In preparation to enabling -Wvla, remove VLA and replace it
with dynamic memory allocation.
From a security viewpoint, the use of Variable Length Arrays can be
a vector for stack overflow attacks. Also, in general, as the code
evolves it is easy to lose track of how big a VLA can get. Thus, we
can end up having segfaults that are hard to debug.
Also, fixed as part of the directive to remove all VLAs from
the kernel: https://lkml.org/lkml/2018/3/7/621
Signed-off-by: Gustavo A. R. Silva <[email protected]>
---
net/netfilter/nfnetlink_cttimeout.c | 26 +++++++++++++++++---------
1 file changed, 17 insertions(+), 9 deletions(-)
diff --git a/net/netfilter/nfnetlink_cttimeout.c b/net/netfilter/nfnetlink_cttimeout.c
index 6819300..dcd7bd3 100644
--- a/net/netfilter/nfnetlink_cttimeout.c
+++ b/net/netfilter/nfnetlink_cttimeout.c
@@ -51,19 +51,27 @@ ctnl_timeout_parse_policy(void *timeouts,
const struct nf_conntrack_l4proto *l4proto,
struct net *net, const struct nlattr *attr)
{
+ struct nlattr **tb;
int ret = 0;
- if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
- struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
+ if (!l4proto->ctnl_timeout.nlattr_to_obj)
+ return 0;
- ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
- attr, l4proto->ctnl_timeout.nla_policy,
- NULL);
- if (ret < 0)
- return ret;
+ tb = kcalloc(l4proto->ctnl_timeout.nlattr_max + 1, sizeof(*tb),
+ GFP_KERNEL);
- ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
- }
+ if (!tb)
+ return -ENOMEM;
+
+ ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max, attr,
+ l4proto->ctnl_timeout.nla_policy, NULL);
+ if (ret < 0)
+ goto err;
+
+ ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
+
+err:
+ kfree(tb);
return ret;
}
--
2.7.4
On Mon, 2018-03-12 at 18:14 -0500, Gustavo A. R. Silva wrote:
> In preparation to enabling -Wvla, remove VLA and replace it
> with dynamic memory allocation.
>
> From a security viewpoint, the use of Variable Length Arrays can be
> a vector for stack overflow attacks. Also, in general, as the code
> evolves it is easy to lose track of how big a VLA can get. Thus, we
> can end up having segfaults that are hard to debug.
>
> Also, fixed as part of the directive to remove all VLAs from
[]
> diff --git a/net/netfilter/nfnetlink_cttimeout.c b/net/netfilter/nfnetlink_cttimeout.c
[]
> @@ -51,19 +51,27 @@ ctnl_timeout_parse_policy(void *timeouts,
> const struct nf_conntrack_l4proto *l4proto,
> struct net *net, const struct nlattr *attr)
> {
> + struct nlattr **tb;
> int ret = 0;
>
> - if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
> - struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
> + if (!l4proto->ctnl_timeout.nlattr_to_obj)
> + return 0;
Why not
if unlikely(!...)
>
> - ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max,
> - attr, l4proto->ctnl_timeout.nla_policy,
> - NULL);
> - if (ret < 0)
> - return ret;
> + tb = kcalloc(l4proto->ctnl_timeout.nlattr_max + 1, sizeof(*tb),
> + GFP_KERNEL);
kmalloc_array?
>
> - ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
> - }
> + if (!tb)
> + return -ENOMEM;
> +
> + ret = nla_parse_nested(tb, l4proto->ctnl_timeout.nlattr_max, attr,
> + l4proto->ctnl_timeout.nla_policy, NULL);
> + if (ret < 0)
> + goto err;
> +
> + ret = l4proto->ctnl_timeout.nlattr_to_obj(tb, net, timeouts);
> +
> +err:
> + kfree(tb);
> return ret;
> }
>
On Mon, Mar 12, 2018 at 04:58:38PM -0700, Joe Perches wrote:
> On Mon, 2018-03-12 at 18:14 -0500, Gustavo A. R. Silva wrote:
> > In preparation to enabling -Wvla, remove VLA and replace it
> > with dynamic memory allocation.
> >
> > From a security viewpoint, the use of Variable Length Arrays can be
> > a vector for stack overflow attacks. Also, in general, as the code
> > evolves it is easy to lose track of how big a VLA can get. Thus, we
> > can end up having segfaults that are hard to debug.
> >
> > Also, fixed as part of the directive to remove all VLAs from
> []
> > diff --git a/net/netfilter/nfnetlink_cttimeout.c b/net/netfilter/nfnetlink_cttimeout.c
> []
> > @@ -51,19 +51,27 @@ ctnl_timeout_parse_policy(void *timeouts,
> > const struct nf_conntrack_l4proto *l4proto,
> > struct net *net, const struct nlattr *attr)
> > {
> > + struct nlattr **tb;
> > int ret = 0;
> >
> > - if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
> > - struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
> > + if (!l4proto->ctnl_timeout.nlattr_to_obj)
> > + return 0;
>
> Why not
> if unlikely(!...)
This is control plane code - not packet path - I think we should just
let the compiler decide on this one, not really need to provide an
explicit hint here.
On Tue, 2018-03-13 at 15:59 +0100, Pablo Neira Ayuso wrote:
> On Mon, Mar 12, 2018 at 04:58:38PM -0700, Joe Perches wrote:
> > On Mon, 2018-03-12 at 18:14 -0500, Gustavo A. R. Silva wrote:
> > > In preparation to enabling -Wvla, remove VLA and replace it
> > > with dynamic memory allocation.
> > >
> > > From a security viewpoint, the use of Variable Length Arrays can be
> > > a vector for stack overflow attacks. Also, in general, as the code
> > > evolves it is easy to lose track of how big a VLA can get. Thus, we
> > > can end up having segfaults that are hard to debug.
> > >
> > > Also, fixed as part of the directive to remove all VLAs from
> >
> > []
> > > diff --git a/net/netfilter/nfnetlink_cttimeout.c b/net/netfilter/nfnetlink_cttimeout.c
> >
> > []
> > > @@ -51,19 +51,27 @@ ctnl_timeout_parse_policy(void *timeouts,
> > > const struct nf_conntrack_l4proto *l4proto,
> > > struct net *net, const struct nlattr *attr)
> > > {
> > > + struct nlattr **tb;
> > > int ret = 0;
> > >
> > > - if (likely(l4proto->ctnl_timeout.nlattr_to_obj)) {
> > > - struct nlattr *tb[l4proto->ctnl_timeout.nlattr_max+1];
> > > + if (!l4proto->ctnl_timeout.nlattr_to_obj)
> > > + return 0;
> >
> > Why not
> > if unlikely(!...)
>
> This is control plane code - not packet path - I think we should just
> let the compiler decide on this one, not really need to provide an
> explicit hint here.
I don't have an issue with that, but it should probably be
mentioned in the changelog as it's unrelated to VLA removal.
On Mon, Mar 12, 2018 at 06:14:42PM -0500, Gustavo A. R. Silva wrote:
> In preparation to enabling -Wvla, remove VLA and replace it
> with dynamic memory allocation.
>
> From a security viewpoint, the use of Variable Length Arrays can be
> a vector for stack overflow attacks. Also, in general, as the code
> evolves it is easy to lose track of how big a VLA can get. Thus, we
> can end up having segfaults that are hard to debug.
>
> Also, fixed as part of the directive to remove all VLAs from
> the kernel: https://lkml.org/lkml/2018/3/7/621
Applied, thanks.