Linux下网口间Ping通信可行性及所需ip/route配置咨询

Alright, let's break down your two scenarios step by step, with exact ip/sysctl commands to force ICMP traffic over the physical cable instead of the kernel's internal loopback (which is the default behavior for same-host IPs).

Part 1: Ping between port0 and port1 of the same NIC #

By default, Linux will route traffic between two IPs on the same host internally (via the loopback interface) instead of sending it out the physical ports. To force the ICMP packets to travel over the cable connecting port0 and port1, follow these steps:

Prerequisites #

• Physically connect port0 and port1 with an Ethernet cable (use a straight-through or crossover cable—most modern NICs auto-sense).
• Replace enp0s0f0 (port0) and enp0s0f1 (port1) with your actual interface names (find them with ip link show).

Configuration Steps #

• Assign IP addresses to both ports (use the same subnet so they can communicate directly):

  # Configure port0
  ip addr add 192.168.10.1/24 dev enp0s0f0
  ip link set enp0s0f0 up
  
  # Configure port1
  ip addr add 192.168.10.2/24 dev enp0s0f1
  ip link set enp0s0f1 up
  

• Disable reverse path filtering (rp_filter):
  Linux's default rp_filter setting will drop packets that don't match expected routing paths, which would block our forced physical route. Disable it for both interfaces and globally:

  sysctl -w net.ipv4.conf.enp0s0f0.rp_filter=0
  sysctl -w net.ipv4.conf.enp0s0f1.rp_filter=0
  sysctl -w net.ipv4.conf.all.rp_filter=0
  

• Add forced host routes to send traffic through the physical ports:
  These routes tell the kernel to send traffic to the other port's IP directly through the physical interface, not internally:

  # Route traffic to port1's IP through port0
  ip route add 192.168.10.2/32 dev enp0s0f0 scope link
  
  # Route traffic to port0's IP through port1
  ip route add 192.168.10.1/32 dev enp0s0f1 scope link
  

Test It #

Run ping -I 192.168.10.1 192.168.10.2—this specifies to use port0's IP as the source, and the traffic will travel over the physical cable to port1.

Part 2: Ping between port0 of first NIC and port0 of second NIC #

This scenario is nearly identical to Part 1—only the interface names change. The core idea is still forcing traffic out the physical port instead of using internal loopback.

Prerequisites #

• Physically connect the two port0s with an Ethernet cable.
• Replace enp0s0f0 (first NIC port0) and enp1s0f0 (second NIC port0) with your actual interface names.

Configuration Steps #

• Assign IP addresses (same subnet):

  # Configure first NIC port0
  ip addr add 192.168.20.1/24 dev enp0s0f0
  ip link set enp0s0f0 up
  
  # Configure second NIC port0
  ip addr add 192.168.20.2/24 dev enp1s0f0
  ip link set enp1s0f0 up
  

• Disable reverse path filtering:

  sysctl -w net.ipv4.conf.enp0s0f0.rp_filter=0
  sysctl -w net.ipv4.conf.enp1s0f0.rp_filter=0
  sysctl -w net.ipv4.conf.all.rp_filter=0
  

• Add forced host routes:

  # Route traffic to second NIC port0's IP through first NIC port0
  ip route add 192.168.20.2/32 dev enp0s0f0 scope link
  
  # Route traffic to first NIC port0's IP through second NIC port0
  ip route add 192.168.20.1/32 dev enp1s0f0 scope link
  

Test It #

Run ping -I 192.168.20.1 192.168.20.2—the ICMP packets will travel over the physical cable between the two separate NICs.

Key Notes #

• To make these settings persistent across reboots:
  • Add the ip addr and ip route commands to a startup script (like /etc/rc.local or a systemd service).
  • Add the sysctl settings to /etc/sysctl.conf (e.g., net.ipv4.conf.enp0s0f0.rp_filter=0) and run sysctl -p to apply.
• If you're using IPv6, the logic is similar—replace ipv4 with ipv6 in the sysctl settings and use ip -6 commands for addresses/routes.

内容的提问来源于stack exchange，提问作者mungayree