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;
Subject: Re: [PATCH v2 bpf 3/5] bpf: bpf_prog_array_free() should take a
 generic non-rcu pointer
To:     Roman Gushchin <guro@fb.com>, netdev@vger.kernel.org
Cc:     linux-kernel@vger.kernel.org, kernel-team@fb.com,
        Alexei Starovoitov <ast@kernel.org>
References: <20180713194114.2711-1-guro@fb.com>
 <20180713194114.2711-3-guro@fb.com>
From:   Daniel Borkmann <daniel@iogearbox.net>
Message-ID: <41b5f919-3155-5bd4-2774-84efd58650e2@iogearbox.net>
Date:   Tue, 17 Jul 2018 00:30:18 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.3.0
MIME-Version: 1.0
In-Reply-To: <20180713194114.2711-3-guro@fb.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 07/13/2018 09:41 PM, Roman Gushchin wrote:
> bpf_prog_array_free() should take a generic non-rcu pointer
> as an argument, as freeing the objects assumes that we're
> holding an exclusive rights on it.
> 
> rcu_access_pointer() can be used to convert a __rcu pointer to
> a generic pointer before passing it to bpf_prog_array_free(),
> if necessary.
> 
> This patch eliminates the following sparse warning:
> kernel/bpf/core.c:1556:9: warning: incorrect type in argument 1 (different address spaces)
> kernel/bpf/core.c:1556:9:    expected struct callback_head *head
> kernel/bpf/core.c:1556:9:    got struct callback_head [noderef] <asn:4>*<noident>
> 
> Fixes: 324bda9e6c5a ("bpf: multi program support for cgroup+bpf")
> Signed-off-by: Roman Gushchin <guro@fb.com>
> Cc: Alexei Starovoitov <ast@kernel.org>
> Cc: Daniel Borkmann <daniel@iogearbox.net>
> ---
>  drivers/media/rc/bpf-lirc.c |  6 +++---
>  include/linux/bpf.h         |  2 +-
>  kernel/bpf/cgroup.c         | 11 ++++++-----
>  kernel/bpf/core.c           |  5 ++---
>  kernel/trace/bpf_trace.c    |  8 ++++----
>  5 files changed, 16 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/media/rc/bpf-lirc.c b/drivers/media/rc/bpf-lirc.c
> index fcfab6635f9c..509b262aa0dc 100644
> --- a/drivers/media/rc/bpf-lirc.c
> +++ b/drivers/media/rc/bpf-lirc.c
> @@ -135,7 +135,7 @@ static int lirc_bpf_attach(struct rc_dev *rcdev, struct bpf_prog *prog)
>  		goto unlock;
>  
>  	rcu_assign_pointer(raw->progs, new_array);
> -	bpf_prog_array_free(old_array);
> +	bpf_prog_array_free(rcu_access_pointer(old_array));

Taking this one as an example, why can't we already do the rcu_dereference() on the
'old_array = raw->progs' where we fetch the old_array initially? Then we also wouldn't
need the rcu_access_pointer() on bpf_prog_array_free() and yet another rcu_dereference()
inside the bpf_prog_array_copy() from your later patch?

Regarding former, rcu_access_pointer() should also only be used for testing the pointer
value, but deeper in bpf_prog_array_free() we also deref it, etc.

>  unlock:
>  	mutex_unlock(&ir_raw_handler_lock);
> @@ -173,7 +173,7 @@ static int lirc_bpf_detach(struct rc_dev *rcdev, struct bpf_prog *prog)
>  		goto unlock;
>  
>  	rcu_assign_pointer(raw->progs, new_array);
> -	bpf_prog_array_free(old_array);
> +	bpf_prog_array_free(rcu_access_pointer(old_array));
>  unlock:
>  	mutex_unlock(&ir_raw_handler_lock);
>  	return ret;
> @@ -204,7 +204,7 @@ void lirc_bpf_free(struct rc_dev *rcdev)
>  	while (*progs)
>  		bpf_prog_put(*progs++);
>  
> -	bpf_prog_array_free(rcdev->raw->progs);
> +	bpf_prog_array_free(rcu_access_pointer(rcdev->raw->progs));
>  }