linux-kernel - Re: [PATCH] net/mlx4_en: fix off by one in error handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <alpine.LFD.2.20.1609141757040.1730@schleppi>
Date:   Wed, 14 Sep 2016 18:08:44 +0200 (CEST)
From:   Sebastian Ott <sebott@...ux.vnet.ibm.com>
To:     Tariq Toukan <ttoukan.linux@...il.com>
cc:     Yishai Hadas <yishaih@...lanox.com>,
        Tariq Toukan <tariqt@...lanox.com>, netdev@...r.kernel.org,
        linux-rdma@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] net/mlx4_en: fix off by one in error handling

On Wed, 14 Sep 2016, Tariq Toukan wrote:
> On 14/09/2016 4:53 PM, Sebastian Ott wrote:
> > On Wed, 14 Sep 2016, Tariq Toukan wrote:
> > > On 14/09/2016 2:09 PM, Sebastian Ott wrote:
> > > > If an error occurs in mlx4_init_eq_table the index used in the
> > > > err_out_unmap label is one too big which results in a panic in
> > > > mlx4_free_eq. This patch fixes the index in the error path.
> > > You are right, but your change below does not cover all cases.
> > > The full solution looks like this:
> > >
> > > @@ -1260,7 +1260,7 @@ int mlx4_init_eq_table(struct mlx4_dev *dev)
> > >                                               eq);
> > >                  }
> > >                  if (err)
> > > -                       goto err_out_unmap;
> > > +                       goto err_out_unmap_excluded;
> > In this case a call to mlx4_create_eq failed. Do you really have to call
> > mlx4_free_eq for this index again?
>
> We agree on this part, that's why here we should goto the _excluded_ label.
> For all other parts, we should not exclude the eq in the highest index, and
> thus we goto the _non_excluded_ label.

But that's exactly what the original patch does. If the failure is within
the for loop at index i, we do the cleanup starting at index i-1. If the
failure is after the for loop then i == dev->caps.num_comp_vectors + 1
and we do the cleanup starting at index i == dev->caps.num_comp_vectors.

In the latter case your patch would have an out of bounds array access.

Regards,
Sebastian