Re: [PATCH net v3 09/11] rxrpc: Fix keyring reference count leak in rxrpc_setsockopt()

[Anderson Nascimento]

On Thu, Mar 26, 2026 at 10:19 AM David Howells <dhowells@xxxxxxxxxx> wrote:
>
> From: Anderson Nascimento <anderson@xxxxxxxxxxxxxxxxxx>
>
> In rxrpc_setsockopt(), the code checks 'rx->key' when handling the
> RXRPC_SECURITY_KEYRING option.  However, this appears to be a logic error.
> The code should be checking 'rx->securities' to determine if a keyring has
> already been defined for the socket.
>
> Currently, if a user calls setsockopt(RXRPC_SECURITY_KEYRING) multiple
> times on the same socket, the check 'if (rx->key)' fails to block
> subsequent calls because 'rx->key' has not been defined by the function.
> This results in a reference count leak on the keyring.
>
> This patch changes the check to 'rx->securities' to correctly identify if
> the socket security keyring has already been configured, returning -EINVAL
> on subsequent attempts.
>
> Before the patch:
>
> It shows the keyring reference counter elevated.
>
> $ cat /proc/keys | grep AFSkeys1
> 27aca8ae I--Q--- 24469721 perm 3f010000  1000  1000 keyring   AFSkeys1: empty
> $
>
> After the patch:
>
> The keyring reference counter remains stable and subsequent calls return an
> error:
>
> $ ./poc
> setsockopt: Invalid argument
> $
>
> Fixes: 17926a79320a ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
> Signed-off-by: Anderson Nascimento <anderson@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: David Howells <dhowells@xxxxxxxxxx>
> Reviewed-by: Jeffrey Altman <jaltman@xxxxxxxxxxxx>
> cc: Marc Dionne <marc.dionne@xxxxxxxxxxxx>
> cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> cc: "David S. Miller" <davem@xxxxxxxxxxxxx>
> cc: Jakub Kicinski <kuba@xxxxxxxxxx>
> cc: Paolo Abeni <pabeni@xxxxxxxxxx>
> cc: Simon Horman <horms@xxxxxxxxxx>
> cc: linux-afs@xxxxxxxxxxxxxxxxxxx
> cc: netdev@xxxxxxxxxxxxxxx
> cc: stable@xxxxxxxxxx
> ---
>  net/rxrpc/af_rxrpc.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/rxrpc/af_rxrpc.c b/net/rxrpc/af_rxrpc.c
> index 0f90272ac254..0b7ed99a3025 100644
> --- a/net/rxrpc/af_rxrpc.c
> +++ b/net/rxrpc/af_rxrpc.c
> @@ -665,7 +665,7 @@ static int rxrpc_setsockopt(struct socket *sock, int level, int optname,
>
>                 case RXRPC_SECURITY_KEYRING:
>                         ret = -EINVAL;
> -                       if (rx->key)
> +                       if (rx->securities)
>                                 goto error;
>                         ret = -EISCONN;
>                         if (rx->sk.sk_state != RXRPC_UNBOUND)
>

I am following this patchset along with sashiko's comments. This time,
it flagged this patch for an API order dependency. The setsockopt
RXRPC_SECURITY_KEY option checks for rx->securities before proceeding,
but after this change, the RXRPC_SECURITY_KEYRING option doesn't check
for rx->key. Consequently, it's possible to set both keys on a socket
depending on the setsockopt call order.

I noticed this and wondered about it before sending the patch, but I
decided to keep the patch as concise as possible rather than assuming
too much and potentially breaking code. To make the logic more
coherent, what if we check if (rx->key || rx->securities) in both
options and remove the rx->securities check from rxrpc_request_key()?

Based on my check, rxrpc_request_key() and rxrpc_server_keyring() are
only called by their setsockopt options. We could also add a comment
to rxrpc_request_key() and rxrpc_server_keyring() noting that key
validation should be handled by the caller. Is the -EINVAL error
appropriate for this case?

-- 
Anderson Nascimento
Allele Security Intelligence
https://www.allelesecurity.com