So, sock_cgroup_data is overloaded with net_prio prioidx and net_cls classid. There are two small patches for those two subsystems in 4.15.0-109 compared with 4.15.0-108. At first look, they appear harmless, but we have little information on the workload that generates this crash.
I tried manipulating net_prio and net_cls while running a socket with a cgroup bpf program attached on ingress, but with little luck reproducint. It seems the workload is more complicated than that.
It might be worth asking the reporter to try with a test kernel with those two commits reverted. On bionic tree, they are:
So, sock_cgroup_data is overloaded with net_prio prioidx and net_cls classid. There are two small patches for those two subsystems in 4.15.0-109 compared with 4.15.0-108. At first look, they appear harmless, but we have little information on the workload that generates this crash.
I tried manipulating net_prio and net_cls while running a socket with a cgroup bpf program attached on ingress, but with little luck reproducint. It seems the workload is more complicated than that.
It might be worth asking the reporter to try with a test kernel with those two commits reverted. On bionic tree, they are:
5eebba2159d707a e9533a52839e1ba 71754c4426 ce9ac4c1036403b 0a6d391f7c
a3e9313430937c4
Cascardo.