Re: [PATCH bpf-next 1/5] bpf: Implement task local storage

From: Song Liu <song@kernel.org>
To: KP Singh <kpsingh@chromium.org>
Cc: open list <linux-kernel@vger.kernel.org>,
	bpf <bpf@vger.kernel.org>, Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Martin KaFai Lau <kafai@fb.com>, Paul Turner <pjt@google.com>,
	Jann Horn <jannh@google.com>, Hao Luo <haoluo@google.com>
Subject: Re: [PATCH bpf-next 1/5] bpf: Implement task local storage
Date: Fri, 30 Oct 2020 17:02:41 -0700	[thread overview]
Message-ID: <CAPhsuW4z4b83AeTB_MUWzVWcZxvQ+hdbN9riL77t4j9uBfDdGA@mail.gmail.com> (raw)
In-Reply-To: <CACYkzJ7kbq0Nq71fJCkHSwEmJfKFKOsvRZos_tT64N1f-aT4-A@mail.gmail.com>

On Fri, Oct 30, 2020 at 4:07 AM KP Singh <kpsingh@chromium.org> wrote:
>
> "
>
> On Fri, Oct 30, 2020 at 12:28 AM Song Liu <song@kernel.org> wrote:
> >
> > On Wed, Oct 28, 2020 at 9:17 AM KP Singh <kpsingh@chromium.org> wrote:
> > >
> > > From: KP Singh <kpsingh@google.com>
> > >
> > > Similar to bpf_local_storage for sockets and inodes add local storage
> > > for task_struct.
> > >
> > > The life-cycle of storage is managed with the life-cycle of the
> > > task_struct.  i.e. the storage is destroyed along with the owning task
> > > with a callback to the bpf_task_storage_free from the task_free LSM
> > > hook.
> >
> > It looks like task local storage is tightly coupled to LSM. As we discussed,
> > it will be great to use task local storage in tracing programs. Would you
> > like to enable it from the beginning? Alternatively, I guess we can also do
> > follow-up patches.
> >
>
> I would prefer if we do it in follow-up patches.
>
> > >
> > > The BPF LSM allocates an __rcu pointer to the bpf_local_storage in
> > > the security blob which are now stackable and can co-exist with other
> > > LSMs.
> > >
> > > The userspace map operations can be done by using a pid fd as a key
> > > passed to the lookup, update and delete operations.
> >
> > While testing task local storage, I noticed a limitation of pid fd:
> >
> > /* Currently, the process identified by
> >  * @pid must be a thread-group leader. This restriction currently exists
> >  * for all aspects of pidfds including pidfd creation (CLONE_PIDFD cannot
> >  * be used with CLONE_THREAD) and pidfd polling (only supports thread group
> >  * leaders).
> >  */
> >
> > This could be a problem for some use cases. How about we try to remove
> > this restriction (maybe with a new flag to pidfd_open) as part of this set?
>
> I would appreciate it if we could also do this in a follow-up patch.
>
> I do see that there is a comment in fork.c:
>
>     "CLONE_THREAD is blocked until someone really needs it."
>
> But I don't understand the requirements well enough and would thus prefer
> to do this in a follow-up series.

Sounds good. Let's work on these in follow-up patches.

Thanks,
Song