s3: Implement global locks in a g_lock tdb

This is the basis to implement global locks in ctdb without depending on a
shared file system. The initial goal is to make ctdb persistent transactions
deterministic without too many timeouts.

11 files changed, 950 insertions, 7 deletions

diff --git a/source3/Makefile.in b/source3/Makefile.in
index dcd133c2d0..aad1bf5d57 100644
--- a/source3/Makefile.in
+++ b/source3/Makefile.in
@@ -266,6 +266,7 @@ EXTRA_ALL_TARGETS = @EXTRA_ALL_TARGETS@
 TDB_LIB_OBJ = lib/util_tdb.o ../lib/util/util_tdb.o \
 	  lib/dbwrap.o lib/dbwrap_tdb.o \
 	  lib/dbwrap_ctdb.o \
+	  lib/g_lock.o \
 	  lib/dbwrap_rbt.o
 
 TDB_VALIDATE_OBJ = lib/tdb_validate.o
@@ -1010,6 +1011,7 @@ NET_OBJ1 = utils/net.o utils/net_ads.o utils/net_help.o \
 	   utils/net_conf.o utils/net_join.o utils/net_user.o \
 	   utils/net_group.o utils/net_file.o utils/net_registry.o \
 	   auth/token_util.o utils/net_dom.o utils/net_share.o \
+	   utils/net_g_lock.o \
 	   utils/net_eventlog.o
 
 # these are not processed by make proto
diff --git a/source3/include/ctdbd_conn.h b/source3/include/ctdbd_conn.h
index d721235336..96e7aa08a6 100644
--- a/source3/include/ctdbd_conn.h
+++ b/source3/include/ctdbd_conn.h
@@ -73,5 +73,7 @@ NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
 			     uint64_t srvid, uint32_t flags, TDB_DATA data, 
 			     TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
 			     int *cstatus);
+NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn);
+NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn);
 
 #endif /* _CTDBD_CONN_H */
diff --git a/source3/include/g_lock.h b/source3/include/g_lock.h
new file mode 100644
index 0000000000..c0eed38d4b
--- /dev/null
+++ b/source3/include/g_lock.h
@@ -0,0 +1,55 @@
+/*
+   Unix SMB/CIFS implementation.
+   global locks based on ctdb
+   Copyright (C) 2009 by Volker Lendecke
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef _G_LOCK_H_
+#define _G_LOCK_H_
+
+#include "dbwrap.h"
+
+struct g_lock_ctx;
+
+enum g_lock_type {
+	G_LOCK_READ = 0,
+	G_LOCK_WRITE = 1,
+};
+
+/*
+ * Or'ed with g_lock_type
+ */
+#define G_LOCK_PENDING (2)
+
+struct g_lock_ctx *g_lock_ctx_init(TALLOC_CTX *mem_ctx,
+				   struct messaging_context *msg);
+
+NTSTATUS g_lock_lock(struct g_lock_ctx *ctx, const char *name,
+		     enum g_lock_type lock_type, struct timeval timeout);
+NTSTATUS g_lock_unlock(struct g_lock_ctx *ctx, const char *name);
+NTSTATUS g_lock_get(struct g_lock_ctx *ctx, const char *name,
+		struct server_id *pid);
+
+int g_lock_locks(struct g_lock_ctx *ctx,
+		 int (*fn)(const char *name, void *private_data),
+		 void *private_data);
+NTSTATUS g_lock_dump(struct g_lock_ctx *ctx, const char *name,
+		     int (*fn)(struct server_id pid,
+			       enum g_lock_type lock_type,
+			       void *private_data),
+		     void *private_data);
+
+#endif
diff --git a/source3/lib/ctdbd_conn.c b/source3/lib/ctdbd_conn.c
index 8ddb12a765..900fa345a1 100644
--- a/source3/lib/ctdbd_conn.c
+++ b/source3/lib/ctdbd_conn.c
@@ -361,10 +361,18 @@ static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
 			goto next_pkt;
 		}
 
-		if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
-			DEBUG(0,("Got cluster reconfigure message in ctdb_read_req\n"));
+		if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
+		    || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
+
+			DEBUG(1, ("ctdb_read_req: Got %s message\n",
+				  (msg->srvid == CTDB_SRVID_RECONFIGURE)
+				  ? "cluster reconfigure" : "SAMBA_NOTIFY"));
+
 			messaging_send(conn->msg_ctx, procid_self(),
 				       MSG_SMB_BRL_VALIDATE, &data_blob_null);
+			messaging_send(conn->msg_ctx, procid_self(),
+				       MSG_DBWRAP_G_LOCK_RETRY,
+				       &data_blob_null);
 			TALLOC_FREE(hdr);
 			goto next_pkt;
 		}
@@ -493,6 +501,11 @@ NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
 		goto fail;
 	}
 
+	status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
+	if (!NT_STATUS_IS_OK(status)) {
+		goto fail;
+	}
+
 	*pconn = conn;
 	return NT_STATUS_OK;
 
@@ -533,15 +546,21 @@ static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
 
 	SMB_ASSERT(conn->msg_ctx != NULL);
 
-	if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
+	if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
+	    || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
 		DEBUG(0,("Got cluster reconfigure message\n"));
 		/*
-		 * when the cluster is reconfigured, we need to clean the brl
-		 * database
+		 * when the cluster is reconfigured or someone of the
+		 * family has passed away (SAMBA_NOTIFY), we need to
+		 * clean the brl database
 		 */
 		messaging_send(conn->msg_ctx, procid_self(),
 			       MSG_SMB_BRL_VALIDATE, &data_blob_null);
 
+		messaging_send(conn->msg_ctx, procid_self(),
+			       MSG_DBWRAP_G_LOCK_RETRY,
+			       &data_blob_null);
+
 		TALLOC_FREE(buf);
 		return NT_STATUS_OK;
 	}
@@ -1302,6 +1321,50 @@ NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
 	return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
 }
 
+NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
+{
+	struct ctdb_client_notify_register reg_data;
+	size_t struct_len;
+	NTSTATUS status;
+	int cstatus;
+
+	reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
+	reg_data.len = 1;
+	reg_data.notify_data[0] = 0;
+
+	struct_len = offsetof(struct ctdb_client_notify_register,
+			      notify_data) + reg_data.len;
+
+	status = ctdbd_control_local(
+		conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
+		make_tdb_data((uint8_t *)&reg_data, struct_len),
+		NULL, NULL, &cstatus);
+	if (!NT_STATUS_IS_OK(status)) {
+		DEBUG(1, ("ctdbd_control_local failed: %s\n",
+			  nt_errstr(status)));
+	}
+	return status;
+}
+
+NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
+{
+	struct ctdb_client_notify_deregister dereg_data;
+	NTSTATUS status;
+	int cstatus;
+
+	dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
+
+	status = ctdbd_control_local(
+		conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
+		make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
+		NULL, NULL, &cstatus);
+	if (!NT_STATUS_IS_OK(status)) {
+		DEBUG(1, ("ctdbd_control_local failed: %s\n",
+			  nt_errstr(status)));
+	}
+	return status;
+}
+
 #else
 
 NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
diff --git a/source3/lib/g_lock.c b/source3/lib/g_lock.c
new file mode 100644
index 0000000000..6508b3995b
--- /dev/null
+++ b/source3/lib/g_lock.c
@@ -0,0 +1,594 @@
+/*
+   Unix SMB/CIFS implementation.
+   global locks based on dbwrap and messaging
+   Copyright (C) 2009 by Volker Lendecke
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "includes.h"
+#include "g_lock.h"
+
+static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
+				    struct server_id pid);
+
+struct g_lock_ctx {
+	struct db_context *db;
+	struct messaging_context *msg;
+};
+
+/*
+ * The "g_lock.tdb" file contains records, indexed by the 0-terminated
+ * lockname. The record contains an array of "struct g_lock_rec"
+ * structures. Waiters have the lock_type with G_LOCK_PENDING or'ed.
+ */
+
+struct g_lock_rec {
+	enum g_lock_type lock_type;
+	struct server_id pid;
+};
+
+struct g_lock_ctx *g_lock_ctx_init(TALLOC_CTX *mem_ctx,
+				   struct messaging_context *msg)
+{
+	struct g_lock_ctx *result;
+
+	result = talloc(mem_ctx, struct g_lock_ctx);
+	if (result == NULL) {
+		return NULL;
+	}
+	result->msg = msg;
+
+	result->db = db_open(result, lock_path("g_lock.tdb"), 0,
+			     TDB_CLEAR_IF_FIRST, O_RDWR|O_CREAT, 0700);
+	if (result->db == NULL) {
+		DEBUG(1, ("g_lock_init: Could not open g_lock.tdb"));
+		TALLOC_FREE(result);
+		return NULL;
+	}
+	return result;
+}
+
+static bool g_lock_conflicts(enum g_lock_type lock_type,
+			     const struct g_lock_rec *rec)
+{
+	enum g_lock_type rec_lock = rec->lock_type;
+
+	if ((rec_lock & G_LOCK_PENDING) != 0) {
+		return false;
+	}
+
+	/*
+	 * Only tested write locks so far. Very likely this routine
+	 * needs to be fixed for read locks....
+	 */
+	if ((lock_type == G_LOCK_READ) && (rec_lock == G_LOCK_READ)) {
+		return false;
+	}
+	return true;
+}
+
+static bool g_lock_parse(TALLOC_CTX *mem_ctx, TDB_DATA data,
+			 int *pnum_locks, struct g_lock_rec **plocks)
+{
+	int i, num_locks;
+	struct g_lock_rec *locks;
+
+	if ((data.dsize % sizeof(struct g_lock_rec)) != 0) {
+		DEBUG(1, ("invalid lock record length %d\n", (int)data.dsize));
+		return false;
+	}
+
+	num_locks = data.dsize / sizeof(struct g_lock_rec);
+	locks = talloc_array(mem_ctx, struct g_lock_rec, num_locks);
+	if (locks == NULL) {
+		DEBUG(1, ("talloc failed\n"));
+		return false;
+	}
+
+	memcpy(locks, data.dptr, data.dsize);
+
+	DEBUG(10, ("locks:\n"));
+	for (i=0; i<num_locks; i++) {
+		DEBUGADD(10, ("%s: %s %s\n",
+			      procid_str(talloc_tos(), &locks[i].pid),
+			      ((locks[i].lock_type & 1) == G_LOCK_READ) ?
+			      "read" : "write",
+			      (locks[i].lock_type & G_LOCK_PENDING) ?
+			      "(pending)" : "(owner)"));
+
+		if (process_exists(locks[i].pid)) {
+			continue;
+		}
+		DEBUGADD(10, ("%s does not exist -- discarding\n",
+			      procid_str(talloc_tos(), &locks[i].pid)));
+
+		if (i < (num_locks-1)) {
+			locks[i] = locks[num_locks-1];
+		}
+		num_locks -= 1;
+	}
+
+	*plocks = locks;
+	*pnum_locks = num_locks;
+	return true;
+}
+
+static struct g_lock_rec *g_lock_addrec(TALLOC_CTX *mem_ctx,
+					struct g_lock_rec *locks,
+					int num_locks,
+					const struct server_id pid,
+					enum g_lock_type lock_type)
+{
+	struct g_lock_rec *result;
+
+	result = talloc_realloc(mem_ctx, locks, struct g_lock_rec,
+				num_locks+1);
+	if (result == NULL) {
+		return NULL;
+	}
+
+	result[num_locks].pid = pid;
+	result[num_locks].lock_type = lock_type;
+	return result;
+}
+
+static void g_lock_got_retry(struct messaging_context *msg,
+			     void *private_data,
+			     uint32_t msg_type,
+			     struct server_id server_id,
+			     DATA_BLOB *data);
+static void g_lock_timedout(struct tevent_context *ev,
+			    struct tevent_timer *te,
+			    struct timeval current_time,
+			    void *private_data);
+
+static NTSTATUS g_lock_trylock(struct g_lock_ctx *ctx, const char *name,
+			       enum g_lock_type lock_type)
+{
+	struct db_record *rec = NULL;
+	struct g_lock_rec *locks = NULL;
+	int i, num_locks;
+	struct server_id self;
+	int our_index;
+	TDB_DATA data;
+	NTSTATUS status = NT_STATUS_OK;
+	NTSTATUS store_status;
+
+again:
+	rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
+				    string_term_tdb_data(name));
+	if (rec == NULL) {
+		DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
+		status = NT_STATUS_LOCK_NOT_GRANTED;
+		goto done;
+	}
+
+	if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
+		DEBUG(10, ("g_lock_parse for %s failed\n", name));
+		status = NT_STATUS_INTERNAL_ERROR;
+		goto done;
+	}
+
+	self = procid_self();
+	our_index = -1;
+
+	for (i=0; i<num_locks; i++) {
+		if (procid_equal(&self, &locks[i].pid)) {
+			if (our_index != -1) {
+				DEBUG(1, ("g_lock_trylock: Added ourself "
+					  "twice!\n"));
+				status = NT_STATUS_INTERNAL_ERROR;
+				goto done;
+			}
+			if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
+				DEBUG(1, ("g_lock_trylock: Found ourself not "
+					  "pending!\n"));
+				status = NT_STATUS_INTERNAL_ERROR;
+				goto done;
+			}
+
+			our_index = i;
+
+			/* never conflict with ourself */
+			continue;
+		}
+		if (g_lock_conflicts(lock_type, &locks[i])) {
+			struct server_id pid = locks[i].pid;
+
+			if (!process_exists(pid)) {
+				TALLOC_FREE(locks);
+				TALLOC_FREE(rec);
+				status = g_lock_force_unlock(ctx, name, pid);
+				if (!NT_STATUS_IS_OK(status)) {
+					DEBUG(1, ("Could not unlock dead lock "
+						  "holder!\n"));
+					goto done;
+				}
+				goto again;
+			}
+			lock_type |= G_LOCK_PENDING;
+		}
+	}
+
+	if (our_index == -1) {
+		/* First round, add ourself */
+
+		locks = g_lock_addrec(talloc_tos(), locks, num_locks,
+				      self, lock_type);
+		if (locks == NULL) {
+			DEBUG(10, ("g_lock_addrec failed\n"));
+			status = NT_STATUS_NO_MEMORY;
+			goto done;
+		}
+	} else {
+		/*
+		 * Retry. We were pending last time. Overwrite the
+		 * stored lock_type with what we calculated, we might
+		 * have acquired the lock this time.
+		 */
+		locks[our_index].lock_type = lock_type;
+	}
+
+	data = make_tdb_data((uint8_t *)locks, talloc_get_size(locks));
+	store_status = rec->store(rec, data, 0);
+	if (!NT_STATUS_IS_OK(store_status)) {
+		DEBUG(1, ("rec->store failed: %s\n",
+			  nt_errstr(store_status)));
+		status = store_status;
+	}
+
+done:
+	TALLOC_FREE(locks);
+	TALLOC_FREE(rec);
+
+	if (NT_STATUS_IS_OK(status) && (lock_type & G_LOCK_PENDING) != 0) {
+		return STATUS_PENDING;
+	}
+
+	return NT_STATUS_OK;
+}
+
+NTSTATUS g_lock_lock(struct g_lock_ctx *ctx, const char *name,
+		     enum g_lock_type lock_type, struct timeval timeout)
+{
+	struct tevent_timer *te = NULL;
+	NTSTATUS status;
+	bool retry = false;
+	bool timedout = false;
+
+	DEBUG(10, ("Trying to acquire lock %d for %s\n", (int)lock_type,
+		   name));
+
+	if (lock_type & ~1) {
+		DEBUG(1, ("Got invalid lock type %d for %s\n",
+			  (int)lock_type, name));
+		return NT_STATUS_INVALID_PARAMETER;
+	}
+
+#ifdef CLUSTER_SUPPORT
+	if (lp_clustering()) {
+		status = ctdb_watch_us(messaging_ctdbd_connection());
+		if (!NT_STATUS_IS_OK(status)) {
+			DEBUG(10, ("could not register retry with ctdb: %s\n",
+				   nt_errstr(status)));
+			goto done;
+		}
+	}
+#endif
+
+	status = messaging_register(ctx->msg, &retry, MSG_DBWRAP_G_LOCK_RETRY,
+				    g_lock_got_retry);
+	if (!NT_STATUS_IS_OK(status)) {
+		DEBUG(10, ("messaging_register failed: %s\n",
+			   nt_errstr(status)));
+		return status;
+	}
+again:
+	retry = false;
+
+	status = g_lock_trylock(ctx, name, lock_type);
+	if (NT_STATUS_IS_OK(status)) {
+		DEBUG(10, ("Got lock %s\n", name));
+		goto done;
+	}
+	if (!NT_STATUS_EQUAL(status, STATUS_PENDING)) {
+		DEBUG(10, ("g_lock_trylock failed: %s\n",
+			   nt_errstr(status)));
+		goto done;
+	}
+
+	if (retry) {
+		goto again;
+	}
+
+	DEBUG(10, ("g_lock_trylock: Did not get lock, waiting...\n"));
+
+	if (te == NULL) {
+		te = tevent_add_timer(
+			ctx->msg->event_ctx, talloc_tos(),
+			timeval_current_ofs(timeout.tv_sec, timeout.tv_usec),
+			g_lock_timedout, &timedout);
+		if (te == NULL) {
+			DEBUG(10, ("tevent_add_timer failed\n"));
+			status = NT_STATUS_NO_MEMORY;
+			goto done;
+		}
+	}
+
+	while (true) {
+		if (tevent_loop_once(ctx->msg->event_ctx) == -1) {
+			DEBUG(1, ("tevent_loop_once failed\n"));
+			status = NT_STATUS_INTERNAL_ERROR;
+			goto done;
+		}
+		if (retry) {
+			goto again;
+		}
+		if (timedout) {
+			DEBUG(10, ("g_lock_lock timed out\n"));
+
+			te = NULL;
+
+			status = NT_STATUS_LOCK_NOT_GRANTED;
+			goto done;
+		}
+	}
+done:
+
+	if (!NT_STATUS_IS_OK(status)) {
+		NTSTATUS unlock_status;
+
+		unlock_status = g_lock_unlock(ctx, name);
+
+		if (!NT_STATUS_IS_OK(unlock_status)) {
+			DEBUG(1, ("Could not remove ourself from the locking "
+				  "db: %s\n", nt_errstr(status)));
+		}
+	}
+
+	messaging_deregister(ctx->msg, MSG_DBWRAP_G_LOCK_RETRY, &retry);
+	TALLOC_FREE(te);
+
+	return status;
+}
+
+static void g_lock_got_retry(struct messaging_context *msg,
+			     void *private_data,
+			     uint32_t msg_type,
+			     struct server_id server_id,
+			     DATA_BLOB *data)
+{
+	bool *pretry = (bool *)private_data;
+
+	DEBUG(10, ("Got retry message from pid %s\n",
+		   procid_str(talloc_tos(), &server_id)));
+
+	*pretry = true;
+}
+
+static void g_lock_timedout(struct tevent_context *ev,
+			    struct tevent_timer *te,
+			    struct timeval current_time,
+			    void *private_data)
+{
+	bool *ptimedout = (bool *)private_data;
+	*ptimedout = true;
+	TALLOC_FREE(te);
+}
+
+static NTSTATUS g_lock_force_unlock(struct g_lock_ctx *ctx, const char *name,
+				    struct server_id pid)
+{
+	struct db_record *rec = NULL;
+	struct g_lock_rec *locks = NULL;
+	int i, num_locks;
+	enum g_lock_type lock_type;
+	NTSTATUS status;
+
+	rec = ctx->db->fetch_locked(ctx->db, talloc_tos(),
+				    string_term_tdb_data(name));
+	if (rec == NULL) {
+		DEBUG(10, ("fetch_locked(\"%s\") failed\n", name));
+		status = NT_STATUS_INTERNAL_ERROR;
+		goto done;
+	}
+
+	if (!g_lock_parse(talloc_tos(), rec->value, &num_locks, &locks)) {
+		DEBUG(10, ("g_lock_parse for %s failed\n", name));
+		status = NT_STATUS_INTERNAL_ERROR;
+		goto done;
+	}
+
+	for (i=0; i<num_locks; i++) {
+		if (procid_equal(&pid, &locks[i].pid)) {
+			break;
+		}
+	}
+
+	if (i == num_locks) {
+		DEBUG(10, ("g_lock_force_unlock: Lock not found\n"));
+		status = NT_STATUS_INTERNAL_ERROR;
+		goto done;
+	}
+
+	lock_type = locks[i].lock_type;
+
+	if (i < (num_locks-1)) {
+		locks[i] = locks[num_locks-1];
+	}
+	num_locks -= 1;
+
+	if (num_locks == 0) {
+		status = rec->delete_rec(rec);
+	} else {
+		TDB_DATA data;
+		data = make_tdb_data((uint8_t *)locks,
+				     sizeof(struct g_lock_rec) * num_locks);
+		status = rec->store(rec, data, 0);
+	}
+
+	if (!NT_STATUS_IS_OK(status)) {
+		DEBUG(1, ("g_lock_force_unlock: Could not store record: %s\n",
+			  nt_errstr(status)));
+		goto done;
+	}
+
+	if ((lock_type & G_LOCK_PENDING) == 0) {
+		/*
+		 * We've been the lock holder. Tell all others to retry.
+		 */
+		for (i=0; i<num_locks; i++) {
+			if ((locks[i].lock_type & G_LOCK_PENDING) == 0) {
+				continue;
+			}
+
+			/*
+			 * Ping all waiters to retry
+			 */
+			status = messaging_send(ctx->msg, locks[i].pid,
+						MSG_DBWRAP_G_LOCK_RETRY,
+						&data_blob_null);
+			if (!NT_STATUS_IS_OK(status)) {
+				DEBUG(1, ("sending retry to %s failed: %s\n",
+					  procid_str(talloc_tos(),
+						     &locks[i].pid),
+					  nt_errstr(status)));
+			}
+		}
+	}
+done:
+
+	TALLOC_FREE(locks);
+	TALLOC_FREE(rec);
+	return status;
+}
+
+NTSTATUS g_lock_unlock(struct g_lock_ctx *ctx, const char *name)
+{
+	NTSTATUS status;
+
+	status = g_lock_force_unlock(ctx, name, procid_self());
+
+#ifdef CLUSTER_SUPPORT
+	if (lp_clustering()) {
+		ctdb_unwatch(messaging_ctdbd_connection());
+	}
+#endif
+	return status;
+}
+
+struct g_lock_locks_state {
+	int (*fn)(const char *name, void *private_data);
+	void *private_data;
+};
+
+static int g_lock_locks_fn(struct db_record *rec, void *priv)
+{
+	struct g_lock_locks_state *state = (struct g_lock_locks_state *)priv;
+
+	if ((rec->key.dsize == 0) || (rec->key.dptr[rec->key.dsize-1] != 0)) {
+		DEBUG(1, ("invalid key in g_lock.tdb, ignoring\n"));
+		return 0;
+	}
+	return state->fn((char *)rec->key.dptr, state->private_data);
+}
+
+int g_lock_locks(struct g_lock_ctx *ctx,
+		 int (*fn)(const char *name, void *private_data),
+		 void *private_data)
+{
+	struct g_lock_locks_state state;
+
+	state.fn = fn;
+	state.private_data = private_data;
+
+	return ctx->db->traverse_read(ctx->db, g_lock_locks_fn, &state);
+}
+
+NTSTATUS g_lock_dump(struct g_lock_ctx *ctx, const char *name,
+		     int (*fn)(struct server_id pid,
+			       enum g_lock_type lock_type,
+			       void *private_data),
+		     void *private_data)
+{
+	TDB_DATA data;
+	int i, num_locks;
+	struct g_lock_rec *locks = NULL;
+	bool ret;
+
+	if (ctx->db->fetch(ctx->db, talloc_tos(), string_term_tdb_data(name),
+			   &data) != 0) {
+		return NT_STATUS_NOT_FOUND;
+	}
+
+	if ((data.dsize == 0) || (data.dptr == NULL)) {
+		return NT_STATUS_OK;
+	}
+
+	ret = g_lock_parse(talloc_tos(), data, &num_locks, &locks);
+
+	TALLOC_FREE(data.dptr);
+
+	if (!ret) {
+		DEBUG(10, ("g_lock_parse for %s failed\n", name));
+		return NT_STATUS_INTERNAL_ERROR;
+	}
+
+	for (i=0; i<num_locks; i++) {
+		if (fn(locks[i].pid, locks[i].lock_type, private_data) != 0) {
+			break;
+		}
+	}
+	TALLOC_FREE(locks);
+	return NT_STATUS_OK;
+}
+
+struct g_lock_get_state {
+	bool found;
+	struct server_id *pid;
+};
+
+static int g_lock_get_fn(struct server_id pid, enum g_lock_type lock_type,
+			 void *priv)
+{
+	struct g_lock_get_state *state = (struct g_lock_get_state *)priv;
+
+	if ((lock_type & G_LOCK_PENDING) != 0) {
+		return 0;
+	}
+
+	state->found = true;
+	*state->pid = pid;
+	return 1;
+}
+
+NTSTATUS g_lock_get(struct g_lock_ctx *ctx, const char *name,
+		    struct server_id *pid)
+{
+	struct g_lock_get_state state;
+	NTSTATUS status;
+
+	state.found = false;
+	state.pid = pid;
+
+	status = g_lock_dump(ctx, name, g_lock_get_fn, &state);
+	if (!NT_STATUS_IS_OK(status)) {
+		return status;
+	}
+	if (!state.found) {
+		return NT_STATUS_NOT_FOUND;
+	}
+	return NT_STATUS_OK;
+}
diff --git a/source3/librpc/gen_ndr/messaging.h b/source3/librpc/gen_ndr/messaging.h
index 225440a8fd..1312c845d4 100644
--- a/source3/librpc/gen_ndr/messaging.h
+++ b/source3/librpc/gen_ndr/messaging.h
@@ -62,7 +62,8 @@ enum messaging_type
 	MSG_WINBIND_VALIDATE_CACHE=(int)(0x0408),
 	MSG_WINBIND_DUMP_DOMAIN_LIST=(int)(0x0409),
 	MSG_DUMP_EVENT_LIST=(int)(0x0500),
-	MSG_DBWRAP_TDB2_CHANGES=(int)(4001)
+	MSG_DBWRAP_TDB2_CHANGES=(int)(4001),
+	MSG_DBWRAP_G_LOCK_RETRY=(int)(4002)
 }
 #else
  { __donnot_use_enum_messaging_type=0x7FFFFFFF}
@@ -118,6 +119,7 @@ enum messaging_type
 #define MSG_WINBIND_DUMP_DOMAIN_LIST ( 0x0409 )
 #define MSG_DUMP_EVENT_LIST ( 0x0500 )
 #define MSG_DBWRAP_TDB2_CHANGES ( 4001 )
+#define MSG_DBWRAP_G_LOCK_RETRY ( 4002 )
 #endif
 ;
 
diff --git a/source3/librpc/gen_ndr/ndr_messaging.c b/source3/librpc/gen_ndr/ndr_messaging.c
index 3e2aa1f3b0..1452630539 100644
--- a/source3/librpc/gen_ndr/ndr_messaging.c
+++ b/source3/librpc/gen_ndr/ndr_messaging.c
@@ -74,6 +74,7 @@ _PUBLIC_ void ndr_print_messaging_type(struct ndr_print *ndr, const char *name,
 		case MSG_WINBIND_DUMP_DOMAIN_LIST: val = "MSG_WINBIND_DUMP_DOMAIN_LIST"; break;
 		case MSG_DUMP_EVENT_LIST: val = "MSG_DUMP_EVENT_LIST"; break;
 		case MSG_DBWRAP_TDB2_CHANGES: val = "MSG_DBWRAP_TDB2_CHANGES"; break;
+		case MSG_DBWRAP_G_LOCK_RETRY: val = "MSG_DBWRAP_G_LOCK_RETRY"; break;
 	}
 	ndr_print_enum(ndr, name, "ENUM", val, r);
 }
diff --git a/source3/librpc/idl/messaging.idl b/source3/librpc/idl/messaging.idl
index 068658587f..08caa59508 100644
--- a/source3/librpc/idl/messaging.idl
+++ b/source3/librpc/idl/messaging.idl
@@ -88,7 +88,8 @@ interface messaging
 		MSG_DUMP_EVENT_LIST		= 0x0500,
 
 		/* dbwrap messages 4001-4999 */
-		MSG_DBWRAP_TDB2_CHANGES		= 4001
+		MSG_DBWRAP_TDB2_CHANGES		= 4001,
+		MSG_DBWRAP_G_LOCK_RETRY		= 4002
 	} messaging_type;
 
 	/* messaging struct sent across the sockets and stored in the tdb */
diff --git a/source3/utils/net.c b/source3/utils/net.c
index 85c3c7dced..0c5f0807ff 100644
--- a/source3/utils/net.c
+++ b/source3/utils/net.c
@@ -612,6 +612,13 @@ static struct functable net_func[] = {
 		N_("  Use 'net help lookup' to get more information about 'net "
 		   "lookup' commands.")
 	},
+	{	"g_lock",
+		net_g_lock,
+		NET_TRANSPORT_LOCAL,
+		N_("Manipulate the global lock table"),
+		N_("  Use 'net help g_lock' to get more information about "
+		   "'net g_lock' commands.")
+	},
 	{	"join",
 		net_join,
 		NET_TRANSPORT_ADS | NET_TRANSPORT_RPC,
diff --git a/source3/utils/net_g_lock.c b/source3/utils/net_g_lock.c
new file mode 100644
index 0000000000..f30ed331be
--- /dev/null
+++ b/source3/utils/net_g_lock.c
@@ -0,0 +1,213 @@
+/*
+ * Samba Unix/Linux SMB client library
+ * Interface to the g_lock facility
+ * Copyright (C) Volker Lendecke 2009
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "includes.h"
+#include "net.h"
+#include "g_lock.h"
+
+static bool net_g_lock_init(TALLOC_CTX *mem_ctx,
+			    struct tevent_context **pev,
+			    struct messaging_context **pmsg,
+			    struct g_lock_ctx **pg_ctx)
+{
+	struct tevent_context *ev = NULL;
+	struct messaging_context *msg = NULL;
+	struct g_lock_ctx *g_ctx = NULL;
+
+	ev = tevent_context_init(talloc_tos());
+	if (ev == NULL) {
+		d_fprintf(stderr, "ERROR: could not init event context\n");
+		goto fail;
+	}
+	msg = messaging_init(talloc_tos(), server_id_self(), ev);
+	if (msg == NULL) {
+		d_fprintf(stderr, "ERROR: could not init messaging context\n");
+		goto fail;
+	}
+	g_ctx = g_lock_ctx_init(talloc_tos(), msg);
+	if (g_ctx == NULL) {
+		d_fprintf(stderr, "ERROR: could not init g_lock context\n");
+		goto fail;
+	}
+
+	*pev = ev;
+	*pmsg = msg;
+	*pg_ctx = g_ctx;
+	return true;
+fail:
+	TALLOC_FREE(g_ctx);
+	TALLOC_FREE(msg);
+	TALLOC_FREE(ev);
+	return false;
+}
+
+
+static int net_g_lock_do(struct net_context *c, int argc, const char **argv)
+{
+	struct tevent_context *ev = NULL;
+	struct messaging_context *msg = NULL;
+	struct g_lock_ctx *g_ctx = NULL;
+	const char *name, *cmd;
+	int timeout, res;
+	bool locked = false;
+	NTSTATUS status;
+	int ret = -1;
+
+	if (argc != 3) {
+		d_printf("Usage: net g_lock do <lockname> <timeout> "
+			 "<command>\n");
+		return -1;
+	}
+	name = argv[0];
+	timeout = atoi(argv[1]);
+	cmd = argv[2];
+
+	if (!net_g_lock_init(talloc_tos(), &ev, &msg, &g_ctx)) {
+		goto done;
+	}
+
+	status = g_lock_lock(g_ctx, name, G_LOCK_WRITE,
+			     timeval_set(timeout / 1000, timeout % 1000));
+	if (!NT_STATUS_IS_OK(status)) {
+		d_fprintf(stderr, "ERROR: Could not get lock: %s\n",
+			  nt_errstr(status));
+		goto done;
+	}
+	locked = true;
+
+	res = system(cmd);
+
+	if (res == -1) {
+		d_fprintf(stderr, "ERROR: system() returned %s\n",
+			  strerror(errno));
+		goto done;
+	}
+	d_fprintf(stderr, "command returned %d\n", res);
+
+	ret = 0;
+
+done:
+	if (locked) {
+		g_lock_unlock(g_ctx, name);
+	}
+	TALLOC_FREE(g_ctx);
+	TALLOC_FREE(msg);
+	TALLOC_FREE(ev);
+	return ret;
+}
+
+static int net_g_lock_dump_fn(struct server_id pid, enum g_lock_type lock_type,
+			      void *private_data)
+{
+	char *pidstr;
+
+	pidstr = procid_str(talloc_tos(), &pid);
+	d_printf("%s: %s (%s)\n", pidstr,
+		 (lock_type & 1) ? "WRITE" : "READ",
+		 (lock_type & G_LOCK_PENDING) ? "pending" : "holder");
+	TALLOC_FREE(pidstr);
+	return 0;
+}
+
+static int net_g_lock_dump(struct net_context *c, int argc, const char **argv)
+{
+	struct tevent_context *ev = NULL;
+	struct messaging_context *msg = NULL;
+	struct g_lock_ctx *g_ctx = NULL;
+	NTSTATUS status;
+	int ret = -1;
+
+	if (argc != 1) {
+		d_printf("Usage: net g_lock dump <lockname>\n");
+		return -1;
+	}
+
+	if (!net_g_lock_init(talloc_tos(), &ev, &msg, &g_ctx)) {
+		goto done;
+	}
+
+	status = g_lock_dump(g_ctx, argv[0], net_g_lock_dump_fn, NULL);
+
+	ret = 0;
+done:
+	TALLOC_FREE(g_ctx);
+	TALLOC_FREE(msg);
+	TALLOC_FREE(ev);
+	return ret;
+}
+
+static int net_g_lock_locks_fn(const char *name, void *private_data)
+{
+	d_printf("%s\n", name);
+	return 0;
+}
+
+static int net_g_lock_locks(struct net_context *c, int argc, const char **argv)
+{
+	struct tevent_context *ev = NULL;
+	struct messaging_context *msg = NULL;
+	struct g_lock_ctx *g_ctx = NULL;
+	int ret = -1;
+
+	if (argc != 0) {
+		d_printf("Usage: net g_lock locks\n");
+		return -1;
+	}
+
+	if (!net_g_lock_init(talloc_tos(), &ev, &msg, &g_ctx)) {
+		goto done;
+	}
+
+	ret = g_lock_locks(g_ctx, net_g_lock_locks_fn, NULL);
+done:
+	TALLOC_FREE(g_ctx);
+	TALLOC_FREE(msg);
+	TALLOC_FREE(ev);
+	return ret;
+}
+
+int net_g_lock(struct net_context *c, int argc, const char **argv)
+{
+	struct functable func[] = {
+		{
+			"do",
+			net_g_lock_do,
+			NET_TRANSPORT_LOCAL,
+			N_("Execute a shell command under a lock"),
+			N_("net g_lock do <lock name> <timeout> <command>\n")
+		},
+		{
+			"locks",
+			net_g_lock_locks,
+			NET_TRANSPORT_LOCAL,
+			N_("List all locknames"),
+			N_("net g_lock locks\n")
+		},
+		{
+			"dump",
+			net_g_lock_dump,
+			NET_TRANSPORT_LOCAL,
+			N_("Dump a g_lock locking table"),
+			N_("net g_lock dump <lock name>\n")
+		},
+		{NULL, NULL, 0, NULL, NULL}
+	};
+
+	return net_run_function(c, argc, argv, "net g_lock", func);
+}
diff --git a/source3/utils/net_proto.h b/source3/utils/net_proto.h
index 098e2a22be..540d31e6b1 100644
--- a/source3/utils/net_proto.h
+++ b/source3/utils/net_proto.h
@@ -497,4 +497,7 @@ NTSTATUS net_lookup_sid_from_name(struct net_context *c, TALLOC_CTX *ctx,
 char *stdin_new_passwd( void);
 char *get_pass( const char *prompt, bool stdin_get);
 
+/* The following definitions come from utils/net_g_lock.c  */
+int net_g_lock(struct net_context *c, int argc, const char **argv);
+
 #endif /*  _NET_PROTO_H_  */