RE: Kernel panic on bpf_skb_pull_data

[John Fastabend <john.fastabend@gmail.com>]

Farbod Shahinfar wrote:
> Hello everyone,
> 
> I am performing some test with BPF SK_SKB and I have encountered a scenario resulting in kernel panic. I use a BPF_SK_SKB_STREAM_PARSER program to parse a request which might be spanning multiple TCP segments. If the end of request is detected in the parser program it returns skb->len, passing the request to the BPF_SK_SKB_STREAM_VERDICT program, and otherwise it returns 0, waiting for more data to be received. You can find the BPF program attached (bpf_test.c). Is there an assumption that the program violates?
> 
> To reproduce the crashing scenario, I use the python script attached (client.py) which sends data in chunks toward the bpf program. Usually, the kernel crashes on the 3rd segment.
> 
> To provide more information, I have attached some crash logs. I have tested this on kernel version 6.1.0 (slightly modified) and version 6.2.8 (unmodified, obtained from kernel.org). It seems that the panic happens when invoking the bpf_skb_pull_data.
> 
> Is this a known issue or is there any information that I can provide to help resolve it?
> 
> Sincerely,
> Farbod Shahinfar
> 
> PhD student at Politecnico di Milano
> https://fshahinfar1.github.io/
> 

I believe I have a fix here for this. I'll check Monday and resend
last set of fixes

Thanks,
John