434 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
2435 0 : stmt->dbname);
2436 :
2437 : /*
2438 : * In order to avoid getting locked out and having to go through
2439 : * standalone mode, we refuse to disallow connections to the database
2440 : * we're currently connected to. Lockout can still happen with concurrent
2441 : * sessions but the likeliness of that is not high enough to worry about.
2442 : */
2443 22 : if (!dballowconnections && dboid == MyDatabaseId)
2444 0 : ereport(ERROR,
2445 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2446 : errmsg("cannot disallow connections for current database")));
2447 :
2448 : /*
2449 : * Build an updated tuple, perusing the information just obtained
2450 : */
2451 22 : if (distemplate)
2452 : {
2453 10 : new_record[Anum_pg_database_datistemplate - 1] = BoolGetDatum(dbistemplate);
2454 10 : new_record_repl[Anum_pg_database_datistemplate - 1] = true;
2455 : }
2456 22 : if (dallowconnections)
2457 : {
2458 12 : new_record[Anum_pg_database_datallowconn - 1] = BoolGetDatum(dballowconnections);
2459 12 : new_record_repl[Anum_pg_database_datallowconn - 1] = true;
2460 : }
2461 22 : if (dconnlimit)
2462 : {
2463 0 : new_record[Anum_pg_database_datconnlimit - 1] = Int32GetDatum(dbconnlimit);
2464 0 : new_record_repl[Anum_pg_database_datconnlimit - 1] = true;
2465 : }
2466 :
2467 22 : newtuple = heap_modify_tuple(tuple, RelationGetDescr(rel), new_record,
2468 : new_record_nulls, new_record_repl);
2469 22 : CatalogTupleUpdate(rel, &tuple->t_self, newtuple);
2470 :
2471 22 : InvokeObjectPostAlterHook(DatabaseRelationId, dboid, 0);
2472 :
2473 22 : systable_endscan(scan);
2474 :
2475 : /* Close pg_database, but keep lock till commit */
2476 22 : table_close(rel, NoLock);
2477 :
2478 22 : return dboid;
2479 : }
2480 :
2481 :
2482 : /*
2483 : * ALTER DATABASE name REFRESH COLLATION VERSION
2484 : */
2485 : ObjectAddress
2486 6 : AlterDatabaseRefreshColl(AlterDatabaseRefreshCollStmt *stmt)
2487 : {
2488 : Relation rel;
2489 : ScanKeyData scankey;
2490 : SysScanDesc scan;
2491 : Oid db_id;
2492 : HeapTuple tuple;
2493 : Form_pg_database datForm;
2494 : ObjectAddress address;
2495 : Datum datum;
2496 : bool isnull;
2497 : char *oldversion;
2498 : char *newversion;
2499 :
2500 6 : rel = table_open(DatabaseRelationId, RowExclusiveLock);
2501 6 : ScanKeyInit(&scankey,
2502 : Anum_pg_database_datname,
2503 : BTEqualStrategyNumber, F_NAMEEQ,
2504 6 : CStringGetDatum(stmt->dbname));
2505 6 : scan = systable_beginscan(rel, DatabaseNameIndexId, true,
2506 : NULL, 1, &scankey);
2507 6 : tuple = systable_getnext(scan);
2508 6 : if (!HeapTupleIsValid(tuple))
2509 0 : ereport(ERROR,
2510 : (errcode(ERRCODE_UNDEFINED_DATABASE),
2511 : errmsg("database \"%s\" does not exist", stmt->dbname)));
2512 :
2513 6 : datForm = (Form_pg_database) GETSTRUCT(tuple);
2514 6 : db_id = datForm->oid;
2515 :
2516 6 : if (!object_ownercheck(DatabaseRelationId, db_id, GetUserId()))
2517 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
2518 0 : stmt->dbname);
2519 :
2520 6 : datum = heap_getattr(tuple, Anum_pg_database_datcollversion, RelationGetDescr(rel), &isnull);
2521 6 : oldversion = isnull ? NULL : TextDatumGetCString(datum);
2522 :
2523 6 : if (datForm->datlocprovider == COLLPROVIDER_LIBC)
2524 : {
2525 4 : datum = heap_getattr(tuple, Anum_pg_database_datcollate, RelationGetDescr(rel), &isnull);
2526 4 : if (isnull)
2527 0 : elog(ERROR, "unexpected null in pg_database");
2528 : }
2529 : else
2530 : {
2531 2 : datum = heap_getattr(tuple, Anum_pg_database_datlocale, RelationGetDescr(rel), &isnull);
2532 2 : if (isnull)
2533 0 : elog(ERROR, "unexpected null in pg_database");
2534 : }
2535 :
2536 6 : newversion = get_collation_actual_version(datForm->datlocprovider,
2537 6 : TextDatumGetCString(datum));
2538 :
2539 : /* cannot change from NULL to non-NULL or vice versa */
2540 6 : if ((!oldversion && newversion) || (oldversion && !newversion))
2541 0 : elog(ERROR, "invalid collation version change");
2542 6 : else if (oldversion && newversion && strcmp(newversion, oldversion) != 0)
2543 0 : {
2544 0 : bool nulls[Natts_pg_database] = {0};
2545 0 : bool replaces[Natts_pg_database] = {0};
2546 0 : Datum values[Natts_pg_database] = {0};
2547 :
2548 0 : ereport(NOTICE,
2549 : (errmsg("changing version from %s to %s",
2550 : oldversion, newversion)));
2551 :
2552 0 : values[Anum_pg_database_datcollversion - 1] = CStringGetTextDatum(newversion);
2553 0 : replaces[Anum_pg_database_datcollversion - 1] = true;
2554 :
2555 0 : tuple = heap_modify_tuple(tuple, RelationGetDescr(rel),
2556 : values, nulls, replaces);
2557 0 : CatalogTupleUpdate(rel, &tuple->t_self, tuple);
2558 0 : heap_freetuple(tuple);
2559 : }
2560 : else
2561 6 : ereport(NOTICE,
2562 : (errmsg("version has not changed")));
2563 :
2564 6 : InvokeObjectPostAlterHook(DatabaseRelationId, db_id, 0);
2565 :
2566 6 : ObjectAddressSet(address, DatabaseRelationId, db_id);
2567 :
2568 6 : systable_endscan(scan);
2569 :
2570 6 : table_close(rel, NoLock);
2571 :
2572 6 : return address;
2573 : }
2574 :
2575 :
2576 : /*
2577 : * ALTER DATABASE name SET ...
2578 : */
2579 : Oid
2580 1058 : AlterDatabaseSet(AlterDatabaseSetStmt *stmt)
2581 : {
2582 1058 : Oid datid = get_database_oid(stmt->dbname, false);
2583 :
2584 : /*
2585 : * Obtain a lock on the database and make sure it didn't go away in the
2586 : * meantime.
2587 : */
2588 1058 : shdepLockAndCheckObject(DatabaseRelationId, datid);
2589 :
2590 1058 : if (!object_ownercheck(DatabaseRelationId, datid, GetUserId()))
2591 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
2592 0 : stmt->dbname);
2593 :
2594 1058 : AlterSetting(datid, InvalidOid, stmt->setstmt);
2595 :
2596 1058 : UnlockSharedObject(DatabaseRelationId, datid, 0, AccessShareLock);
2597 :
2598 1058 : return datid;
2599 : }
2600 :
2601 :
2602 : /*
2603 : * ALTER DATABASE name OWNER TO newowner
2604 : */
2605 : ObjectAddress
2606 44 : AlterDatabaseOwner(const char *dbname, Oid newOwnerId)
2607 : {
2608 : Oid db_id;
2609 : HeapTuple tuple;
2610 : Relation rel;
2611 : ScanKeyData scankey;
2612 : SysScanDesc scan;
2613 : Form_pg_database datForm;
2614 : ObjectAddress address;
2615 :
2616 : /*
2617 : * Get the old tuple. We don't need a lock on the database per se,
2618 : * because we're not going to do anything that would mess up incoming
2619 : * connections.
2620 : */
2621 44 : rel = table_open(DatabaseRelationId, RowExclusiveLock);
2622 44 : ScanKeyInit(&scankey,
2623 : Anum_pg_database_datname,
2624 : BTEqualStrategyNumber, F_NAMEEQ,
2625 : CStringGetDatum(dbname));
2626 44 : scan = systable_beginscan(rel, DatabaseNameIndexId, true,
2627 : NULL, 1, &scankey);
2628 44 : tuple = systable_getnext(scan);
2629 44 : if (!HeapTupleIsValid(tuple))
2630 0 : ereport(ERROR,
2631 : (errcode(ERRCODE_UNDEFINED_DATABASE),
2632 : errmsg("database \"%s\" does not exist", dbname)));
2633 :
2634 44 : datForm = (Form_pg_database) GETSTRUCT(tuple);
2635 44 : db_id = datForm->oid;
2636 :
2637 : /*
2638 : * If the new owner is the same as the existing owner, consider the
2639 : * command to have succeeded. This is to be consistent with other
2640 : * objects.
2641 : */
2642 44 : if (datForm->datdba != newOwnerId)
2643 : {
2644 : Datum repl_val[Natts_pg_database];
2645 24 : bool repl_null[Natts_pg_database] = {0};
2646 24 : bool repl_repl[Natts_pg_database] = {0};
2647 : Acl *newAcl;
2648 : Datum aclDatum;
2649 : bool isNull;
2650 : HeapTuple newtuple;
2651 :
2652 : /* Otherwise, must be owner of the existing object */
2653 24 : if (!object_ownercheck(DatabaseRelationId, db_id, GetUserId()))
2654 0 : aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
2655 : dbname);
2656 :
2657 : /* Must be able to become new owner */
2658 24 : check_can_set_role(GetUserId(), newOwnerId);
2659 :
2660 : /*
2661 : * must have createdb rights
2662 : *
2663 : * NOTE: This is different from other alter-owner checks in that the
2664 : * current user is checked for createdb privileges instead of the
2665 : * destination owner. This is consistent with the CREATE case for
2666 : * databases. Because superusers will always have this right, we need
2667 : * no special case for them.
2668 : */
2669 24 : if (!have_createdb_privilege())
2670 0 : ereport(ERROR,
2671 : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2672 : errmsg("permission denied to change owner of database")));
2673 :
2674 24 : repl_repl[Anum_pg_database_datdba - 1] = true;
2675 24 : repl_val[Anum_pg_database_datdba - 1] = ObjectIdGetDatum(newOwnerId);
2676 :
2677 : /*
2678 : * Determine the modified ACL for the new owner. This is only
2679 : * necessary when the ACL is non-null.
2680 : */
2681 24 : aclDatum = heap_getattr(tuple,
2682 : Anum_pg_database_datacl,
2683 : RelationGetDescr(rel),
2684 : &isNull);
2685 24 : if (!isNull)
2686 : {
2687 0 : newAcl = aclnewowner(DatumGetAclP(aclDatum),
2688 : datForm->datdba, newOwnerId);
2689 0 : repl_repl[Anum_pg_database_datacl - 1] = true;
2690 0 : repl_val[Anum_pg_database_datacl - 1] = PointerGetDatum(newAcl);
2691 : }
2692 :
2693 24 : newtuple = heap_modify_tuple(tuple, RelationGetDescr(rel), repl_val, repl_null, repl_repl);
2694 24 : CatalogTupleUpdate(rel, &newtuple->t_self, newtuple);
2695 :
2696 24 : heap_freetuple(newtuple);
2697 :
2698 : /* Update owner dependency reference */
2699 24 : changeDependencyOnOwner(DatabaseRelationId, db_id, newOwnerId);
2700 : }
2701 :
2702 44 : InvokeObjectPostAlterHook(DatabaseRelationId, db_id, 0);
2703 :
2704 44 : ObjectAddressSet(address, DatabaseRelationId, db_id);
2705 :
2706 44 : systable_endscan(scan);
2707 :
2708 : /* Close pg_database, but keep lock till commit */
2709 44 : table_close(rel, NoLock);
2710 :
2711 44 : return address;
2712 : }
2713 :
2714 :
2715 : Datum
2716 76 : pg_database_collation_actual_version(PG_FUNCTION_ARGS)
2717 : {
2718 76 : Oid dbid = PG_GETARG_OID(0);
2719 : HeapTuple tp;
2720 : char datlocprovider;
2721 : Datum datum;
2722 : char *version;
2723 :
2724 76 : tp = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(dbid));
2725 76 : if (!HeapTupleIsValid(tp))
2726 0 : ereport(ERROR,
2727 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2728 : errmsg("database with OID %u does not exist", dbid)));
2729 :
2730 76 : datlocprovider = ((Form_pg_database) GETSTRUCT(tp))->datlocprovider;
2731 :
2732 76 : if (datlocprovider == COLLPROVIDER_LIBC)
2733 62 : datum = SysCacheGetAttrNotNull(DATABASEOID, tp, Anum_pg_database_datcollate);
2734 : else
2735 14 : datum = SysCacheGetAttrNotNull(DATABASEOID, tp, Anum_pg_database_datlocale);
2736 :
2737 76 : version = get_collation_actual_version(datlocprovider,
2738 76 : TextDatumGetCString(datum));
2739 :
2740 76 : ReleaseSysCache(tp);
2741 :
2742 76 : if (version)
2743 48 : PG_RETURN_TEXT_P(cstring_to_text(version));
2744 : else
2745 28 : PG_RETURN_NULL();
2746 : }
2747 :
2748 :
2749 : /*
2750 : * Helper functions
2751 : */
2752 :
2753 : /*
2754 : * Look up info about the database named "name". If the database exists,
2755 : * obtain the specified lock type on it, fill in any of the remaining
2756 : * parameters that aren't NULL, and return true. If no such database,
2757 : * return false.
2758 : */
2759 : static bool
2760 732 : get_db_info(const char *name, LOCKMODE lockmode,
2761 : Oid *dbIdP, Oid *ownerIdP,
2762 : int *encodingP, bool *dbIsTemplateP, bool *dbAllowConnP, bool *dbHasLoginEvtP,
2763 : TransactionId *dbFrozenXidP, MultiXactId *dbMinMultiP,
2764 : Oid *dbTablespace, char **dbCollate, char **dbCtype, char **dbLocale,
2765 : char **dbIcurules,
2766 : char *dbLocProvider,
2767 : char **dbCollversion)
2768 : {
2769 732 : bool result = false;
2770 : Relation relation;
2771 :
2772 : Assert(name);
2773 :
2774 : /* Caller may wish to grab a better lock on pg_database beforehand... */
2775 732 : relation = table_open(DatabaseRelationId, AccessShareLock);
2776 :
2777 : /*
2778 : * Loop covers the rare case where the database is renamed before we can
2779 : * lock it. We try again just in case we can find a new one of the same
2780 : * name.
2781 : */
2782 : for (;;)
2783 0 : {
2784 : ScanKeyData scanKey;
2785 : SysScanDesc scan;
2786 : HeapTuple tuple;
2787 : Oid dbOid;
2788 :
2789 : /*
2790 : * there's no syscache for database-indexed-by-name, so must do it the
2791 : * hard way
2792 : */
2793 732 : ScanKeyInit(&scanKey,
2794 : Anum_pg_database_datname,
2795 : BTEqualStrategyNumber, F_NAMEEQ,
2796 : CStringGetDatum(name));
2797 :
2798 732 : scan = systable_beginscan(relation, DatabaseNameIndexId, true,
2799 : NULL, 1, &scanKey);
2800 :
2801 732 : tuple = systable_getnext(scan);
2802 :
2803 732 : if (!HeapTupleIsValid(tuple))
2804 : {
2805 : /* definitely no database of that name */
2806 32 : systable_endscan(scan);
2807 32 : break;
2808 : }
2809 :
2810 700 : dbOid = ((Form_pg_database) GETSTRUCT(tuple))->oid;
2811 :
2812 700 : systable_endscan(scan);
2813 :
2814 : /*
2815 : * Now that we have a database OID, we can try to lock the DB.
2816 : */
2817 700 : if (lockmode != NoLock)
2818 700 : LockSharedObject(DatabaseRelationId, dbOid, 0, lockmode);
2819 :
2820 : /*
2821 : * And now, re-fetch the tuple by OID. If it's still there and still
2822 : * the same name, we win; else, drop the lock and loop back to try
2823 : * again.
2824 : */
2825 700 : tuple = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(dbOid));
2826 700 : if (HeapTupleIsValid(tuple))
2827 : {
2828 700 : Form_pg_database dbform = (Form_pg_database) GETSTRUCT(tuple);
2829 :
2830 700 : if (strcmp(name, NameStr(dbform->datname)) == 0)
2831 : {
2832 : Datum datum;
2833 : bool isnull;
2834 :
2835 : /* oid of the database */
2836 700 : if (dbIdP)
2837 700 : *dbIdP = dbOid;
2838 : /* oid of the owner */
2839 700 : if (ownerIdP)
2840 628 : *ownerIdP = dbform->datdba;
2841 : /* character encoding */
2842 700 : if (encodingP)
2843 628 : *encodingP = dbform->encoding;
2844 : /* allowed as template? */
2845 700 : if (dbIsTemplateP)
2846 690 : *dbIsTemplateP = dbform->datistemplate;
2847 : /* Has on login event trigger? */
2848 700 : if (dbHasLoginEvtP)
2849 628 : *dbHasLoginEvtP = dbform->dathasloginevt;
2850 : /* allowing connections? */
2851 700 : if (dbAllowConnP)
2852 628 : *dbAllowConnP = dbform->datallowconn;
2853 : /* limit of frozen XIDs */
2854 700 : if (dbFrozenXidP)
2855 628 : *dbFrozenXidP = dbform->datfrozenxid;
2856 : /* minimum MultiXactId */
2857 700 : if (dbMinMultiP)
2858 628 : *dbMinMultiP = dbform->datminmxid;
2859 : /* default tablespace for this database */
2860 700 : if (dbTablespace)
2861 638 : *dbTablespace = dbform->dattablespace;
2862 : /* default locale settings for this database */
2863 700 : if (dbLocProvider)
2864 628 : *dbLocProvider = dbform->datlocprovider;
2865 700 : if (dbCollate)
2866 : {
2867 628 : datum = SysCacheGetAttrNotNull(DATABASEOID, tuple, Anum_pg_database_datcollate);
2868 628 : *dbCollate = TextDatumGetCString(datum);
2869 : }
2870 700 : if (dbCtype)
2871 : {
2872 628 : datum = SysCacheGetAttrNotNull(DATABASEOID, tuple, Anum_pg_database_datctype);
2873 628 : *dbCtype = TextDatumGetCString(datum);
2874 : }
2875 700 : if (dbLocale)
2876 : {
2877 628 : datum = SysCacheGetAttr(DATABASEOID, tuple, Anum_pg_database_datlocale, &isnull);
2878 628 : if (isnull)
2879 572 : *dbLocale = NULL;
2880 : else
2881 56 : *dbLocale = TextDatumGetCString(datum);
2882 : }
2883 700 : if (dbIcurules)
2884 : {
2885 628 : datum = SysCacheGetAttr(DATABASEOID, tuple, Anum_pg_database_daticurules, &isnull);
2886 628 : if (isnull)
2887 628 : *dbIcurules = NULL;
2888 : else
2889 0 : *dbIcurules = TextDatumGetCString(datum);
2890 : }
2891 700 : if (dbCollversion)
2892 : {
2893 628 : datum = SysCacheGetAttr(DATABASEOID, tuple, Anum_pg_database_datcollversion, &isnull);
2894 628 : if (isnull)
2895 386 : *dbCollversion = NULL;
2896 : else
2897 242 : *dbCollversion = TextDatumGetCString(datum);
2898 : }
2899 700 : ReleaseSysCache(tuple);
2900 700 : result = true;
2901 700 : break;
2902 : }
2903 : /* can only get here if it was just renamed */
2904 0 : ReleaseSysCache(tuple);
2905 : }
2906 :
2907 0 : if (lockmode != NoLock)
2908 0 : UnlockSharedObject(DatabaseRelationId, dbOid, 0, lockmode);
2909 : }
2910 :
2911 732 : table_close(relation, AccessShareLock);
2912 :
2913 732 : return result;
2914 : }
2915 :
2916 : /* Check if current user has createdb privileges */
2917 : bool
2918 688 : have_createdb_privilege(void)
2919 : {
2920 688 : bool result = false;
2921 : HeapTuple utup;
2922 :
2923 : /* Superusers can always do everything */
2924 688 : if (superuser())
2925 652 : return true;
2926 :
2927 36 : utup = SearchSysCache1(AUTHOID, ObjectIdGetDatum(GetUserId()));
2928 36 : if (HeapTupleIsValid(utup))
2929 : {
2930 36 : result = ((Form_pg_authid) GETSTRUCT(utup))->rolcreatedb;
2931 36 : ReleaseSysCache(utup);
2932 : }
2933 36 : return result;
2934 : }
2935 :
2936 : /*
2937 : * Remove tablespace directories
2938 : *
2939 : * We don't know what tablespaces db_id is using, so iterate through all
2940 : * tablespaces removing <tablespace>/db_id
2941 : */
2942 : static void
2943 60 : remove_dbtablespaces(Oid db_id)
2944 : {
2945 : Relation rel;
2946 : TableScanDesc scan;
2947 : HeapTuple tuple;
2948 60 : List *ltblspc = NIL;
2949 : ListCell *cell;
2950 : int ntblspc;
2951 : int i;
2952 : Oid *tablespace_ids;
2953 :
2954 60 : rel = table_open(TableSpaceRelationId, AccessShareLock);
2955 60 : scan = table_beginscan_catalog(rel, 0, NULL);
2956 222 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
2957 : {
2958 162 : Form_pg_tablespace spcform = (Form_pg_tablespace) GETSTRUCT(tuple);
2959 162 : Oid dsttablespace = spcform->oid;
2960 : char *dstpath;
2961 : struct stat st;
2962 :
2963 : /* Don't mess with the global tablespace */
2964 162 : if (dsttablespace == GLOBALTABLESPACE_OID)
2965 102 : continue;
2966 :
2967 102 : dstpath = GetDatabasePath(db_id, dsttablespace);
2968 :
2969 102 : if (lstat(dstpath, &st) < 0 || !S_ISDIR(st.st_mode))
2970 : {
2971 : /* Assume we can ignore it */
2972 42 : pfree(dstpath);
2973 42 : continue;
2974 : }
2975 :
2976 60 : if (!rmtree(dstpath, true))
2977 0 : ereport(WARNING,
2978 : (errmsg("some useless files may be left behind in old database directory \"%s\"",
2979 : dstpath)));
2980 :
2981 60 : ltblspc = lappend_oid(ltblspc, dsttablespace);
2982 60 : pfree(dstpath);
2983 : }
2984 :
2985 60 : ntblspc = list_length(ltblspc);
2986 60 : if (ntblspc == 0)
2987 : {
2988 0 : table_endscan(scan);
2989 0 : table_close(rel, AccessShareLock);
2990 0 : return;
2991 : }
2992 :
2993 60 : tablespace_ids = (Oid *) palloc(ntblspc * sizeof(Oid));
2994 60 : i = 0;
2995 120 : foreach(cell, ltblspc)
2996 60 : tablespace_ids[i++] = lfirst_oid(cell);
2997 :
2998 : /* Record the filesystem change in XLOG */
2999 : {
3000 : xl_dbase_drop_rec xlrec;
3001 :
3002 60 : xlrec.db_id = db_id;
3003 60 : xlrec.ntablespaces = ntblspc;
3004 :
3005 60 : XLogBeginInsert();
3006 60 : XLogRegisterData((char *) &xlrec, MinSizeOfDbaseDropRec);
3007 60 : XLogRegisterData((char *) tablespace_ids, ntblspc * sizeof(Oid));
3008 :
3009 60 : (void) XLogInsert(RM_DBASE_ID,
3010 : XLOG_DBASE_DROP | XLR_SPECIAL_REL_UPDATE);
3011 : }
3012 :
3013 60 : list_free(ltblspc);
3014 60 : pfree(tablespace_ids);
3015 :
3016 60 : table_endscan(scan);
3017 60 : table_close(rel, AccessShareLock);
3018 : }
3019 :
3020 : /*
3021 : * Check for existing files that conflict with a proposed new DB OID;
3022 : * return true if there are any
3023 : *
3024 : * If there were a subdirectory in any tablespace matching the proposed new
3025 : * OID, we'd get a create failure due to the duplicate name ... and then we'd
3026 : * try to remove that already-existing subdirectory during the cleanup in
3027 : * remove_dbtablespaces. Nuking existing files seems like a bad idea, so
3028 : * instead we make this extra check before settling on the OID of the new
3029 : * database. This exactly parallels what GetNewRelFileNumber() does for table
3030 : * relfilenumber values.
3031 : */
3032 : static bool
3033 602 : check_db_file_conflict(Oid db_id)
3034 : {
3035 602 : bool result = false;
3036 : Relation rel;
3037 : TableScanDesc scan;
3038 : HeapTuple tuple;
3039 :
3040 602 : rel = table_open(TableSpaceRelationId, AccessShareLock);
3041 602 : scan = table_beginscan_catalog(rel, 0, NULL);
3042 1896 : while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
3043 : {
3044 1294 : Form_pg_tablespace spcform = (Form_pg_tablespace) GETSTRUCT(tuple);
3045 1294 : Oid dsttablespace = spcform->oid;
3046 : char *dstpath;
3047 : struct stat st;
3048 :
3049 : /* Don't mess with the global tablespace */
3050 1294 : if (dsttablespace == GLOBALTABLESPACE_OID)
3051 602 : continue;
3052 :
3053 692 : dstpath = GetDatabasePath(db_id, dsttablespace);
3054 :
3055 692 : if (lstat(dstpath, &st) == 0)
3056 : {
3057 : /* Found a conflicting file (or directory, whatever) */
3058 0 : pfree(dstpath);
3059 0 : result = true;
3060 0 : break;
3061 : }
3062 :
3063 692 : pfree(dstpath);
3064 : }
3065 :
3066 602 : table_endscan(scan);
3067 602 : table_close(rel, AccessShareLock);
3068 :
3069 602 : return result;
3070 : }
3071 :
3072 : /*
3073 : * Issue a suitable errdetail message for a busy database
3074 : */
3075 : static int
3076 0 : errdetail_busy_db(int notherbackends, int npreparedxacts)
3077 : {
3078 0 : if (notherbackends > 0 && npreparedxacts > 0)
3079 :
3080 : /*
3081 : * We don't deal with singular versus plural here, since gettext
3082 : * doesn't support multiple plurals in one string.
3083 : */
3084 0 : errdetail("There are %d other session(s) and %d prepared transaction(s) using the database.",
3085 : notherbackends, npreparedxacts);
3086 0 : else if (notherbackends > 0)
3087 0 : errdetail_plural("There is %d other session using the database.",
3088 : "There are %d other sessions using the database.",
3089 : notherbackends,
3090 : notherbackends);
3091 : else
3092 0 : errdetail_plural("There is %d prepared transaction using the database.",
3093 : "There are %d prepared transactions using the database.",
3094 : npreparedxacts,
3095 : npreparedxacts);
3096 0 : return 0; /* just to keep ereport macro happy */
3097 : }
3098 :
3099 : /*
3100 : * get_database_oid - given a database name, look up the OID
3101 : *
3102 : * If missing_ok is false, throw an error if database name not found. If
3103 : * true, just return InvalidOid.
3104 : */
3105 : Oid
3106 2352 : get_database_oid(const char *dbname, bool missing_ok)
3107 : {
3108 : Relation pg_database;
3109 : ScanKeyData entry[1];
3110 : SysScanDesc scan;
3111 : HeapTuple dbtuple;
3112 : Oid oid;
3113 :
3114 : /*
3115 : * There's no syscache for pg_database indexed by name, so we must look
3116 : * the hard way.
3117 : */
3118 2352 : pg_database = table_open(DatabaseRelationId, AccessShareLock);
3119 2352 : ScanKeyInit(&entry[0],
3120 : Anum_pg_database_datname,
3121 : BTEqualStrategyNumber, F_NAMEEQ,
3122 : CStringGetDatum(dbname));
3123 2352 : scan = systable_beginscan(pg_database, DatabaseNameIndexId, true,
3124 : NULL, 1, entry);
3125 :
3126 2352 : dbtuple = systable_getnext(scan);
3127 :
3128 : /* We assume that there can be at most one matching tuple */
3129 2352 : if (HeapTupleIsValid(dbtuple))
3130 1744 : oid = ((Form_pg_database) GETSTRUCT(dbtuple))->oid;
3131 : else
3132 608 : oid = InvalidOid;
3133 :
3134 2352 : systable_endscan(scan);
3135 2352 : table_close(pg_database, AccessShareLock);
3136 :
3137 2352 : if (!OidIsValid(oid) && !missing_ok)
3138 6 : ereport(ERROR,
3139 : (errcode(ERRCODE_UNDEFINED_DATABASE),
3140 : errmsg("database \"%s\" does not exist",
3141 : dbname)));
3142 :
3143 2346 : return oid;
3144 : }
3145 :
3146 :
3147 : /*
3148 : * get_database_name - given a database OID, look up the name
3149 : *
3150 : * Returns a palloc'd string, or NULL if no such database.
3151 : */
3152 : char *
3153 26792 : get_database_name(Oid dbid)
3154 : {
3155 : HeapTuple dbtuple;
3156 : char *result;
3157 :
3158 26792 : dbtuple = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(dbid));
3159 26792 : if (HeapTupleIsValid(dbtuple))
3160 : {
3161 26592 : result = pstrdup(NameStr(((Form_pg_database) GETSTRUCT(dbtuple))->datname));
3162 26592 : ReleaseSysCache(dbtuple);
3163 : }
3164 : else
3165 200 : result = NULL;
3166 :
3167 26792 : return result;
3168 : }
3169 :
3170 :
3171 : /*
3172 : * While dropping a database the pg_database row is marked invalid, but the
3173 : * catalog contents still exist. Connections to such a database are not
3174 : * allowed.
3175 : */
3176 : bool
3177 26960 : database_is_invalid_form(Form_pg_database datform)
3178 : {
3179 26960 : return datform->datconnlimit == DATCONNLIMIT_INVALID_DB;
3180 : }
3181 :
3182 :
3183 : /*
3184 : * Convenience wrapper around database_is_invalid_form()
3185 : */
3186 : bool
3187 628 : database_is_invalid_oid(Oid dboid)
3188 : {
3189 : HeapTuple dbtup;
3190 : Form_pg_database dbform;
3191 : bool invalid;
3192 :
3193 628 : dbtup = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(dboid));
3194 628 : if (!HeapTupleIsValid(dbtup))
3195 0 : elog(ERROR, "cache lookup failed for database %u", dboid);
3196 628 : dbform = (Form_pg_database) GETSTRUCT(dbtup);
3197 :
3198 628 : invalid = database_is_invalid_form(dbform);
3199 :
3200 628 : ReleaseSysCache(dbtup);
3201 :
3202 628 : return invalid;
3203 : }
3204 :
3205 :
3206 : /*
3207 : * recovery_create_dbdir()
3208 : *
3209 : * During recovery, there's a case where we validly need to recover a missing
3210 : * tablespace directory so that recovery can continue. This happens when
3211 : * recovery wants to create a database but the holding tablespace has been
3212 : * removed before the server stopped. Since we expect that the directory will
3213 : * be gone before reaching recovery consistency, and we have no knowledge about
3214 : * the tablespace other than its OID here, we create a real directory under
3215 : * pg_tblspc here instead of restoring the symlink.
3216 : *
3217 : * If only_tblspc is true, then the requested directory must be in pg_tblspc/
3218 : */
3219 : static void
3220 40 : recovery_create_dbdir(char *path, bool only_tblspc)
3221 : {
3222 : struct stat st;
3223 :
3224 : Assert(RecoveryInProgress());
3225 :
3226 40 : if (stat(path, &st) == 0)
3227 40 : return;
3228 :
3229 0 : if (only_tblspc && strstr(path, "pg_tblspc/") == NULL)
3230 0 : elog(PANIC, "requested to created invalid directory: %s", path);
3231 :
3232 0 : if (reachedConsistency && !allow_in_place_tablespaces)
3233 0 : ereport(PANIC,
3234 : errmsg("missing directory \"%s\"", path));
3235 :
3236 0 : elog(reachedConsistency ? WARNING : DEBUG1,
3237 : "creating missing directory: %s", path);
3238 :
3239 0 : if (pg_mkdir_p(path, pg_dir_create_mode) != 0)
3240 0 : ereport(PANIC,
3241 : errmsg("could not create missing directory \"%s\": %m", path));
3242 : }
3243 :
3244 :
3245 : /*
3246 : * DATABASE resource manager's routines
3247 : */
3248 : void
3249 68 : dbase_redo(XLogReaderState *record)
3250 : {
3251 68 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
3252 :
3253 : /* Backup blocks are not used in dbase records */
3254 : Assert(!XLogRecHasAnyBlockRefs(record));
3255 :
3256 68 : if (info == XLOG_DBASE_CREATE_FILE_COPY)
3257 : {
3258 6 : xl_dbase_create_file_copy_rec *xlrec =
3259 6 : (xl_dbase_create_file_copy_rec *) XLogRecGetData(record);
3260 : char *src_path;
3261 : char *dst_path;
3262 : char *parent_path;
3263 : struct stat st;
3264 :
3265 6 : src_path = GetDatabasePath(xlrec->src_db_id, xlrec->src_tablespace_id);
3266 6 : dst_path = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id);
3267 :
3268 : /*
3269 : * Our theory for replaying a CREATE is to forcibly drop the target
3270 : * subdirectory if present, then re-copy the source data. This may be
3271 : * more work than needed, but it is simple to implement.
3272 : */
3273 6 : if (stat(dst_path, &st) == 0 && S_ISDIR(st.st_mode))
3274 : {
3275 0 : if (!rmtree(dst_path, true))
3276 : /* If this failed, copydir() below is going to error. */
3277 0 : ereport(WARNING,
3278 : (errmsg("some useless files may be left behind in old database directory \"%s\"",
3279 : dst_path)));
3280 : }
3281 :
3282 : /*
3283 : * If the parent of the target path doesn't exist, create it now. This
3284 : * enables us to create the target underneath later.
3285 : */
3286 6 : parent_path = pstrdup(dst_path);
3287 6 : get_parent_directory(parent_path);
3288 6 : if (stat(parent_path, &st) < 0)
3289 : {
3290 0 : if (errno != ENOENT)
3291 0 : ereport(FATAL,
3292 : errmsg("could not stat directory \"%s\": %m",
3293 : dst_path));
3294 :
3295 : /* create the parent directory if needed and valid */
3296 0 : recovery_create_dbdir(parent_path, true);
3297 : }
3298 6 : pfree(parent_path);
3299 :
3300 : /*
3301 : * There's a case where the copy source directory is missing for the
3302 : * same reason above. Create the empty source directory so that
3303 : * copydir below doesn't fail. The directory will be dropped soon by
3304 : * recovery.
3305 : */
3306 6 : if (stat(src_path, &st) < 0 && errno == ENOENT)
3307 0 : recovery_create_dbdir(src_path, false);
3308 :
3309 : /*
3310 : * Force dirty buffers out to disk, to ensure source database is
3311 : * up-to-date for the copy.
3312 : */
3313 6 : FlushDatabaseBuffers(xlrec->src_db_id);
3314 :
3315 : /* Close all smgr fds in all backends. */
3316 6 : WaitForProcSignalBarrier(EmitProcSignalBarrier(PROCSIGNAL_BARRIER_SMGRRELEASE));
3317 :
3318 : /*
3319 : * Copy this subdirectory to the new location
3320 : *
3321 : * We don't need to copy subdirectories
3322 : */
3323 6 : copydir(src_path, dst_path, false);
3324 :
3325 6 : pfree(src_path);
3326 6 : pfree(dst_path);
3327 : }
3328 62 : else if (info == XLOG_DBASE_CREATE_WAL_LOG)
3329 : {
3330 40 : xl_dbase_create_wal_log_rec *xlrec =
3331 40 : (xl_dbase_create_wal_log_rec *) XLogRecGetData(record);
3332 : char *dbpath;
3333 : char *parent_path;
3334 :
3335 40 : dbpath = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id);
3336 :
3337 : /* create the parent directory if needed and valid */
3338 40 : parent_path = pstrdup(dbpath);
3339 40 : get_parent_directory(parent_path);
3340 40 : recovery_create_dbdir(parent_path, true);
3341 :
3342 : /* Create the database directory with the version file. */
3343 40 : CreateDirAndVersionFile(dbpath, xlrec->db_id, xlrec->tablespace_id,
3344 : true);
3345 40 : pfree(dbpath);
3346 : }
3347 22 : else if (info == XLOG_DBASE_DROP)
3348 : {
3349 22 : xl_dbase_drop_rec *xlrec = (xl_dbase_drop_rec *) XLogRecGetData(record);
3350 : char *dst_path;
3351 : int i;
3352 :
3353 22 : if (InHotStandby)
3354 : {
3355 : /*
3356 : * Lock database while we resolve conflicts to ensure that
3357 : * InitPostgres() cannot fully re-execute concurrently. This
3358 : * avoids backends re-connecting automatically to same database,
3359 : * which can happen in some cases.
3360 : *
3361 : * This will lock out walsenders trying to connect to db-specific
3362 : * slots for logical decoding too, so it's safe for us to drop
3363 : * slots.
3364 : */
3365 22 : LockSharedObjectForSession(DatabaseRelationId, xlrec->db_id, 0, AccessExclusiveLock);
3366 22 : ResolveRecoveryConflictWithDatabase(xlrec->db_id);
3367 : }
3368 :
3369 : /* Drop any database-specific replication slots */
3370 22 : ReplicationSlotsDropDBSlots(xlrec->db_id);
3371 :
3372 : /* Drop pages for this database that are in the shared buffer cache */
3373 22 : DropDatabaseBuffers(xlrec->db_id);
3374 :
3375 : /* Also, clean out any fsync requests that might be pending in md.c */
3376 22 : ForgetDatabaseSyncRequests(xlrec->db_id);
3377 :
3378 : /* Clean out the xlog relcache too */
3379 22 : XLogDropDatabase(xlrec->db_id);
3380 :
3381 : /* Close all smgr fds in all backends. */
3382 22 : WaitForProcSignalBarrier(EmitProcSignalBarrier(PROCSIGNAL_BARRIER_SMGRRELEASE));
3383 :
3384 44 : for (i = 0; i < xlrec->ntablespaces; i++)
3385 : {
3386 22 : dst_path = GetDatabasePath(xlrec->db_id, xlrec->tablespace_ids[i]);
3387 :
3388 : /* And remove the physical files */
3389 22 : if (!rmtree(dst_path, true))
3390 0 : ereport(WARNING,
3391 : (errmsg("some useless files may be left behind in old database directory \"%s\"",
3392 : dst_path)));
3393 22 : pfree(dst_path);
3394 : }
3395 :
3396 22 : if (InHotStandby)
3397 : {
3398 : /*
3399 : * Release locks prior to commit. XXX There is a race condition
3400 : * here that may allow backends to reconnect, but the window for
3401 : * this is small because the gap between here and commit is mostly
3402 : * fairly small and it is unlikely that people will be dropping
3403 : * databases that we are trying to connect to anyway.
3404 : */
3405 22 : UnlockSharedObjectForSession(DatabaseRelationId, xlrec->db_id, 0, AccessExclusiveLock);
3406 : }
3407 : }
3408 : else
3409 0 : elog(PANIC, "dbase_redo: unknown op code %u", info);
3410 68 : }
|