diff mbox series

sctp: initialize endpoint LSM labels also on the client side

Message ID 20211021153846.745289-1-omosnace@redhat.com (mailing list archive)
State New, archived
Headers show
Series sctp: initialize endpoint LSM labels also on the client side | expand

Commit Message

Ondrej Mosnacek Oct. 21, 2021, 3:38 p.m. UTC
The secid* fields in struct sctp_endpoint are used to initialize the
labels of a peeloff socket created from the given association. Currently
they are initialized properly when a new association is created on the
server side (upon receiving an INIT packet), but not on the client side.

As a result, when the client obtains a peeloff socket via
sctp_peeloff(3) under SELinux, it ends up unlabeled, leading to
unexpected denials.

Fix this by calling the security_sctp_assoc_request() hook also upon
receiving a valid INIT-ACK response from the server, so that the
endpoint labels are properly initialized also on the client side.

Fixes: 2277c7cd75e3 ("sctp: Add LSM hooks")
Cc: Richard Haines <richard_c_haines@btinternet.com>
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
---
 include/net/sctp/structs.h | 11 ++++++-----
 net/sctp/sm_statefuns.c    |  5 +++++
 2 files changed, 11 insertions(+), 5 deletions(-)

Comments

Ondrej Mosnacek Oct. 21, 2021, 3:52 p.m. UTC | #1
On Thu, Oct 21, 2021 at 5:38 PM Ondrej Mosnacek <omosnace@redhat.com> wrote:
> The secid* fields in struct sctp_endpoint are used to initialize the
> labels of a peeloff socket created from the given association. Currently
> they are initialized properly when a new association is created on the
> server side (upon receiving an INIT packet), but not on the client side.
>
> As a result, when the client obtains a peeloff socket via
> sctp_peeloff(3) under SELinux, it ends up unlabeled, leading to
> unexpected denials.
>
> Fix this by calling the security_sctp_assoc_request() hook also upon
> receiving a valid INIT-ACK response from the server, so that the
> endpoint labels are properly initialized also on the client side.
>
> Fixes: 2277c7cd75e3 ("sctp: Add LSM hooks")
> Cc: Richard Haines <richard_c_haines@btinternet.com>
> Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
> ---
>  include/net/sctp/structs.h | 11 ++++++-----
>  net/sctp/sm_statefuns.c    |  5 +++++
>  2 files changed, 11 insertions(+), 5 deletions(-)

See also the selinux-testsuite [1] patch [2] that verifies this fix
(i.e. the new tests fail without this patch and pass with it). Not
being very familiar with SCTP, I'm not 100% sure if this fix is
correct or complete, so reviews are very much welcome.

[1] https://github.com/SELinuxProject/selinux-testsuite/
[2] https://patchwork.kernel.org/project/selinux/patch/20211021144543.740762-1-omosnace@redhat.com/

>
> diff --git a/include/net/sctp/structs.h b/include/net/sctp/structs.h
> index 651bba654d77..033a955592dd 100644
> --- a/include/net/sctp/structs.h
> +++ b/include/net/sctp/structs.h
> @@ -1356,11 +1356,12 @@ struct sctp_endpoint {
>
>         __u8  strreset_enable;
>
> -       /* Security identifiers from incoming (INIT). These are set by
> -        * security_sctp_assoc_request(). These will only be used by
> -        * SCTP TCP type sockets and peeled off connections as they
> -        * cause a new socket to be generated. security_sctp_sk_clone()
> -        * will then plug these into the new socket.
> +       /* Security identifiers from incoming (INIT/INIT-ACK). These
> +        * are set by security_sctp_assoc_request(). These will only
> +        * be used by SCTP TCP type sockets and peeled off connections
> +        * as they cause a new socket to be generated.
> +        * security_sctp_sk_clone() will then plug these into the new
> +        * socket.
>          */
>
>         u32 secid;
> diff --git a/net/sctp/sm_statefuns.c b/net/sctp/sm_statefuns.c
> index 32df65f68c12..cb291c7f5fb7 100644
> --- a/net/sctp/sm_statefuns.c
> +++ b/net/sctp/sm_statefuns.c
> @@ -521,6 +521,11 @@ enum sctp_disposition sctp_sf_do_5_1C_ack(struct net *net,
>         if (!sctp_vtag_verify(chunk, asoc))
>                 return sctp_sf_pdiscard(net, ep, asoc, type, arg, commands);
>
> +       /* Update socket peer label if first association. */
> +       if (security_sctp_assoc_request((struct sctp_endpoint *)ep,
> +                                       chunk->skb))
> +               return sctp_sf_pdiscard(net, ep, asoc, type, arg, commands);
> +
>         /* 6.10 Bundling
>          * An endpoint MUST NOT bundle INIT, INIT ACK or
>          * SHUTDOWN COMPLETE with any other chunks.
> --
> 2.31.1
>
Marcelo Ricardo Leitner Oct. 21, 2021, 3:55 p.m. UTC | #2
On Thu, Oct 21, 2021 at 05:38:46PM +0200, Ondrej Mosnacek wrote:
> The secid* fields in struct sctp_endpoint are used to initialize the
> labels of a peeloff socket created from the given association. Currently
> they are initialized properly when a new association is created on the
> server side (upon receiving an INIT packet), but not on the client side.

+Cc Xin
Xin Long Oct. 22, 2021, 6:32 a.m. UTC | #3
On Thu, Oct 21, 2021 at 11:55 PM Marcelo Ricardo Leitner
<marcelo.leitner@gmail.com> wrote:
>
> On Thu, Oct 21, 2021 at 05:38:46PM +0200, Ondrej Mosnacek wrote:
> > The secid* fields in struct sctp_endpoint are used to initialize the
> > labels of a peeloff socket created from the given association. Currently
> > they are initialized properly when a new association is created on the
> > server side (upon receiving an INIT packet), but not on the client side.
>
> +Cc Xin
Thanks Marcelo,

security_sctp_assoc_request() is not supposed to call on the client side,
as we can see on TCP. The client side's labels should be set to the
connection by selinux_inet_conn_request(). But we can't do it based
on the current hooks.

The root problem is that the current hooks incorrectly treat sctp_endpoint
in SCTP as request_sock in TCP, while it should've been sctp_association.
We need a bigger change on the current security sctp code.

I will post the patch series in hand, please take a look.
Ondrej Mosnacek Oct. 25, 2021, 7:55 a.m. UTC | #4
On Fri, Oct 22, 2021 at 8:33 AM Xin Long <lucien.xin@gmail.com> wrote:
> On Thu, Oct 21, 2021 at 11:55 PM Marcelo Ricardo Leitner
> <marcelo.leitner@gmail.com> wrote:
> >
> > On Thu, Oct 21, 2021 at 05:38:46PM +0200, Ondrej Mosnacek wrote:
> > > The secid* fields in struct sctp_endpoint are used to initialize the
> > > labels of a peeloff socket created from the given association. Currently
> > > they are initialized properly when a new association is created on the
> > > server side (upon receiving an INIT packet), but not on the client side.
> >
> > +Cc Xin
> Thanks Marcelo,
>
> security_sctp_assoc_request() is not supposed to call on the client side,
> as we can see on TCP. The client side's labels should be set to the
> connection by selinux_inet_conn_request(). But we can't do it based
> on the current hooks.
>
> The root problem is that the current hooks incorrectly treat sctp_endpoint
> in SCTP as request_sock in TCP, while it should've been sctp_association.
> We need a bigger change on the current security sctp code.
>
> I will post the patch series in hand, please take a look.

Thanks, your patches indeed seem to do the right thing and they also
do pass selinux-testsuite with the added client peeloff tests (as also
confirmed by Richard already). I have just a few minor comments, which
I'll send as replies to the individual patches.

--
Ondrej Mosnacek
Software Engineer, Linux Security - SELinux kernel
Red Hat, Inc.
diff mbox series

Patch

diff --git a/include/net/sctp/structs.h b/include/net/sctp/structs.h
index 651bba654d77..033a955592dd 100644
--- a/include/net/sctp/structs.h
+++ b/include/net/sctp/structs.h
@@ -1356,11 +1356,12 @@  struct sctp_endpoint {
 
 	__u8  strreset_enable;
 
-	/* Security identifiers from incoming (INIT). These are set by
-	 * security_sctp_assoc_request(). These will only be used by
-	 * SCTP TCP type sockets and peeled off connections as they
-	 * cause a new socket to be generated. security_sctp_sk_clone()
-	 * will then plug these into the new socket.
+	/* Security identifiers from incoming (INIT/INIT-ACK). These
+	 * are set by security_sctp_assoc_request(). These will only
+	 * be used by SCTP TCP type sockets and peeled off connections
+	 * as they cause a new socket to be generated.
+	 * security_sctp_sk_clone() will then plug these into the new
+	 * socket.
 	 */
 
 	u32 secid;
diff --git a/net/sctp/sm_statefuns.c b/net/sctp/sm_statefuns.c
index 32df65f68c12..cb291c7f5fb7 100644
--- a/net/sctp/sm_statefuns.c
+++ b/net/sctp/sm_statefuns.c
@@ -521,6 +521,11 @@  enum sctp_disposition sctp_sf_do_5_1C_ack(struct net *net,
 	if (!sctp_vtag_verify(chunk, asoc))
 		return sctp_sf_pdiscard(net, ep, asoc, type, arg, commands);
 
+	/* Update socket peer label if first association. */
+	if (security_sctp_assoc_request((struct sctp_endpoint *)ep,
+					chunk->skb))
+		return sctp_sf_pdiscard(net, ep, asoc, type, arg, commands);
+
 	/* 6.10 Bundling
 	 * An endpoint MUST NOT bundle INIT, INIT ACK or
 	 * SHUTDOWN COMPLETE with any other chunks.