Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 25 Jul 2023 10:14:05 -0700
From:   Jakub Kicinski <kuba@kernel.org>
To:     Leon Romanovsky <leon@kernel.org>
Cc:     Lin Ma <linma@zju.edu.cn>, jgg@ziepe.ca, markzhang@nvidia.com,
        michaelgur@nvidia.com, ohartoov@nvidia.com,
        chenzhongjin@huawei.com, yuancan@huawei.com,
        linux-rdma@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v1] RDMA/nldev: Add length check for
 IFLA_BOND_ARP_IP_TARGET parsing
Message-ID: <20230725101405.4cd51059@kernel.org>
In-Reply-To: <20230725052557.GI11388@unreal>
References: <20230723074504.3706691-1-linma@zju.edu.cn>
        <20230724174707.GB11388@unreal>
        <3c0760b5.e264b.1898a6368f8.Coremail.linma@zju.edu.cn>
        <20230725052557.GI11388@unreal>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Tue, 25 Jul 2023 08:25:57 +0300 Leon Romanovsky wrote:
> > Yeah I have seen that. Just as Jakub said, empty netlink attributes are valid 
> > (they are viewed as flag). The point is that different attribute has different
> > length requirement. For this specific code, the RDMA_NLDEV_ATTR_STAT_HWCOUNTERS
> > attribute is a nested one whose inner attributes should be NLA_U32. But as you
> > can see in variable nldev_policy, the description does not use nested policy to
> > enfore that, which results in the bug discussed in my commit message.
> > 
> >  [RDMA_NLDEV_ATTR_STAT_HWCOUNTERS]       = { .type = NLA_NESTED },
> > 
> > The elegant fix could be add the nested policy description to nldev_policy while
> > this is toublesome as no existing nla_attr has been given to this nested nlattr.
> > Hence, add the length check is the simplest solution and you can see such nla_len
> > check code all over the kernel.  
> 
> Right, and this is what bothers me.
> 
> I would more than happy to change nla_for_each_nested() to be something
> like nla_for_each_nested_type(...., sizeof(u32)), which will skip empty
> lines, for code which can't have them.

In general the idea of auto-skipping stuff kernel doesn't recognize
is a bit old school. Better direction would be extending the policy
validation to cover use cases for such loops.