Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210122123003.46125-1-mikko.ylinen@linux.intel.com>
 <CACYkzJ6sMBvZ_ZG9++jwpQ+JQL3PL02okhD0O5Ftz4Hd7jEC3Q@mail.gmail.com>
 <CACYkzJ5i6DNxY3D3SqiO_LO2HBbA-EgPFdK8ZBcFNmTVWweucg@mail.gmail.com> <YA5rXLwy4mcgcvLx@outtakka>
In-Reply-To: <YA5rXLwy4mcgcvLx@outtakka>
From:   KP Singh <kpsingh@kernel.org>
Date:   Mon, 25 Jan 2021 18:49:26 +0100
Message-ID: <CACYkzJ6q53qwEvTjvNP_d5O+ytQM=jFVRQ9PV2+ie6Ah5VDnRA@mail.gmail.com>
Subject: Re: [PATCH] bpf: Drop disabled LSM hooks from the sleepable set
To:     Mikko Ylinen <mikko.ylinen@linux.intel.com>
Cc:     bpf <bpf@vger.kernel.org>,
        open list <linux-kernel@vger.kernel.org>,
        KP Singh <kpsingh@google.com>,
        Daniel Borkmann <daniel@iogearbox.net>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, Jan 25, 2021 at 7:55 AM Mikko Ylinen
<mikko.ylinen@linux.intel.com> wrote:
>
> On Sat, Jan 23, 2021 at 12:50:21AM +0100, KP Singh wrote:
> > On Fri, Jan 22, 2021 at 11:33 PM KP Singh <kpsingh@kernel.org> wrote:
> > >
> > > On Fri, Jan 22, 2021 at 1:32 PM Mikko Ylinen
> > > <mikko.ylinen@linux.intel.com> wrote:
> > > >
> > > > Networking LSM hooks are conditionally enabled and when building the new
> > > > sleepable BPF LSM hooks with the networking LSM hooks disabled, the
> > > > following build error occurs:
> > > >
> > > > BTFIDS  vmlinux
> > > > FAILED unresolved symbol bpf_lsm_socket_socketpair
> > > >

[...]

>
> Agree, a way to get the set automatically created makes sense. But the
> extra parameter to LSM_HOOK macro would be BPF specific, right?
>

The information about whether the hook "must not sleep" has been
mentioned sporadically in comments and

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/linux/lsm_hooks.h#n920
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/linux/lsm_hooks.h#n594

I think it would be generally useful for the framework to actually provide this
in the definition in the hook and then ensure (by calling
might_sleep() for hooks
that can sleep).

- KP

> -- Regards, Mikko