| 1 | /*------------------------------------------------------------------------- |
| 2 | * |
| 3 | * tablespace.c |
| 4 | * Commands to manipulate table spaces |
| 5 | * |
| 6 | * Tablespaces in PostgreSQL are designed to allow users to determine |
| 7 | * where the data file(s) for a given database object reside on the file |
| 8 | * system. |
| 9 | * |
| 10 | * A tablespace represents a directory on the file system. At tablespace |
| 11 | * creation time, the directory must be empty. To simplify things and |
| 12 | * remove the possibility of having file name conflicts, we isolate |
| 13 | * files within a tablespace into database-specific subdirectories. |
| 14 | * |
| 15 | * To support file access via the information given in RelFileNode, we |
| 16 | * maintain a symbolic-link map in $PGDATA/pg_tblspc. The symlinks are |
| 17 | * named by tablespace OIDs and point to the actual tablespace directories. |
| 18 | * There is also a per-cluster version directory in each tablespace. |
| 19 | * Thus the full path to an arbitrary file is |
| 20 | * $PGDATA/pg_tblspc/spcoid/PG_MAJORVER_CATVER/dboid/relfilenode |
| 21 | * e.g. |
| 22 | * $PGDATA/pg_tblspc/20981/PG_9.0_201002161/719849/83292814 |
| 23 | * |
| 24 | * There are two tablespaces created at initdb time: pg_global (for shared |
| 25 | * tables) and pg_default (for everything else). For backwards compatibility |
| 26 | * and to remain functional on platforms without symlinks, these tablespaces |
| 27 | * are accessed specially: they are respectively |
| 28 | * $PGDATA/global/relfilenode |
| 29 | * $PGDATA/base/dboid/relfilenode |
| 30 | * |
| 31 | * To allow CREATE DATABASE to give a new database a default tablespace |
| 32 | * that's different from the template database's default, we make the |
| 33 | * provision that a zero in pg_class.reltablespace means the database's |
| 34 | * default tablespace. Without this, CREATE DATABASE would have to go in |
| 35 | * and munge the system catalogs of the new database. |
| 36 | * |
| 37 | * |
| 38 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
| 39 | * Portions Copyright (c) 1994, Regents of the University of California |
| 40 | * |
| 41 | * |
| 42 | * IDENTIFICATION |
| 43 | * src/backend/commands/tablespace.c |
| 44 | * |
| 45 | *------------------------------------------------------------------------- |
| 46 | */ |
| 47 | #include "postgres.h" |
| 48 | |
| 49 | #include <unistd.h> |
| 50 | #include <dirent.h> |
| 51 | #include <sys/stat.h> |
| 52 | |
| 53 | #include "access/heapam.h" |
| 54 | #include "access/reloptions.h" |
| 55 | #include "access/htup_details.h" |
| 56 | #include "access/sysattr.h" |
| 57 | #include "access/tableam.h" |
| 58 | #include "access/xact.h" |
| 59 | #include "access/xlog.h" |
| 60 | #include "access/xloginsert.h" |
| 61 | #include "catalog/catalog.h" |
| 62 | #include "catalog/dependency.h" |
| 63 | #include "catalog/indexing.h" |
| 64 | #include "catalog/namespace.h" |
| 65 | #include "catalog/objectaccess.h" |
| 66 | #include "catalog/pg_namespace.h" |
| 67 | #include "catalog/pg_tablespace.h" |
| 68 | #include "commands/comment.h" |
| 69 | #include "commands/seclabel.h" |
| 70 | #include "commands/tablecmds.h" |
| 71 | #include "commands/tablespace.h" |
| 72 | #include "common/file_perm.h" |
| 73 | #include "miscadmin.h" |
| 74 | #include "postmaster/bgwriter.h" |
| 75 | #include "storage/fd.h" |
| 76 | #include "storage/lmgr.h" |
| 77 | #include "storage/standby.h" |
| 78 | #include "utils/acl.h" |
| 79 | #include "utils/builtins.h" |
| 80 | #include "utils/fmgroids.h" |
| 81 | #include "utils/guc.h" |
| 82 | #include "utils/lsyscache.h" |
| 83 | #include "utils/memutils.h" |
| 84 | #include "utils/rel.h" |
| 85 | #include "utils/varlena.h" |
| 86 | |
| 87 | |
| 88 | /* GUC variables */ |
| 89 | char *default_tablespace = NULL; |
| 90 | char *temp_tablespaces = NULL; |
| 91 | |
| 92 | |
| 93 | static void create_tablespace_directories(const char *location, |
| 94 | const Oid tablespaceoid); |
| 95 | static bool destroy_tablespace_directories(Oid tablespaceoid, bool redo); |
| 96 | |
| 97 | |
| 98 | /* |
| 99 | * Each database using a table space is isolated into its own name space |
| 100 | * by a subdirectory named for the database OID. On first creation of an |
| 101 | * object in the tablespace, create the subdirectory. If the subdirectory |
| 102 | * already exists, fall through quietly. |
| 103 | * |
| 104 | * isRedo indicates that we are creating an object during WAL replay. |
| 105 | * In this case we will cope with the possibility of the tablespace |
| 106 | * directory not being there either --- this could happen if we are |
| 107 | * replaying an operation on a table in a subsequently-dropped tablespace. |
| 108 | * We handle this by making a directory in the place where the tablespace |
| 109 | * symlink would normally be. This isn't an exact replay of course, but |
| 110 | * it's the best we can do given the available information. |
| 111 | * |
| 112 | * If tablespaces are not supported, we still need it in case we have to |
| 113 | * re-create a database subdirectory (of $PGDATA/base) during WAL replay. |
| 114 | */ |
| 115 | void |
| 116 | TablespaceCreateDbspace(Oid spcNode, Oid dbNode, bool isRedo) |
| 117 | { |
| 118 | struct stat st; |
| 119 | char *dir; |
| 120 | |
| 121 | /* |
| 122 | * The global tablespace doesn't have per-database subdirectories, so |
| 123 | * nothing to do for it. |
| 124 | */ |
| 125 | if (spcNode == GLOBALTABLESPACE_OID) |
| 126 | return; |
| 127 | |
| 128 | Assert(OidIsValid(spcNode)); |
| 129 | Assert(OidIsValid(dbNode)); |
| 130 | |
| 131 | dir = GetDatabasePath(dbNode, spcNode); |
| 132 | |
| 133 | if (stat(dir, &st) < 0) |
| 134 | { |
| 135 | /* Directory does not exist? */ |
| 136 | if (errno == ENOENT) |
| 137 | { |
| 138 | /* |
| 139 | * Acquire TablespaceCreateLock to ensure that no DROP TABLESPACE |
| 140 | * or TablespaceCreateDbspace is running concurrently. |
| 141 | */ |
| 142 | LWLockAcquire(TablespaceCreateLock, LW_EXCLUSIVE); |
| 143 | |
| 144 | /* |
| 145 | * Recheck to see if someone created the directory while we were |
| 146 | * waiting for lock. |
| 147 | */ |
| 148 | if (stat(dir, &st) == 0 && S_ISDIR(st.st_mode)) |
| 149 | { |
| 150 | /* Directory was created */ |
| 151 | } |
| 152 | else |
| 153 | { |
| 154 | /* Directory creation failed? */ |
| 155 | if (MakePGDirectory(dir) < 0) |
| 156 | { |
| 157 | char *parentdir; |
| 158 | |
| 159 | /* Failure other than not exists or not in WAL replay? */ |
| 160 | if (errno != ENOENT || !isRedo) |
| 161 | ereport(ERROR, |
| 162 | (errcode_for_file_access(), |
| 163 | errmsg("could not create directory \"%s\": %m" , |
| 164 | dir))); |
| 165 | |
| 166 | /* |
| 167 | * Parent directories are missing during WAL replay, so |
| 168 | * continue by creating simple parent directories rather |
| 169 | * than a symlink. |
| 170 | */ |
| 171 | |
| 172 | /* create two parents up if not exist */ |
| 173 | parentdir = pstrdup(dir); |
| 174 | get_parent_directory(parentdir); |
| 175 | get_parent_directory(parentdir); |
| 176 | /* Can't create parent and it doesn't already exist? */ |
| 177 | if (MakePGDirectory(parentdir) < 0 && errno != EEXIST) |
| 178 | ereport(ERROR, |
| 179 | (errcode_for_file_access(), |
| 180 | errmsg("could not create directory \"%s\": %m" , |
| 181 | parentdir))); |
| 182 | pfree(parentdir); |
| 183 | |
| 184 | /* create one parent up if not exist */ |
| 185 | parentdir = pstrdup(dir); |
| 186 | get_parent_directory(parentdir); |
| 187 | /* Can't create parent and it doesn't already exist? */ |
| 188 | if (MakePGDirectory(parentdir) < 0 && errno != EEXIST) |
| 189 | ereport(ERROR, |
| 190 | (errcode_for_file_access(), |
| 191 | errmsg("could not create directory \"%s\": %m" , |
| 192 | parentdir))); |
| 193 | pfree(parentdir); |
| 194 | |
| 195 | /* Create database directory */ |
| 196 | if (MakePGDirectory(dir) < 0) |
| 197 | ereport(ERROR, |
| 198 | (errcode_for_file_access(), |
| 199 | errmsg("could not create directory \"%s\": %m" , |
| 200 | dir))); |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | LWLockRelease(TablespaceCreateLock); |
| 205 | } |
| 206 | else |
| 207 | { |
| 208 | ereport(ERROR, |
| 209 | (errcode_for_file_access(), |
| 210 | errmsg("could not stat directory \"%s\": %m" , dir))); |
| 211 | } |
| 212 | } |
| 213 | else |
| 214 | { |
| 215 | /* Is it not a directory? */ |
| 216 | if (!S_ISDIR(st.st_mode)) |
| 217 | ereport(ERROR, |
| 218 | (errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 219 | errmsg("\"%s\" exists but is not a directory" , |
| 220 | dir))); |
| 221 | } |
| 222 | |
| 223 | pfree(dir); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * Create a table space |
| 228 | * |
| 229 | * Only superusers can create a tablespace. This seems a reasonable restriction |
| 230 | * since we're determining the system layout and, anyway, we probably have |
| 231 | * root if we're doing this kind of activity |
| 232 | */ |
| 233 | Oid |
| 234 | CreateTableSpace(CreateTableSpaceStmt *stmt) |
| 235 | { |
| 236 | #ifdef HAVE_SYMLINK |
| 237 | Relation rel; |
| 238 | Datum values[Natts_pg_tablespace]; |
| 239 | bool nulls[Natts_pg_tablespace]; |
| 240 | HeapTuple tuple; |
| 241 | Oid tablespaceoid; |
| 242 | char *location; |
| 243 | Oid ownerId; |
| 244 | Datum newOptions; |
| 245 | |
| 246 | /* Must be super user */ |
| 247 | if (!superuser()) |
| 248 | ereport(ERROR, |
| 249 | (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), |
| 250 | errmsg("permission denied to create tablespace \"%s\"" , |
| 251 | stmt->tablespacename), |
| 252 | errhint("Must be superuser to create a tablespace." ))); |
| 253 | |
| 254 | /* However, the eventual owner of the tablespace need not be */ |
| 255 | if (stmt->owner) |
| 256 | ownerId = get_rolespec_oid(stmt->owner, false); |
| 257 | else |
| 258 | ownerId = GetUserId(); |
| 259 | |
| 260 | /* Unix-ify the offered path, and strip any trailing slashes */ |
| 261 | location = pstrdup(stmt->location); |
| 262 | canonicalize_path(location); |
| 263 | |
| 264 | /* disallow quotes, else CREATE DATABASE would be at risk */ |
| 265 | if (strchr(location, '\'')) |
| 266 | ereport(ERROR, |
| 267 | (errcode(ERRCODE_INVALID_NAME), |
| 268 | errmsg("tablespace location cannot contain single quotes" ))); |
| 269 | |
| 270 | /* |
| 271 | * Allowing relative paths seems risky |
| 272 | * |
| 273 | * this also helps us ensure that location is not empty or whitespace |
| 274 | */ |
| 275 | if (!is_absolute_path(location)) |
| 276 | ereport(ERROR, |
| 277 | (errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 278 | errmsg("tablespace location must be an absolute path" ))); |
| 279 | |
| 280 | /* |
| 281 | * Check that location isn't too long. Remember that we're going to append |
| 282 | * 'PG_XXX/<dboid>/<relid>_<fork>.<nnn>'. FYI, we never actually |
| 283 | * reference the whole path here, but MakePGDirectory() uses the first two |
| 284 | * parts. |
| 285 | */ |
| 286 | if (strlen(location) + 1 + strlen(TABLESPACE_VERSION_DIRECTORY) + 1 + |
| 287 | OIDCHARS + 1 + OIDCHARS + 1 + FORKNAMECHARS + 1 + OIDCHARS > MAXPGPATH) |
| 288 | ereport(ERROR, |
| 289 | (errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 290 | errmsg("tablespace location \"%s\" is too long" , |
| 291 | location))); |
| 292 | |
| 293 | /* Warn if the tablespace is in the data directory. */ |
| 294 | if (path_is_prefix_of_path(DataDir, location)) |
| 295 | ereport(WARNING, |
| 296 | (errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 297 | errmsg("tablespace location should not be inside the data directory" ))); |
| 298 | |
| 299 | /* |
| 300 | * Disallow creation of tablespaces named "pg_xxx"; we reserve this |
| 301 | * namespace for system purposes. |
| 302 | */ |
| 303 | if (!allowSystemTableMods && IsReservedName(stmt->tablespacename)) |
| 304 | ereport(ERROR, |
| 305 | (errcode(ERRCODE_RESERVED_NAME), |
| 306 | errmsg("unacceptable tablespace name \"%s\"" , |
| 307 | stmt->tablespacename), |
| 308 | errdetail("The prefix \"pg_\" is reserved for system tablespaces." ))); |
| 309 | |
| 310 | /* |
| 311 | * If built with appropriate switch, whine when regression-testing |
| 312 | * conventions for tablespace names are violated. |
| 313 | */ |
| 314 | #ifdef ENFORCE_REGRESSION_TEST_NAME_RESTRICTIONS |
| 315 | if (strncmp(stmt->tablespacename, "regress_" , 8) != 0) |
| 316 | elog(WARNING, "tablespaces created by regression test cases should have names starting with \"regress_\"" ); |
| 317 | #endif |
| 318 | |
| 319 | /* |
| 320 | * Check that there is no other tablespace by this name. (The unique |
| 321 | * index would catch this anyway, but might as well give a friendlier |
| 322 | * message.) |
| 323 | */ |
| 324 | if (OidIsValid(get_tablespace_oid(stmt->tablespacename, true))) |
| 325 | ereport(ERROR, |
| 326 | (errcode(ERRCODE_DUPLICATE_OBJECT), |
| 327 | errmsg("tablespace \"%s\" already exists" , |
| 328 | stmt->tablespacename))); |
| 329 | |
| 330 | /* |
| 331 | * Insert tuple into pg_tablespace. The purpose of doing this first is to |
| 332 | * lock the proposed tablename against other would-be creators. The |
| 333 | * insertion will roll back if we find problems below. |
| 334 | */ |
| 335 | rel = table_open(TableSpaceRelationId, RowExclusiveLock); |
| 336 | |
| 337 | MemSet(nulls, false, sizeof(nulls)); |
| 338 | |
| 339 | tablespaceoid = GetNewOidWithIndex(rel, TablespaceOidIndexId, |
| 340 | Anum_pg_tablespace_oid); |
| 341 | values[Anum_pg_tablespace_oid - 1] = ObjectIdGetDatum(tablespaceoid); |
| 342 | values[Anum_pg_tablespace_spcname - 1] = |
| 343 | DirectFunctionCall1(namein, CStringGetDatum(stmt->tablespacename)); |
| 344 | values[Anum_pg_tablespace_spcowner - 1] = |
| 345 | ObjectIdGetDatum(ownerId); |
| 346 | nulls[Anum_pg_tablespace_spcacl - 1] = true; |
| 347 | |
| 348 | /* Generate new proposed spcoptions (text array) */ |
| 349 | newOptions = transformRelOptions((Datum) 0, |
| 350 | stmt->options, |
| 351 | NULL, NULL, false, false); |
| 352 | (void) tablespace_reloptions(newOptions, true); |
| 353 | if (newOptions != (Datum) 0) |
| 354 | values[Anum_pg_tablespace_spcoptions - 1] = newOptions; |
| 355 | else |
| 356 | nulls[Anum_pg_tablespace_spcoptions - 1] = true; |
| 357 | |
| 358 | tuple = heap_form_tuple(rel->rd_att, values, nulls); |
| 359 | |
| 360 | CatalogTupleInsert(rel, tuple); |
| 361 | |
| 362 | heap_freetuple(tuple); |
| 363 | |
| 364 | /* Record dependency on owner */ |
| 365 | recordDependencyOnOwner(TableSpaceRelationId, tablespaceoid, ownerId); |
| 366 | |
| 367 | /* Post creation hook for new tablespace */ |
| 368 | InvokeObjectPostCreateHook(TableSpaceRelationId, tablespaceoid, 0); |
| 369 | |
| 370 | create_tablespace_directories(location, tablespaceoid); |
| 371 | |
| 372 | /* Record the filesystem change in XLOG */ |
| 373 | { |
| 374 | xl_tblspc_create_rec xlrec; |
| 375 | |
| 376 | xlrec.ts_id = tablespaceoid; |
| 377 | |
| 378 | XLogBeginInsert(); |
| 379 | XLogRegisterData((char *) &xlrec, |
| 380 | offsetof(xl_tblspc_create_rec, ts_path)); |
| 381 | XLogRegisterData((char *) location, strlen(location) + 1); |
| 382 | |
| 383 | (void) XLogInsert(RM_TBLSPC_ID, XLOG_TBLSPC_CREATE); |
| 384 | } |
| 385 | |
| 386 | /* |
| 387 | * Force synchronous commit, to minimize the window between creating the |
| 388 | * symlink on-disk and marking the transaction committed. It's not great |
| 389 | * that there is any window at all, but definitely we don't want to make |
| 390 | * it larger than necessary. |
| 391 | */ |
| 392 | ForceSyncCommit(); |
| 393 | |
| 394 | pfree(location); |
| 395 | |
| 396 | /* We keep the lock on pg_tablespace until commit */ |
| 397 | table_close(rel, NoLock); |
| 398 | |
| 399 | return tablespaceoid; |
| 400 | #else /* !HAVE_SYMLINK */ |
| 401 | ereport(ERROR, |
| 402 | (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 403 | errmsg("tablespaces are not supported on this platform" ))); |
| 404 | return InvalidOid; /* keep compiler quiet */ |
| 405 | #endif /* HAVE_SYMLINK */ |
| 406 | } |
| 407 | |
| 408 | /* |
| 409 | * Drop a table space |
| 410 | * |
| 411 | * Be careful to check that the tablespace is empty. |
| 412 | */ |
| 413 | void |
| 414 | DropTableSpace(DropTableSpaceStmt *stmt) |
| 415 | { |
| 416 | #ifdef HAVE_SYMLINK |
| 417 | char *tablespacename = stmt->tablespacename; |
| 418 | TableScanDesc scandesc; |
| 419 | Relation rel; |
| 420 | HeapTuple tuple; |
| 421 | Form_pg_tablespace spcform; |
| 422 | ScanKeyData entry[1]; |
| 423 | Oid tablespaceoid; |
| 424 | |
| 425 | /* |
| 426 | * Find the target tuple |
| 427 | */ |
| 428 | rel = table_open(TableSpaceRelationId, RowExclusiveLock); |
| 429 | |
| 430 | ScanKeyInit(&entry[0], |
| 431 | Anum_pg_tablespace_spcname, |
| 432 | BTEqualStrategyNumber, F_NAMEEQ, |
| 433 | CStringGetDatum(tablespacename)); |
| 434 | scandesc = table_beginscan_catalog(rel, 1, entry); |
| 435 | tuple = heap_getnext(scandesc, ForwardScanDirection); |
| 436 | |
| 437 | if (!HeapTupleIsValid(tuple)) |
| 438 | { |
| 439 | if (!stmt->missing_ok) |
| 440 | { |
| 441 | ereport(ERROR, |
| 442 | (errcode(ERRCODE_UNDEFINED_OBJECT), |
| 443 | errmsg("tablespace \"%s\" does not exist" , |
| 444 | tablespacename))); |
| 445 | } |
| 446 | else |
| 447 | { |
| 448 | ereport(NOTICE, |
| 449 | (errmsg("tablespace \"%s\" does not exist, skipping" , |
| 450 | tablespacename))); |
| 451 | /* XXX I assume I need one or both of these next two calls */ |
| 452 | table_endscan(scandesc); |
| 453 | table_close(rel, NoLock); |
| 454 | } |
| 455 | return; |
| 456 | } |
| 457 | |
| 458 | spcform = (Form_pg_tablespace) GETSTRUCT(tuple); |
| 459 | tablespaceoid = spcform->oid; |
| 460 | |
| 461 | /* Must be tablespace owner */ |
| 462 | if (!pg_tablespace_ownercheck(tablespaceoid, GetUserId())) |
| 463 | aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_TABLESPACE, |
| 464 | tablespacename); |
| 465 | |
| 466 | /* Disallow drop of the standard tablespaces, even by superuser */ |
| 467 | if (tablespaceoid == GLOBALTABLESPACE_OID || |
| 468 | tablespaceoid == DEFAULTTABLESPACE_OID) |
| 469 | aclcheck_error(ACLCHECK_NO_PRIV, OBJECT_TABLESPACE, |
| 470 | tablespacename); |
| 471 | |
| 472 | /* DROP hook for the tablespace being removed */ |
| 473 | InvokeObjectDropHook(TableSpaceRelationId, tablespaceoid, 0); |
| 474 | |
| 475 | /* |
| 476 | * Remove the pg_tablespace tuple (this will roll back if we fail below) |
| 477 | */ |
| 478 | CatalogTupleDelete(rel, &tuple->t_self); |
| 479 | |
| 480 | table_endscan(scandesc); |
| 481 | |
| 482 | /* |
| 483 | * Remove any comments or security labels on this tablespace. |
| 484 | */ |
| 485 | DeleteSharedComments(tablespaceoid, TableSpaceRelationId); |
| 486 | DeleteSharedSecurityLabel(tablespaceoid, TableSpaceRelationId); |
| 487 | |
| 488 | /* |
| 489 | * Remove dependency on owner. |
| 490 | */ |
| 491 | deleteSharedDependencyRecordsFor(TableSpaceRelationId, tablespaceoid, 0); |
| 492 | |
| 493 | /* |
| 494 | * Acquire TablespaceCreateLock to ensure that no TablespaceCreateDbspace |
| 495 | * is running concurrently. |
| 496 | */ |
| 497 | LWLockAcquire(TablespaceCreateLock, LW_EXCLUSIVE); |
| 498 | |
| 499 | /* |
| 500 | * Try to remove the physical infrastructure. |
| 501 | */ |
| 502 | if (!destroy_tablespace_directories(tablespaceoid, false)) |
| 503 | { |
| 504 | /* |
| 505 | * Not all files deleted? However, there can be lingering empty files |
| 506 | * in the directories, left behind by for example DROP TABLE, that |
| 507 | * have been scheduled for deletion at next checkpoint (see comments |
| 508 | * in mdunlink() for details). We could just delete them immediately, |
| 509 | * but we can't tell them apart from important data files that we |
| 510 | * mustn't delete. So instead, we force a checkpoint which will clean |
| 511 | * out any lingering files, and try again. |
| 512 | * |
| 513 | * XXX On Windows, an unlinked file persists in the directory listing |
| 514 | * until no process retains an open handle for the file. The DDL |
| 515 | * commands that schedule files for unlink send invalidation messages |
| 516 | * directing other PostgreSQL processes to close the files. DROP |
| 517 | * TABLESPACE should not give up on the tablespace becoming empty |
| 518 | * until all relevant invalidation processing is complete. |
| 519 | */ |
| 520 | RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT); |
| 521 | if (!destroy_tablespace_directories(tablespaceoid, false)) |
| 522 | { |
| 523 | /* Still not empty, the files must be important then */ |
| 524 | ereport(ERROR, |
| 525 | (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 526 | errmsg("tablespace \"%s\" is not empty" , |
| 527 | tablespacename))); |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | /* Record the filesystem change in XLOG */ |
| 532 | { |
| 533 | xl_tblspc_drop_rec xlrec; |
| 534 | |
| 535 | xlrec.ts_id = tablespaceoid; |
| 536 | |
| 537 | XLogBeginInsert(); |
| 538 | XLogRegisterData((char *) &xlrec, sizeof(xl_tblspc_drop_rec)); |
| 539 | |
| 540 | (void) XLogInsert(RM_TBLSPC_ID, XLOG_TBLSPC_DROP); |
| 541 | } |
| 542 | |
| 543 | /* |
| 544 | * Note: because we checked that the tablespace was empty, there should be |
| 545 | * no need to worry about flushing shared buffers or free space map |
| 546 | * entries for relations in the tablespace. |
| 547 | */ |
| 548 | |
| 549 | /* |
| 550 | * Force synchronous commit, to minimize the window between removing the |
| 551 | * files on-disk and marking the transaction committed. It's not great |
| 552 | * that there is any window at all, but definitely we don't want to make |
| 553 | * it larger than necessary. |
| 554 | */ |
| 555 | ForceSyncCommit(); |
| 556 | |
| 557 | /* |
| 558 | * Allow TablespaceCreateDbspace again. |
| 559 | */ |
| 560 | LWLockRelease(TablespaceCreateLock); |
| 561 | |
| 562 | /* We keep the lock on pg_tablespace until commit */ |
| 563 | table_close(rel, NoLock); |
| 564 | #else /* !HAVE_SYMLINK */ |
| 565 | ereport(ERROR, |
| 566 | (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 567 | errmsg("tablespaces are not supported on this platform" ))); |
| 568 | #endif /* HAVE_SYMLINK */ |
| 569 | } |
| 570 | |
| 571 | |
| 572 | /* |
| 573 | * create_tablespace_directories |
| 574 | * |
| 575 | * Attempt to create filesystem infrastructure linking $PGDATA/pg_tblspc/ |
| 576 | * to the specified directory |
| 577 | */ |
| 578 | static void |
| 579 | create_tablespace_directories(const char *location, const Oid tablespaceoid) |
| 580 | { |
| 581 | char *linkloc; |
| 582 | char *location_with_version_dir; |
| 583 | struct stat st; |
| 584 | |
| 585 | linkloc = psprintf("pg_tblspc/%u" , tablespaceoid); |
| 586 | location_with_version_dir = psprintf("%s/%s" , location, |
| 587 | TABLESPACE_VERSION_DIRECTORY); |
| 588 | |
| 589 | /* |
| 590 | * Attempt to coerce target directory to safe permissions. If this fails, |
| 591 | * it doesn't exist or has the wrong owner. |
| 592 | */ |
| 593 | if (chmod(location, pg_dir_create_mode) != 0) |
| 594 | { |
| 595 | if (errno == ENOENT) |
| 596 | ereport(ERROR, |
| 597 | (errcode(ERRCODE_UNDEFINED_FILE), |
| 598 | errmsg("directory \"%s\" does not exist" , location), |
| 599 | InRecovery ? errhint("Create this directory for the tablespace before " |
| 600 | "restarting the server." ) : 0)); |
| 601 | else |
| 602 | ereport(ERROR, |
| 603 | (errcode_for_file_access(), |
| 604 | errmsg("could not set permissions on directory \"%s\": %m" , |
| 605 | location))); |
| 606 | } |
| 607 | |
| 608 | if (InRecovery) |
| 609 | { |
| 610 | /* |
| 611 | * Our theory for replaying a CREATE is to forcibly drop the target |
| 612 | * subdirectory if present, and then recreate it. This may be more |
| 613 | * work than needed, but it is simple to implement. |
| 614 | */ |
| 615 | if (stat(location_with_version_dir, &st) == 0 && S_ISDIR(st.st_mode)) |
| 616 | { |
| 617 | if (!rmtree(location_with_version_dir, true)) |
| 618 | /* If this failed, MakePGDirectory() below is going to error. */ |
| 619 | ereport(WARNING, |
| 620 | (errmsg("some useless files may be left behind in old database directory \"%s\"" , |
| 621 | location_with_version_dir))); |
| 622 | } |
| 623 | } |
| 624 | |
| 625 | /* |
| 626 | * The creation of the version directory prevents more than one tablespace |
| 627 | * in a single location. |
| 628 | */ |
| 629 | if (MakePGDirectory(location_with_version_dir) < 0) |
| 630 | { |
| 631 | if (errno == EEXIST) |
| 632 | ereport(ERROR, |
| 633 | (errcode(ERRCODE_OBJECT_IN_USE), |
| 634 | errmsg("directory \"%s\" already in use as a tablespace" , |
| 635 | location_with_version_dir))); |
| 636 | else |
| 637 | ereport(ERROR, |
| 638 | (errcode_for_file_access(), |
| 639 | errmsg("could not create directory \"%s\": %m" , |
| 640 | location_with_version_dir))); |
| 641 | } |
| 642 | |
| 643 | /* |
| 644 | * In recovery, remove old symlink, in case it points to the wrong place. |
| 645 | */ |
| 646 | if (InRecovery) |
| 647 | remove_tablespace_symlink(linkloc); |
| 648 | |
| 649 | /* |
| 650 | * Create the symlink under PGDATA |
| 651 | */ |
| 652 | if (symlink(location, linkloc) < 0) |
| 653 | ereport(ERROR, |
| 654 | (errcode_for_file_access(), |
| 655 | errmsg("could not create symbolic link \"%s\": %m" , |
| 656 | linkloc))); |
| 657 | |
| 658 | pfree(linkloc); |
| 659 | pfree(location_with_version_dir); |
| 660 | } |
| 661 | |
| 662 | |
| 663 | /* |
| 664 | * destroy_tablespace_directories |
| 665 | * |
| 666 | * Attempt to remove filesystem infrastructure for the tablespace. |
| 667 | * |
| 668 | * 'redo' indicates we are redoing a drop from XLOG; in that case we should |
| 669 | * not throw an ERROR for problems, just LOG them. The worst consequence of |
| 670 | * not removing files here would be failure to release some disk space, which |
| 671 | * does not justify throwing an error that would require manual intervention |
| 672 | * to get the database running again. |
| 673 | * |
| 674 | * Returns true if successful, false if some subdirectory is not empty |
| 675 | */ |
| 676 | static bool |
| 677 | destroy_tablespace_directories(Oid tablespaceoid, bool redo) |
| 678 | { |
| 679 | char *linkloc; |
| 680 | char *linkloc_with_version_dir; |
| 681 | DIR *dirdesc; |
| 682 | struct dirent *de; |
| 683 | char *subfile; |
| 684 | struct stat st; |
| 685 | |
| 686 | linkloc_with_version_dir = psprintf("pg_tblspc/%u/%s" , tablespaceoid, |
| 687 | TABLESPACE_VERSION_DIRECTORY); |
| 688 | |
| 689 | /* |
| 690 | * Check if the tablespace still contains any files. We try to rmdir each |
| 691 | * per-database directory we find in it. rmdir failure implies there are |
| 692 | * still files in that subdirectory, so give up. (We do not have to worry |
| 693 | * about undoing any already completed rmdirs, since the next attempt to |
| 694 | * use the tablespace from that database will simply recreate the |
| 695 | * subdirectory via TablespaceCreateDbspace.) |
| 696 | * |
| 697 | * Since we hold TablespaceCreateLock, no one else should be creating any |
| 698 | * fresh subdirectories in parallel. It is possible that new files are |
| 699 | * being created within subdirectories, though, so the rmdir call could |
| 700 | * fail. Worst consequence is a less friendly error message. |
| 701 | * |
| 702 | * If redo is true then ENOENT is a likely outcome here, and we allow it |
| 703 | * to pass without comment. In normal operation we still allow it, but |
| 704 | * with a warning. This is because even though ProcessUtility disallows |
| 705 | * DROP TABLESPACE in a transaction block, it's possible that a previous |
| 706 | * DROP failed and rolled back after removing the tablespace directories |
| 707 | * and/or symlink. We want to allow a new DROP attempt to succeed at |
| 708 | * removing the catalog entries (and symlink if still present), so we |
| 709 | * should not give a hard error here. |
| 710 | */ |
| 711 | dirdesc = AllocateDir(linkloc_with_version_dir); |
| 712 | if (dirdesc == NULL) |
| 713 | { |
| 714 | if (errno == ENOENT) |
| 715 | { |
| 716 | if (!redo) |
| 717 | ereport(WARNING, |
| 718 | (errcode_for_file_access(), |
| 719 | errmsg("could not open directory \"%s\": %m" , |
| 720 | linkloc_with_version_dir))); |
| 721 | /* The symlink might still exist, so go try to remove it */ |
| 722 | goto remove_symlink; |
| 723 | } |
| 724 | else if (redo) |
| 725 | { |
| 726 | /* in redo, just log other types of error */ |
| 727 | ereport(LOG, |
| 728 | (errcode_for_file_access(), |
| 729 | errmsg("could not open directory \"%s\": %m" , |
| 730 | linkloc_with_version_dir))); |
| 731 | pfree(linkloc_with_version_dir); |
| 732 | return false; |
| 733 | } |
| 734 | /* else let ReadDir report the error */ |
| 735 | } |
| 736 | |
| 737 | while ((de = ReadDir(dirdesc, linkloc_with_version_dir)) != NULL) |
| 738 | { |
| 739 | if (strcmp(de->d_name, "." ) == 0 || |
| 740 | strcmp(de->d_name, ".." ) == 0) |
| 741 | continue; |
| 742 | |
| 743 | subfile = psprintf("%s/%s" , linkloc_with_version_dir, de->d_name); |
| 744 | |
| 745 | /* This check is just to deliver a friendlier error message */ |
| 746 | if (!redo && !directory_is_empty(subfile)) |
| 747 | { |
| 748 | FreeDir(dirdesc); |
| 749 | pfree(subfile); |
| 750 | pfree(linkloc_with_version_dir); |
| 751 | return false; |
| 752 | } |
| 753 | |
| 754 | /* remove empty directory */ |
| 755 | if (rmdir(subfile) < 0) |
| 756 | ereport(redo ? LOG : ERROR, |
| 757 | (errcode_for_file_access(), |
| 758 | errmsg("could not remove directory \"%s\": %m" , |
| 759 | subfile))); |
| 760 | |
| 761 | pfree(subfile); |
| 762 | } |
| 763 | |
| 764 | FreeDir(dirdesc); |
| 765 | |
| 766 | /* remove version directory */ |
| 767 | if (rmdir(linkloc_with_version_dir) < 0) |
| 768 | { |
| 769 | ereport(redo ? LOG : ERROR, |
| 770 | (errcode_for_file_access(), |
| 771 | errmsg("could not remove directory \"%s\": %m" , |
| 772 | linkloc_with_version_dir))); |
| 773 | pfree(linkloc_with_version_dir); |
| 774 | return false; |
| 775 | } |
| 776 | |
| 777 | /* |
| 778 | * Try to remove the symlink. We must however deal with the possibility |
| 779 | * that it's a directory instead of a symlink --- this could happen during |
| 780 | * WAL replay (see TablespaceCreateDbspace), and it is also the case on |
| 781 | * Windows where junction points lstat() as directories. |
| 782 | * |
| 783 | * Note: in the redo case, we'll return true if this final step fails; |
| 784 | * there's no point in retrying it. Also, ENOENT should provoke no more |
| 785 | * than a warning. |
| 786 | */ |
| 787 | remove_symlink: |
| 788 | linkloc = pstrdup(linkloc_with_version_dir); |
| 789 | get_parent_directory(linkloc); |
| 790 | if (lstat(linkloc, &st) < 0) |
| 791 | { |
| 792 | int saved_errno = errno; |
| 793 | |
| 794 | ereport(redo ? LOG : (saved_errno == ENOENT ? WARNING : ERROR), |
| 795 | (errcode_for_file_access(), |
| 796 | errmsg("could not stat file \"%s\": %m" , |
| 797 | linkloc))); |
| 798 | } |
| 799 | else if (S_ISDIR(st.st_mode)) |
| 800 | { |
| 801 | if (rmdir(linkloc) < 0) |
| 802 | { |
| 803 | int saved_errno = errno; |
| 804 | |
| 805 | ereport(redo ? LOG : (saved_errno == ENOENT ? WARNING : ERROR), |
| 806 | (errcode_for_file_access(), |
| 807 | errmsg("could not remove directory \"%s\": %m" , |
| 808 | linkloc))); |
| 809 | } |
| 810 | } |
| 811 | #ifdef S_ISLNK |
| 812 | else if (S_ISLNK(st.st_mode)) |
| 813 | { |
| 814 | if (unlink(linkloc) < 0) |
| 815 | { |
| 816 | int saved_errno = errno; |
| 817 | |
| 818 | ereport(redo ? LOG : (saved_errno == ENOENT ? WARNING : ERROR), |
| 819 | (errcode_for_file_access(), |
| 820 | errmsg("could not remove symbolic link \"%s\": %m" , |
| 821 | linkloc))); |
| 822 | } |
| 823 | } |
| 824 | #endif |
| 825 | else |
| 826 | { |
| 827 | /* Refuse to remove anything that's not a directory or symlink */ |
| 828 | ereport(redo ? LOG : ERROR, |
| 829 | (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 830 | errmsg("\"%s\" is not a directory or symbolic link" , |
| 831 | linkloc))); |
| 832 | } |
| 833 | |
| 834 | pfree(linkloc_with_version_dir); |
| 835 | pfree(linkloc); |
| 836 | |
| 837 | return true; |
| 838 | } |
| 839 | |
| 840 | |
| 841 | /* |
| 842 | * Check if a directory is empty. |
| 843 | * |
| 844 | * This probably belongs somewhere else, but not sure where... |
| 845 | */ |
| 846 | bool |
| 847 | directory_is_empty(const char *path) |
| 848 | { |
| 849 | DIR *dirdesc; |
| 850 | struct dirent *de; |
| 851 | |
| 852 | dirdesc = AllocateDir(path); |
| 853 | |
| 854 | while ((de = ReadDir(dirdesc, path)) != NULL) |
| 855 | { |
| 856 | if (strcmp(de->d_name, "." ) == 0 || |
| 857 | strcmp(de->d_name, ".." ) == 0) |
| 858 | continue; |
| 859 | FreeDir(dirdesc); |
| 860 | return false; |
| 861 | } |
| 862 | |
| 863 | FreeDir(dirdesc); |
| 864 | return true; |
| 865 | } |
| 866 | |
| 867 | /* |
| 868 | * remove_tablespace_symlink |
| 869 | * |
| 870 | * This function removes symlinks in pg_tblspc. On Windows, junction points |
| 871 | * act like directories so we must be able to apply rmdir. This function |
| 872 | * works like the symlink removal code in destroy_tablespace_directories, |
| 873 | * except that failure to remove is always an ERROR. But if the file doesn't |
| 874 | * exist at all, that's OK. |
| 875 | */ |
| 876 | void |
| 877 | remove_tablespace_symlink(const char *linkloc) |
| 878 | { |
| 879 | struct stat st; |
| 880 | |
| 881 | if (lstat(linkloc, &st) < 0) |
| 882 | { |
| 883 | if (errno == ENOENT) |
| 884 | return; |
| 885 | ereport(ERROR, |
| 886 | (errcode_for_file_access(), |
| 887 | errmsg("could not stat file \"%s\": %m" , linkloc))); |
| 888 | } |
| 889 | |
| 890 | if (S_ISDIR(st.st_mode)) |
| 891 | { |
| 892 | /* |
| 893 | * This will fail if the directory isn't empty, but not if it's a |
| 894 | * junction point. |
| 895 | */ |
| 896 | if (rmdir(linkloc) < 0 && errno != ENOENT) |
| 897 | ereport(ERROR, |
| 898 | (errcode_for_file_access(), |
| 899 | errmsg("could not remove directory \"%s\": %m" , |
| 900 | linkloc))); |
| 901 | } |
| 902 | #ifdef S_ISLNK |
| 903 | else if (S_ISLNK(st.st_mode)) |
| 904 | { |
| 905 | if (unlink(linkloc) < 0 && errno != ENOENT) |
| 906 | ereport(ERROR, |
| 907 | (errcode_for_file_access(), |
| 908 | errmsg("could not remove symbolic link \"%s\": %m" , |
| 909 | linkloc))); |
| 910 | } |
| 911 | #endif |
| 912 | else |
| 913 | { |
| 914 | /* Refuse to remove anything that's not a directory or symlink */ |
| 915 | ereport(ERROR, |
| 916 | (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 917 | errmsg("\"%s\" is not a directory or symbolic link" , |
| 918 | linkloc))); |
| 919 | } |
| 920 | } |
| 921 | |
| 922 | /* |
| 923 | * Rename a tablespace |
| 924 | */ |
| 925 | ObjectAddress |
| 926 | RenameTableSpace(const char *oldname, const char *newname) |
| 927 | { |
| 928 | Oid tspId; |
| 929 | Relation rel; |
| 930 | ScanKeyData entry[1]; |
| 931 | TableScanDesc scan; |
| 932 | HeapTuple tup; |
| 933 | HeapTuple newtuple; |
| 934 | Form_pg_tablespace newform; |
| 935 | ObjectAddress address; |
| 936 | |
| 937 | /* Search pg_tablespace */ |
| 938 | rel = table_open(TableSpaceRelationId, RowExclusiveLock); |
| 939 | |
| 940 | ScanKeyInit(&entry[0], |
| 941 | Anum_pg_tablespace_spcname, |
| 942 | BTEqualStrategyNumber, F_NAMEEQ, |
| 943 | CStringGetDatum(oldname)); |
| 944 | scan = table_beginscan_catalog(rel, 1, entry); |
| 945 | tup = heap_getnext(scan, ForwardScanDirection); |
| 946 | if (!HeapTupleIsValid(tup)) |
| 947 | ereport(ERROR, |
| 948 | (errcode(ERRCODE_UNDEFINED_OBJECT), |
| 949 | errmsg("tablespace \"%s\" does not exist" , |
| 950 | oldname))); |
| 951 | |
| 952 | newtuple = heap_copytuple(tup); |
| 953 | newform = (Form_pg_tablespace) GETSTRUCT(newtuple); |
| 954 | tspId = newform->oid; |
| 955 | |
| 956 | table_endscan(scan); |
| 957 | |
| 958 | /* Must be owner */ |
| 959 | if (!pg_tablespace_ownercheck(tspId, GetUserId())) |
| 960 | aclcheck_error(ACLCHECK_NO_PRIV, OBJECT_TABLESPACE, oldname); |
| 961 | |
| 962 | /* Validate new name */ |
| 963 | if (!allowSystemTableMods && IsReservedName(newname)) |
| 964 | ereport(ERROR, |
| 965 | (errcode(ERRCODE_RESERVED_NAME), |
| 966 | errmsg("unacceptable tablespace name \"%s\"" , newname), |
| 967 | errdetail("The prefix \"pg_\" is reserved for system tablespaces." ))); |
| 968 | |
| 969 | /* |
| 970 | * If built with appropriate switch, whine when regression-testing |
| 971 | * conventions for tablespace names are violated. |
| 972 | */ |
| 973 | #ifdef ENFORCE_REGRESSION_TEST_NAME_RESTRICTIONS |
| 974 | if (strncmp(newname, "regress_" , 8) != 0) |
| 975 | elog(WARNING, "tablespaces created by regression test cases should have names starting with \"regress_\"" ); |
| 976 | #endif |
| 977 | |
| 978 | /* Make sure the new name doesn't exist */ |
| 979 | ScanKeyInit(&entry[0], |
| 980 | Anum_pg_tablespace_spcname, |
| 981 | BTEqualStrategyNumber, F_NAMEEQ, |
| 982 | CStringGetDatum(newname)); |
| 983 | scan = table_beginscan_catalog(rel, 1, entry); |
| 984 | tup = heap_getnext(scan, ForwardScanDirection); |
| 985 | if (HeapTupleIsValid(tup)) |
| 986 | ereport(ERROR, |
| 987 | (errcode(ERRCODE_DUPLICATE_OBJECT), |
| 988 | errmsg("tablespace \"%s\" already exists" , |
| 989 | newname))); |
| 990 | |
| 991 | table_endscan(scan); |
| 992 | |
| 993 | /* OK, update the entry */ |
| 994 | namestrcpy(&(newform->spcname), newname); |
| 995 | |
| 996 | CatalogTupleUpdate(rel, &newtuple->t_self, newtuple); |
| 997 | |
| 998 | InvokeObjectPostAlterHook(TableSpaceRelationId, tspId, 0); |
| 999 | |
| 1000 | ObjectAddressSet(address, TableSpaceRelationId, tspId); |
| 1001 | |
| 1002 | table_close(rel, NoLock); |
| 1003 | |
| 1004 | return address; |
| 1005 | } |
| 1006 | |
| 1007 | /* |
| 1008 | * Alter table space options |
| 1009 | */ |
| 1010 | Oid |
| 1011 | AlterTableSpaceOptions(AlterTableSpaceOptionsStmt *stmt) |
| 1012 | { |
| 1013 | Relation rel; |
| 1014 | ScanKeyData entry[1]; |
| 1015 | TableScanDesc scandesc; |
| 1016 | HeapTuple tup; |
| 1017 | Oid tablespaceoid; |
| 1018 | Datum datum; |
| 1019 | Datum newOptions; |
| 1020 | Datum repl_val[Natts_pg_tablespace]; |
| 1021 | bool isnull; |
| 1022 | bool repl_null[Natts_pg_tablespace]; |
| 1023 | bool repl_repl[Natts_pg_tablespace]; |
| 1024 | HeapTuple newtuple; |
| 1025 | |
| 1026 | /* Search pg_tablespace */ |
| 1027 | rel = table_open(TableSpaceRelationId, RowExclusiveLock); |
| 1028 | |
| 1029 | ScanKeyInit(&entry[0], |
| 1030 | Anum_pg_tablespace_spcname, |
| 1031 | BTEqualStrategyNumber, F_NAMEEQ, |
| 1032 | CStringGetDatum(stmt->tablespacename)); |
| 1033 | scandesc = table_beginscan_catalog(rel, 1, entry); |
| 1034 | tup = heap_getnext(scandesc, ForwardScanDirection); |
| 1035 | if (!HeapTupleIsValid(tup)) |
| 1036 | ereport(ERROR, |
| 1037 | (errcode(ERRCODE_UNDEFINED_OBJECT), |
| 1038 | errmsg("tablespace \"%s\" does not exist" , |
| 1039 | stmt->tablespacename))); |
| 1040 | |
| 1041 | tablespaceoid = ((Form_pg_tablespace) GETSTRUCT(tup))->oid; |
| 1042 | |
| 1043 | /* Must be owner of the existing object */ |
| 1044 | if (!pg_tablespace_ownercheck(tablespaceoid, GetUserId())) |
| 1045 | aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_TABLESPACE, |
| 1046 | stmt->tablespacename); |
| 1047 | |
| 1048 | /* Generate new proposed spcoptions (text array) */ |
| 1049 | datum = heap_getattr(tup, Anum_pg_tablespace_spcoptions, |
| 1050 | RelationGetDescr(rel), &isnull); |
| 1051 | newOptions = transformRelOptions(isnull ? (Datum) 0 : datum, |
| 1052 | stmt->options, NULL, NULL, false, |
| 1053 | stmt->isReset); |
| 1054 | (void) tablespace_reloptions(newOptions, true); |
| 1055 | |
| 1056 | /* Build new tuple. */ |
| 1057 | memset(repl_null, false, sizeof(repl_null)); |
| 1058 | memset(repl_repl, false, sizeof(repl_repl)); |
| 1059 | if (newOptions != (Datum) 0) |
| 1060 | repl_val[Anum_pg_tablespace_spcoptions - 1] = newOptions; |
| 1061 | else |
| 1062 | repl_null[Anum_pg_tablespace_spcoptions - 1] = true; |
| 1063 | repl_repl[Anum_pg_tablespace_spcoptions - 1] = true; |
| 1064 | newtuple = heap_modify_tuple(tup, RelationGetDescr(rel), repl_val, |
| 1065 | repl_null, repl_repl); |
| 1066 | |
| 1067 | /* Update system catalog. */ |
| 1068 | CatalogTupleUpdate(rel, &newtuple->t_self, newtuple); |
| 1069 | |
| 1070 | InvokeObjectPostAlterHook(TableSpaceRelationId, tablespaceoid, 0); |
| 1071 | |
| 1072 | heap_freetuple(newtuple); |
| 1073 | |
| 1074 | /* Conclude heap scan. */ |
| 1075 | table_endscan(scandesc); |
| 1076 | table_close(rel, NoLock); |
| 1077 | |
| 1078 | return tablespaceoid; |
| 1079 | } |
| 1080 | |
| 1081 | /* |
| 1082 | * Routines for handling the GUC variable 'default_tablespace'. |
| 1083 | */ |
| 1084 | |
| 1085 | /* check_hook: validate new default_tablespace */ |
| 1086 | bool |
| 1087 | check_default_tablespace(char **newval, void **, GucSource source) |
| 1088 | { |
| 1089 | /* |
| 1090 | * If we aren't inside a transaction, or connected to a database, we |
| 1091 | * cannot do the catalog accesses necessary to verify the name. Must |
| 1092 | * accept the value on faith. |
| 1093 | */ |
| 1094 | if (IsTransactionState() && MyDatabaseId != InvalidOid) |
| 1095 | { |
| 1096 | if (**newval != '\0' && |
| 1097 | !OidIsValid(get_tablespace_oid(*newval, true))) |
| 1098 | { |
| 1099 | /* |
| 1100 | * When source == PGC_S_TEST, don't throw a hard error for a |
| 1101 | * nonexistent tablespace, only a NOTICE. See comments in guc.h. |
| 1102 | */ |
| 1103 | if (source == PGC_S_TEST) |
| 1104 | { |
| 1105 | ereport(NOTICE, |
| 1106 | (errcode(ERRCODE_UNDEFINED_OBJECT), |
| 1107 | errmsg("tablespace \"%s\" does not exist" , |
| 1108 | *newval))); |
| 1109 | } |
| 1110 | else |
| 1111 | { |
| 1112 | GUC_check_errdetail("Tablespace \"%s\" does not exist." , |
| 1113 | *newval); |
| 1114 | return false; |
| 1115 | } |
| 1116 | } |
| 1117 | } |
| 1118 | |
| 1119 | return true; |
| 1120 | } |
| 1121 | |
| 1122 | /* |
| 1123 | * GetDefaultTablespace -- get the OID of the current default tablespace |
| 1124 | * |
| 1125 | * Temporary objects have different default tablespaces, hence the |
| 1126 | * relpersistence parameter must be specified. Also, for partitioned tables, |
| 1127 | * we disallow specifying the database default, so that needs to be specified |
| 1128 | * too. |
| 1129 | * |
| 1130 | * May return InvalidOid to indicate "use the database's default tablespace". |
| 1131 | * |
| 1132 | * Note that caller is expected to check appropriate permissions for any |
| 1133 | * result other than InvalidOid. |
| 1134 | * |
| 1135 | * This exists to hide (and possibly optimize the use of) the |
| 1136 | * default_tablespace GUC variable. |
| 1137 | */ |
| 1138 | Oid |
| 1139 | GetDefaultTablespace(char relpersistence, bool partitioned) |
| 1140 | { |
| 1141 | Oid result; |
| 1142 | |
| 1143 | /* The temp-table case is handled elsewhere */ |
| 1144 | if (relpersistence == RELPERSISTENCE_TEMP) |
| 1145 | { |
| 1146 | PrepareTempTablespaces(); |
| 1147 | return GetNextTempTableSpace(); |
| 1148 | } |
| 1149 | |
| 1150 | /* Fast path for default_tablespace == "" */ |
| 1151 | if (default_tablespace == NULL || default_tablespace[0] == '\0') |
| 1152 | return InvalidOid; |
| 1153 | |
| 1154 | /* |
| 1155 | * It is tempting to cache this lookup for more speed, but then we would |
| 1156 | * fail to detect the case where the tablespace was dropped since the GUC |
| 1157 | * variable was set. Note also that we don't complain if the value fails |
| 1158 | * to refer to an existing tablespace; we just silently return InvalidOid, |
| 1159 | * causing the new object to be created in the database's tablespace. |
| 1160 | */ |
| 1161 | result = get_tablespace_oid(default_tablespace, true); |
| 1162 | |
| 1163 | /* |
| 1164 | * Allow explicit specification of database's default tablespace in |
| 1165 | * default_tablespace without triggering permissions checks. Don't allow |
| 1166 | * specifying that when creating a partitioned table, however, since the |
| 1167 | * result is confusing. |
| 1168 | */ |
| 1169 | if (result == MyDatabaseTableSpace) |
| 1170 | { |
| 1171 | if (partitioned) |
| 1172 | ereport(ERROR, |
| 1173 | (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 1174 | errmsg("cannot specify default tablespace for partitioned relations" ))); |
| 1175 | result = InvalidOid; |
| 1176 | } |
| 1177 | return result; |
| 1178 | } |
| 1179 | |
| 1180 | |
| 1181 | /* |
| 1182 | * Routines for handling the GUC variable 'temp_tablespaces'. |
| 1183 | */ |
| 1184 | |
| 1185 | typedef struct |
| 1186 | { |
| 1187 | int numSpcs; |
| 1188 | Oid tblSpcs[FLEXIBLE_ARRAY_MEMBER]; |
| 1189 | } ; |
| 1190 | |
| 1191 | /* check_hook: validate new temp_tablespaces */ |
| 1192 | bool |
| 1193 | check_temp_tablespaces(char **newval, void **, GucSource source) |
| 1194 | { |
| 1195 | char *rawname; |
| 1196 | List *namelist; |
| 1197 | |
| 1198 | /* Need a modifiable copy of string */ |
| 1199 | rawname = pstrdup(*newval); |
| 1200 | |
| 1201 | /* Parse string into list of identifiers */ |
| 1202 | if (!SplitIdentifierString(rawname, ',', &namelist)) |
| 1203 | { |
| 1204 | /* syntax error in name list */ |
| 1205 | GUC_check_errdetail("List syntax is invalid." ); |
| 1206 | pfree(rawname); |
| 1207 | list_free(namelist); |
| 1208 | return false; |
| 1209 | } |
| 1210 | |
| 1211 | /* |
| 1212 | * If we aren't inside a transaction, or connected to a database, we |
| 1213 | * cannot do the catalog accesses necessary to verify the name. Must |
| 1214 | * accept the value on faith. Fortunately, there's then also no need to |
| 1215 | * pass the data to fd.c. |
| 1216 | */ |
| 1217 | if (IsTransactionState() && MyDatabaseId != InvalidOid) |
| 1218 | { |
| 1219 | temp_tablespaces_extra *; |
| 1220 | Oid *tblSpcs; |
| 1221 | int numSpcs; |
| 1222 | ListCell *l; |
| 1223 | |
| 1224 | /* temporary workspace until we are done verifying the list */ |
| 1225 | tblSpcs = (Oid *) palloc(list_length(namelist) * sizeof(Oid)); |
| 1226 | numSpcs = 0; |
| 1227 | foreach(l, namelist) |
| 1228 | { |
| 1229 | char *curname = (char *) lfirst(l); |
| 1230 | Oid curoid; |
| 1231 | AclResult aclresult; |
| 1232 | |
| 1233 | /* Allow an empty string (signifying database default) */ |
| 1234 | if (curname[0] == '\0') |
| 1235 | { |
| 1236 | tblSpcs[numSpcs++] = InvalidOid; |
| 1237 | continue; |
| 1238 | } |
| 1239 | |
| 1240 | /* |
| 1241 | * In an interactive SET command, we ereport for bad info. When |
| 1242 | * source == PGC_S_TEST, don't throw a hard error for a |
| 1243 | * nonexistent tablespace, only a NOTICE. See comments in guc.h. |
| 1244 | */ |
| 1245 | curoid = get_tablespace_oid(curname, source <= PGC_S_TEST); |
| 1246 | if (curoid == InvalidOid) |
| 1247 | { |
| 1248 | if (source == PGC_S_TEST) |
| 1249 | ereport(NOTICE, |
| 1250 | (errcode(ERRCODE_UNDEFINED_OBJECT), |
| 1251 | errmsg("tablespace \"%s\" does not exist" , |
| 1252 | curname))); |
| 1253 | continue; |
| 1254 | } |
| 1255 | |
| 1256 | /* |
| 1257 | * Allow explicit specification of database's default tablespace |
| 1258 | * in temp_tablespaces without triggering permissions checks. |
| 1259 | */ |
| 1260 | if (curoid == MyDatabaseTableSpace) |
| 1261 | { |
| 1262 | tblSpcs[numSpcs++] = InvalidOid; |
| 1263 | continue; |
| 1264 | } |
| 1265 | |
| 1266 | /* Check permissions, similarly complaining only if interactive */ |
| 1267 | aclresult = pg_tablespace_aclcheck(curoid, GetUserId(), |
| 1268 | ACL_CREATE); |
| 1269 | if (aclresult != ACLCHECK_OK) |
| 1270 | { |
| 1271 | if (source >= PGC_S_INTERACTIVE) |
| 1272 | aclcheck_error(aclresult, OBJECT_TABLESPACE, curname); |
| 1273 | continue; |
| 1274 | } |
| 1275 | |
| 1276 | tblSpcs[numSpcs++] = curoid; |
| 1277 | } |
| 1278 | |
| 1279 | /* Now prepare an "extra" struct for assign_temp_tablespaces */ |
| 1280 | myextra = malloc(offsetof(temp_tablespaces_extra, tblSpcs) + |
| 1281 | numSpcs * sizeof(Oid)); |
| 1282 | if (!myextra) |
| 1283 | return false; |
| 1284 | myextra->numSpcs = numSpcs; |
| 1285 | memcpy(myextra->tblSpcs, tblSpcs, numSpcs * sizeof(Oid)); |
| 1286 | *extra = (void *) myextra; |
| 1287 | |
| 1288 | pfree(tblSpcs); |
| 1289 | } |
| 1290 | |
| 1291 | pfree(rawname); |
| 1292 | list_free(namelist); |
| 1293 | |
| 1294 | return true; |
| 1295 | } |
| 1296 | |
| 1297 | /* assign_hook: do extra actions as needed */ |
| 1298 | void |
| 1299 | assign_temp_tablespaces(const char *newval, void *) |
| 1300 | { |
| 1301 | temp_tablespaces_extra * = (temp_tablespaces_extra *) extra; |
| 1302 | |
| 1303 | /* |
| 1304 | * If check_temp_tablespaces was executed inside a transaction, then pass |
| 1305 | * the list it made to fd.c. Otherwise, clear fd.c's list; we must be |
| 1306 | * still outside a transaction, or else restoring during transaction exit, |
| 1307 | * and in either case we can just let the next PrepareTempTablespaces call |
| 1308 | * make things sane. |
| 1309 | */ |
| 1310 | if (myextra) |
| 1311 | SetTempTablespaces(myextra->tblSpcs, myextra->numSpcs); |
| 1312 | else |
| 1313 | SetTempTablespaces(NULL, 0); |
| 1314 | } |
| 1315 | |
| 1316 | /* |
| 1317 | * PrepareTempTablespaces -- prepare to use temp tablespaces |
| 1318 | * |
| 1319 | * If we have not already done so in the current transaction, parse the |
| 1320 | * temp_tablespaces GUC variable and tell fd.c which tablespace(s) to use |
| 1321 | * for temp files. |
| 1322 | */ |
| 1323 | void |
| 1324 | PrepareTempTablespaces(void) |
| 1325 | { |
| 1326 | char *rawname; |
| 1327 | List *namelist; |
| 1328 | Oid *tblSpcs; |
| 1329 | int numSpcs; |
| 1330 | ListCell *l; |
| 1331 | |
| 1332 | /* No work if already done in current transaction */ |
| 1333 | if (TempTablespacesAreSet()) |
| 1334 | return; |
| 1335 | |
| 1336 | /* |
| 1337 | * Can't do catalog access unless within a transaction. This is just a |
| 1338 | * safety check in case this function is called by low-level code that |
| 1339 | * could conceivably execute outside a transaction. Note that in such a |
| 1340 | * scenario, fd.c will fall back to using the current database's default |
| 1341 | * tablespace, which should always be OK. |
| 1342 | */ |
| 1343 | if (!IsTransactionState()) |
| 1344 | return; |
| 1345 | |
| 1346 | /* Need a modifiable copy of string */ |
| 1347 | rawname = pstrdup(temp_tablespaces); |
| 1348 | |
| 1349 | /* Parse string into list of identifiers */ |
| 1350 | if (!SplitIdentifierString(rawname, ',', &namelist)) |
| 1351 | { |
| 1352 | /* syntax error in name list */ |
| 1353 | SetTempTablespaces(NULL, 0); |
| 1354 | pfree(rawname); |
| 1355 | list_free(namelist); |
| 1356 | return; |
| 1357 | } |
| 1358 | |
| 1359 | /* Store tablespace OIDs in an array in TopTransactionContext */ |
| 1360 | tblSpcs = (Oid *) MemoryContextAlloc(TopTransactionContext, |
| 1361 | list_length(namelist) * sizeof(Oid)); |
| 1362 | numSpcs = 0; |
| 1363 | foreach(l, namelist) |
| 1364 | { |
| 1365 | char *curname = (char *) lfirst(l); |
| 1366 | Oid curoid; |
| 1367 | AclResult aclresult; |
| 1368 | |
| 1369 | /* Allow an empty string (signifying database default) */ |
| 1370 | if (curname[0] == '\0') |
| 1371 | { |
| 1372 | tblSpcs[numSpcs++] = InvalidOid; |
| 1373 | continue; |
| 1374 | } |
| 1375 | |
| 1376 | /* Else verify that name is a valid tablespace name */ |
| 1377 | curoid = get_tablespace_oid(curname, true); |
| 1378 | if (curoid == InvalidOid) |
| 1379 | { |
| 1380 | /* Skip any bad list elements */ |
| 1381 | continue; |
| 1382 | } |
| 1383 | |
| 1384 | /* |
| 1385 | * Allow explicit specification of database's default tablespace in |
| 1386 | * temp_tablespaces without triggering permissions checks. |
| 1387 | */ |
| 1388 | if (curoid == MyDatabaseTableSpace) |
| 1389 | { |
| 1390 | tblSpcs[numSpcs++] = InvalidOid; |
| 1391 | continue; |
| 1392 | } |
| 1393 | |
| 1394 | /* Check permissions similarly */ |
| 1395 | aclresult = pg_tablespace_aclcheck(curoid, GetUserId(), |
| 1396 | ACL_CREATE); |
| 1397 | if (aclresult != ACLCHECK_OK) |
| 1398 | continue; |
| 1399 | |
| 1400 | tblSpcs[numSpcs++] = curoid; |
| 1401 | } |
| 1402 | |
| 1403 | SetTempTablespaces(tblSpcs, numSpcs); |
| 1404 | |
| 1405 | pfree(rawname); |
| 1406 | list_free(namelist); |
| 1407 | } |
| 1408 | |
| 1409 | |
| 1410 | /* |
| 1411 | * get_tablespace_oid - given a tablespace name, look up the OID |
| 1412 | * |
| 1413 | * If missing_ok is false, throw an error if tablespace name not found. If |
| 1414 | * true, just return InvalidOid. |
| 1415 | */ |
| 1416 | Oid |
| 1417 | get_tablespace_oid(const char *tablespacename, bool missing_ok) |
| 1418 | { |
| 1419 | Oid result; |
| 1420 | Relation rel; |
| 1421 | TableScanDesc scandesc; |
| 1422 | HeapTuple tuple; |
| 1423 | ScanKeyData entry[1]; |
| 1424 | |
| 1425 | /* |
| 1426 | * Search pg_tablespace. We use a heapscan here even though there is an |
| 1427 | * index on name, on the theory that pg_tablespace will usually have just |
| 1428 | * a few entries and so an indexed lookup is a waste of effort. |
| 1429 | */ |
| 1430 | rel = table_open(TableSpaceRelationId, AccessShareLock); |
| 1431 | |
| 1432 | ScanKeyInit(&entry[0], |
| 1433 | Anum_pg_tablespace_spcname, |
| 1434 | BTEqualStrategyNumber, F_NAMEEQ, |
| 1435 | CStringGetDatum(tablespacename)); |
| 1436 | scandesc = table_beginscan_catalog(rel, 1, entry); |
| 1437 | tuple = heap_getnext(scandesc, ForwardScanDirection); |
| 1438 | |
| 1439 | /* We assume that there can be at most one matching tuple */ |
| 1440 | if (HeapTupleIsValid(tuple)) |
| 1441 | result = ((Form_pg_tablespace) GETSTRUCT(tuple))->oid; |
| 1442 | else |
| 1443 | result = InvalidOid; |
| 1444 | |
| 1445 | table_endscan(scandesc); |
| 1446 | table_close(rel, AccessShareLock); |
| 1447 | |
| 1448 | if (!OidIsValid(result) && !missing_ok) |
| 1449 | ereport(ERROR, |
| 1450 | (errcode(ERRCODE_UNDEFINED_OBJECT), |
| 1451 | errmsg("tablespace \"%s\" does not exist" , |
| 1452 | tablespacename))); |
| 1453 | |
| 1454 | return result; |
| 1455 | } |
| 1456 | |
| 1457 | /* |
| 1458 | * get_tablespace_name - given a tablespace OID, look up the name |
| 1459 | * |
| 1460 | * Returns a palloc'd string, or NULL if no such tablespace. |
| 1461 | */ |
| 1462 | char * |
| 1463 | get_tablespace_name(Oid spc_oid) |
| 1464 | { |
| 1465 | char *result; |
| 1466 | Relation rel; |
| 1467 | TableScanDesc scandesc; |
| 1468 | HeapTuple tuple; |
| 1469 | ScanKeyData entry[1]; |
| 1470 | |
| 1471 | /* |
| 1472 | * Search pg_tablespace. We use a heapscan here even though there is an |
| 1473 | * index on oid, on the theory that pg_tablespace will usually have just a |
| 1474 | * few entries and so an indexed lookup is a waste of effort. |
| 1475 | */ |
| 1476 | rel = table_open(TableSpaceRelationId, AccessShareLock); |
| 1477 | |
| 1478 | ScanKeyInit(&entry[0], |
| 1479 | Anum_pg_tablespace_oid, |
| 1480 | BTEqualStrategyNumber, F_OIDEQ, |
| 1481 | ObjectIdGetDatum(spc_oid)); |
| 1482 | scandesc = table_beginscan_catalog(rel, 1, entry); |
| 1483 | tuple = heap_getnext(scandesc, ForwardScanDirection); |
| 1484 | |
| 1485 | /* We assume that there can be at most one matching tuple */ |
| 1486 | if (HeapTupleIsValid(tuple)) |
| 1487 | result = pstrdup(NameStr(((Form_pg_tablespace) GETSTRUCT(tuple))->spcname)); |
| 1488 | else |
| 1489 | result = NULL; |
| 1490 | |
| 1491 | table_endscan(scandesc); |
| 1492 | table_close(rel, AccessShareLock); |
| 1493 | |
| 1494 | return result; |
| 1495 | } |
| 1496 | |
| 1497 | |
| 1498 | /* |
| 1499 | * TABLESPACE resource manager's routines |
| 1500 | */ |
| 1501 | void |
| 1502 | tblspc_redo(XLogReaderState *record) |
| 1503 | { |
| 1504 | uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK; |
| 1505 | |
| 1506 | /* Backup blocks are not used in tblspc records */ |
| 1507 | Assert(!XLogRecHasAnyBlockRefs(record)); |
| 1508 | |
| 1509 | if (info == XLOG_TBLSPC_CREATE) |
| 1510 | { |
| 1511 | xl_tblspc_create_rec *xlrec = (xl_tblspc_create_rec *) XLogRecGetData(record); |
| 1512 | char *location = xlrec->ts_path; |
| 1513 | |
| 1514 | create_tablespace_directories(location, xlrec->ts_id); |
| 1515 | } |
| 1516 | else if (info == XLOG_TBLSPC_DROP) |
| 1517 | { |
| 1518 | xl_tblspc_drop_rec *xlrec = (xl_tblspc_drop_rec *) XLogRecGetData(record); |
| 1519 | |
| 1520 | /* |
| 1521 | * If we issued a WAL record for a drop tablespace it implies that |
| 1522 | * there were no files in it at all when the DROP was done. That means |
| 1523 | * that no permanent objects can exist in it at this point. |
| 1524 | * |
| 1525 | * It is possible for standby users to be using this tablespace as a |
| 1526 | * location for their temporary files, so if we fail to remove all |
| 1527 | * files then do conflict processing and try again, if currently |
| 1528 | * enabled. |
| 1529 | * |
| 1530 | * Other possible reasons for failure include bollixed file |
| 1531 | * permissions on a standby server when they were okay on the primary, |
| 1532 | * etc etc. There's not much we can do about that, so just remove what |
| 1533 | * we can and press on. |
| 1534 | */ |
| 1535 | if (!destroy_tablespace_directories(xlrec->ts_id, true)) |
| 1536 | { |
| 1537 | ResolveRecoveryConflictWithTablespace(xlrec->ts_id); |
| 1538 | |
| 1539 | /* |
| 1540 | * If we did recovery processing then hopefully the backends who |
| 1541 | * wrote temp files should have cleaned up and exited by now. So |
| 1542 | * retry before complaining. If we fail again, this is just a LOG |
| 1543 | * condition, because it's not worth throwing an ERROR for (as |
| 1544 | * that would crash the database and require manual intervention |
| 1545 | * before we could get past this WAL record on restart). |
| 1546 | */ |
| 1547 | if (!destroy_tablespace_directories(xlrec->ts_id, true)) |
| 1548 | ereport(LOG, |
| 1549 | (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 1550 | errmsg("directories for tablespace %u could not be removed" , |
| 1551 | xlrec->ts_id), |
| 1552 | errhint("You can remove the directories manually if necessary." ))); |
| 1553 | } |
| 1554 | } |
| 1555 | else |
| 1556 | elog(PANIC, "tblspc_redo: unknown op code %u" , info); |
| 1557 | } |
| 1558 | |