char *
get_relation_name(Oid relid)
{
return quote_qualified_identifier(
get_namespace_name(get_rel_namespace(relid)),
get_rel_name(relid));
}
/*
* Auto-stats employs this sub-routine to issue an analyze on a specific relation.
*/
static void
autostats_issue_analyze(Oid relationOid)
{
VacuumStmt *analyzeStmt = NULL;
RangeVar *relation = NULL;
/*
* If this user does not own the table, then auto-stats will not issue the
* analyze.
*/
if (!(pg_class_ownercheck(relationOid, GetUserId()) ||
(pg_database_ownercheck(MyDatabaseId, GetUserId()) && !IsSharedRelation(relationOid))))
{
elog(DEBUG3, "Auto-stats did not issue ANALYZE on tableoid %d since the user does not have table-owner level permissions.",
relationOid);
return;
}
relation = makeRangeVar(get_namespace_name(get_rel_namespace(relationOid)), get_rel_name(relationOid), -1);
analyzeStmt = makeNode(VacuumStmt);
/* Set up command parameters */
analyzeStmt->vacuum = false;
analyzeStmt->full = false;
analyzeStmt->analyze = true;
analyzeStmt->freeze_min_age = -1;
analyzeStmt->verbose = false;
analyzeStmt->rootonly = false;
analyzeStmt->relation = relation; /* not used since we pass relids list */
analyzeStmt->va_cols = NIL;
vacuum(analyzeStmt, NIL, NULL, false, false);
pfree(analyzeStmt);
}
/*
* LockTagIsTemp
* Determine whether a locktag is for a lock on a temporary object
*
* We need this because 2PC cannot deal with temp objects
*/
bool
LockTagIsTemp(const LOCKTAG *tag)
{
switch (tag->locktag_type)
{
case LOCKTAG_RELATION:
case LOCKTAG_RELATION_EXTEND:
case LOCKTAG_PAGE:
case LOCKTAG_TUPLE:
case LOCKTAG_RELATION_RESYNCHRONIZE:
case LOCKTAG_RELATION_APPENDONLY_SEGMENT_FILE:
/* check for lock on a temp relation */
/* field1 is dboid, field2 is reloid for all of these */
if ((Oid) tag->locktag_field1 == InvalidOid)
return false; /* shared, so not temp */
if (isTempNamespace(get_rel_namespace((Oid) tag->locktag_field2)))
return true;
break;
case LOCKTAG_TRANSACTION:
/* there are no temp transactions */
break;
case LOCKTAG_OBJECT:
/* there are currently no non-table temp objects */
break;
case LOCKTAG_USERLOCK:
case LOCKTAG_ADVISORY:
/* assume these aren't temp */
break;
}
return false; /* default case */
}
/*
* GetTableCreationCommands takes in a relationId, and returns the list of DDL
* commands needed to reconstruct the relation, excluding indexes and
* constraints.
*/
List *
GetTableCreationCommands(Oid relationId, bool includeSequenceDefaults)
{
List *tableDDLEventList = NIL;
char tableType = 0;
char *tableSchemaDef = NULL;
char *tableColumnOptionsDef = NULL;
char *createSchemaCommand = NULL;
Oid schemaId = InvalidOid;
/*
* Set search_path to NIL so that all objects outside of pg_catalog will be
* schema-prefixed. pg_catalog will be added automatically when we call
* PushOverrideSearchPath(), since we set addCatalog to true;
*/
OverrideSearchPath *overridePath = GetOverrideSearchPath(CurrentMemoryContext);
overridePath->schemas = NIL;
overridePath->addCatalog = true;
PushOverrideSearchPath(overridePath);
/* if foreign table, fetch extension and server definitions */
tableType = get_rel_relkind(relationId);
if (tableType == RELKIND_FOREIGN_TABLE)
{
char *extensionDef = pg_get_extensiondef_string(relationId);
char *serverDef = pg_get_serverdef_string(relationId);
if (extensionDef != NULL)
{
tableDDLEventList = lappend(tableDDLEventList, extensionDef);
}
tableDDLEventList = lappend(tableDDLEventList, serverDef);
}
/* create schema if the table is not in the default namespace (public) */
schemaId = get_rel_namespace(relationId);
createSchemaCommand = CreateSchemaDDLCommand(schemaId);
if (createSchemaCommand != NULL)
{
tableDDLEventList = lappend(tableDDLEventList, createSchemaCommand);
}
/* fetch table schema and column option definitions */
tableSchemaDef = pg_get_tableschemadef_string(relationId, includeSequenceDefaults);
tableColumnOptionsDef = pg_get_tablecolumnoptionsdef_string(relationId);
tableDDLEventList = lappend(tableDDLEventList, tableSchemaDef);
if (tableColumnOptionsDef != NULL)
{
tableDDLEventList = lappend(tableDDLEventList, tableColumnOptionsDef);
}
/* revert back to original search_path */
PopOverrideSearchPath();
return tableDDLEventList;
}
Datum
pg_relation_size(PG_FUNCTION_ARGS)
{
Oid relOid = PG_GETARG_OID(0);
text *forkName = PG_GETARG_TEXT_P(1);
Relation rel;
int64 size = 0;
/**
* This function is peculiar in that it does its own dispatching.
* It does not work on entry db since we do not support dispatching
* from entry-db currently.
*/
if (Gp_role == GP_ROLE_EXECUTE && Gp_segment == -1)
elog(ERROR, "This query is not currently supported by GPDB.");
rel = try_relation_open(relOid, AccessShareLock, false);
/*
* While we scan pg_class with an MVCC snapshot,
* someone else might drop the table. It's better to return NULL for
* already-dropped tables than throw an error and abort the whole query.
*/
if (!RelationIsValid(rel))
PG_RETURN_NULL();
if (relOid == 0 || rel->rd_node.relNode == 0)
size = 0;
else
size = calculate_relation_size(rel,
forkname_to_number(text_to_cstring(forkName)));
if (Gp_role == GP_ROLE_DISPATCH)
{
StringInfoData buffer;
char *schemaName;
char *relName;
schemaName = get_namespace_name(get_rel_namespace(relOid));
if (schemaName == NULL)
elog(ERROR, "Cannot find schema for oid %d", relOid);
relName = get_rel_name(relOid);
if (relName == NULL)
elog(ERROR, "Cannot find relation for oid %d", relOid);
initStringInfo(&buffer);
appendStringInfo(&buffer, "select sum(pg_relation_size('%s.%s'))::int8 from gp_dist_random('gp_id');", quote_identifier(schemaName), quote_identifier(relName));
size += get_size_from_segDBs(buffer.data);
}
relation_close(rel, AccessShareLock);
PG_RETURN_INT64(size);
}
/*
* Print a relation name into the StringInfo provided by caller.
*/
static void
print_relname(StringInfo s, Relation rel)
{
Form_pg_class class_form = RelationGetForm(rel);
appendStringInfoString(s,
quote_qualified_identifier(
get_namespace_name(
get_rel_namespace(RelationGetRelid(rel))),
NameStr(class_form->relname)));
}
/*
* GenerateAttachShardPartitionCommand generates command to attach a child table
* table to its parent in a partitioning hierarchy.
*/
char *
GenerateAttachShardPartitionCommand(ShardInterval *shardInterval)
{
Oid schemaId = get_rel_namespace(shardInterval->relationId);
char *schemaName = get_namespace_name(schemaId);
char *escapedSchemaName = quote_literal_cstr(schemaName);
char *command = GenerateAlterTableAttachPartitionCommand(shardInterval->relationId);
char *escapedCommand = quote_literal_cstr(command);
int shardIndex = ShardIndex(shardInterval);
Oid parentSchemaId = InvalidOid;
char *parentSchemaName = NULL;
char *escapedParentSchemaName = NULL;
uint64 parentShardId = INVALID_SHARD_ID;
StringInfo attachPartitionCommand = makeStringInfo();
Oid parentRelationId = PartitionParentOid(shardInterval->relationId);
if (parentRelationId == InvalidOid)
{
ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("cannot attach partition"),
errdetail("Referenced relation cannot be found.")));
}
parentSchemaId = get_rel_namespace(parentRelationId);
parentSchemaName = get_namespace_name(parentSchemaId);
escapedParentSchemaName = quote_literal_cstr(parentSchemaName);
parentShardId = ColocatedShardIdInRelation(parentRelationId, shardIndex);
appendStringInfo(attachPartitionCommand,
WORKER_APPLY_INTER_SHARD_DDL_COMMAND, parentShardId,
escapedParentSchemaName, shardInterval->shardId,
escapedSchemaName, escapedCommand);
return attachPartitionCommand->data;
}
/* Gpmon helpers. */
char * GetScanRelNameGpmon(Oid relid, char schema_rel_name[SCAN_REL_NAME_BUF_SIZE])
{
if (relid > 0)
{
char *relname = get_rel_name(relid);
char *schemaname = get_namespace_name(get_rel_namespace(relid));
snprintf(schema_rel_name, SCAN_REL_NAME_BUF_SIZE, "%s.%s", schemaname, relname);
if (relname)
pfree(relname);
if (schemaname)
pfree(schemaname);
}
return schema_rel_name;
}
static char *
get_relation_name(Oid relid)
{
Oid nsp = get_rel_namespace(relid);
char *nspname;
/* Qualify the name if not visible in search path */
if (RelationIsVisible(relid))
nspname = NULL;
else
nspname = get_namespace_name(nsp);
return quote_qualified_identifier(nspname, get_rel_name(relid));
}
static char *
get_relation_name(Oid relid)
{
Oid nsp = get_rel_namespace(relid);
char *nspname;
char *strver;
int ver;
/* Get the version of the running server (PG_VERSION_NUM would return
* the version we compiled the extension with) */
strver = GetConfigOptionByName("server_version_num", NULL
#if PG_VERSION_NUM >= 90600
, false /* missing_ok */
#endif
);
ver = atoi(strver);
pfree(strver);
/*
* Relation names given by PostgreSQL core are always
* qualified since some minor releases. Note that this change
* wasn't introduced in PostgreSQL 9.2 and 9.1 releases.
*/
if ((ver >= 100000 && ver < 100003) ||
(ver >= 90600 && ver < 90608) ||
(ver >= 90500 && ver < 90512) ||
(ver >= 90400 && ver < 90417) ||
(ver >= 90300 && ver < 90322) ||
(ver >= 90200 && ver < 90300) ||
(ver >= 90100 && ver < 90200))
{
/* Qualify the name if not visible in search path */
if (RelationIsVisible(relid))
nspname = NULL;
else
nspname = get_namespace_name(nsp);
}
else
{
/* Always qualify the name */
if (OidIsValid(nsp))
nspname = get_namespace_name(nsp);
else
nspname = NULL;
}
return quote_qualified_identifier(nspname, get_rel_name(relid));
}
/*
* CheckTableSchemaNameForDrop errors out if the current user does not
* have permission to undistribute the given relation, taking into
* account that it may be called from the drop trigger. If the table exists,
* the function rewrites the given table and schema name.
*/
void
CheckTableSchemaNameForDrop(Oid relationId, char **schemaName, char **tableName)
{
char *tempTableName = get_rel_name(relationId);
if (tempTableName != NULL)
{
/* ensure proper values are used if the table exists */
Oid schemaId = get_rel_namespace(relationId);
(*schemaName) = get_namespace_name(schemaId);
(*tableName) = tempTableName;
EnsureTableOwner(relationId);
}
}
/**
* Method determines if a relation is master-only or distributed among segments.
* Input:
* relationOid
* Output:
* true if masteronly
*/
bool
isMasterOnly(Oid relationOid)
{
Assert(relationOid != InvalidOid);
Oid schemaOid = get_rel_namespace(relationOid);
GpPolicy *distributionPolicy = GpPolicyFetch(CurrentMemoryContext, relationOid);
bool masterOnly = (Gp_role == GP_ROLE_UTILITY
|| IsSystemNamespace(schemaOid)
|| IsToastNamespace(schemaOid)
|| IsAoSegmentNamespace(schemaOid)
|| (distributionPolicy == NULL)
|| (distributionPolicy->ptype == POLICYTYPE_ENTRY));
return masterOnly;
}
/*
* truncate relation
*/
void
TruncateTable(Oid relid)
{
TruncateStmt stmt;
RangeVar *heap;
Assert(OidIsValid(relid));
heap = makeRangeVar(get_namespace_name(get_rel_namespace(relid)),
get_rel_name(relid), -1);
memset(&stmt, 0, sizeof(stmt));
stmt.type = T_TruncateStmt;
stmt.relations = list_make1(heap);
stmt.behavior = DROP_RESTRICT;
ExecuteTruncate(&stmt);
CommandCounterIncrement();
}
/*
* master_apply_delete_command takes in a delete command, finds shards that
* match the criteria defined in the delete command, drops the found shards from
* the worker nodes, and updates the corresponding metadata on the master node.
* This function drops a shard if and only if all rows in the shard satisfy
* the conditions in the delete command. Note that this function only accepts
* conditions on the partition key and if no condition is provided then all
* shards are deleted.
*
* We mark shard placements that we couldn't drop as to be deleted later. If a
* shard satisfies the given conditions, we delete it from shard metadata table
* even though related shard placements are not deleted.
*/
Datum
master_apply_delete_command(PG_FUNCTION_ARGS)
{
text *queryText = PG_GETARG_TEXT_P(0);
char *queryString = text_to_cstring(queryText);
char *relationName = NULL;
char *schemaName = NULL;
Oid relationId = InvalidOid;
List *shardIntervalList = NIL;
List *deletableShardIntervalList = NIL;
List *queryTreeList = NIL;
Query *deleteQuery = NULL;
Node *whereClause = NULL;
Node *deleteCriteria = NULL;
Node *queryTreeNode = NULL;
DeleteStmt *deleteStatement = NULL;
int droppedShardCount = 0;
LOCKMODE lockMode = 0;
char partitionMethod = 0;
bool failOK = false;
#if (PG_VERSION_NUM >= 100000)
RawStmt *rawStmt = (RawStmt *) ParseTreeRawStmt(queryString);
queryTreeNode = rawStmt->stmt;
#else
queryTreeNode = ParseTreeNode(queryString);
#endif
EnsureCoordinator();
CheckCitusVersion(ERROR);
if (!IsA(queryTreeNode, DeleteStmt))
{
ereport(ERROR, (errmsg("query \"%s\" is not a delete statement",
queryString)));
}
deleteStatement = (DeleteStmt *) queryTreeNode;
schemaName = deleteStatement->relation->schemaname;
relationName = deleteStatement->relation->relname;
/*
* We take an exclusive lock while dropping shards to prevent concurrent
* writes. We don't want to block SELECTs, which means queries might fail
* if they access a shard that has just been dropped.
*/
lockMode = ExclusiveLock;
relationId = RangeVarGetRelid(deleteStatement->relation, lockMode, failOK);
/* schema-prefix if it is not specified already */
if (schemaName == NULL)
{
Oid schemaId = get_rel_namespace(relationId);
schemaName = get_namespace_name(schemaId);
}
CheckDistributedTable(relationId);
EnsureTablePermissions(relationId, ACL_DELETE);
#if (PG_VERSION_NUM >= 100000)
queryTreeList = pg_analyze_and_rewrite(rawStmt, queryString, NULL, 0, NULL);
#else
queryTreeList = pg_analyze_and_rewrite(queryTreeNode, queryString, NULL, 0);
#endif
deleteQuery = (Query *) linitial(queryTreeList);
CheckTableCount(deleteQuery);
/* get where clause and flatten it */
whereClause = (Node *) deleteQuery->jointree->quals;
deleteCriteria = eval_const_expressions(NULL, whereClause);
partitionMethod = PartitionMethod(relationId);
if (partitionMethod == DISTRIBUTE_BY_HASH)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot delete from hash distributed table with this "
"command"),
errdetail("Delete statements on hash-partitioned tables "
"are not supported with master_apply_delete_command."),
errhint("Use master_modify_multiple_shards command instead.")));
}
else if (partitionMethod == DISTRIBUTE_BY_NONE)
{
ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot delete from distributed table"),
errdetail("Delete statements on reference tables "
"are not supported.")));
}
CheckDeleteCriteria(deleteCriteria);
CheckPartitionColumn(relationId, deleteCriteria);
shardIntervalList = LoadShardIntervalList(relationId);
/* drop all shards if where clause is not present */
if (deleteCriteria == NULL)
{
deletableShardIntervalList = shardIntervalList;
ereport(DEBUG2, (errmsg("dropping all shards for \"%s\"", relationName)));
}
else
{
deletableShardIntervalList = ShardsMatchingDeleteCriteria(relationId,
shardIntervalList,
deleteCriteria);
}
droppedShardCount = DropShards(relationId, schemaName, relationName,
deletableShardIntervalList);
PG_RETURN_INT32(droppedShardCount);
}
/*
* exec_object_restorecon
*
* This routine is a helper called by sepgsql_restorecon; it set up
* initial security labels of database objects within the supplied
* catalog OID.
*/
static void
exec_object_restorecon(struct selabel_handle *sehnd, Oid catalogId)
{
Relation rel;
SysScanDesc sscan;
HeapTuple tuple;
char *database_name = get_database_name(MyDatabaseId);
char *namespace_name;
Oid namespace_id;
char *relation_name;
/*
* Open the target catalog. We don't want to allow writable
* accesses by other session during initial labeling.
*/
rel = heap_open(catalogId, AccessShareLock);
sscan = systable_beginscan(rel, InvalidOid, false,
SnapshotNow, 0, NULL);
while (HeapTupleIsValid(tuple = systable_getnext(sscan)))
{
Form_pg_namespace nspForm;
Form_pg_class relForm;
Form_pg_attribute attForm;
Form_pg_proc proForm;
char *objname;
int objtype = 1234;
ObjectAddress object;
security_context_t context;
/*
* The way to determine object name depends on object classes.
* So, any branches set up `objtype', `objname' and `object' here.
*/
switch (catalogId)
{
case NamespaceRelationId:
nspForm = (Form_pg_namespace) GETSTRUCT(tuple);
objtype = SELABEL_DB_SCHEMA;
objname = quote_object_name(database_name,
NameStr(nspForm->nspname),
NULL, NULL);
object.classId = NamespaceRelationId;
object.objectId = HeapTupleGetOid(tuple);
object.objectSubId = 0;
break;
case RelationRelationId:
relForm = (Form_pg_class) GETSTRUCT(tuple);
if (relForm->relkind == RELKIND_RELATION)
objtype = SELABEL_DB_TABLE;
else if (relForm->relkind == RELKIND_SEQUENCE)
objtype = SELABEL_DB_SEQUENCE;
else if (relForm->relkind == RELKIND_VIEW)
objtype = SELABEL_DB_VIEW;
else
continue; /* no need to assign security label */
namespace_name = get_namespace_name(relForm->relnamespace);
objname = quote_object_name(database_name,
namespace_name,
NameStr(relForm->relname),
NULL);
pfree(namespace_name);
object.classId = RelationRelationId;
object.objectId = HeapTupleGetOid(tuple);
object.objectSubId = 0;
break;
case AttributeRelationId:
attForm = (Form_pg_attribute) GETSTRUCT(tuple);
if (get_rel_relkind(attForm->attrelid) != RELKIND_RELATION)
continue; /* no need to assign security label */
objtype = SELABEL_DB_COLUMN;
namespace_id = get_rel_namespace(attForm->attrelid);
namespace_name = get_namespace_name(namespace_id);
relation_name = get_rel_name(attForm->attrelid);
objname = quote_object_name(database_name,
namespace_name,
relation_name,
NameStr(attForm->attname));
pfree(namespace_name);
pfree(relation_name);
object.classId = RelationRelationId;
object.objectId = attForm->attrelid;
object.objectSubId = attForm->attnum;
break;
case ProcedureRelationId:
proForm = (Form_pg_proc) GETSTRUCT(tuple);
objtype = SELABEL_DB_PROCEDURE;
namespace_name = get_namespace_name(proForm->pronamespace);
objname = quote_object_name(database_name,
namespace_name,
NameStr(proForm->proname),
NULL);
pfree(namespace_name);
object.classId = ProcedureRelationId;
object.objectId = HeapTupleGetOid(tuple);
object.objectSubId = 0;
break;
default:
elog(ERROR, "unexpected catalog id: %u", catalogId);
objname = NULL; /* for compiler quiet */
break;
}
if (selabel_lookup_raw(sehnd, &context, objname, objtype) == 0)
{
PG_TRY();
{
/*
* Check SELinux permission to relabel the fetched object,
* then do the actual relabeling.
*/
sepgsql_object_relabel(&object, context);
SetSecurityLabel(&object, SEPGSQL_LABEL_TAG, context);
}
PG_CATCH();
{
freecon(context);
PG_RE_THROW();
}
PG_END_TRY();
freecon(context);
}
else if (errno == ENOENT)
ereport(WARNING,
(errmsg("SELinux: no initial label assigned for %s (type=%d), skipping",
objname, objtype)));
else
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("SELinux: could not determine initial security label for %s (type=%d): %m", objname, objtype)));
pfree(objname);
}
systable_endscan(sscan);
heap_close(rel, NoLock);
}
/*
* sepgsql_relation_drop
*
* It checks privileges to drop the supplied relation.
*/
void
sepgsql_relation_drop(Oid relOid)
{
ObjectAddress object;
char *audit_name;
uint16_t tclass;
char relkind;
relkind = get_rel_relkind(relOid);
switch (relkind)
{
case RELKIND_RELATION:
tclass = SEPG_CLASS_DB_TABLE;
break;
case RELKIND_SEQUENCE:
tclass = SEPG_CLASS_DB_SEQUENCE;
break;
case RELKIND_VIEW:
tclass = SEPG_CLASS_DB_VIEW;
break;
case RELKIND_INDEX:
/* ignore indexes on toast tables */
if (get_rel_namespace(relOid) == PG_TOAST_NAMESPACE)
return;
/* other indexes are handled specially below; no need for tclass */
break;
default:
/* ignore other relkinds */
return;
}
/*
* check db_schema:{remove_name} permission
*/
object.classId = NamespaceRelationId;
object.objectId = get_rel_namespace(relOid);
object.objectSubId = 0;
audit_name = getObjectDescription(&object);
sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_SCHEMA,
SEPG_DB_SCHEMA__REMOVE_NAME,
audit_name,
true);
pfree(audit_name);
/* deal with indexes specially */
if (relkind == RELKIND_INDEX)
{
sepgsql_index_modify(relOid);
return;
}
/*
* check db_table/sequence/view:{drop} permission
*/
object.classId = RelationRelationId;
object.objectId = relOid;
object.objectSubId = 0;
audit_name = getObjectDescription(&object);
sepgsql_avc_check_perms(&object,
tclass,
SEPG_DB_TABLE__DROP,
audit_name,
true);
pfree(audit_name);
/*
* check db_column:{drop} permission
*/
if (relkind == RELKIND_RELATION)
{
Form_pg_attribute attForm;
CatCList *attrList;
HeapTuple atttup;
int i;
attrList = SearchSysCacheList1(ATTNUM, ObjectIdGetDatum(relOid));
for (i = 0; i < attrList->n_members; i++)
{
atttup = &attrList->members[i]->tuple;
attForm = (Form_pg_attribute) GETSTRUCT(atttup);
if (attForm->attisdropped)
continue;
object.classId = RelationRelationId;
object.objectId = relOid;
object.objectSubId = attForm->attnum;
audit_name = getObjectDescription(&object);
sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_COLUMN,
SEPG_DB_COLUMN__DROP,
audit_name,
true);
pfree(audit_name);
}
ReleaseCatCacheList(attrList);
}
}
/*
* sepgsql_relation_drop
*
* It checks privileges to drop the supplied relation.
*/
void
sepgsql_relation_drop(Oid relOid)
{
ObjectAddress object;
char *audit_name;
uint16_t tclass = 0;
char relkind;
relkind = get_rel_relkind(relOid);
if (relkind == RELKIND_RELATION)
tclass = SEPG_CLASS_DB_TABLE;
else if (relkind == RELKIND_SEQUENCE)
tclass = SEPG_CLASS_DB_SEQUENCE;
else if (relkind == RELKIND_VIEW)
tclass = SEPG_CLASS_DB_VIEW;
else
return;
/*
* check db_schema:{remove_name} permission
*/
object.classId = NamespaceRelationId;
object.objectId = get_rel_namespace(relOid);
object.objectSubId = 0;
audit_name = getObjectDescription(&object);
sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_SCHEMA,
SEPG_DB_SCHEMA__REMOVE_NAME,
audit_name,
true);
pfree(audit_name);
/*
* check db_table/sequence/view:{drop} permission
*/
object.classId = RelationRelationId;
object.objectId = relOid;
object.objectSubId = 0;
audit_name = getObjectDescription(&object);
sepgsql_avc_check_perms(&object,
tclass,
SEPG_DB_TABLE__DROP,
audit_name,
true);
pfree(audit_name);
/*
* check db_column:{drop} permission
*/
if (relkind == RELKIND_RELATION)
{
Form_pg_attribute attForm;
CatCList *attrList;
HeapTuple atttup;
int i;
attrList = SearchSysCacheList1(ATTNUM, ObjectIdGetDatum(relOid));
for (i = 0; i < attrList->n_members; i++)
{
atttup = &attrList->members[i]->tuple;
attForm = (Form_pg_attribute) GETSTRUCT(atttup);
if (attForm->attisdropped)
continue;
object.classId = RelationRelationId;
object.objectId = relOid;
object.objectSubId = attForm->attnum;
audit_name = getObjectDescription(&object);
sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_COLUMN,
SEPG_DB_COLUMN__DROP,
audit_name,
true);
pfree(audit_name);
}
ReleaseCatCacheList(attrList);
}
}
static const char *
get_quoted_nspname(Oid oid)
{
return quote_identifier(get_namespace_name(get_rel_namespace(oid)));
}
/*
* GetTableDDLEvents takes in a relationId, and returns the list of DDL commands
* needed to reconstruct the relation. These DDL commands are all palloced; and
* include the table's schema definition, optional column storage and statistics
* definitions, and index and constraint defitions.
*/
List *
GetTableDDLEvents(Oid relationId)
{
List *tableDDLEventList = NIL;
char tableType = 0;
List *sequenceIdlist = getOwnedSequences(relationId);
ListCell *sequenceIdCell;
char *tableSchemaDef = NULL;
char *tableColumnOptionsDef = NULL;
char *schemaName = NULL;
Oid schemaId = InvalidOid;
Relation pgIndex = NULL;
SysScanDesc scanDescriptor = NULL;
ScanKeyData scanKey[1];
int scanKeyCount = 1;
HeapTuple heapTuple = NULL;
/*
* Set search_path to NIL so that all objects outside of pg_catalog will be
* schema-prefixed. pg_catalog will be added automatically when we call
* PushOverrideSearchPath(), since we set addCatalog to true;
*/
OverrideSearchPath *overridePath = GetOverrideSearchPath(CurrentMemoryContext);
overridePath->schemas = NIL;
overridePath->addCatalog = true;
PushOverrideSearchPath(overridePath);
/* if foreign table, fetch extension and server definitions */
tableType = get_rel_relkind(relationId);
if (tableType == RELKIND_FOREIGN_TABLE)
{
char *extensionDef = pg_get_extensiondef_string(relationId);
char *serverDef = pg_get_serverdef_string(relationId);
if (extensionDef != NULL)
{
tableDDLEventList = lappend(tableDDLEventList, extensionDef);
}
tableDDLEventList = lappend(tableDDLEventList, serverDef);
}
/* create schema if the table is not in the default namespace (public) */
schemaId = get_rel_namespace(relationId);
schemaName = get_namespace_name(schemaId);
if (strncmp(schemaName, "public", NAMEDATALEN) != 0)
{
StringInfo schemaNameDef = makeStringInfo();
appendStringInfo(schemaNameDef, CREATE_SCHEMA_COMMAND, schemaName);
tableDDLEventList = lappend(tableDDLEventList, schemaNameDef->data);
}
/* create sequences if needed */
foreach(sequenceIdCell, sequenceIdlist)
{
Oid sequenceRelid = lfirst_oid(sequenceIdCell);
char *sequenceDef = pg_get_sequencedef_string(sequenceRelid);
tableDDLEventList = lappend(tableDDLEventList, sequenceDef);
}
/*
* check_relation_privileges
*
* It actually checks required permissions on a certain relation
* and its columns.
*/
static bool
check_relation_privileges(Oid relOid,
Bitmapset *selected,
Bitmapset *inserted,
Bitmapset *updated,
uint32 required,
bool abort_on_violation)
{
ObjectAddress object;
char *audit_name;
Bitmapset *columns;
int index;
char relkind = get_rel_relkind(relOid);
bool result = true;
/*
* Hardwired Policies: SE-PostgreSQL enforces - clients cannot modify
* system catalogs using DMLs - clients cannot reference/modify toast
* relations using DMLs
*/
if (sepgsql_getenforce() > 0)
{
Oid relnamespace = get_rel_namespace(relOid);
if (IsSystemNamespace(relnamespace) &&
(required & (SEPG_DB_TABLE__UPDATE |
SEPG_DB_TABLE__INSERT |
SEPG_DB_TABLE__DELETE)) != 0)
ereport(ERROR,
(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
errmsg("SELinux: hardwired security policy violation")));
if (relkind == RELKIND_TOASTVALUE)
ereport(ERROR,
(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
errmsg("SELinux: hardwired security policy violation")));
}
/*
* Check permissions on the relation
*/
object.classId = RelationRelationId;
object.objectId = relOid;
object.objectSubId = 0;
audit_name = getObjectIdentity(&object);
switch (relkind)
{
case RELKIND_RELATION:
case RELKIND_PARTITIONED_TABLE:
result = sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_TABLE,
required,
audit_name,
abort_on_violation);
break;
case RELKIND_SEQUENCE:
Assert((required & ~SEPG_DB_TABLE__SELECT) == 0);
if (required & SEPG_DB_TABLE__SELECT)
result = sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_SEQUENCE,
SEPG_DB_SEQUENCE__GET_VALUE,
audit_name,
abort_on_violation);
break;
case RELKIND_VIEW:
result = sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_VIEW,
SEPG_DB_VIEW__EXPAND,
audit_name,
abort_on_violation);
break;
default:
/* nothing to be checked */
break;
}
pfree(audit_name);
/*
* Only columns owned by relations shall be checked
*/
if (relkind != RELKIND_RELATION && relkind != RELKIND_PARTITIONED_TABLE)
return true;
/*
* Check permissions on the columns
*/
selected = fixup_whole_row_references(relOid, selected);
inserted = fixup_whole_row_references(relOid, inserted);
updated = fixup_whole_row_references(relOid, updated);
columns = bms_union(selected, bms_union(inserted, updated));
while ((index = bms_first_member(columns)) >= 0)
{
AttrNumber attnum;
uint32 column_perms = 0;
if (bms_is_member(index, selected))
column_perms |= SEPG_DB_COLUMN__SELECT;
if (bms_is_member(index, inserted))
{
if (required & SEPG_DB_TABLE__INSERT)
column_perms |= SEPG_DB_COLUMN__INSERT;
}
if (bms_is_member(index, updated))
{
if (required & SEPG_DB_TABLE__UPDATE)
column_perms |= SEPG_DB_COLUMN__UPDATE;
}
if (column_perms == 0)
continue;
/* obtain column's permission */
attnum = index + FirstLowInvalidHeapAttributeNumber;
object.classId = RelationRelationId;
object.objectId = relOid;
object.objectSubId = attnum;
audit_name = getObjectDescription(&object);
result = sepgsql_avc_check_perms(&object,
SEPG_CLASS_DB_COLUMN,
column_perms,
audit_name,
abort_on_violation);
pfree(audit_name);
if (!result)
return result;
}
return true;
}
/*
* Creates a sample table with data from a PXF table.
* We need to create a copy of the PXF table, in order to pass the sampling
* parameters pxf_sample_ratio and pxf_max_fragments as attributes,
* and to create a segment reject limit of 25 percent.
*
* The new PXF table is sampled and the results are saved in the returned sample table.
* Note that ANALYZE can be executed only by the database owner.
* It is safe to assume that the database owner has permissions to create temp tables.
* The sampling is done by uniformly sampling pxf_sample_ratio records of each fragments,
* up to pxf_max_fragments.
*
* Input:
* relationOid - relation to be sampled
* sampleTableName - sample table name, moderately unique
* lAttributeNames - attributes to be included in the sample
* relTuples - estimated size of relation
* relFrags - estimated number of fragments in relation
* requestedSampleSize - as determined by attribute statistics requirements.
* sampleTableRelTuples - limit on size of the sample.
* Output:
* sampleTableRelTuples - number of tuples in the sample table created.
*/
Oid buildPxfSampleTable(Oid relationOid,
char* sampleTableName,
List *lAttributeNames,
float4 relTuples,
float4 relFrags,
float4 requestedSampleSize,
float4 *sampleTableRelTuples)
{
const char *schemaName = get_namespace_name(get_rel_namespace(relationOid)); /* must be pfreed */
const char *tableName = get_rel_name(relationOid); /* must be pfreed */
char *sampleSchemaName = pstrdup("pg_temp");
char *pxfSampleTable = temporarySampleTableName(relationOid, "pg_analyze_pxf"); /* must be pfreed */
Oid sampleTableOid = InvalidOid;
Oid pxfSampleTableOid = InvalidOid;
RangeVar *rangeVar = NULL;
float4 pxfSamplingRatio = 0.0;
Assert(requestedSampleSize > 0.0);
Assert(relTuples > 0.0);
Assert(relFrags > 0.0);
/* calculate pxf_sample_ratio */
pxfSamplingRatio = calculateSamplingRatio(relTuples, relFrags, requestedSampleSize);
/* build copy of original pxf table */
buildPxfTableCopy(relationOid,
pxfSamplingRatio,
pxf_stat_max_fragments,
schemaName, tableName,
sampleSchemaName, pxfSampleTable);
rangeVar = makeRangeVar(NULL /*catalogname*/, sampleSchemaName, pxfSampleTable, -1);
pxfSampleTableOid = RangeVarGetRelid(rangeVar, true /* failOK */, false /*allowHcatalog*/);
buildSampleFromPxf(sampleSchemaName, sampleTableName, pxfSampleTable,
lAttributeNames, sampleTableRelTuples);
rangeVar = makeRangeVar(NULL /*catalogname*/, sampleSchemaName, sampleTableName, -1);
sampleTableOid = RangeVarGetRelid(rangeVar, true /* failOK */, false /*allowHcatalog*/);
Assert(sampleTableOid != InvalidOid);
/**
* MPP-10723: Very rarely, we may be unlucky and generate an empty sample table. We error out in this case rather than
* generate bad statistics.
*/
if (*sampleTableRelTuples < 1.0)
{
elog(ERROR, "ANALYZE unable to generate accurate statistics on table %s.%s. Try lowering gp_analyze_relative_error",
quote_identifier(schemaName),
quote_identifier(tableName));
}
if (pxfSampleTableOid != InvalidOid)
{
elog(DEBUG2, "ANALYZE dropping PXF sample table");
dropSampleTable(pxfSampleTableOid, true);
}
pfree((void *) rangeVar);
pfree((void *) pxfSampleTable);
pfree((void *) tableName);
pfree((void *) schemaName);
pfree((void *) sampleSchemaName);
return sampleTableOid;
}
/*
* Executes an ALTER OBJECT / RENAME TO statement. Based on the object
* type, the function appropriate to that type is executed.
*/
void
ExecRenameStmt(RenameStmt *stmt)
{
switch (stmt->renameType)
{
case OBJECT_AGGREGATE:
RenameAggregate(stmt->object, stmt->objarg, stmt->newname);
break;
case OBJECT_CONVERSION:
RenameConversion(stmt->object, stmt->newname);
break;
case OBJECT_DATABASE:
RenameDatabase(stmt->subname, stmt->newname);
break;
case OBJECT_EXTPROTOCOL:
RenameExtProtocol(stmt->subname, stmt->newname);
break;
case OBJECT_FUNCTION:
RenameFunction(stmt->object, stmt->objarg, stmt->newname);
break;
case OBJECT_LANGUAGE:
RenameLanguage(stmt->subname, stmt->newname);
break;
case OBJECT_OPCLASS:
RenameOpClass(stmt->object, stmt->subname, stmt->newname);
break;
case OBJECT_OPFAMILY:
RenameOpFamily(stmt->object, stmt->subname, stmt->newname);
break;
case OBJECT_ROLE:
RenameRole(stmt->subname, stmt->newname);
break;
case OBJECT_SCHEMA:
RenameSchema(stmt->subname, stmt->newname);
break;
case OBJECT_TABLESPACE:
RenameTableSpace(stmt->subname, stmt->newname);
break;
case OBJECT_FILESPACE:
RenameFileSpace(stmt->subname, stmt->newname);
break;
case OBJECT_TABLE:
case OBJECT_SEQUENCE:
case OBJECT_VIEW:
case OBJECT_INDEX:
{
if (Gp_role == GP_ROLE_DISPATCH)
{
CheckRelationOwnership(stmt->relation, true);
stmt->objid = RangeVarGetRelid(stmt->relation, false);
}
/*
* RENAME TABLE requires that we (still) hold
* CREATE rights on the containing namespace, as
* well as ownership of the table.
*/
Oid namespaceId = get_rel_namespace(stmt->objid);
AclResult aclresult;
aclresult = pg_namespace_aclcheck(namespaceId,
GetUserId(),
ACL_CREATE);
if (aclresult != ACLCHECK_OK)
aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
get_namespace_name(namespaceId));
renamerel(stmt->objid, stmt->newname, stmt->renameType, stmt);
break;
}
case OBJECT_COLUMN:
case OBJECT_TRIGGER:
{
Oid relid;
CheckRelationOwnership(stmt->relation, true);
relid = RangeVarGetRelid(stmt->relation, false);
switch (stmt->renameType)
{
case OBJECT_COLUMN:
renameatt(relid,
stmt->subname, /* old att name */
stmt->newname, /* new att name */
interpretInhOption(stmt->relation->inhOpt), /* recursive? */
false); /* recursing already? */
break;
case OBJECT_TRIGGER:
renametrig(relid,
stmt->subname, /* old att name */
stmt->newname); /* new att name */
break;
default:
/* can't happen */ ;
}
break;
}
case OBJECT_TSPARSER:
RenameTSParser(stmt->object, stmt->newname);
break;
case OBJECT_TSDICTIONARY:
RenameTSDictionary(stmt->object, stmt->newname);
break;
case OBJECT_TSTEMPLATE:
RenameTSTemplate(stmt->object, stmt->newname);
break;
case OBJECT_TSCONFIGURATION:
RenameTSConfiguration(stmt->object, stmt->newname);
break;
default:
elog(ERROR, "unrecognized rename stmt type: %d",
(int) stmt->renameType);
}
if (Gp_role == GP_ROLE_DISPATCH)
{
CdbDispatchUtilityStatement((Node *) stmt, "ExecRenameStmt");
}
}
static const char *
get_quoted_nspname(Oid oid)
{
const char *nspname = get_namespace_name(get_rel_namespace(oid));
return (nspname ? quote_identifier(nspname) : NULL);
}
static void pg_decode_change(LogicalDecodingContext* ctx, ReorderBufferTXN* txn, Relation relation, ReorderBufferChange* change) {
DecodingJsonData* data;
Form_pg_class class_form;
TupleDesc tupdesc;
HeapTuple tuple;
MemoryContext old;
data = ctx->output_plugin_private;
data->xact_wrote_changes = true;
class_form = RelationGetForm(relation);
tupdesc = RelationGetDescr(relation);
old = MemoryContextSwitchTo(data->context);
OutputPluginPrepareWrite(ctx, true);
appendStringInfoString(ctx->out, "{\"type\":\"table\"");
appendStringInfo(
ctx->out,
",\"schema\":\"%s\"",
get_namespace_name(
get_rel_namespace(
RelationGetRelid(relation)
)
)
);
appendStringInfo(ctx->out, ",\"name\":\"%s\"", NameStr(class_form->relname));
appendStringInfo(
ctx->out,
",\"change\":\"%s\"",
change->action == REORDER_BUFFER_CHANGE_INSERT
? "INSERT"
: change->action == REORDER_BUFFER_CHANGE_UPDATE
? "UPDATE"
: change->action == REORDER_BUFFER_CHANGE_DELETE
? "DELETE"
: "FIXME"
);
if (change->action == REORDER_BUFFER_CHANGE_UPDATE || change->action == REORDER_BUFFER_CHANGE_DELETE) {
appendStringInfoString(ctx->out, ",\"key\":{");
RelationGetIndexList(relation);
if (OidIsValid(relation->rd_replidindex)) {
int i;
Relation index = index_open(relation->rd_replidindex, ShareLock);
tuple =
change->data.tp.oldtuple
? &change->data.tp.oldtuple->tuple
: &change->data.tp.newtuple->tuple;
for (i = 0; i < index->rd_index->indnatts; i++) {
int j = index->rd_index->indkey.values[i];
Form_pg_attribute attr = tupdesc->attrs[j - 1];
if (i > 0) appendStringInfoChar(ctx->out, ',');
appendStringInfo(ctx->out, "\"%s\":", NameStr(attr->attname));
print_value(ctx->out, tupdesc, tuple, j - 1);
}
index_close(index, NoLock);
} else {
appendStringInfoString(ctx->out, "\"***FIXME***\"");
}
appendStringInfoChar(ctx->out, '}');
}
if (change->action == REORDER_BUFFER_CHANGE_UPDATE || change->action == REORDER_BUFFER_CHANGE_INSERT) {
appendStringInfoString(ctx->out, ",\"data\":{");
tuple_to_stringinfo(ctx->out, tupdesc, &change->data.tp.newtuple->tuple, false);
appendStringInfoChar(ctx->out, '}');
}
appendStringInfoChar(ctx->out, '}');
MemoryContextSwitchTo(old);
MemoryContextReset(data->context);
OutputPluginWrite(ctx, true);
}
/*
* Executes an ALTER OBJECT / RENAME TO statement. Based on the object
* type, the function appropriate to that type is executed.
*/
void
ExecRenameStmt(RenameStmt *stmt)
{
switch (stmt->renameType)
{
case OBJECT_AGGREGATE:
RenameAggregate(stmt->object, stmt->objarg, stmt->newname);
break;
case OBJECT_CONVERSION:
RenameConversion(stmt->object, stmt->newname);
break;
case OBJECT_DATABASE:
RenameDatabase(stmt->subname, stmt->newname);
break;
case OBJECT_FUNCTION:
RenameFunction(stmt->object, stmt->objarg, stmt->newname);
break;
case OBJECT_LANGUAGE:
RenameLanguage(stmt->subname, stmt->newname);
break;
case OBJECT_OPCLASS:
RenameOpClass(stmt->object, stmt->subname, stmt->newname);
break;
case OBJECT_ROLE:
RenameRole(stmt->subname, stmt->newname);
break;
case OBJECT_SCHEMA:
RenameSchema(stmt->subname, stmt->newname);
break;
case OBJECT_TABLESPACE:
RenameTableSpace(stmt->subname, stmt->newname);
break;
case OBJECT_TABLE:
case OBJECT_INDEX:
case OBJECT_COLUMN:
case OBJECT_TRIGGER:
{
Oid relid;
CheckRelationOwnership(stmt->relation, true);
relid = RangeVarGetRelid(stmt->relation, false);
switch (stmt->renameType)
{
case OBJECT_TABLE:
case OBJECT_INDEX:
{
/*
* RENAME TABLE requires that we (still) hold
* CREATE rights on the containing namespace, as
* well as ownership of the table.
*/
Oid namespaceId = get_rel_namespace(relid);
AclResult aclresult;
aclresult = pg_namespace_aclcheck(namespaceId,
GetUserId(),
ACL_CREATE);
if (aclresult != ACLCHECK_OK)
aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
get_namespace_name(namespaceId));
renamerel(relid, stmt->newname);
break;
}
case OBJECT_COLUMN:
renameatt(relid,
stmt->subname, /* old att name */
stmt->newname, /* new att name */
interpretInhOption(stmt->relation->inhOpt), /* recursive? */
false); /* recursing already? */
break;
case OBJECT_TRIGGER:
renametrig(relid,
stmt->subname, /* old att name */
stmt->newname); /* new att name */
break;
default:
/* can't happen */
;
}
break;
}
default:
elog(ERROR, "unrecognized rename stmt type: %d",
(int) stmt->renameType);
}
}
/*
* master_append_table_to_shard appends the given table's contents to the given
* shard, and updates shard metadata on the master node. If the function fails
* to append table data to all shard placements, it doesn't update any metadata
* and errors out. Else if the function fails to append table data to some of
* the shard placements, it marks those placements as invalid. These invalid
* placements will get cleaned up during shard rebalancing.
*/
Datum
master_append_table_to_shard(PG_FUNCTION_ARGS)
{
uint64 shardId = PG_GETARG_INT64(0);
text *sourceTableNameText = PG_GETARG_TEXT_P(1);
text *sourceNodeNameText = PG_GETARG_TEXT_P(2);
uint32 sourceNodePort = PG_GETARG_UINT32(3);
char *sourceTableName = text_to_cstring(sourceTableNameText);
char *sourceNodeName = text_to_cstring(sourceNodeNameText);
Oid shardSchemaOid = 0;
char *shardSchemaName = NULL;
char *shardTableName = NULL;
char *shardQualifiedName = NULL;
List *shardPlacementList = NIL;
List *succeededPlacementList = NIL;
List *failedPlacementList = NIL;
ListCell *shardPlacementCell = NULL;
ListCell *failedPlacementCell = NULL;
uint64 newShardSize = 0;
uint64 shardMaxSizeInBytes = 0;
float4 shardFillLevel = 0.0;
char partitionMethod = 0;
ShardInterval *shardInterval = LoadShardInterval(shardId);
Oid relationId = shardInterval->relationId;
bool cstoreTable = CStoreTable(relationId);
char storageType = shardInterval->storageType;
EnsureTablePermissions(relationId, ACL_INSERT);
if (storageType != SHARD_STORAGE_TABLE && !cstoreTable)
{
ereport(ERROR, (errmsg("cannot append to shardId " UINT64_FORMAT, shardId),
errdetail("The underlying shard is not a regular table")));
}
partitionMethod = PartitionMethod(relationId);
if (partitionMethod == DISTRIBUTE_BY_HASH)
{
ereport(ERROR, (errmsg("cannot append to shardId " UINT64_FORMAT, shardId),
errdetail("We currently don't support appending to shards "
"in hash-partitioned tables")));
}
/*
* We lock on the shardId, but do not unlock. When the function returns, and
* the transaction for this function commits, this lock will automatically
* be released. This ensures appends to a shard happen in a serial manner.
*/
LockShardResource(shardId, AccessExclusiveLock);
/* get schame name of the target shard */
shardSchemaOid = get_rel_namespace(relationId);
shardSchemaName = get_namespace_name(shardSchemaOid);
/* Build shard table name. */
shardTableName = get_rel_name(relationId);
AppendShardIdToName(&shardTableName, shardId);
shardQualifiedName = quote_qualified_identifier(shardSchemaName, shardTableName);
shardPlacementList = FinalizedShardPlacementList(shardId);
if (shardPlacementList == NIL)
{
ereport(ERROR, (errmsg("could not find any shard placements for shardId "
UINT64_FORMAT, shardId),
errhint("Try running master_create_empty_shard() first")));
}
/* issue command to append table to each shard placement */
foreach(shardPlacementCell, shardPlacementList)
{
ShardPlacement *shardPlacement = (ShardPlacement *) lfirst(shardPlacementCell);
char *workerName = shardPlacement->nodeName;
uint32 workerPort = shardPlacement->nodePort;
List *queryResultList = NIL;
StringInfo workerAppendQuery = makeStringInfo();
appendStringInfo(workerAppendQuery, WORKER_APPEND_TABLE_TO_SHARD,
quote_literal_cstr(shardQualifiedName),
quote_literal_cstr(sourceTableName),
quote_literal_cstr(sourceNodeName), sourceNodePort);
/* inserting data should be performed by the current user */
queryResultList = ExecuteRemoteQuery(workerName, workerPort, NULL,
workerAppendQuery);
if (queryResultList != NIL)
{
succeededPlacementList = lappend(succeededPlacementList, shardPlacement);
}
else
{
failedPlacementList = lappend(failedPlacementList, shardPlacement);
}
}