Re: [PATCH] net: sched: fix possible OOB write in fl_set_geneve_opt()

[Hangyu Hua]

On 31/5/2023 16:04, Simon Horman wrote:
> On Wed, May 31, 2023 at 01:38:49PM +0800, Hangyu Hua wrote:
> > On 30/5/2023 19:36, Simon Horman wrote:
> > > [Updated Pieter's email address, dropped old email address of mine]
> > >
> > > On Mon, May 29, 2023 at 12:36:15PM +0800, Hangyu Hua wrote:
> > > > If we send two TCA_FLOWER_KEY_ENC_OPTS_GENEVE packets and their total
> > > > size is 252 bytes(key->enc_opts.len = 252) then
> > > > key->enc_opts.len = opt->length = data_len / 4 = 0 when the third
> > > > TCA_FLOWER_KEY_ENC_OPTS_GENEVE packet enters fl_set_geneve_opt. This
> > > > bypasses the next bounds check and results in an out-of-bounds.
> > > >
> > > > Fixes: 0a6e77784f49 ("net/sched: allow flower to match tunnel options")
> > > > Signed-off-by: Hangyu Hua <hbh25y@xxxxxxxxx>
> > >
> > > Hi Hangyu Hua,
> > >
> > > Thanks. I think I see the problem too.
> > > But I do wonder, is this more general than Geneve options?
> > > That is, can this occur with any sequence of options, that
> > > consume space in enc_opts (configured in fl_set_key()) that
> > > in total are more than 256 bytes?
> >
> > I think you are right. It is a good idea to add check in fl_set_vxlan_opt
> > and fl_set_erspan_opt and fl_set_gtp_opt too.
> > But they should be submitted as other patches. fl_set_geneve_opt has already
> > check this with the following code:
> >
> > static int fl_set_geneve_opt(const struct nlattr *nla, struct fl_flow_key
> > *key,
> > 			     int depth, int option_len,
> > 			     struct netlink_ext_ack *extack)
> > {
> > ...
> > 		if (new_len > FLOW_DIS_TUN_OPTS_MAX) {
> > 			NL_SET_ERR_MSG(extack, "Tunnel options exceeds max size");
> > 			return -ERANGE;
> > 		}
> > ...
> > }
> >
> > This bug will only be triggered under this special
> > condition(key->enc_opts.len = 252). So I think it will be better understood
> > by submitting this patch independently.
>
> A considered approach sounds good to me.
>
> I do wonder, could the bounds checks be centralised in the caller?
> Maybe not if it doesn't know the length that will be consumed.

This may make code more complex. I am not sure if it is necessary to do 
this.

> > By the way, I think memset's third param should be option_len in
> > fl_set_vxlan_opt and fl_set_erspan_opt. Do I need to submit another patch to
> > fix all these issues?
>
> I think that in general one fix per patch is best.

I see. I will try to handle these issues.

> Some minor nits.
>
> 1. As this is a fix for networking code it is probably targeted
>     at the net, as opposed to net-next, tree. This should be indicated
>     in the patch subject.
>
> 	 Subject: [PATCH net v2] ...
>
> 2. I think the usual patch prefix for this file, of late,
>     has been 'net/sched: flower: '
>
> 	 Subject: [PATCH net v2]  net/sched: flower: ...

Get it. I will send a v2 later.