Hello,
I am trying to write a kernel module using netfilter to be able to stall
and store TCP packets on data structures of my own and then release them
on a signal. At first I used the nf_queue+nf_reinject system, but now I
would rather be able to manage packets and data structures more freely
in kernel space.
I am using kernel version 5.10.113 and using only a NF_INET_POST_ROUTING
hook set to INT_MAX priority. Trying to copy what nf_queue/nf_reinject
do, this is what I do for storing, before returning NF_STOLEN in the
hook function:
pkt.skb = skb;
pkt.state = *state;
rcu_read_lock();
if (state->in)
dev_hold(state->in);
if (state->out)
dev_hold(state->out);
if (state->sk)
sock_hold(state->sk);
rcu_read_unlock();
And this is what I do for releasing:
struct nf_hook_state *state = &pkt.state;
local_bh_disable();
(state->okfn)(state->net, state->sk, pkt.skb);
local_bh_enable();
rcu_read_lock();
if (state->in)
dev_put(state->in);
if (state->out)
dev_put(state->out);
if (state->sk)
sock_put(state->sk);
rcu_read_unlock();
And this stalling/releasing works well until at some point it does not
and my VM freezes after calling the okfn (which in my case always is
ip_finish_output if I understood correctly). I am debugging by
outputting dmesg prints from guest to host and the kernel throws no
error, it is just frozen. I'm also not getting any feedback from KASAN.
So I guess I might be doing something wrong resulting in a deadlock of
some sort? Does anybody have any pointer whether there is anything else
I should be looking out for when stalling and then releasing a packet
this way?
Thank you for your time,
Federico
