diff --git a/modules/tm/defs.h b/modules/tm/defs.h
index 1e08b950786..b712b1d3c5d 100644
--- a/modules/tm/defs.h
+++ b/modules/tm/defs.h
@@ -30,5 +30,9 @@
 #ifndef _TM_DEFS_H
 #define _TM_DEFS_H
 
+/* CANCEL_REASON_SUPPORT on by default */
+#ifndef NO_CANCEL_REASON_SUPPORT
+#define CANCEL_REASON_SUPPORT
+#endif /* NO_CANCEL_REASON_SUPPORT */
 
 #endif
diff --git a/modules/tm/h_table.h b/modules/tm/h_table.h
index 52b6d2cf4ac..18382453c1f 100644
--- a/modules/tm/h_table.h
+++ b/modules/tm/h_table.h
@@ -75,7 +75,6 @@ struct async_state;
 #include "../../mem/shm_mem.h"
 #include "lock.h"
 #include "sip_msg.h"
-#include "t_reply.h"
 #include "t_hooks.h"
 #ifdef USE_DNS_FAILOVER
 #include "../../dns_cache.h"
@@ -428,7 +427,12 @@ typedef struct cell
 #endif
 
 	/* protection against concurrent reply processing */
-	ser_lock_t   reply_mutex;
+	ser_lock_t reply_mutex;
+	/* pid of the process that holds the reply lock */
+	atomic_t reply_locker_pid;
+	/* recursive reply lock count */
+	int reply_rec_lock_level;
+
 	/* protect against concurrent async continues */
 	ser_lock_t   async_mutex;
 		
diff --git a/modules/tm/t_append_branches.c b/modules/tm/t_append_branches.c
index ab6f298b8b7..b276d9e9182 100644
--- a/modules/tm/t_append_branches.c
+++ b/modules/tm/t_append_branches.c
@@ -43,6 +43,7 @@
 #include "t_msgbuilder.h"
 #include "t_lookup.h"
 #include "t_fwd.h"
+#include "t_reply.h"
 #include "t_append_branches.h"
 
 int t_append_branches(void) {
diff --git a/modules/tm/t_funcs.c b/modules/tm/t_funcs.c
index 8e04de6faa0..cec58e03a65 100644
--- a/modules/tm/t_funcs.c
+++ b/modules/tm/t_funcs.c
@@ -36,6 +36,7 @@
 #include "t_funcs.h"
 #include "t_fwd.h"
 #include "t_lookup.h"
+#include "t_reply.h"
 #include "config.h"
 #include "t_stats.h"
 
diff --git a/modules/tm/t_fwd.c b/modules/tm/t_fwd.c
index 9fcf3813eea..9cd002f891a 100644
--- a/modules/tm/t_fwd.c
+++ b/modules/tm/t_fwd.c
@@ -44,6 +44,8 @@
 #include "t_cancel.h"
 #include "t_lookup.h"
 #include "t_fwd.h"
+#include "t_reply.h"
+#include "h_table.h"
 #include "../../fix_lumps.h"
 #include "config.h"
 #ifdef USE_DNS_FAILOVER
diff --git a/modules/tm/t_msgbuilder.h b/modules/tm/t_msgbuilder.h
index ae09f31d566..4ede95b900a 100644
--- a/modules/tm/t_msgbuilder.h
+++ b/modules/tm/t_msgbuilder.h
@@ -27,6 +27,7 @@
 #include "defs.h"
 #include "dlg.h"
 #include "h_table.h"
+#include "t_reply.h"
 
 
 #define CSEQ "CSeq: "
diff --git a/modules/tm/t_reply.c b/modules/tm/t_reply.c
index b5909c73702..5dfbfd0dbda 100644
--- a/modules/tm/t_reply.c
+++ b/modules/tm/t_reply.c
@@ -2775,3 +2775,34 @@ void rpc_reply(rpc_t* rpc, void* c)
 		return;
 	}
 }
+
+/**
+ * re-entrant locking of reply mutex
+ */
+void tm_reply_mutex_lock(tm_cell_t *t)
+{
+	int mypid;
+
+	mypid = my_pid();
+	if (likely(atomic_get(&t->reply_locker_pid) != mypid)) {
+		lock(&t->reply_mutex);
+		atomic_set(&t->reply_locker_pid, mypid);
+	} else {
+		/* locked within the same process that called us*/
+		t->reply_rec_lock_level++;
+	}
+}
+
+/**
+ * re-entrant unlocking of reply mutex
+ */
+void tm_reply_mutex_unlock(tm_cell_t *t)
+{
+	if (likely(t->reply_rec_lock_level == 0)) {
+		atomic_set(&t->reply_locker_pid, 0);
+		unlock(&t->reply_mutex);
+	} else  {
+		/* recursive locked => decrease rec. lock count */
+		t->reply_rec_lock_level--;
+	}
+}
diff --git a/modules/tm/t_reply.h b/modules/tm/t_reply.h
index 2151e4a7d68..91b2ac839ce 100644
--- a/modules/tm/t_reply.h
+++ b/modules/tm/t_reply.h
@@ -23,11 +23,6 @@
 #ifndef _T_REPLY_H
 #define _T_REPLY_H
 
-/* CANCEL_REASON_SUPPORT on by default */
-#ifndef NO_CANCEL_REASON_SUPPORT
-#define CANCEL_REASON_SUPPORT
-#endif /* NO_CANCEL_REASON_SUPPORT */
-
 #include "defs.h"
 #include "../../rpc.h"
 #include "../../tags.h"
@@ -139,8 +134,11 @@ int w_t_reply_wrp(struct sip_msg *m, unsigned int code, char *txt);
 typedef int (*tget_reply_totag_f)(struct sip_msg *, str *);
 int t_get_reply_totag(struct sip_msg *msg, str *totag);
 
-#define LOCK_REPLIES(_t) lock(&(_t)->reply_mutex )
-#define UNLOCK_REPLIES(_t) unlock(&(_t)->reply_mutex )
+void tm_reply_mutex_lock(tm_cell_t *t);
+void tm_reply_mutex_unlock(tm_cell_t *t);
+
+#define LOCK_REPLIES(_t) tm_reply_mutex_lock((_t))
+#define UNLOCK_REPLIES(_t) tm_reply_mutex_unlock((_t))
 
 /* This function is called whenever a reply for our module is received;
  * we need to register this function on module initialization;
diff --git a/modules/tm/t_serial.c b/modules/tm/t_serial.c
index 0bc671d3e01..b3207599d39 100644
--- a/modules/tm/t_serial.c
+++ b/modules/tm/t_serial.c
@@ -34,6 +34,7 @@
 #include "../../ut.h"
 #include "config.h"
 #include "t_funcs.h"
+#include "t_reply.h"
 #include "t_lookup.h"
 
 /* usr_avp flag for sequential forking */