tcp: add asserts and safety in case no CC module is attached

[rscheff]

Summary

While further troubleshooting PR276761, i encountered one instance
where the CC algo was gone already, when a retransmission timeout
timer fired. While that shouldn't happen, i think it's prudent
to not blindly follow potential NULL pointers when calling CC
functions.

The impact of checking the pointer should be minimal in all these instances.

[2475.559032] calltrap() at calltrap+0x8/frame 0xfffffe00ce8aac80
[2475.581285] --- trap 0xc, rip = 0xffffffff80d474c0, rsp = 0xfffffe00ce8aad50, rbp = 0xfffffe00ce8aad70 ---
[2475.603529] cc_cong_signal() at cc_cong_signal+0x1e0/frame 0xfffffe00ce8aad70
[2475.628033] tcp_timer_rexmt() at tcp_timer_rexmt+0x66b/frame 0xfffffe00ce8aadd0
[2475.651720] tcp_timer_enter() at tcp_timer_enter+0x15e/frame 0xfffffe00ce8aae10

(tp->t_cc was NULL)

[tuexen]

I think we should find the cause, where the CC module gone...

[melifaro]

I agree we should rootcase the case. However - why do we need to perform these conditional checks everywhere?
Why can’t t_cc refer to the dummy cc structure? Why can’t the non-implemented/non-needed function point to no-op functions?

[glebius]

Yes, we need to find the root cause instead. The patch can be used as a temporary solution for the bug report user affected.

[rscheff]

Root cause was found, no need to keep this around