Re: [PATCH net-next 0/4] Add getsockopt(SO_PEERCGROUPID) and fdinfo API to retreive socket's peer cgroup id

[Michal Koutný]

Hello.

On Sun, Mar 09, 2025 at 02:28:11PM +0100, Alexander Mikhalitsyn <aleksandr.mikhalitsyn@xxxxxxxxxxxxx> wrote:
> As we don't add any new state to the socket itself there is no potential locking issues
> or performance problems. We use already existing sk->sk_cgrp_data.

This is the cgroup where the socket was created in. If such a socket is
fd-passed to another cgroup, SO_PEERCGROUPID may not be what's expected.

> We already have analogical interfaces to retrieve this
> information:
> - inet_diag: INET_DIAG_CGROUP_ID
> - eBPF: bpf_sk_cgroup_id
> 
> Having getsockopt() interface makes sense for many applications, because using eBPF is
> not always an option, while inet_diag has obvious complexety and performance drawbacks
> if we only want to get this specific info for one specific socket.

I'm not that familiar with INET_DIAG_CGROUP_ID but that one sounds like
fit for the purpose of obtaining socket creator's cgroup whereas what is
desired here is slightly different thing -- cgroup of actual sender
through the socket.

> Idea comes from UAPI kernel group:
> https://uapi-group.org/kernel-features/

In theory shortlived (sending) program may reside in shortlived cgroup
and the consumer of SO_PEERCGROUPID (even if it is real sender) would
only have that number to work with.
It doesn't guarantee existence of original cgroup or stable translation
to cgroup path. I think having something like this could be useful but
consumers must still be aware of limitations (and possible switch from
path-oriented to id-oriented work with cgroups).

Thanks,
Michal

Attachment: signature.asc
Description: PGP signature