diff --git a/include/nuttx/net/netdev.h b/include/nuttx/net/netdev.h
index 23a3f2f5e5d8a..f4a08cf6eeca6 100644
--- a/include/nuttx/net/netdev.h
+++ b/include/nuttx/net/netdev.h
@@ -442,6 +442,7 @@ struct net_driver_s
   /* Remember the outgoing fragments waiting to be sent */
 
 #ifdef CONFIG_NET_IPFRAG
+  uint8_t d_ipfrag_reassembled; /* Current packet is reassembled by IPFRAG */
   struct iob_queue_s d_fragout;
 #endif
 
diff --git a/net/ipforward/ipv6_forward.c b/net/ipforward/ipv6_forward.c
index 18d7a941ad763..f8b53eaef2807 100644
--- a/net/ipforward/ipv6_forward.c
+++ b/net/ipforward/ipv6_forward.c
@@ -390,11 +390,17 @@ static int ipv6_dev_forward(FAR struct net_driver_s *dev,
     }
   else if (ret == PACKET_NOT_FORWARDED)
     {
-      /* Verify that the full packet will fit within the forwarding devices
-       * MTU.  We provide no support for fragmenting forwarded packets.
+      /* Verify that the full packet will fit within the forwarding device's
+       * MTU.  Normal IPv6 forwarding does not fragment oversized packets,
+       * but packets reassembled locally for NAT66/IPFILTER may be split
+       * again by the egress IPFRAG path.
        */
 
-      if (NET_LL_HDRLEN(fwddev) + dev->d_len > NETDEV_PKTSIZE(fwddev))
+      if (NET_LL_HDRLEN(fwddev) + dev->d_len > NETDEV_PKTSIZE(fwddev)
+#ifdef CONFIG_NET_IPFRAG
+          && !dev->d_ipfrag_reassembled
+#endif
+         )
         {
           nwarn("WARNING: Packet > MTU... Dropping\n");
           ret = -EFBIG;
diff --git a/net/ipfrag/ipv6_frag.c b/net/ipfrag/ipv6_frag.c
index 5ae861cd9a1a4..c27f731bb0214 100644
--- a/net/ipfrag/ipv6_frag.c
+++ b/net/ipfrag/ipv6_frag.c
@@ -469,6 +469,8 @@ int32_t ipv6_fragin(FAR struct net_driver_s *dev)
   FAR struct ip_fragsnode_s *node = NULL;
   FAR struct ip_fraglink_s *fraginfo = NULL;
   bool restartwdog;
+  uint8_t reassembled;
+  int32_t ret;
 
   if (dev->d_len != dev->d_iob->io_pktlen)
     {
@@ -521,7 +523,12 @@ int32_t ipv6_fragin(FAR struct net_driver_s *dev)
 
       kmm_free(node);
 
-      return ipv6_input(dev);
+      reassembled = dev->d_ipfrag_reassembled;
+      dev->d_ipfrag_reassembled = true;
+      ret = ipv6_input(dev);
+      dev->d_ipfrag_reassembled = reassembled;
+
+      return ret;
     }
 
   nxmutex_unlock(&g_ipfrag_lock);
diff --git a/net/netdev/netdev_register.c b/net/netdev/netdev_register.c
index e92e0d992b1f3..3bbe6f10bbbee 100644
--- a/net/netdev/netdev_register.c
+++ b/net/netdev/netdev_register.c
@@ -427,6 +427,10 @@ int netdev_register(FAR struct net_driver_s *dev, enum net_lltype_e lltype)
 
       dev->d_polltype = 0;
 
+#ifdef CONFIG_NET_IPFRAG
+      dev->d_ipfrag_reassembled = false;
+#endif
+
       nxrmutex_init(&dev->d_lock);
 
       /* We need exclusive access for the following operations */