Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 26 Mar 2020 20:12:56 -0700
From:   Alexei Starovoitov <alexei.starovoitov@gmail.com>
To:     KP Singh <kpsingh@chromium.org>
Cc:     linux-kernel@vger.kernel.org, bpf@vger.kernel.org,
        linux-security-module@vger.kernel.org,
        Brendan Jackman <jackmanb@google.com>,
        Florent Revest <revest@google.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        James Morris <jmorris@namei.org>,
        Kees Cook <keescook@chromium.org>,
        Paul Turner <pjt@google.com>, Jann Horn <jannh@google.com>,
        Florent Revest <revest@chromium.org>,
        Brendan Jackman <jackmanb@chromium.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Subject: Re: [PATCH bpf-next v7 4/8] bpf: lsm: Implement attach, detach and
 execution
Message-ID: <20200327031256.vhk2luomxgex3ui4@ast-mbp>
References: <20200326142823.26277-1-kpsingh@chromium.org>
 <20200326142823.26277-5-kpsingh@chromium.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200326142823.26277-5-kpsingh@chromium.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Mar 26, 2020 at 03:28:19PM +0100, KP Singh wrote:
>  
>  	if (arg == nr_args) {
> -		if (prog->expected_attach_type == BPF_TRACE_FEXIT) {
> +		/* BPF_LSM_MAC programs only have int and void functions they
> +		 * can be attached to. When they are attached to a void function
> +		 * they result in the creation of an FEXIT trampoline and when
> +		 * to a function that returns an int, a MODIFY_RETURN
> +		 * trampoline.
> +		 */
> +		if (prog->expected_attach_type == BPF_TRACE_FEXIT ||
> +		    prog->expected_attach_type == BPF_LSM_MAC) {
>  			if (!t)
>  				return true;
>  			t = btf_type_by_id(btf, t->type);

Could you add a comment here that though BPF_MODIFY_RETURN-like check
if (ret_type != 'int') return -EINVAL;
is _not_ done here. It is still safe, since LSM hooks have only
void and int return types.

> +	case BPF_LSM_MAC:
> +		if (!prog->aux->attach_func_proto->type)
> +			/* The function returns void, we cannot modify its
> +			 * return value.
> +			 */
> +			return BPF_TRAMP_FEXIT;
> +		else
> +			return BPF_TRAMP_MODIFY_RETURN;

I was thinking whether it would help performance significantly enough
if we add a flavor of BPF_TRAMP_FEXIT that doesn't have
BPF_TRAMP_F_CALL_ORIG.
That will save the cost of nop call, but I guess indirect call due
to lsm infra is slow enough, so this extra few cycles won't be noticeable.
So I'm fine with it as-is. When lsm hooks will get rid of indirect call
we can optimize it further.