Physical Hardware Labs: Blasting .pcap Captures onto Real Copper & Fiber (tcpreplay)
Wire .frasm -> .pcap captures into SPAN ports, optical TAPs, two-NIC firewall hairpins (tcpprep + tcpreplay), and managed switch trunks without NIC drivers eating your VLAN tags.
- FrameAssembly never touches raw sockets: it writes
.pcapfiles in user space, andtcpreplay/ Bittwist / Ostinato blasts those bytes onto real copper or SFP+ fiber. - Two-NIC hairpins (
eth1 -> Trust,eth2 <- Untrust) let one Linux box impersonate both client and server across a hardware firewall viatcpprep+tcpreplay. - Turn off Linux NIC VLAN-stripping offloads (
ethtool -K eth1 rxvlan off txvlan off) before replaying 802.1Q, QinQ, or deliberately weird anomaly packets.
1. SPAN Ports, Optical TAPs & Passive NDR / OT Sensors
Standing up a hardware IDS/NDR box (Suricata, Zeek, Corelight, ExtraHop) or an industrial OT sensor (Claroty, Dragos, Nozomi) hanging off a SPAN port or passive fiber TAP? Plug one spare lab NIC straight into the monitor port and feed it a classic .pcap.
Because passive sensors never talk back, your .frasm script has to carry *both* sides (-> and <-) of the TCP handshake so the stream reassembler marks the flow established before inspecting TLS SNI, HTTP, SMB, or Modbus.
// Bidirectional VLAN 120 trace -> physical SPAN / optical TAP sensor porthost branch_ws { ip 10.20.15.42 mac 00:50:56:aa:15:42 }host shadow_edge { ip 198.51.100.84 mac 00:1c:73:00:00:01 }compile { // Full 3-way handshake -> stream reassembler marks flow established branch_ws -> shadow_edge tcp syn srcport 54210 dstport 443 win 65535 mss 1460 vlan 120 wait 1ms branch_ws <- shadow_edge tcp syn ack win 65535 mss 1460 vlan 120 wait 1ms branch_ws -> shadow_edge tcp ack vlan 120 wait 1ms // TLS 1.3 ClientHello SNI + HTTP/3 QUIC Initial probe branch_ws -> shadow_edge tls clienthello sni "unmanaged-transfer.example.net" alpn "h2" vlan 120 wait 2ms branch_ws -> shadow_edge quic initial version 1 dcid 0x1122334455667788 scid 0xaabbccddeeff0011 pktnum 1 vlan 120}# 1. Compile classic microsecond .pcap (what tcpreplay expects)
frameassembly span_sensor_audit.frasm span_sensor_audit.pcap
# 2. Bring up your spare lab NIC and fire away
sudo ip link set dev eth1 up
sudo tcpreplay --intf1=eth1 --pps=500 span_sensor_audit.pcap2. Two-NIC Hardware Firewall Hairpin (eth1 -> NGFW <- eth2)
Blast a two-way .pcap out of *one* NIC into a stateful Layer 3 hardware firewall (Palo Alto, FortiGate, SRX, Firepower) and watch every server reply die horribly on the Trust port thanks to Reverse Path Forwarding (RPF) and zone checks.
The bench trick: cable two NICs on your Linux tester (eth1 -> Firewall Trust, eth2 -> Firewall Untrust/DMZ). Toss two gratuitous arp frames at the top so the firewall doesn’t stall ARPing for fake hosts, set dstmac to each ingress interface MAC, and let tcpprep + tcpreplay split client <-> server across both cables:
// Two-NIC inline L3 firewall bench:// eth1 -> Firewall Trust: 10.10.1.1 mac 00:1b:17:00:01:01// eth2 -> Firewall Untrust: 172.16.2.1 mac 00:1b:17:00:02:01host trust_client { ip 10.10.1.50 mac 02:00:00:10:01:50 }host dmz_server { ip 172.16.2.80 mac 02:00:00:16:02:80 }host fw_trust { ip 10.10.1.1 mac 00:1b:17:00:01:01 }host fw_untrust { ip 172.16.2.1 mac 00:1b:17:00:02:01 }compile { // 1. Seed firewall ARP tables on Trust (eth1) and Untrust (eth2) trust_client -> fw_trust arp gratuitous wait 5ms dmz_server -> fw_untrust arp gratuitous wait 5ms // 2. Stateful TCP + HTTP flow addressed to each firewall ingress MAC trust_client -> dmz_server tcp syn srcport 52100 dstport 8080 win 65535 mss 1460 dstmac 00:1b:17:00:01:01 wait 1ms dmz_server -> trust_client tcp syn ack srcport 8080 dstport 52100 seq 5000 acknum 1001 win 65535 mss 1460 dstmac 00:1b:17:00:02:01 wait 1ms trust_client -> dmz_server tcp ack srcport 52100 dstport 8080 seq 1001 acknum 5001 dstmac 00:1b:17:00:01:01 wait 1ms trust_client -> dmz_server http get uri "/api/v1/health" hosthdr "dmz-api.lab.local" srcport 52100 dstport 8080 seq 1001 acknum 5001 dstmac 00:1b:17:00:01:01 wait 2ms dmz_server -> trust_client http status 200 reason "OK" body "{\"zone\":\"dmz\"}" srcport 8080 dstport 52100 seq 5001 acknum 1072 dstmac 00:1b:17:00:02:01}# 1. Compile classic .pcap
frameassembly inline_ngfw_hairpin.frasm inline_ngfw_hairpin.pcap
# 2. Split client (10.10.1.0/24 -> eth1) vs server (everything else -> eth2)
tcpprep --cidr=10.10.1.0/24 --pcap=inline_ngfw_hairpin.pcap --cachefile=hairpin.cache
# 3. Replay synchronously across Trust (eth1) <-> Untrust (eth2)
sudo tcpreplay --cachefile=hairpin.cache --intf1=eth1 --intf2=eth2 inline_ngfw_hairpin.pcapvwire) Mode Is Even LazierPut your lab firewall in Layer 2 Virtual-Wire / transparent bridge mode between eth1 and eth2 and you can skip all the dstmac overrides — plain -> and <- arrows work out of the box.
3. Managed Switch Ports: LLDP, 802.1X EAPOL & QinQ Trunks
Cabling a lab NIC straight into a Cisco Catalyst/Nexus, Arista, Juniper, or MikroTik port lets you test show lldp neighbors, 802.1X eapol port auth, Dynamic ARP Inspection (DAI), and carrier 802.1ad QinQ trunks.
Because all 8 pure Layer 2 protocols (arp, vlan, qinq, lldp, stp, lacp, eapol, mpls) accept MAC-only host blocks, you don’t even need dummy IPs for L2 bring-up:
// Direct switch-port poke: LLDP + 802.1X EAPOL + QinQ double-taghost lab_nic { mac 00:1b:21:aa:bb:01 }host tor_sw_01 { mac 00:1c:73:00:09:99 }host cpe_edge { ip 172.20.40.10 mac 00:1b:21:aa:bb:02 }host pe_core { ip 172.20.40.1 mac 00:1c:73:00:09:98 }compile { // 1. Light up 'show lldp neighbors' lab_nic -> tor_sw_01 lldp chassisid "lab-tester-01" portid "eth1" ttl 120 sysname "packet-bench.lab" wait 5ms // 2. Wake up 802.1X authenticator on the switch port lab_nic -> tor_sw_01 eapol start version 2 wait 5ms lab_nic -> tor_sw_01 eapol eappacket version 2 eapcode 2 eapid 1 eapmethod 1 identity "host/bench01.corp.local" wait 5ms // 3. Test 802.1ad QinQ S-VLAN 2100 -> C-VLAN 40 handoff cpe_edge -> pe_core icmp echorequest id 77 seq 1 qinq 2100 innervlan 40 pcp 5}4. Five Ways a Physical NIC or Switch Will Quietly Sabotage You
Before blaming your .frasm script when packets vanish on copper, check these five hardware traps:
# Turn off Linux NIC VLAN-tag eating & segmentation offloads on your replay port
sudo ethtool -K eth1 rxvlan off txvlan off tx off rx off tso off gso off gro off
# Bump MTU so QinQ (+8B) or VXLAN (+50B) doesn't hit errno=90 (Message too long)
sudo ip link set dev eth1 mtu 1558| Hardware Trap | What Goes Sideways | Fix |
|---|---|---|
NIC VLAN offload (txvlan on) | Outer vlan or qinq tags vanish or mutate on the wire | sudo ethtool -K eth1 txvlan off rxvlan off before tcpreplay. |
1500 byte NIC MTU | tcpreplay complains Message too long (errno=90) on QinQ / VXLAN frames | Raise replay NIC + switch port MTU (1508 QinQ, 1558+ VXLAN, or 9000 jumbo). |
Switch bpduguard | Sending one stp BPDU instantly slams your switch port into err-disabled | Only fire stp frames into isolated lab ports where BPDU Guard is what you are testing! |
Burst rate vs wait timers | Hardware policer drops zero-gap bursts | Default tcpreplay honors .frasm wait deltas; add --pps=100 for gentle pacing. |
| Windows laptops | No native AF_PACKET socket for tcpreplay | Use Npcap (with raw 802.1Q enabled) + Ostinato/Bittwist, or pass a USB NIC into Linux. |