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

[Simon Horman <simon.horman@corigine.com>]

On Wed, May 31, 2023 at 06:05:29PM +0800, Hangyu Hua wrote:
> 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@gmail.com>
> > > > 
> > > > 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.

Understood. I agree that complex seems undesirable.