about summary refs log tree commit diff
path: root/db2/xa
diff options
context:
space:
mode:
Diffstat (limited to 'db2/xa')
-rw-r--r--db2/xa/xa.c682
-rw-r--r--db2/xa/xa_db.c308
-rw-r--r--db2/xa/xa_map.c305
3 files changed, 1295 insertions, 0 deletions
diff --git a/db2/xa/xa.c b/db2/xa/xa.c
new file mode 100644
index 0000000000..94a96e7e09
--- /dev/null
+++ b/db2/xa/xa.c
@@ -0,0 +1,682 @@
+/*-
+ * See the file LICENSE for redistribution information.
+ *
+ * Copyright (c) 1998
+ *	Sleepycat Software.  All rights reserved.
+ */
+
+/* XXX Remove the global transaction and hang it off the environment. */
+#include "config.h"
+
+#ifndef lint
+static const char sccsid[] = "@(#)xa.c	10.4 (Sleepycat) 10/11/98";
+#endif /* not lint */
+
+#ifndef NO_SYSTEM_INCLUDES
+#include <sys/types.h>
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#endif
+
+#include "db_int.h"
+#include "db_page.h"
+#include "shqueue.h"
+#include "log.h"
+#include "txn.h"
+#include "db_auto.h"
+#include "db_ext.h"
+#include "db_dispatch.h"
+
+static int  __db_xa_close __P((char *, int, long));
+static int  __db_xa_commit __P((XID *, int, long));
+static int  __db_xa_complete __P((int *, int *, int, long));
+static int  __db_xa_end __P((XID *, int, long));
+static int  __db_xa_forget __P((XID *, int, long));
+static int  __db_xa_open __P((char *, int, long));
+static int  __db_xa_prepare __P((XID *, int, long));
+static int  __db_xa_recover __P((XID *, long, int, long));
+static int  __db_xa_rollback __P((XID *, int, long));
+static int  __db_xa_start __P((XID *, int, long));
+static void __xa_txn_end __P((DB_ENV *));
+static void __xa_txn_init __P((DB_ENV *, TXN_DETAIL *, size_t));
+
+/*
+ * Possible flag values:
+ *	Dynamic registration	0 => no dynamic registration
+ *				TMREGISTER => dynamic registration
+ *	Asynchronous operation	0 => no support for asynchrony
+ *				TMUSEASYNC => async support
+ *	Migration support	0 => migration of transactions across
+ *				     threads is possible
+ *				TMNOMIGRATE => no migration across threads
+ */
+const struct xa_switch_t db_xa_switch = {
+	 "Berkeley DB",		/* name[RMNAMESZ] */
+	 TMNOMIGRATE,		/* flags */
+	 0,			/* version */
+	 __db_xa_open,		/* xa_open_entry */
+	 __db_xa_close,		/* xa_close_entry */
+	 __db_xa_start,		/* xa_start_entry */
+	 __db_xa_end,		/* xa_end_entry */
+	 __db_xa_rollback,	/* xa_rollback_entry */
+	 __db_xa_prepare,	/* xa_prepare_entry */
+	 __db_xa_commit,	/* xa_commit_entry */
+	 __db_xa_recover,	/* xa_recover_entry */
+	 __db_xa_forget,	/* xa_forget_entry */
+	 __db_xa_complete	/* xa_complete_entry */
+};
+
+/*
+ * __db_xa_open --
+ *	The open call in the XA protocol.  The rmid field is an id number
+ * that the TM assigned us and will pass us on every xa call.  We need to
+ * map that rmid number into a dbenv structure that we create during
+ * initialization.  Since this id number is thread specific, we do not
+ * need to store it in shared memory.  The file xa_map.c implements all
+ * such xa->db mappings.
+ *	The xa_info field is instance specific information.  We require
+ * that the value of DB_HOME be passed in xa_info.  Since xa_info is the
+ * only thing that we get to pass to db_appinit, any config information
+ * will have to be done via a config file instead of via the db_appinit
+ * call.
+ */
+static int
+__db_xa_open(xa_info, rmid, flags)
+	char *xa_info;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+	if (flags != TMNOFLAGS)
+		return (XAER_INVAL);
+
+	/* Verify if we already have this environment open. */
+	if (__db_rmid_to_env(rmid, &env, 0) == 0)
+		return (XA_OK);
+
+	/*
+	 * Since we cannot tell whether the environment is OK or not,
+	 * we can't actually do the db_appinit in xa_open.  Instead,
+	 * we save the mapping between the rmid and the xa_info.  If
+	 * we next get a call to __xa_recover, we do the db_appinit
+	 * with DB_RECOVER set.  If we get any other call, then we
+	 * do the db_appinit.
+	 */
+	return (__db_map_rmid_name(rmid, xa_info));
+}
+
+/*
+ * __db_xa_close --
+ *	The close call of the XA protocol.  The only trickiness here
+ * is that if there are any active transactions, we must fail.  It is
+ * *not* an error to call close on an environment that has already been
+ * closed (I am interpreting that to mean it's OK to call close on an
+ * environment that has never been opened).
+ */
+static int
+__db_xa_close(xa_info, rmid, flags)
+	char *xa_info;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	int ret, t_ret;
+
+	COMPQUIET(xa_info, NULL);
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+	if (flags != TMNOFLAGS)
+		return (XAER_INVAL);
+
+	/* If the environment is closed, then we're done. */
+	if (__db_rmid_to_env(rmid, &env, 0) != 0)
+		return (XA_OK);
+
+	/* Check if there are any pending transactions. */
+	if (env->xa_txn != NULL && env->xa_txn->txnid != TXN_INVALID)
+		return (XAER_PROTO);
+
+	/* Now, destroy the mapping and close the environment. */
+	ret = __db_unmap_rmid(rmid);
+	if ((t_ret = db_appexit(env)) != 0 && ret == 0)
+		ret = t_ret;
+
+	__os_free(env, sizeof(DB_ENV));
+
+	return (ret == 0 ? XA_OK : XAER_RMERR);
+}
+
+/*
+ * __db_xa_start --
+ *	Begin a transaction for the current resource manager.
+ */
+static int
+__db_xa_start(xid, rmid, flags)
+	XID *xid;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	TXN_DETAIL *td;
+	size_t off;
+	int is_known;
+
+#define	OK_FLAGS	(TMJOIN | TMRESUME | TMNOWAIT | TMASYNC | TMNOFLAGS)
+	if (LF_ISSET(~OK_FLAGS))
+		return (XAER_INVAL);
+
+	if (LF_ISSET(TMJOIN) && LF_ISSET(TMRESUME))
+		return (XAER_INVAL);
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+
+	if (__db_rmid_to_env(rmid, &env, 1) != 0)
+		return (XAER_PROTO);
+
+	is_known = __db_xid_to_txn(env, xid, &off) == 0;
+
+	if (is_known && !LF_ISSET(TMRESUME) && !LF_ISSET(TMJOIN))
+		return (XAER_DUPID);
+
+	if (!is_known && LF_ISSET(TMRESUME | TMJOIN))
+		return (XAER_NOTA);
+
+	/*
+	 * This can't block, so we can ignore TMNOWAIT.
+	 *
+	 * Other error conditions: RMERR, RMFAIL, OUTSIDE, PROTO, RB*
+	 */
+	if (is_known) {
+		td = (TXN_DETAIL *)((u_int8_t *)env->tx_info->region + off);
+		if (td->xa_status == TXN_XA_SUSPENDED && !LF_SET(TMRESUME))
+			return (XAER_PROTO);
+		if (td->xa_status == TXN_XA_DEADLOCKED)
+			return (XA_RBDEADLOCK);
+		if (td->xa_status == TXN_XA_ABORTED)
+			return (XA_RBOTHER);
+
+		/* Now, fill in the global transaction structure. */
+		__xa_txn_init(env, td, off);
+		td->xa_status = TXN_XA_STARTED;
+	} else {
+		if (__txn_xa_begin(env, env->xa_txn) != 0)
+			return (XAER_RMERR);
+		(void)__db_map_xid(env, xid, env->xa_txn->off);
+		td = (TXN_DETAIL *)
+		    ((u_int8_t *)env->tx_info->region + env->xa_txn->off);
+		td->xa_status = TXN_XA_STARTED;
+	}
+	return (XA_OK);
+}
+
+/*
+ * __db_xa_end --
+ *	Disassociate the current transaction from the current process.
+ */
+static int
+__db_xa_end(xid, rmid, flags)
+	XID *xid;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	DB_TXN *txn;
+	TXN_DETAIL *td;
+	size_t off;
+
+	if (flags != TMNOFLAGS && !LF_ISSET(TMSUSPEND | TMSUCCESS | TMFAIL))
+		return (XAER_INVAL);
+
+	if (__db_rmid_to_env(rmid, &env, 0) != 0)
+		return (XAER_PROTO);
+
+	if (__db_xid_to_txn(env, xid, &off) != 0)
+		return (XAER_NOTA);
+
+	txn = env->xa_txn;
+	if (off != txn->off)
+		return (XAER_PROTO);
+
+	td = (TXN_DETAIL *)((u_int8_t *)env->tx_info->region + off);
+	if (td->xa_status == TXN_XA_DEADLOCKED)
+		return (XA_RBDEADLOCK);
+
+	if (td->status == TXN_ABORTED)
+		return (XA_RBOTHER);
+
+	if (td->xa_status != TXN_XA_STARTED)
+		return (XAER_PROTO);
+
+	/* Update the shared memory last_lsn field */
+	td->last_lsn = txn->last_lsn;
+
+	/*
+	 * If we ever support XA migration, we cannot keep SUSPEND/END
+	 * status in the shared region; it would have to be process local.
+	 */
+	if (LF_ISSET(TMSUSPEND))
+		td->xa_status = TXN_XA_SUSPENDED;
+	else
+		td->xa_status = TXN_XA_ENDED;
+
+	txn->txnid = TXN_INVALID;
+	return (XA_OK);
+}
+
+/*
+ * __db_xa_prepare --
+ *	Sync the log to disk so we can guarantee recoverability.
+ */
+static int
+__db_xa_prepare(xid, rmid, flags)
+	XID *xid;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	TXN_DETAIL *td;
+	size_t off;
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+	if (flags != TMNOFLAGS)
+		return (XAER_INVAL);
+
+	/*
+	 * We need to know if we've ever called prepare on this.
+	 * As part of the prepare, we set the xa_status field to
+	 * reflect that fact that prepare has been called, and if
+	 * it's ever called again, it's an error.
+	 */
+	if (__db_rmid_to_env(rmid, &env, 1) != 0)
+		return (XAER_PROTO);
+
+	if (__db_xid_to_txn(env, xid, &off) != 0)
+		return (XAER_NOTA);
+
+	td = (TXN_DETAIL *)((u_int8_t *)env->tx_info->region + off);
+
+	if (td->xa_status == TXN_XA_DEADLOCKED)
+		return (XA_RBDEADLOCK);
+
+	if (td->xa_status != TXN_XA_ENDED && td->xa_status != TXN_XA_SUSPENDED)
+		return (XAER_PROTO);
+
+	/* Now, fill in the global transaction structure. */
+	__xa_txn_init(env, td, off);
+
+	if (txn_prepare(env->xa_txn) != 0)
+		return (XAER_RMERR);
+
+	td->xa_status = TXN_XA_PREPARED;
+
+	/* No fatal value that would require an XAER_RMFAIL. */
+	__xa_txn_end(env);
+	return (XA_OK);
+}
+
+/*
+ * __db_xa_commit --
+ *	Commit the transaction
+ */
+static int
+__db_xa_commit(xid, rmid, flags)
+	XID *xid;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	TXN_DETAIL *td;
+	size_t off;
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+#undef	OK_FLAGS
+#define	OK_FLAGS	(TMNOFLAGS | TMNOWAIT | TMONEPHASE)
+	if (LF_ISSET(~OK_FLAGS))
+		return (XAER_INVAL);
+
+	/*
+	 * We need to know if we've ever called prepare on this.
+	 * We can verify this by examining the xa_status field.
+	 */
+	if (__db_rmid_to_env(rmid, &env, 1) != 0)
+		return (XAER_PROTO);
+
+	if (__db_xid_to_txn(env, xid, &off) != 0)
+		return (XAER_NOTA);
+
+	td = (TXN_DETAIL *)((u_int8_t *)env->tx_info->region + off);
+
+	if (td->xa_status == TXN_XA_DEADLOCKED)
+		return (XA_RBDEADLOCK);
+
+	if (td->xa_status == TXN_XA_ABORTED)
+		return (XA_RBOTHER);
+
+	if (LF_SET(TMONEPHASE) &&
+	    td->xa_status != TXN_XA_ENDED && td->xa_status != TXN_XA_SUSPENDED)
+		return (XAER_PROTO);
+
+	if (!LF_SET(TMONEPHASE) && td->xa_status != TXN_XA_PREPARED)
+		return (XAER_PROTO);
+
+	/* Now, fill in the global transaction structure. */
+	__xa_txn_init(env, td, off);
+
+	if (txn_commit(env->xa_txn) != 0)
+		return (XAER_RMERR);
+
+	/* No fatal value that would require an XAER_RMFAIL. */
+	__xa_txn_end(env);
+	return (XA_OK);
+}
+
+/*
+ * __db_xa_recover --
+ *	Returns a list of prepared and heuristically completed transactions.
+ *
+ * The return value is the number of xids placed into the xid array (less
+ * than or equal to the count parameter).  The flags are going to indicate
+ * whether we are starting a scan or continuing one.
+ */
+static int
+__db_xa_recover(xids, count, rmid, flags)
+	XID *xids;
+	long count, flags;
+	int rmid;
+{
+	__txn_xa_regop_args *argp;
+	DBT data;
+	DB_ENV *env;
+	DB_LOG *log;
+	XID *xidp;
+	char *dbhome;
+	int err, ret;
+	u_int32_t rectype, txnid;
+
+	ret = 0;
+	xidp = xids;
+
+
+	/*
+	 * If we are starting a scan, then we need to open the environment
+	 * and run recovery.  This recovery puts us in a state where we can
+	 * either commit or abort any transactions that were prepared but not
+	 * yet committed.  Once we've done that, we need to figure out where
+	 * to begin checking for such transactions.  If we are not starting
+	 * a scan, then the environment had better have already been recovered
+	 * and we'll start from * wherever the log cursor is.  Since XA apps
+	 * cannot be threaded, we don't have to worry about someone else
+	 * having moved it.
+	 */
+	if (LF_ISSET(TMSTARTRSCAN)) {
+		/* If the environment is open, we have a problem. */
+		if (__db_rmid_to_env(rmid, &env, 0) == XA_OK)
+			return (XAER_PROTO);
+
+		if ((ret = __os_calloc(1, sizeof(DB_ENV), &env)) != 0)
+			return (XAER_RMERR);
+
+		if (__db_rmid_to_name(rmid, &dbhome) != 0)
+			goto err1;
+
+#undef XA_FLAGS
+#define	XA_FLAGS DB_RECOVER | \
+	DB_CREATE | DB_INIT_LOCK | DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_TXN
+		if ((ret = db_appinit(dbhome, NULL, env, XA_FLAGS)) != 0)
+			goto err1;
+
+		if (__db_map_rmid(rmid, env) != 0)
+			goto err2;
+
+		/* Now figure out from where to begin scan. */
+		log = env->lg_info;
+		if ((err = __log_findckp(log, &log->xa_first)) == DB_NOTFOUND) {
+			/*
+			 * If there were no log files, then we have no
+			 * transactions to return, so we simply return 0.
+			 */
+			return (0);
+		}
+		if ((err = __db_txnlist_init(&log->xa_info)) != 0)
+			goto err3;
+	} else {
+		/* We had better already know about this rmid. */
+		if (__db_rmid_to_env(rmid, &env, 0) != 0)
+			return (XAER_PROTO);
+		/*
+		 * If we are not starting a scan, the log cursor had
+		 * better be set.
+		 */
+		log = env->lg_info;
+		if (IS_ZERO_LSN(log->xa_lsn))
+			return (XAER_PROTO);
+	}
+
+	/*
+	 * At this point log->xa_first contains the point in the log
+	 * to which we need to roll back.  If we are starting a scan,
+	 * we'll start at the last record; if we're continuing a scan,
+	 * we'll have to start at log->xa_lsn.
+	 */
+
+	memset(&data, 0, sizeof(data));
+	for (err = log_get(log, &log->xa_lsn, &data,
+	    LF_ISSET(TMSTARTRSCAN) ? DB_LAST : DB_SET);
+	    err == 0 && log_compare(&log->xa_lsn, &log->xa_first) > 0;
+	    err = log_get(log, &log->xa_lsn, &data, DB_PREV)) {
+		memcpy(&rectype, data.data, sizeof(rectype));
+
+		/*
+		 * The only record type we care about is an DB_txn_xa_regop.
+		 * If it's a commit, we have to add it to a txnlist.  If it's
+		 * a prepare, and we don't have a commit, then we return it.
+		 * We are redoing some of what's in the xa_regop_recovery
+		 * code, but we have to do it here so we can get at the xid
+		 * in the record.
+		 */
+		if (rectype != DB_txn_xa_regop && rectype != DB_txn_regop)
+			continue;
+
+		memcpy(&txnid, (u_int8_t *)data.data + sizeof(rectype),
+		    sizeof(txnid));
+		err = __db_txnlist_find(log->xa_info, txnid);
+		switch (rectype) {
+		case DB_txn_regop:
+			if (err == DB_NOTFOUND)
+				__db_txnlist_add(log->xa_info, txnid);
+			err = 0;
+			break;
+		case DB_txn_xa_regop:
+			/*
+			 * This transaction is commited, so we needn't read
+			 * the record and do anything.
+			 */
+			if (err == 0)
+				break;
+			if ((err =
+			    __txn_xa_regop_read(data.data, &argp)) != 0) {
+				ret = XAER_RMERR;
+				goto out;
+			}
+
+			xidp->formatID = argp->formatID;
+			xidp->gtrid_length = argp->gtrid;
+			xidp->bqual_length = argp->bqual;
+			memcpy(xidp->data, argp->xid.data, argp->xid.size);
+			ret++;
+			xidp++;
+			__os_free(argp, sizeof(*argp));
+			if (ret == count)
+				goto done;
+			break;
+		}
+	}
+
+	if (err != 0 && err != DB_NOTFOUND)
+		goto out;
+
+done:	if (LF_ISSET(TMENDRSCAN)) {
+		ZERO_LSN(log->xa_lsn);
+		ZERO_LSN(log->xa_first);
+
+out:		__db_txnlist_end(log->xa_info);
+		log->xa_info = NULL;
+	}
+	return (ret);
+
+err3:	(void)__db_unmap_rmid(rmid);
+err2:	(void)db_appexit(env);
+err1:	__os_free(env, sizeof(DB_ENV));
+	return (XAER_RMERR);
+}
+
+/*
+ * __db_xa_rollback
+ *	Abort an XA transaction.
+ */
+static int
+__db_xa_rollback(xid, rmid, flags)
+	XID *xid;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	TXN_DETAIL *td;
+	size_t off;
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+	if (flags != TMNOFLAGS)
+		return (XAER_INVAL);
+
+	if (__db_rmid_to_env(rmid, &env, 1) != 0)
+		return (XAER_PROTO);
+
+	if (__db_xid_to_txn(env, xid, &off) != 0)
+		return (XAER_NOTA);
+
+	td = (TXN_DETAIL *)((u_int8_t *)env->tx_info->region + off);
+
+	if (td->xa_status == TXN_XA_DEADLOCKED)
+		return (XA_RBDEADLOCK);
+
+	if (td->xa_status == TXN_XA_ABORTED)
+		return (XA_RBOTHER);
+
+	if (LF_SET(TMONEPHASE) &&
+	    td->xa_status != TXN_XA_ENDED && td->xa_status != TXN_XA_SUSPENDED)
+		return (XAER_PROTO);
+
+	if (!LF_SET(TMONEPHASE) && td->xa_status != TXN_XA_PREPARED)
+		return (XAER_PROTO);
+
+	/* Now, fill in the global transaction structure. */
+	__xa_txn_init(env, td, off);
+	if (txn_abort(env->xa_txn) != 0)
+		return (XAER_RMERR);
+
+	/* No fatal value that would require an XAER_RMFAIL. */
+	__xa_txn_end(env);
+	return (XA_OK);
+}
+
+/*
+ * __db_xa_forget --
+ *	Forget about an XID for a transaction that was heuristically
+ * completed.  Since we do not heuristically complete anything, I
+ * don't think we have to do anything here, but we should make sure
+ * that we reclaim the slots in the txnid table.
+ */
+static int
+__db_xa_forget(xid, rmid, flags)
+	XID *xid;
+	int rmid;
+	long flags;
+{
+	DB_ENV *env;
+	size_t off;
+
+	if (LF_ISSET(TMASYNC))
+		return (XAER_ASYNC);
+	if (flags != TMNOFLAGS)
+		return (XAER_INVAL);
+
+	if (__db_rmid_to_env(rmid, &env, 1) != 0)
+		return (XAER_PROTO);
+
+	/*
+	 * If mapping is gone, then we're done.
+	 */
+	if (__db_xid_to_txn(env, xid, &off) != 0)
+		return (XA_OK);
+
+	__db_unmap_xid(env, xid, off);
+
+	/* No fatal value that would require an XAER_RMFAIL. */
+	return (XA_OK);
+}
+
+/*
+ * __db_xa_complete --
+ *	Used to wait for asynchronous operations to complete.  Since we're
+ *	not doing asynch, this is an invalid operation.
+ */
+static int
+__db_xa_complete(handle, retval, rmid, flags)
+	int *handle, *retval, rmid;
+	long flags;
+{
+	COMPQUIET(handle, NULL);
+	COMPQUIET(retval, NULL);
+	COMPQUIET(rmid, 0);
+	COMPQUIET(flags, 0);
+
+	return (XAER_INVAL);
+}
+
+/*
+ * __xa_txn_init --
+ * 	Fill in the fields of the local transaction structure given
+ *	the detail transaction structure.
+ */
+static void
+__xa_txn_init(env, td, off)
+	DB_ENV *env;
+	TXN_DETAIL *td;
+	size_t off;
+{
+	DB_TXN *txn;
+
+	txn = env->xa_txn;
+	txn->mgrp = env->tx_info;
+	txn->parent = NULL;
+	txn->last_lsn = td->last_lsn;
+	txn->txnid = td->txnid;
+	txn->off = off;
+	txn->flags = 0;
+}
+
+/*
+ * __xa_txn_end --
+ * 	Invalidate a transaction structure that was generated by xa_txn_init.
+ */
+static void
+__xa_txn_end(env)
+	DB_ENV *env;
+{
+	DB_TXN *txn;
+
+	txn = env->xa_txn;
+	if (txn != NULL)
+		txn->txnid = TXN_INVALID;
+}
+
diff --git a/db2/xa/xa_db.c b/db2/xa/xa_db.c
new file mode 100644
index 0000000000..4aaaeff108
--- /dev/null
+++ b/db2/xa/xa_db.c
@@ -0,0 +1,308 @@
+/*-
+ * See the file LICENSE for redistribution information.
+ *
+ * Copyright (c) 1998
+ *	Sleepycat Software.  All rights reserved.
+ */
+
+#include "config.h"
+
+#ifndef lint
+static const char sccsid[] = "@(#)xa_db.c	10.6 (Sleepycat) 12/19/98";
+#endif /* not lint */
+
+#ifndef NO_SYSTEM_INCLUDES
+#include <sys/types.h>
+
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#endif
+
+#undef stat
+
+#include "db_int.h"
+#include "db_page.h"
+#include "xa.h"
+#include "xa_ext.h"
+#include "db_am.h"
+#include "db_ext.h"
+#include "common_ext.h"
+
+static int __xa_c_close __P((DBC *));
+static int __xa_c_del __P((DBC *, u_int32_t));
+static int __xa_c_get __P((DBC *, DBT *, DBT *, u_int32_t));
+static int __xa_c_put __P((DBC *, DBT *, DBT *, u_int32_t));
+static int __xa_close __P((DB *, u_int32_t));
+static int __xa_cursor __P((DB *, DB_TXN *, DBC **, u_int32_t));
+static int __xa_del __P((DB *, DB_TXN *, DBT *, u_int32_t));
+static int __xa_fd __P((DB *, int *));
+static int __xa_get __P((DB *, DB_TXN *, DBT *, DBT *, u_int32_t));
+static int __xa_put __P((DB *, DB_TXN *, DBT *, DBT *, u_int32_t));
+static int __xa_stat __P((DB *, void *, void *(*)(size_t), u_int32_t));
+static int __xa_sync __P((DB *, u_int32_t));
+
+int
+db_xa_open(fname, type, flags, mode, dbinfo, dbpp)
+	const char *fname;
+	DBTYPE type;
+	u_int32_t flags;
+	int mode;
+	DB_INFO *dbinfo;
+	DB **dbpp;
+{
+	DB *dbp, *real_dbp;
+	DB_ENV *dbenv;
+	struct __rmname *rp;
+	int ret;
+
+	/*
+	 * First try to open up the underlying DB.
+	 *
+	 * !!!
+	 * The dbenv argument is taken from the global list of environments.
+	 * When the transaction manager called xa_start() (__db_xa_start()),
+	 * the "current" DB environment was moved to the start of the list.
+	 * However, if we were called in a tpsvrinit function (which is
+	 * entirely plausible), then it's possible that xa_open was called
+	 * (which simply recorded the name of the environment to open) and
+	 * this is the next call into DB.  In that case, we still have to
+	 * open the environment.
+	 *
+	 * The way that we know that xa_open and nothing else was called
+	 * is because the nameq is not NULL.
+	 */
+	if ((rp = TAILQ_FIRST(&DB_GLOBAL(db_nameq))) != NULL &&
+	    (ret = __db_rmid_to_env(rp->rmid, &dbenv, 1)) != 0)
+		return (ret);
+
+	dbenv = TAILQ_FIRST(&DB_GLOBAL(db_envq));
+	if ((ret = db_open(fname,
+	    type, flags, mode, dbenv, dbinfo, &real_dbp)) != 0)
+		return (ret);
+
+	/*
+	 * Allocate our own DB handle, and copy the exported fields and
+	 * function pointers into it.  The internal pointer references
+	 * the real underlying DB handle.
+	 */
+	if ((ret = __os_calloc(1, sizeof(DB), &dbp)) != 0) {
+		(void)real_dbp->close(real_dbp, 0);
+		return (ret);
+	}
+	dbp->type = real_dbp->type;
+	dbp->byteswapped = real_dbp->byteswapped;
+	dbp->dbenv = dbenv;
+	dbp->internal = real_dbp;
+	TAILQ_INIT(&dbp->active_queue);
+	TAILQ_INIT(&dbp->free_queue);
+	dbp->close = __xa_close;
+	dbp->cursor = __xa_cursor;
+	dbp->del = __xa_del;
+	dbp->fd = __xa_fd;
+	dbp->get = __xa_get;
+	dbp->join = real_dbp->join;
+	dbp->put = __xa_put;
+	dbp->stat = __xa_stat;
+	dbp->sync = __xa_sync;
+
+	*dbpp = dbp;
+	return (0);
+}
+
+static int
+__xa_close(dbp, flags)
+	DB *dbp;
+	u_int32_t flags;
+{
+	DB *real_dbp;
+	DBC *dbc;
+	int ret;
+
+	/* Close any associated cursors. */
+	while ((dbc = TAILQ_FIRST(&dbp->active_queue)) != NULL)
+		(void)dbc->c_close(dbc);
+
+	/* Close the DB handle. */
+	real_dbp = (DB *)dbp->internal;
+	ret = real_dbp->close(real_dbp, flags);
+
+	__os_free(dbp, sizeof(DB));
+	return (ret);
+}
+
+static int
+__xa_cursor(dbp, txn, dbcp, flags)
+	DB *dbp;
+	DB_TXN *txn;
+	DBC **dbcp;
+	u_int32_t flags;
+{
+	DB *real_dbp;
+	DBC *real_dbc, *dbc;
+	int ret;
+
+	real_dbp = (DB *)dbp->internal;
+	txn = dbp->dbenv->xa_txn;
+
+	if ((ret = real_dbp->cursor(real_dbp, txn, &real_dbc, flags)) != 0)
+		return (ret);
+
+	/*
+	 * Allocate our own DBC handle, and copy the exported fields and
+	 * function pointers into it.  The internal pointer references
+	 * the real underlying DBC handle.
+	 */
+	if ((ret = __os_calloc(1, sizeof(DBC), &dbc)) != 0) {
+		(void)real_dbc->c_close(real_dbc);
+		return (ret);
+	}
+	dbc->dbp = dbp;
+	dbc->c_close = __xa_c_close;
+	dbc->c_del = __xa_c_del;
+	dbc->c_get = __xa_c_get;
+	dbc->c_put = __xa_c_put;
+	dbc->internal = real_dbc;
+	TAILQ_INSERT_TAIL(&dbp->active_queue, dbc, links);
+
+	*dbcp = dbc;
+	return (0);
+}
+
+static int
+__xa_fd(dbp, fdp)
+	DB *dbp;
+	int *fdp;
+{
+	DB *real_dbp;
+
+	COMPQUIET(fdp, NULL);
+
+	real_dbp = (DB *)dbp->internal;
+	return (__db_eopnotsup(real_dbp->dbenv));
+}
+
+static int
+__xa_del(dbp, txn, key, flags)
+	DB *dbp;
+	DB_TXN *txn;
+	DBT *key;
+	u_int32_t flags;
+{
+	DB *real_dbp;
+
+	real_dbp = (DB *)dbp->internal;
+	txn = dbp->dbenv->xa_txn;
+
+	return (real_dbp->del(real_dbp, txn, key, flags));
+}
+
+static int
+__xa_get(dbp, txn, key, data, flags)
+	DB *dbp;
+	DB_TXN *txn;
+	DBT *key;
+	DBT *data;
+	u_int32_t flags;
+{
+	DB *real_dbp;
+
+	real_dbp = (DB *)dbp->internal;
+	txn = dbp->dbenv->xa_txn;
+
+	return (real_dbp->get(real_dbp, txn, key, data, flags));
+}
+
+static int
+__xa_put(dbp, txn, key, data, flags)
+	DB *dbp;
+	DB_TXN *txn;
+	DBT *key;
+	DBT *data;
+	u_int32_t flags;
+{
+	DB *real_dbp;
+
+	real_dbp = (DB *)dbp->internal;
+	txn = dbp->dbenv->xa_txn;
+
+	return (real_dbp->put(real_dbp, txn, key, data, flags));
+}
+
+static int
+__xa_stat(dbp, spp, db_malloc, flags)
+	DB *dbp;
+	void *spp;
+	void *(*db_malloc) __P((size_t));
+	u_int32_t flags;
+{
+	DB *real_dbp;
+
+	real_dbp = (DB *)dbp->internal;
+	return (real_dbp->stat(real_dbp, spp, db_malloc, flags));
+}
+
+static int
+__xa_sync(dbp, flags)
+	DB *dbp;
+	u_int32_t flags;
+{
+	DB *real_dbp;
+
+	real_dbp = (DB *)dbp->internal;
+	return (real_dbp->sync(real_dbp, flags));
+}
+
+static int
+__xa_c_close(dbc)
+	DBC *dbc;
+{
+	DBC *real_dbc;
+	int ret;
+
+	real_dbc = (DBC *)dbc->internal;
+
+	ret = real_dbc->c_close(real_dbc);
+
+	TAILQ_REMOVE(&dbc->dbp->active_queue, dbc, links);
+	__os_free(dbc, sizeof(DBC));
+
+	return (ret);
+}
+
+static int
+__xa_c_del(dbc, flags)
+	DBC *dbc;
+	u_int32_t flags;
+{
+	DBC *real_dbc;
+
+	real_dbc = (DBC *)dbc->internal;
+	return (real_dbc->c_del(real_dbc, flags));
+}
+
+static int
+__xa_c_get(dbc, key, data, flags)
+	DBC *dbc;
+	DBT *key;
+	DBT *data;
+	u_int32_t flags;
+{
+	DBC *real_dbc;
+
+	real_dbc = (DBC *)dbc->internal;
+	return (real_dbc->c_get(real_dbc, key, data, flags));
+}
+
+static int
+__xa_c_put(dbc, key, data, flags)
+	DBC *dbc;
+	DBT *key;
+	DBT *data;
+	u_int32_t flags;
+{
+	DBC *real_dbc;
+
+	real_dbc = (DBC *)dbc->internal;
+	return (real_dbc->c_put(real_dbc, key, data, flags));
+}
diff --git a/db2/xa/xa_map.c b/db2/xa/xa_map.c
new file mode 100644
index 0000000000..d4ebbae22f
--- /dev/null
+++ b/db2/xa/xa_map.c
@@ -0,0 +1,305 @@
+/*-
+ * See the file LICENSE for redistribution information.
+ *
+ * Copyright (c) 1996, 1997, 1998
+ *	Sleepycat Software.  All rights reserved.
+ */
+
+#include "config.h"
+
+#ifndef lint
+static const char sccsid[] = "@(#)xa_map.c	10.4 (Sleepycat) 10/20/98";
+#endif /* not lint */
+
+#ifndef NO_SYSTEM_INCLUDES
+#include <sys/types.h>
+
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#endif
+
+#include "db_int.h"
+#include "shqueue.h"
+#include "txn.h"
+
+/*
+ * This file contains all the mapping information that we need to support
+ * the DB/XA interface.
+ */
+
+/*
+ * __db_rmid_to_env
+ *	Return the environment associated with a given XA rmid.
+ *
+ * PUBLIC: int __db_rmid_to_env __P((int rmid, DB_ENV **envp, int open_ok));
+ */
+int
+__db_rmid_to_env(rmid, envp, open_ok)
+	int rmid;
+	DB_ENV **envp;
+	int open_ok;
+{
+	DB_ENV *env;
+	char *dbhome;
+
+	env = TAILQ_FIRST(&DB_GLOBAL(db_envq));
+	if (env != NULL && env->xa_rmid == rmid) {
+		*envp = env;
+		return (0);
+	}
+
+	/*
+	 * When we map an rmid, move that environment to be the first one in
+	 * the list of environments, so we pass the correct environment from
+	 * the upcoming db_xa_open() call into db_open().
+	 */
+	for (; env != NULL; env = TAILQ_NEXT(env, links))
+		if (env->xa_rmid == rmid) {
+			TAILQ_REMOVE(&DB_GLOBAL(db_envq), env, links);
+			TAILQ_INSERT_HEAD(&DB_GLOBAL(db_envq), env, links);
+			*envp = env;
+			return (0);
+		}
+
+	/*
+	 * We have not found the rmid on the environment list.  If we
+	 * are allowed to do an open, search for the rmid on the name
+	 * list and, if we find it, then open it.
+	 */
+	if (!open_ok)
+		return (1);
+
+	if (__db_rmid_to_name(rmid, &dbhome) != 0)
+		return (1);
+#undef XA_FLAGS
+#define	XA_FLAGS \
+	DB_CREATE | DB_INIT_LOCK | DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_TXN
+
+	if (__os_calloc(1, sizeof(DB_ENV), &env) != 0)
+		return (1);
+
+	if (db_appinit(dbhome, NULL, env, XA_FLAGS) != 0) 
+		goto err;
+
+	if (__db_map_rmid(rmid, env) != 0)
+		goto err1;
+
+	__db_unmap_rmid_name(rmid);
+
+	*envp = env;
+	return (0);
+
+err1:	(void)db_appexit(env);
+err:	__os_free(env, sizeof(DB_ENV));
+	return (1);
+}
+
+/*
+ * __db_xid_to_txn
+ *	Return the txn that corresponds to this XID.
+ *
+ * PUBLIC: int __db_xid_to_txn __P((DB_ENV *, XID *, size_t *));
+ */
+int
+__db_xid_to_txn(dbenv, xid, offp)
+	DB_ENV *dbenv;
+	XID *xid;
+	size_t *offp;
+{
+	DB_TXNREGION *tmr;
+	struct __txn_detail *td;
+
+	/*
+	 * Search the internal active transaction table to find the
+	 * matching xid.  If this is a performance hit, then we
+	 * can create a hash table, but I doubt it's worth it.
+	 */
+	tmr = dbenv->tx_info->region;
+
+	LOCK_TXNREGION(dbenv->tx_info);
+	for (td = SH_TAILQ_FIRST(&tmr->active_txn, __txn_detail);
+	    td != NULL;
+	    td = SH_TAILQ_NEXT(td, links, __txn_detail))
+		if (memcmp(xid->data, td->xid, XIDDATASIZE) == 0)
+			break;
+	UNLOCK_TXNREGION(dbenv->tx_info);
+
+	if (td == NULL)
+		return (EINVAL);
+
+	*offp = (u_int8_t *)td - (u_int8_t *)tmr;
+	return (0);
+}
+
+/*
+ * __db_map_rmid
+ *	Create a mapping between the specified rmid and environment.
+ *
+ * PUBLIC: int __db_map_rmid __P((int, DB_ENV *));
+ */
+int
+__db_map_rmid(rmid, env)
+	int rmid;
+	DB_ENV *env;
+{
+	if (__os_calloc(1, sizeof(DB_TXN), &env->xa_txn) != 0)
+		return (XAER_RMERR);
+	env->xa_txn->txnid = TXN_INVALID;
+	env->xa_rmid = rmid;
+	TAILQ_INSERT_HEAD(&DB_GLOBAL(db_envq), env, links);
+	return (XA_OK);
+}
+
+/*
+ * __db_unmap_rmid
+ *	Destroy the mapping for the given rmid.
+ *
+ * PUBLIC: int __db_unmap_rmid __P((int));
+ */
+int
+__db_unmap_rmid(rmid)
+	int rmid;
+{
+	DB_ENV *e;
+
+	for (e = TAILQ_FIRST(&DB_GLOBAL(db_envq));
+	    e->xa_rmid != rmid;
+	    e = TAILQ_NEXT(e, links));
+
+	if (e == NULL)
+		return (EINVAL);
+
+	TAILQ_REMOVE(&DB_GLOBAL(db_envq), e, links);
+	if (e->xa_txn != NULL)
+		__os_free(e->xa_txn, sizeof(DB_TXN));
+	return (0);
+}
+
+/*
+ * __db_map_xid
+ *	Create a mapping between this XID and the transaction at
+ *	"off" in the shared region.
+ *
+ * PUBLIC: int __db_map_xid __P((DB_ENV *, XID *, size_t));
+ */
+int
+__db_map_xid(env, xid, off)
+	DB_ENV *env;
+	XID *xid;
+	size_t off;
+{
+	DB_TXNMGR *tm;
+	TXN_DETAIL *td;
+
+	tm = env->tx_info;
+	td = (TXN_DETAIL *)((u_int8_t *)tm->region + off);
+
+	LOCK_TXNREGION(tm);
+	memcpy(td->xid, xid->data, XIDDATASIZE);
+	UNLOCK_TXNREGION(tm);
+
+	return (0);
+}
+
+/*
+ * __db_unmap_xid
+ *	Destroy the mapping for the specified XID.
+ *
+ * PUBLIC: void __db_unmap_xid __P((DB_ENV *, XID *, size_t));
+ */
+
+void
+__db_unmap_xid(env, xid, off)
+	DB_ENV *env;
+	XID *xid;
+	size_t off;
+{
+	TXN_DETAIL *td;
+
+	COMPQUIET(xid, NULL);
+
+	td = (TXN_DETAIL *)((u_int8_t *)env->tx_info->region + off);
+	memset(td->xid, 0, sizeof(td->xid));
+}
+
+/*
+ * __db_map_rmid_name --
+ * 	Create a mapping from an rmid to a name (the xa_info argument).
+ * We use this during create and then at some later point when we are
+ * trying to map an rmid, we might indicate that it's OK to do an open
+ * in which case, we'll get the xa_info parameter from here and then
+ * free it up.
+ *
+ * PUBLIC: int __db_map_rmid_name __P((int, char *));
+ */
+
+int
+__db_map_rmid_name(rmid, dbhome)
+	int rmid;
+	char *dbhome;
+{
+	struct __rmname *entry;
+	int ret;
+
+	if ((ret = __os_malloc(sizeof(struct __rmname), NULL, &entry)) != 0)
+		return (ret);
+
+	if ((ret = __os_strdup(dbhome, &entry->dbhome)) != 0) {
+		__os_free(entry, sizeof(struct __rmname));
+		return (ret);
+	}
+
+	entry->rmid = rmid;
+
+	TAILQ_INSERT_HEAD(&DB_GLOBAL(db_nameq), entry, links);
+	return (0);
+}
+
+/*
+ * __db_rmid_to_name --
+ *	Given an rmid, return the name of the home directory for that
+ * rmid.
+ *
+ * PUBLIC: int __db_rmid_to_name __P((int, char **));
+ */
+int
+__db_rmid_to_name(rmid, dbhomep)
+	int rmid;
+	char **dbhomep;
+{
+	struct __rmname *np;
+
+	for (np = TAILQ_FIRST(&DB_GLOBAL(db_nameq)); np != NULL;
+	    np = TAILQ_NEXT(np, links)) {
+		if (np->rmid == rmid) {
+			*dbhomep = np->dbhome;
+			return (0);
+		}
+	}
+	return (1);
+}
+
+/*
+ * __db_unmap_rmid_name --
+ *	Given an rmid, remove its entry from the name list.
+ *
+ * PUBLIC:  void __db_unmap_rmid_name __P((int));
+ */
+void
+__db_unmap_rmid_name(rmid)
+	int rmid;
+{
+	struct __rmname *np, *next;
+
+	for (np = TAILQ_FIRST(&DB_GLOBAL(db_nameq)); np != NULL; np = next) {
+		next = TAILQ_NEXT(np, links);
+		if (np->rmid == rmid) {
+			TAILQ_REMOVE(&DB_GLOBAL(db_nameq), np, links);
+			__os_freestr(np->dbhome);
+			__os_free(np, sizeof(struct __rmname));
+			return;
+		}
+	}
+	return;
+}