Re: [PATCH 4/4] cgroup/bpf: Honor cgroup NS in cgroup_iter for ancestors

From: "Michal Koutný" <mkoutny@suse.com>
To: Yosry Ahmed <yosryahmed@google.com>
Cc: Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Cgroups <cgroups@vger.kernel.org>, bpf <bpf@vger.kernel.org>,
	Tejun Heo <tj@kernel.org>, Aditya Kali <adityakali@google.com>,
	Serge Hallyn <serge.hallyn@canonical.com>,
	Roman Gushchin <roman.gushchin@linux.dev>,
	Yonghong Song <yhs@fb.com>,
	Muneendra Kumar <muneendra.kumar@broadcom.com>,
	Hao Luo <haoluo@google.com>
Subject: Re: [PATCH 4/4] cgroup/bpf: Honor cgroup NS in cgroup_iter for ancestors
Date: Mon, 29 Aug 2022 14:59:57 +0200	[thread overview]
Message-ID: <20220829125957.GB3579@blackbody.suse.cz> (raw)
In-Reply-To: <CAJD7tkZZ6j6mPfwwFDy_ModYux5447HFP=oPwa6MFA_NYAZ9-g@mail.gmail.com>

[-- Attachment #1: Type: text/plain, Size: 918 bytes --]

On Fri, Aug 26, 2022 at 10:41:37AM -0700, Yosry Ahmed <yosryahmed@google.com> wrote:
> I understand that currently cgroup_iter is the only user of this, but
> for future use cases, is it safe to assume that cgrp will always be
> inside ns? Would it be safer to do something like:

I preferred the simpler root_cgrp comparison to avoid pointer
arithmetics in cgroup_is_descendant. But I also made the assumption of
cgrp in ns.

Thanks, I'll likely adjust cgroup_path_ns to make it more robust for
an external cgrp.

I'd like to clarify, if a process A in a broad cgroup ns sets up a BPF
cgroup iterator, exposes it via bpffs and than a process B in a narrowed
cgroup ns (which excludes the origin cgroup) wants to traverse the
iterator, should it fail straight ahead (regardless of iter order)?
The alternative would be to allow self-dereference but prohibit any
iterator moves (regardless of order).

Thanks,
Michal

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 228 bytes --]