path: root/contrib/postgres_fdw/connection.c

/*-------------------------------------------------------------------------
 *
 * connection.c
 *		  Connection management functions for postgres_fdw
 *
 * Portions Copyright (c) 2012-2013, PostgreSQL Global Development Group
 *
 * IDENTIFICATION
 *		  contrib/postgres_fdw/connection.c
 *
 *-------------------------------------------------------------------------
 */
#include "postgres.h"

#include "postgres_fdw.h"

#include "access/htup_details.h"
#include "catalog/pg_user_mapping.h"
#include "access/xact.h"
#include "mb/pg_wchar.h"
#include "miscadmin.h"
#include "storage/latch.h"
#include "storage/proc.h"
#include "utils/hsearch.h"
#include "utils/memutils.h"
#include "utils/syscache.h"
#include "utils/timestamp.h"


/*
 * Connection cache hash table entry
 *
 * The lookup key in this hash table is the foreign server OID plus the user
 * mapping OID.  (We use just one connection per user per foreign server,
 * so that we can ensure all scans use the same snapshot during a query.)
 *
 * The "conn" pointer can be NULL if we don't currently have a live connection.
 * When we do have a connection, xact_depth tracks the current depth of
 * transactions and subtransactions open on the remote side.  We need to issue
 * commands at the same nesting depth on the remote as we're executing at
 * ourselves, so that rolling back a subtransaction will kill the right
 * queries and not the wrong ones.
 */
typedef struct ConnCacheKey
{
	Oid			serverid;		/* OID of foreign server */
	Oid			userid;			/* OID of local user whose mapping we use */
} ConnCacheKey;

typedef struct ConnCacheEntry
{
	ConnCacheKey key;			/* hash key (must be first) */
	PGconn	   *conn;			/* connection to foreign server, or NULL */
	int			xact_depth;		/* 0 = no xact open, 1 = main xact open, 2 =
								 * one level of subxact open, etc */
	bool		have_prep_stmt; /* have we prepared any stmts in this xact? */
	bool		have_error;		/* have any subxacts aborted in this xact? */
	bool		changing_xact_state;	/* xact state change in process */
} ConnCacheEntry;

/*
 * Connection cache (initialized on first use)
 */
static HTAB *ConnectionHash = NULL;

/* for assigning cursor numbers and prepared statement numbers */
static unsigned int cursor_number = 0;
static unsigned int prep_stmt_number = 0;

/* tracks whether any work is needed in callback functions */
static bool xact_got_connection = false;

/* prototypes of private functions */
static PGconn *connect_pg_server(ForeignServer *server, UserMapping *user);
static void check_conn_params(const char **keywords, const char **values);
static void configure_remote_session(PGconn *conn);
static void do_sql_command(PGconn *conn, const char *sql);
static void begin_remote_xact(ConnCacheEntry *entry);
static void pgfdw_xact_callback(XactEvent event, void *arg);
static void pgfdw_subxact_callback(SubXactEvent event,
					   SubTransactionId mySubid,
					   SubTransactionId parentSubid,
					   void *arg);
static void pgfdw_reject_incomplete_xact_state_change(ConnCacheEntry *entry);
static bool pgfdw_cancel_query(PGconn *conn);
static bool pgfdw_exec_cleanup_query(PGconn *conn, const char *query,
						 bool ignore_errors);
static bool pgfdw_get_cleanup_result(PGconn *conn, TimestampTz endtime,
						 PGresult **result);


/*
 * Get a PGconn which can be used to execute queries on the remote PostgreSQL
 * server with the user's authorization.  A new connection is established
 * if we don't already have a suitable one, and a transaction is opened at
 * the right subtransaction nesting depth if we didn't do that already.
 *
 * will_prep_stmt must be true if caller intends to create any prepared
 * statements.  Since those don't go away automatically at transaction end
 * (not even on error), we need this flag to cue manual cleanup.
 *
 * XXX Note that caching connections theoretically requires a mechanism to
 * detect change of FDW objects to invalidate already established connections.
 * We could manage that by watching for invalidation events on the relevant
 * syscaches.  For the moment, though, it's not clear that this would really
 * be useful and not mere pedantry.  We could not flush any active connections
 * mid-transaction anyway.
 */
PGconn *
GetConnection(ForeignServer *server, UserMapping *user,
			  bool will_prep_stmt)
{
	bool		found;
	ConnCacheEntry *entry;
	ConnCacheKey key;

	/* First time through, initialize connection cache hashtable */
	if (ConnectionHash == NULL)
	{
		HASHCTL		ctl;

		MemSet(&ctl, 0, sizeof(ctl));
		ctl.keysize = sizeof(ConnCacheKey);
		ctl.entrysize = sizeof(ConnCacheEntry);
		ctl.hash = tag_hash;
		/* allocate ConnectionHash in the cache context */
		ctl.hcxt = CacheMemoryContext;
		ConnectionHash = hash_create("postgres_fdw connections", 8,
									 &ctl,
								   HASH_ELEM | HASH_FUNCTION | HASH_CONTEXT);

		/*
		 * Register some callback functions that manage connection cleanup.
		 * This should be done just once in each backend.
		 */
		RegisterXactCallback(pgfdw_xact_callback, NULL);
		RegisterSubXactCallback(pgfdw_subxact_callback, NULL);
	}

	/* Set flag that we did GetConnection during the current transaction */
	xact_got_connection = true;

	/* Create hash key for the entry.  Assume no pad bytes in key struct */
	key.serverid = server->serverid;
	key.userid = user->userid;

	/*
	 * Find or create cached entry for requested connection.
	 */
	entry = hash_search(ConnectionHash, &key, HASH_ENTER, &found);
	if (!found)
	{
		/* initialize new hashtable entry (key is already filled in) */
		entry->conn = NULL;
		entry->xact_depth = 0;
		entry->have_prep_stmt = false;
		entry->have_error = false;
		entry->changing_xact_state = false;
	}

	/* Reject further use of connections which failed abort cleanup. */
	pgfdw_reject_incomplete_xact_state_change(entry);

	/*
	 * We don't check the health of cached connection here, because it would
	 * require some overhead.  Broken connection will be detected when the
	 * connection is actually used.
	 */

	/*
	 * If cache entry doesn't have a connection, we have to establish a new
	 * connection.  (If connect_pg_server throws an error, the cache entry
	 * will be left in a valid empty state.)
	 */
	if (entry->conn == NULL)
	{
		entry->xact_depth = 0;	/* just to be sure */
		entry->have_prep_stmt = false;
		entry->have_error = false;
		entry->conn = connect_pg_server(server, user);
		elog(DEBUG3, "new postgres_fdw connection %p for server \"%s\"",
			 entry->conn, server->servername);
	}

	/*
	 * Start a new transaction or subtransaction if needed.
	 */
	begin_remote_xact(entry);

	/* Remember if caller will prepare statements */
	entry->have_prep_stmt |= will_prep_stmt;

	return entry->conn;
}

/*
 * Connect to remote server using specified server and user mapping properties.
 */
static PGconn *
connect_pg_server(ForeignServer *server, UserMapping *user)
{
	PGconn	   *volatile conn = NULL;

	/*
	 * Use PG_TRY block to ensure closing connection on error.
	 */
	PG_TRY();
	{
		const char **keywords;
		const char **values;
		int			n;

		/*
		 * Construct connection params from generic options of ForeignServer
		 * and UserMapping.  (Some of them might not be libpq options, in
		 * which case we'll just waste a few array slots.)  Add 3 extra slots
		 * for fallback_application_name, client_encoding, end marker.
		 */
		n = list_length(server->options) + list_length(user->options) + 3;
		keywords = (const char **) palloc(n * sizeof(char *));
		values = (const char **) palloc(n * sizeof(char *));

		n = 0;
		n += ExtractConnectionOptions(server->options,
									  keywords + n, values + n);
		n += ExtractConnectionOptions(user->options,
									  keywords + n, values + n);

		/* Use "postgres_fdw" as fallback_application_name. */
		keywords[n] = "fallback_application_name";
		values[n] = "postgres_fdw";
		n++;

		/* Set client_encoding so that libpq can convert encoding properly. */
		keywords[n] = "client_encoding";
		values[n] = GetDatabaseEncodingName();
		n++;

		keywords[n] = values[n] = NULL;

		/* verify connection parameters and make connection */
		check_conn_params(keywords, values);

		conn = PQconnectdbParams(keywords, values, false);
		if (!conn || PQstatus(conn) != CONNECTION_OK)
		{
			char	   *connmessage;
			int			msglen;

			/* libpq typically appends a newline, strip that */
			connmessage = pstrdup(PQerrorMessage(conn));
			msglen = strlen(connmessage);
			if (msglen > 0 && connmessage[msglen - 1] == '\n')
				connmessage[msglen - 1] = '\0';
			ereport(ERROR,
			   (errcode(ERRCODE_SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION),
				errmsg("could not connect to server \"%s\"",
					   server->servername),
				errdetail_internal("%s", connmessage)));
		}

		/*
		 * Check that non-superuser has used password to establish connection;
		 * otherwise, he's piggybacking on the postgres server's user
		 * identity. See also dblink_security_check() in contrib/dblink.
		 */
		if (!superuser() && !PQconnectionUsedPassword(conn))
			ereport(ERROR,
				  (errcode(ERRCODE_S_R_E_PROHIBITED_SQL_STATEMENT_ATTEMPTED),
				   errmsg("password is required"),
				   errdetail("Non-superuser cannot connect if the server does not request a password."),
				   errhint("Target server's authentication method must be changed.")));

		/* Prepare new session for use */
		configure_remote_session(conn);

		pfree(keywords);
		pfree(values);
	}
	PG_CATCH();
	{
		/* Release PGconn data structure if we managed to create one */
		if (conn)
			PQfinish(conn);
		PG_RE_THROW();
	}
	PG_END_TRY();

	return conn;
}

/*
 * For non-superusers, insist that the connstr specify a password.  This
 * prevents a password from being picked up from .pgpass, a service file,
 * the environment, etc.  We don't want the postgres user's passwords
 * to be accessible to non-superusers.  (See also dblink_connstr_check in
 * contrib/dblink.)
 */
static void
check_conn_params(const char **keywords, const char **values)
{
	int			i;

	/* no check required if superuser */
	if (superuser())
		return;

	/* ok if params contain a non-empty password */
	for (i = 0; keywords[i] != NULL; i++)
	{
		if (strcmp(keywords[i], "password") == 0 && values[i][0] != '\0')
			return;
	}

	ereport(ERROR,
			(errcode(ERRCODE_S_R_E_PROHIBITED_SQL_STATEMENT_ATTEMPTED),
			 errmsg("password is required"),
			 errdetail("Non-superusers must provide a password in the user mapping.")));
}

/*
 * Issue SET commands to make sure remote session is configured properly.
 *
 * We do this just once at connection, assuming nothing will change the
 * values later.  Since we'll never send volatile function calls to the
 * remote, there shouldn't be any way to break this assumption from our end.
 * It's possible to think of ways to break it at the remote end, eg making
 * a foreign table point to a view that includes a set_config call ---
 * but once you admit the possibility of a malicious view definition,
 * there are any number of ways to break things.
 */
static void
configure_remote_session(PGconn *conn)
{
	int			remoteversion = PQserverVersion(conn);

	/* Force the search path to contain only pg_catalog (see deparse.c) */
	do_sql_command(conn, "SET search_path = pg_catalog");

	/*
	 * Set remote timezone; this is basically just cosmetic, since all
	 * transmitted and returned timestamptzs should specify a zone explicitly
	 * anyway.  However it makes the regression test outputs more predictable.
	 *
	 * We don't risk setting remote zone equal to ours, since the remote
	 * server might use a different timezone database.  Instead, use UTC
	 * (quoted, because very old servers are picky about case).
	 */
	do_sql_command(conn, "SET timezone = 'UTC'");

	/*
	 * Set values needed to ensure unambiguous data output from remote.  (This
	 * logic should match what pg_dump does.  See also set_transmission_modes
	 * in postgres_fdw.c.)
	 */
	do_sql_command(conn, "SET datestyle = ISO");
	if (remoteversion >= 80400)
		do_sql_command(conn, "SET intervalstyle = postgres");
	if (remoteversion >= 90000)
		do_sql_command(conn, "SET extra_float_digits = 3");
	else
		do_sql_command(conn, "SET extra_float_digits = 2");
}

/*
 * Convenience subroutine to issue a non-data-returning SQL command to remote
 */
static void
do_sql_command(PGconn *conn, const char *sql)
{
	PGresult   *res;

	if (!PQsendQuery(conn, sql))
		pgfdw_report_error(ERROR, NULL, conn, false, sql);
	res = pgfdw_get_result(conn, sql);
	if (PQresultStatus(res) != PGRES_COMMAND_OK)
		pgfdw_report_error(ERROR, res, conn, true, sql);
	PQclear(res);
}

/*
 * Start remote transaction or subtransaction, if needed.
 *
 * Note that we always use at least REPEATABLE READ in the remote session.
 * This is so that, if a query initiates multiple scans of the same or
 * different foreign tables, we will get snapshot-consistent results from
 * those scans.  A disadvantage is that we can't provide sane emulation of
 * READ COMMITTED behavior --- it would be nice if we had some other way to
 * control which remote queries share a snapshot.
 */
static void
begin_remote_xact(ConnCacheEntry *entry)
{
	int			curlevel = GetCurrentTransactionNestLevel();

	/* Start main transaction if we haven't yet */
	if (entry->xact_depth <= 0)
	{
		const char *sql;

		elog(DEBUG3, "starting remote transaction on connection %p",
			 entry->conn);

		if (IsolationIsSerializable())
			sql = "START TRANSACTION ISOLATION LEVEL SERIALIZABLE";
		else
			sql = "START TRANSACTION ISOLATION LEVEL REPEATABLE READ";
		entry->changing_xact_state = true;
		do_sql_command(entry->conn, sql);
		entry->xact_depth = 1;
		entry->changing_xact_state = false;
	}

	/*
	 * If we're in a subtransaction, stack up savepoints to match our level.
	 * This ensures we can rollback just the desired effects when a
	 * subtransaction aborts.
	 */
	while (entry->xact_depth < curlevel)
	{
		char		sql[64];

		snprintf(sql, sizeof(sql), "SAVEPOINT s%d", entry->xact_depth + 1);
		entry->changing_xact_state = true;
		do_sql_command(entry->conn, sql);
		entry->xact_depth++;
		entry->changing_xact_state = false;
	}
}

/*
 * Release connection reference count created by calling GetConnection.
 */
void
ReleaseConnection(PGconn *conn)
{
	/*
	 * Currently, we don't actually track connection references because all
	 * cleanup is managed on a transaction or subtransaction basis instead. So
	 * there's nothing to do here.
	 */
}

/*
 * Assign a "unique" number for a cursor.
 *
 * These really only need to be unique per connection within a transaction.
 * For the moment we ignore the per-connection point and assign them across
 * all connections in the transaction, but we ask for the connection to be
 * supplied in case we want to refine that.
 *
 * Note that even if wraparound happens in a very long transaction, actual
 * collisions are highly improbable; just be sure to use %u not %d to print.
 */
unsigned int
GetCursorNumber(PGconn *conn)
{
	return ++cursor_number;
}

/*
 * Assign a "unique" number for a prepared statement.
 *
 * This works much like GetCursorNumber, except that we never reset the counter
 * within a session.  That's because we can't be 100% sure we've gotten rid
 * of all prepared statements on all connections, and it's not really worth
 * increasing the risk of prepared-statement name collisions by resetting.
 */
unsigned int
GetPrepStmtNumber(PGconn *conn)
{
	return ++prep_stmt_number;
}

/*
 * Submit a query and wait for the result.
 *
 * This function is interruptible by signals.
 *
 * Caller is responsible for the error handling on the result.
 */
PGresult *
pgfdw_exec_query(PGconn *conn, const char *query)
{
	/*
	 * Submit a query.  Since we don't use non-blocking mode, this also can
	 * block.  But its risk is relatively small, so we ignore that for now.
	 */
	if (!PQsendQuery(conn, query))
		pgfdw_report_error(ERROR, NULL, conn, false, query);

	/* Wait for the result. */
	return pgfdw_get_result(conn, query);
}

/*
 * Wait for the result from a prior asynchronous execution function call.
 *
 * This function offers quick responsiveness by checking for any interruptions.
 *
 * This function emulates the PQexec()'s behavior of returning the last result
 * when there are many.
 *
 * Caller is responsible for the error handling on the result.
 */
PGresult *
pgfdw_get_result(PGconn *conn, const char *query)
{
	PGresult   *last_res = NULL;

	for (;;)
	{
		PGresult *res;

		while (PQisBusy(conn))
		{
			int		wc;

			/* Sleep until there's something to do */
			wc = WaitLatchOrSocket(&MyProc->procLatch,
								   WL_LATCH_SET | WL_SOCKET_READABLE,
								   PQsocket(conn),
								   -1L);
			ResetLatch(&MyProc->procLatch);

			CHECK_FOR_INTERRUPTS();

			/* Data available in socket */
			if (wc & WL_SOCKET_READABLE)
			{
				if (!PQconsumeInput(conn))
					pgfdw_report_error(ERROR, NULL, conn, false, query);
			}
		}

		res = PQgetResult(conn);
		if (res == NULL)
			break;				/* query is complete */

		PQclear(last_res);
		last_res = res;
	}

	return last_res;
}

/*
 * Report an error we got from the remote server.
 *
 * elevel: error level to use (typically ERROR, but might be less)
 * res: PGresult containing the error
 * conn: connection we did the query on
 * clear: if true, PQclear the result (otherwise caller will handle it)
 * sql: NULL, or text of remote command we tried to execute
 *
 * Note: callers that choose not to throw ERROR for a remote error are
 * responsible for making sure that the associated ConnCacheEntry gets
 * marked with have_error = true.
 */
void
pgfdw_report_error(int elevel, PGresult *res, PGconn *conn,
				   bool clear, const char *sql)
{
	/* If requested, PGresult must be released before leaving this function. */
	PG_TRY();
	{
		char	   *diag_sqlstate = PQresultErrorField(res, PG_DIAG_SQLSTATE);
		char	   *message_primary = PQresultErrorField(res, PG_DIAG_MESSAGE_PRIMARY);
		char	   *message_detail = PQresultErrorField(res, PG_DIAG_MESSAGE_DETAIL);
		char	   *message_hint = PQresultErrorField(res, PG_DIAG_MESSAGE_HINT);
		char	   *message_context = PQresultErrorField(res, PG_DIAG_CONTEXT);
		int			sqlstate;

		if (diag_sqlstate)
			sqlstate = MAKE_SQLSTATE(diag_sqlstate[0],
									 diag_sqlstate[1],
									 diag_sqlstate[2],
									 diag_sqlstate[3],
									 diag_sqlstate[4]);
		else
			sqlstate = ERRCODE_CONNECTION_FAILURE;

		/*
		 * If we don't get a message from the PGresult, try the PGconn.  This
		 * is needed because for connection-level failures, PQexec may just
		 * return NULL, not a PGresult at all.
		 */
		if (message_primary == NULL)
			message_primary = PQerrorMessage(conn);

		ereport(elevel,
				(errcode(sqlstate),
				 message_primary ? errmsg_internal("%s", message_primary) :
				 errmsg("could not obtain message string for remote error"),
			   message_detail ? errdetail_internal("%s", message_detail) : 0,
				 message_hint ? errhint("%s", message_hint) : 0,
				 message_context ? errcontext("%s", message_context) : 0,
				 sql ? errcontext("Remote SQL command: %s", sql) : 0));
	}
	PG_CATCH();
	{
		if (clear)
			PQclear(res);
		PG_RE_THROW();
	}
	PG_END_TRY();
	if (clear)
		PQclear(res);
}

/*
 * pgfdw_xact_callback --- cleanup at main-transaction end.
 */
static void
pgfdw_xact_callback(XactEvent event, void *arg)
{
	HASH_SEQ_STATUS scan;
	ConnCacheEntry *entry;

	/* Quick exit if no connections were touched in this transaction. */
	if (!xact_got_connection)
		return;

	/*
	 * Scan all connection cache entries to find open remote transactions, and
	 * close them.
	 */
	hash_seq_init(&scan, ConnectionHash);
	while ((entry = (ConnCacheEntry *) hash_seq_search(&scan)))
	{
		PGresult   *res;

		/* Ignore cache entry if no open connection right now */
		if (entry->conn == NULL)
			continue;

		/* If it has an open remote transaction, try to close it */
		if (entry->xact_depth > 0)
		{
			bool		abort_cleanup_failure = false;

			elog(DEBUG3, "closing remote transaction on connection %p",
				 entry->conn);

			switch (event)
			{
				case XACT_EVENT_PRE_COMMIT:

					/*
					 * If abort cleanup previously failed for this connection,
					 * we can't issue any more commands against it.
					 */
					pgfdw_reject_incomplete_xact_state_change(entry);

					/* Commit all remote transactions during pre-commit */
					entry->changing_xact_state = true;
					do_sql_command(entry->conn, "COMMIT TRANSACTION");
					entry->changing_xact_state = false;

					/*
					 * If there were any errors in subtransactions, and we
					 * made prepared statements, do a DEALLOCATE ALL to make
					 * sure we get rid of all prepared statements. This is
					 * annoying and not terribly bulletproof, but it's
					 * probably not worth trying harder.
					 *
					 * DEALLOCATE ALL only exists in 8.3 and later, so this
					 * constrains how old a server postgres_fdw can
					 * communicate with.  We intentionally ignore errors in
					 * the DEALLOCATE, so that we can hobble along to some
					 * extent with older servers (leaking prepared statements
					 * as we go; but we don't really support update operations
					 * pre-8.3 anyway).
					 */
					if (entry->have_prep_stmt && entry->have_error)
					{
						res = PQexec(entry->conn, "DEALLOCATE ALL");
						PQclear(res);
					}
					entry->have_prep_stmt = false;
					entry->have_error = false;
					break;
				case XACT_EVENT_PRE_PREPARE:

					/*
					 * We disallow remote transactions that modified anything,
					 * since it's not very reasonable to hold them open until
					 * the prepared transaction is committed.  For the moment,
					 * throw error unconditionally; later we might allow
					 * read-only cases.  Note that the error will cause us to
					 * come right back here with event == XACT_EVENT_ABORT, so
					 * we'll clean up the connection state at that point.
					 */
					ereport(ERROR,
							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
							 errmsg("cannot prepare a transaction that modified remote tables")));
					break;
				case XACT_EVENT_COMMIT:
				case XACT_EVENT_PREPARE:
					/* Pre-commit should have closed the open transaction */
					elog(ERROR, "missed cleaning up connection during pre-commit");
					break;
				case XACT_EVENT_ABORT:

					/*
					 * Don't try to clean up the connection if we're already
					 * in error recursion trouble.
					 */
					if (in_error_recursion_trouble())
						entry->changing_xact_state = true;

					/*
					 * If connection is already unsalvageable, don't touch it
					 * further.
					 */
					if (entry->changing_xact_state)
						break;

					/*
					 * Mark this connection as in the process of changing
					 * transaction state.
					 */
					entry->changing_xact_state = true;

					/* Assume we might have lost track of prepared statements */
					entry->have_error = true;

					/*
					 * If a command has been submitted to the remote server by
					 * using an asynchronous execution function, the command
					 * might not have yet completed.  Check to see if a command
					 * is still being processed by the remote server, and if so,
					 * request cancellation of the command.
					 */
					if (PQtransactionStatus(entry->conn) == PQTRANS_ACTIVE &&
						!pgfdw_cancel_query(entry->conn))
					{
						/* Unable to cancel running query. */
						abort_cleanup_failure = true;
					}
					else if (!pgfdw_exec_cleanup_query(entry->conn,
													   "ABORT TRANSACTION",
													   false))
					{
						/* Unable to abort remote transaction. */
						abort_cleanup_failure = true;
					}
					else if (entry->have_prep_stmt && entry->have_error &&
							 !pgfdw_exec_cleanup_query(entry->conn,
													   "DEALLOCATE ALL",
													   true))
					{
						/* Trouble clearing prepared statements. */
						abort_cleanup_failure = true;
					}
					else
					{
						entry->have_prep_stmt = false;
						entry->have_error = false;
					}

					/* Disarm changing_xact_state if it all worked. */
					entry->changing_xact_state = abort_cleanup_failure;
					break;
			}
		}

		/* Reset state to show we're out of a transaction */
		entry->xact_depth = 0;

		/*
		 * If the connection isn't in a good idle state, discard it to
		 * recover. Next GetConnection will open a new connection.
		 */
		if (PQstatus(entry->conn) != CONNECTION_OK ||
			PQtransactionStatus(entry->conn) != PQTRANS_IDLE ||
			entry->changing_xact_state)
		{
			elog(DEBUG3, "discarding connection %p", entry->conn);
			PQfinish(entry->conn);
			entry->conn = NULL;
			entry->changing_xact_state = false;
		}
	}

	/*
	 * Regardless of the event type, we can now mark ourselves as out of the
	 * transaction.  (Note: if we are here during PRE_COMMIT or PRE_PREPARE,
	 * this saves a useless scan of the hashtable during COMMIT or PREPARE.)
	 */
	xact_got_connection = false;

	/* Also reset cursor numbering for next transaction */
	cursor_number = 0;
}

/*
 * pgfdw_subxact_callback --- cleanup at subtransaction end.
 */
static void
pgfdw_subxact_callback(SubXactEvent event, SubTransactionId mySubid,
					   SubTransactionId parentSubid, void *arg)
{
	HASH_SEQ_STATUS scan;
	ConnCacheEntry *entry;
	int			curlevel;

	/* Nothing to do at subxact start, nor after commit. */
	if (!(event == SUBXACT_EVENT_PRE_COMMIT_SUB ||
		  event == SUBXACT_EVENT_ABORT_SUB))
		return;

	/* Quick exit if no connections were touched in this transaction. */
	if (!xact_got_connection)
		return;

	/*
	 * Scan all connection cache entries to find open remote subtransactions
	 * of the current level, and close them.
	 */
	curlevel = GetCurrentTransactionNestLevel();
	hash_seq_init(&scan, ConnectionHash);
	while ((entry = (ConnCacheEntry *) hash_seq_search(&scan)))
	{
		char		sql[100];

		/*
		 * We only care about connections with open remote subtransactions of
		 * the current level.
		 */
		if (entry->conn == NULL || entry->xact_depth < curlevel)
			continue;

		if (entry->xact_depth > curlevel)
			elog(ERROR, "missed cleaning up remote subtransaction at level %d",
				 entry->xact_depth);

		if (event == SUBXACT_EVENT_PRE_COMMIT_SUB)
		{
			/*
			 * If abort cleanup previously failed for this connection, we
			 * can't issue any more commands against it.
			 */
			pgfdw_reject_incomplete_xact_state_change(entry);

			/* Commit all remote subtransactions during pre-commit */
			snprintf(sql, sizeof(sql), "RELEASE SAVEPOINT s%d", curlevel);
			entry->changing_xact_state = true;
			do_sql_command(entry->conn, sql);
			entry->changing_xact_state = false;
		}
		else if (in_error_recursion_trouble())
		{
			/*
			 * Don't try to clean up the connection if we're already in error
			 * recursion trouble.
			 */
			entry->changing_xact_state = true;
		}
		else if (!entry->changing_xact_state)
		{
			bool		abort_cleanup_failure = false;

			/* Remember that abort cleanup is in progress. */
			entry->changing_xact_state = true;

			/* Assume we might have lost track of prepared statements */
			entry->have_error = true;

			/*
			 * If a command has been submitted to the remote server by using an
			 * asynchronous execution function, the command might not have yet
			 * completed.  Check to see if a command is still being processed by
			 * the remote server, and if so, request cancellation of the
			 * command.
			 */
			if (PQtransactionStatus(entry->conn) == PQTRANS_ACTIVE &&
				!pgfdw_cancel_query(entry->conn))
				abort_cleanup_failure = true;
			else
			{
				/* Rollback all remote subtransactions during abort */
				snprintf(sql, sizeof(sql),
						 "ROLLBACK TO SAVEPOINT s%d; RELEASE SAVEPOINT s%d",
						 curlevel, curlevel);
				if (!pgfdw_exec_cleanup_query(entry->conn, sql, false))
					abort_cleanup_failure = true;
			}

			/* Disarm changing_xact_state if it all worked. */
			entry->changing_xact_state = abort_cleanup_failure;
		}

		/* OK, we're outta that level of subtransaction */
		entry->xact_depth--;
	}
}

/*
 * Raise an error if the given connection cache entry is marked as being
 * in the middle of an xact state change.  This should be called at which no
 * such change is expected to be in progress; if one is found to be in
 * progress, it means that we aborted in the middle of a previous state change
 * and now don't know what the remote transaction state actually is.
 * Such connections can't safely be further used.  Re-establishing the
 * connection would change the snapshot and roll back any writes already
 * performed, so that's not an option, either. Thus, we must abort.
 */
static void
pgfdw_reject_incomplete_xact_state_change(ConnCacheEntry *entry)
{
	ForeignServer *server;

	if (!entry->changing_xact_state)
		return;

	server = GetForeignServer(entry->key.serverid);
	ereport(ERROR,
			(errcode(ERRCODE_CONNECTION_EXCEPTION),
			 errmsg("connection to server \"%s\" was lost",
					server->servername)));
}

/*
 * Cancel the currently-in-progress query (whose query text we do not have)
 * and ignore the result.  Returns true if we successfully cancel the query
 * and discard any pending result, and false if not.
 */
static bool
pgfdw_cancel_query(PGconn *conn)
{
	PGcancel   *cancel;
	char		errbuf[256];
	PGresult   *result = NULL;
	TimestampTz endtime;

	/*
	 * If it takes too long to cancel the query and discard the result, assume
	 * the connection is dead.
	 */
	endtime = TimestampTzPlusMilliseconds(GetCurrentTimestamp(), 30000);

	/*
	 * Issue cancel request.  Unfortunately, there's no good way to limit the
	 * amount of time that we might block inside PQgetCancel().
	 */
	if ((cancel = PQgetCancel(conn)))
	{
		if (!PQcancel(cancel, errbuf, sizeof(errbuf)))
		{
			ereport(WARNING,
					(errcode(ERRCODE_CONNECTION_FAILURE),
					 errmsg("could not send cancel request: %s",
							errbuf)));
			PQfreeCancel(cancel);
			return false;
		}
		PQfreeCancel(cancel);
	}

	/* Get and discard the result of the query. */
	if (pgfdw_get_cleanup_result(conn, endtime, &result))
		return false;
	PQclear(result);

	return true;
}

/*
 * Submit a query during (sub)abort cleanup and wait up to 30 seconds for the
 * result.  If the query is executed without error, the return value is true.
 * If the query is executed successfully but returns an error, the return
 * value is true if and only if ignore_errors is set.  If the query can't be
 * sent or times out, the return value is false.
 */
static bool
pgfdw_exec_cleanup_query(PGconn *conn, const char *query, bool ignore_errors)
{
	PGresult   *result = NULL;
	TimestampTz endtime;

	/*
	 * If it takes too long to execute a cleanup query, assume the connection
	 * is dead.  It's fairly likely that this is why we aborted in the first
	 * place (e.g. statement timeout, user cancel), so the timeout shouldn't
	 * be too long.
	 */
	endtime = TimestampTzPlusMilliseconds(GetCurrentTimestamp(), 30000);

	/*
	 * Submit a query.  Since we don't use non-blocking mode, this also can
	 * block.  But its risk is relatively small, so we ignore that for now.
	 */
	if (!PQsendQuery(conn, query))
	{
		pgfdw_report_error(WARNING, NULL, conn, false, query);
		return false;
	}

	/* Get the result of the query. */
	if (pgfdw_get_cleanup_result(conn, endtime, &result))
		return false;

	/* Issue a warning if not successful. */
	if (PQresultStatus(result) != PGRES_COMMAND_OK)
	{
		pgfdw_report_error(WARNING, result, conn, true, query);
		return ignore_errors;
	}

	return true;
}

/*
 * Get, during abort cleanup, the result of a query that is in progress.  This
 * might be a query that is being interrupted by transaction abort, or it might
 * be a query that was initiated as part of transaction abort to get the remote
 * side back to the appropriate state.
 *
 * It's not a huge problem if we throw an ERROR here, but if we get into error
 * recursion trouble, we'll end up slamming the connection shut, which will
 * necessitate failing the entire toplevel transaction even if subtransactions
 * were used.  Try to use WARNING where we can.
 *
 * endtime is the time at which we should give up and assume the remote
 * side is dead.  Returns true if the timeout expired, otherwise false.
 * Sets *result except in case of a timeout.
 */
static bool
pgfdw_get_cleanup_result(PGconn *conn, TimestampTz endtime, PGresult **result)
{
	PGresult   *last_res = NULL;

	for (;;)
	{
		PGresult   *res;

		while (PQisBusy(conn))
		{
			int			wc;
			TimestampTz now = GetCurrentTimestamp();
			long		secs;
			int			microsecs;
			long		cur_timeout;

			/* If timeout has expired, give up, else get sleep time. */
			if (now >= endtime)
				return true;
			TimestampDifference(now, endtime, &secs, &microsecs);

			/* To protect against clock skew, limit sleep to one minute. */
			cur_timeout = Min(60000, secs * USECS_PER_SEC + microsecs);

			/* Sleep until there's something to do */
			wc = WaitLatchOrSocket(&MyProc->procLatch,
							  WL_LATCH_SET | WL_SOCKET_READABLE | WL_TIMEOUT,
								   PQsocket(conn),
								   cur_timeout);
			ResetLatch(&MyProc->procLatch);

			CHECK_FOR_INTERRUPTS();

			/* Data available in socket */
			if (wc & WL_SOCKET_READABLE)
			{
				if (!PQconsumeInput(conn))
				{
					*result = NULL;
					return false;
				}
			}
		}

		res = PQgetResult(conn);
		if (res == NULL)
			break;				/* query is complete */

		PQclear(last_res);
		last_res = res;
	}

	*result = last_res;
	return false;
}