RE: [PATCH bpf 0/8] Fixes to bpf_msg_push/pop_data and test_sockmap

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

zijianzhang@ wrote:
> From: Zijian Zhang <zijianzhang@bytedance.com>
> 
> Several fixes to test_sockmap and added push/pop logic for msg_verify_data
> Before the fixes, some of the tests in test_sockmap are problematic,
> resulting in pseudo-correct result.
> 
> 1. txmsg_pass is not set in some tests, as a result, no eBPF program is
> attached to the sockmap.
> 2. In SENDPAGE, a wrong iov_length in test_send_large may result in some
> test skippings and failures.
> 3. The calculation of total_bytes in msg_loop_rx is wrong, which may cause
> msg_loop_rx end early and skip some data tests.
> 
> Besides, for msg_verify_data, I added push/pop checking logic to function
> msg_verify_data and added more tests for different cases.

Thanks! Yep I think push/pop are not widely used anywhere unfortunately.
There are some interesting uses for push/pop to add/edit headers, but
I've not gotten there yet clearly.

> 
> After that, I found that there are some bugs in bpf_msg_push_data,
> bpf_msg_pop_data and sk_msg_reset_curr, and fix them. I guess the reason
> why they have not been exposed is that because of the above problems, they
> will not be triggered.

Good. I'll review these quickly tonight/tomorrow and run some testing.
We don't currently have any longer running tests with push/pop.

> 
> With the fixes, we can pass the sockmap test with data integrity test now.
> However, the fixes to test_sockmap expose more problems in sockhash test
> with SENDPAGE and ktls with SENDPAGE.
> 
> The problem I observed,

Thanks for digging into these.

[...]