Possible bug that causes xdpfwd exit 255 , need help

[cxw620]

When I enable FORWARDED PPS logging, it seems that the memory consumption will go high and cause xdpfwd exit, which breaks the network (all inbound traffic will be forwarded to one of dests added).

Oct 03 11:16:56 CN-GZ-Tencent xdpfwd[2146]: [4] Retrieved 2 CPUs on host.
Oct 03 11:28:16 CN-GZ-Tencent xdpfwd[2146]: [28K blob data]
Oct 03 11:29:22 CN-GZ-Tencent systemd[1]: Stopping xdpfwd.service - XDP Forward program....
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [2.9K blob data]
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [3B blob data]
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [2] Cleaning up...
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [3B blob data]
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [2] Un-pinning BPF maps from file system...
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [3B blob data]
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [1] [WARNING] Failed to un-pin BPF map 'map_block' from file system (-2).
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [3B blob data]
Oct 03 11:29:22 CN-GZ-Tencent xdpfwd[2146]: [1] Exiting.
Oct 03 11:29:22 CN-GZ-Tencent bash[2334]: Error: either "dev" is duplicate, or "off" is garbage.
Oct 03 11:29:22 CN-GZ-Tencent systemd[1]: xdpfwd.service: Control process exited, code=exited, status=255/EXCEPTION
Oct 03 11:29:22 CN-GZ-Tencent systemd[1]: xdpfwd.service: Consumed 148ms CPU time, 203.4M memory peak.

I also notice that if I used xdpfwd-add to save config and did not pass in arg interface, xdpfwd-add will not warn me and just write to the config, making restart failed.

I'm not quite familiar with C and could not figure out the solution, so I'm here for help.

[gamemann]

Hey, sorry for the late reply!

Would you be able to provide your Linux distro and kernel?

You can retrieve this information with the following commands.

cat /etc/*-release

uname -r 

Also, how much total RAM does your server have?

I also notice that if I used xdpfwd-add to save config and did not pass in arg interface, xdpfwd-add will not warn me and just write to the config, making restart failed.

Just to clarify, does it write the wrong interface to the config file and then restart the program? I'll try to look into this issue specifically soon! The CLI utilities are somewhat new, so there may be some bugs I missed.

Thank you!

[cxw620]

I use Debian 13, and Xanmod kernel (6.16.9). The system is fresh installed. I've tested on Debian 12 and other version of Xanmod kernel before and got the same issue. The server usually has 1G RAM, swap not enabled. XDP mode is xdpdrv.

Just to clarify, does it write the wrong interface to the config file and then restart the program?

No. When the network was down, I logined with VNC console and found that there's no interface item in the config. The restart attempts was done by systemd since xdpfwd exited abruptly.

---

It seems this issue only occurs when forwarded packet logging (uncomment #define STATS_COUNT_FWD_BACK) is enabled. Here're steps that may reproduce the problem.

• Prepare a 2C 1G VM.
• Install Debian 12. I use Linux reinstall script (https://github.com/leitbogioro/Tools) to replace the preinstall one. (optional?)
• Upgrade to Debian 13. (optional?)
• Install Xanmod kernel. (optional?)
• Install xdpfwd: ./install.sh --libxdp
• Enable systemd profile and start xdpfwd.
• Test with STATS_COUNT_FWD_BACK disabled, it should work fine.
• Try to add rules with xdpfwd-add at runtime without specifying interface, and save to config with -s, check the config file and interface would be not present.
• Uncomment #define STATS_COUNT_FWD_BACK, and run ./install.sh --libxdp again.
• systemctl restart
• Test and use xdpfwd to monitor the realtime forwarded packet count. A few seconds (or minutes?) later, the SSH connection would be down and cannot connect to the VM any more.

---

Add:

• The network was bidirectionally down.
• The VPSs I've tested were all behind NAT (?) (has a public IP, but the provider assigns the machine an internal IP address.)