2020-06-16 02:15:44

by Xidong Wang

Subject: [PATCH 1/1] openvswitch: fix infoleak in conntrack

From: xidongwang <[email protected]>

The stack object “zone_limit” has 3 members. In function
ovs_ct_limit_get_default_limit(), the member "count" is
not initialized and sent out via “nla_put_nohdr”.

Signed-off-by: xidongwang <[email protected]>
---
net/openvswitch/conntrack.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
index 4340f25..1b7820a 100644
--- a/net/openvswitch/conntrack.c
+++ b/net/openvswitch/conntrack.c
@@ -2020,6 +2020,7 @@ static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
{
struct ovs_zone_limit zone_limit;
int err;
+ memset(&zone_limit, 0, sizeof(zone_limit));

zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
zone_limit.limit = info->default_limit;
--
2.7.4

2020-06-16 07:19:55

by Pravin Shelar

Subject: Re: [PATCH 1/1] openvswitch: fix infoleak in conntrack

On Mon, Jun 15, 2020 at 7:13 PM Xidong Wang <[email protected]> wrote:
>
> From: xidongwang <[email protected]>
>
> The stack object “zone_limit” has 3 members. In function
> ovs_ct_limit_get_default_limit(), the member "count" is
> not initialized and sent out via “nla_put_nohdr”.
>
> Signed-off-by: xidongwang <[email protected]>

Looks good.
Acked-by: Pravin B Shelar <[email protected]>

2020-06-16 07:49:34

by Tonghao Zhang

Subject: Re: [ovs-dev] [PATCH 1/1] openvswitch: fix infoleak in conntrack

On Tue, Jun 16, 2020 at 10:13 AM Xidong Wang <[email protected]> wrote:
>
> From: xidongwang <[email protected]>
>
> The stack object “zone_limit” has 3 members. In function
> ovs_ct_limit_get_default_limit(), the member "count" is
> not initialized and sent out via “nla_put_nohdr”.
>
> Signed-off-by: xidongwang <[email protected]>
> ---
> net/openvswitch/conntrack.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
> index 4340f25..1b7820a 100644
> --- a/net/openvswitch/conntrack.c
> +++ b/net/openvswitch/conntrack.c
> @@ -2020,6 +2020,7 @@ static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
> {
> struct ovs_zone_limit zone_limit;
> int err;
> + memset(&zone_limit, 0, sizeof(zone_limit));
why not init zone.count == 0, instead of memset, because zone_id/limit
will be inited later.
memset uses more cpu cycles.
> zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
> zone_limit.limit = info->default_limit;
> --
> 2.7.4
>
> _______________________________________________
> dev mailing list
> [email protected]
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev



--
Best regards, Tonghao

2020-06-16 08:45:11

by Simon Horman

Subject: Re: [ovs-dev] [PATCH 1/1] openvswitch: fix infoleak in conntrack

On Mon, Jun 15, 2020 at 07:13:01PM -0700, Xidong Wang wrote:
> From: xidongwang <[email protected]>
>
> The stack object “zone_limit” has 3 members. In function
> ovs_ct_limit_get_default_limit(), the member "count" is
> not initialized and sent out via “nla_put_nohdr”.

Hi Xidong,

thanks for your patch.

It appears that the patch is a fix. So I think that subject should be
targeted at the net tree and thus the subject should include
"[PATCH net]". (The other option being to target the net-next tree
in which case the subject should include "[PATCH net-next]".)

Also, as a fix it would be useful to include a fixes tag that references
the patch that introduced the problem. This is to facilitate backporting
to -stable branches of released kernels. In this case the following seems
appropriate.

Fixes: 11efd5cb04a1 ("openvswitch: Support conntrack zone limit")

> Signed-off-by: xidongwang <[email protected]>
> ---
> net/openvswitch/conntrack.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/net/openvswitch/conntrack.c b/net/openvswitch/conntrack.c
> index 4340f25..1b7820a 100644
> --- a/net/openvswitch/conntrack.c
> +++ b/net/openvswitch/conntrack.c
> @@ -2020,6 +2020,7 @@ static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
> {
> struct ovs_zone_limit zone_limit;
> int err;

There should be a blank line here.

> + memset(&zone_limit, 0, sizeof(zone_limit));

Moreover, initializing the entire structure to zero only to overwrite
most of its fields immediately below seems a bit inefficient.

Perhaps it would be better to just initialise count.

> zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
> zone_limit.limit = info->default_limit;
zone_limit.count = 0;

> --
> 2.7.4
>
> _______________________________________________
> dev mailing list
> [email protected]
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev