net: Introduce struct sockaddr_unspec

[PATCH v3 2/9] net/l2tp: Add missing sa_family validation in pppol2tp_sockaddr_get_info

Posted by Kees Cook 3 months, 2 weeks ago

While reviewing the struct proto_ops connect() and bind() callback
implementations, I noticed that there doesn't appear to be any
validation that AF_PPPOX sockaddr structures actually have sa_family set
to AF_PPPOX. The pppol2tp_sockaddr_get_info() checks only look at the
sizes.

I don't see any way that this might actually cause problems as specific
info fields are being populated, for which the existing size checks are
correct, but it stood out as a missing address family check.

Add the check and return -EAFNOSUPPORT on mismatch.

Signed-off-by: Kees Cook <kees@kernel.org>
---
 net/l2tp/l2tp_ppp.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/net/l2tp/l2tp_ppp.c b/net/l2tp/l2tp_ppp.c
index 5e12e7ce17d8..b7a9c224520f 100644
--- a/net/l2tp/l2tp_ppp.c
+++ b/net/l2tp/l2tp_ppp.c
@@ -535,6 +535,13 @@ struct l2tp_connect_info {
 static int pppol2tp_sockaddr_get_info(const void *sa, int sa_len,
 				      struct l2tp_connect_info *info)
 {
+	const struct sockaddr_unspec *sockaddr = sa;
+
+	if (sa_len < offsetofend(struct sockaddr, sa_family))
+		return -EINVAL;
+	if (sockaddr->sa_family != AF_PPPOX)
+		return -EAFNOSUPPORT;
+
 	switch (sa_len) {
 	case sizeof(struct sockaddr_pppol2tp):
 	{
-- 
2.34.1

Re: [PATCH v3 2/9] net/l2tp: Add missing sa_family validation in pppol2tp_sockaddr_get_info

Posted by Paolo Abeni 3 months, 2 weeks ago

On 10/20/25 11:26 PM, Kees Cook wrote:
> While reviewing the struct proto_ops connect() and bind() callback
> implementations, I noticed that there doesn't appear to be any
> validation that AF_PPPOX sockaddr structures actually have sa_family set
> to AF_PPPOX. The pppol2tp_sockaddr_get_info() checks only look at the
> sizes.
> 
> I don't see any way that this might actually cause problems as specific
> info fields are being populated, for which the existing size checks are
> correct, but it stood out as a missing address family check.
> 
> Add the check and return -EAFNOSUPPORT on mismatch.
> 
> Signed-off-by: Kees Cook <kees@kernel.org>
> ---
>  net/l2tp/l2tp_ppp.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/net/l2tp/l2tp_ppp.c b/net/l2tp/l2tp_ppp.c
> index 5e12e7ce17d8..b7a9c224520f 100644
> --- a/net/l2tp/l2tp_ppp.c
> +++ b/net/l2tp/l2tp_ppp.c
> @@ -535,6 +535,13 @@ struct l2tp_connect_info {
>  static int pppol2tp_sockaddr_get_info(const void *sa, int sa_len,
>  				      struct l2tp_connect_info *info)
>  {
> +	const struct sockaddr_unspec *sockaddr = sa;
> +
> +	if (sa_len < offsetofend(struct sockaddr, sa_family))
> +		return -EINVAL;
> +	if (sockaddr->sa_family != AF_PPPOX)
> +		return -EAFNOSUPPORT;

I fear we can't introduce this check, as it could break existing
user-space application currently passing random data into sa_family but
still able to connect successfully.

/P

Re: [PATCH v3 2/9] net/l2tp: Add missing sa_family validation in pppol2tp_sockaddr_get_info

Posted by Kees Cook 3 months, 2 weeks ago

On Thu, Oct 23, 2025 at 12:47:32PM +0200, Paolo Abeni wrote:
> On 10/20/25 11:26 PM, Kees Cook wrote:
> > While reviewing the struct proto_ops connect() and bind() callback
> > implementations, I noticed that there doesn't appear to be any
> > validation that AF_PPPOX sockaddr structures actually have sa_family set
> > to AF_PPPOX. The pppol2tp_sockaddr_get_info() checks only look at the
> > sizes.
> > 
> > I don't see any way that this might actually cause problems as specific
> > info fields are being populated, for which the existing size checks are
> > correct, but it stood out as a missing address family check.
> > 
> > Add the check and return -EAFNOSUPPORT on mismatch.
> > 
> > Signed-off-by: Kees Cook <kees@kernel.org>
> > ---
> >  net/l2tp/l2tp_ppp.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> > 
> > diff --git a/net/l2tp/l2tp_ppp.c b/net/l2tp/l2tp_ppp.c
> > index 5e12e7ce17d8..b7a9c224520f 100644
> > --- a/net/l2tp/l2tp_ppp.c
> > +++ b/net/l2tp/l2tp_ppp.c
> > @@ -535,6 +535,13 @@ struct l2tp_connect_info {
> >  static int pppol2tp_sockaddr_get_info(const void *sa, int sa_len,
> >  				      struct l2tp_connect_info *info)
> >  {
> > +	const struct sockaddr_unspec *sockaddr = sa;
> > +
> > +	if (sa_len < offsetofend(struct sockaddr, sa_family))
> > +		return -EINVAL;
> > +	if (sockaddr->sa_family != AF_PPPOX)
> > +		return -EAFNOSUPPORT;
> 
> I fear we can't introduce this check, as it could break existing
> user-space application currently passing random data into sa_family but
> still able to connect successfully.

Isn't sa_family kind of the critical determining factor on how the
network stack decides to handle sockaddr stuff? I'll drop it for now,
I guess, but that's surprising to me.

-Kees

-- 
Kees Cook