On 3/10/24 12:23 PM, Jakub Sitnicki wrote:
On Fri, Mar 08, 2024 at 04:59 PM -08, Andrii Nakryiko wrote:
On Tue, Mar 5, 2024 at 12:22 PM Yonghong Song <yonghong.song@xxxxxxxxx> wrote:
Add link support for sk_msg program. This will make user space
easy to manage as most common used programs have alrady have
link support.
So we have:
SEC("sk_skb/stream_parser") mapping to SK_SKB/BPF_SK_SKB_STREAM_PARSER.
SEC("sk_skb/stream_verdict") mapping to SK_SKB/BPF_SK_SKB_STREAM_VERDICT.
SEC("sk_msg") mapping to SK_MSG/BPF_SK_MSG_VERDICT.
Are those all kind of in the same category and should we support link
for both SK_MSG and SK_SKB? I'm not too familiar, maybe John or
someone else can clarify.
We also have the most recent SK_SKB/BPF_SK_SKB_VERDICT [1], which is
what one would use instead of SK_SKB/BPF_SK_SKB_STREAM_* prog pair when
chopping the stream of bytes into frames is not needed. For instance,
because we're redirecting from a UDP socket.
I have a cheatsheet that clarifies in which redirect-with-sockmap
configuration what program type + attach type is used [2].
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=a7ba4558e69a3c2ae4ca521f015832ef44799538
[2] https://github.com/jsitnicki/srecon-2023-sockmap/blob/main/sockmap-cheatsheet.png
Thanks. I will take a look. Yes, agree, we only need to implement
the recommended one, which is SK_SKB/BPF_SK_SKB_VERDICT if I understand correctly.
