[PATCH net-next 0/2] kcm: fix two syzcaller issues

[PATCH net-next 0/2] kcm: fix two syzcaller issues

[Tom Herbert]

In this patch set:

- Don't allow attaching non-TCP or listener sockets to a KCM mux.
- In kcm_attach Check if sk_user_data is already set. This is
  under lock to avoid race conditions. More work is need to make
  all of the users of sk_user_data to use the same locking.

Tom Herbert (2):
  kcm: Only allow TCP sockets to be attached to a KCM mux
  kcm: Check if sk_user_data already set in kcm_attach

 net/kcm/kcmsock.c | 25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)

-- 
2.11.0

[PATCH net-next 1/2] kcm: Only allow TCP sockets to be attached to a KCM mux

[Tom Herbert]

TCP sockets for IPv4 and IPv6 that are not listeners or in closed
stated are allowed to be attached to a KCM mux.

Fixes: ab7ac4eb9832 ("kcm: Kernel Connection Multiplexor module")
Reported-by: syzbot+8865eaff7f9acd593945@syzkaller.appspotmail.com
Signed-off-by: Tom Herbert <tom@quantonium.net>
---
 net/kcm/kcmsock.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/net/kcm/kcmsock.c b/net/kcm/kcmsock.c
index d4e98f20fc2a..474ea2200592 100644
--- a/net/kcm/kcmsock.c
+++ b/net/kcm/kcmsock.c
@@ -1391,6 +1391,15 @@ static int kcm_attach(struct socket *sock, struct socket *csock,
 	if (csk->sk_family == PF_KCM)
 		return -EOPNOTSUPP;
 
+	/* Only allow TCP sockets to be attached for now */
+	if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
+	    csk->sk_protocol != IPPROTO_TCP)
+		return -EOPNOTSUPP;
+
+	/* Don't allow listeners or closed sockets */
+	if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE)
+		return -EOPNOTSUPP;
+
 	psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
 	if (!psock)
 		return -ENOMEM;
-- 
2.11.0

[PATCH net-next 2/2] kcm: Check if sk_user_data already set in kcm_attach

[Tom Herbert]

This is needed to prevent sk_user_data being overwritten.
The check is done under the callback lock. This should prevent
a socket from being attached twice to a KCM mux. It also prevents
a socket from being attached for other use cases of sk_user_data
as long as the other cases set sk_user_data under the lock.
Followup work is needed to unify all the use cases of sk_user_data
to use the same locking.

Reported-by: syzbot+114b15f2be420a8886c3@syzkaller.appspotmail.com
Fixes: ab7ac4eb9832 ("kcm: Kernel Connection Multiplexor module")
Signed-off-by: Tom Herbert <tom@quantonium.net>
---
 net/kcm/kcmsock.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/net/kcm/kcmsock.c b/net/kcm/kcmsock.c
index 474ea2200592..5ae02a5b0838 100644
--- a/net/kcm/kcmsock.c
+++ b/net/kcm/kcmsock.c
@@ -1414,9 +1414,18 @@ static int kcm_attach(struct socket *sock, struct socket *csock,
 		return err;
 	}
 
-	sock_hold(csk);
-
 	write_lock_bh(&csk->sk_callback_lock);
+
+	/* Check if sk_user_data is aready by KCM or someone else.
+	 * Must be done under lock to prevent race conditions.
+	 */
+	if (csk->sk_user_data) {
+		write_unlock_bh(&csk->sk_callback_lock);
+		strp_done(&psock->strp);
+		kmem_cache_free(kcm_psockp, psock);
+		return -EALREADY;
+	}
+
 	psock->save_data_ready = csk->sk_data_ready;
 	psock->save_write_space = csk->sk_write_space;
 	psock->save_state_change = csk->sk_state_change;
@@ -1424,8 +1433,11 @@ static int kcm_attach(struct socket *sock, struct socket *csock,
 	csk->sk_data_ready = psock_data_ready;
 	csk->sk_write_space = psock_write_space;
 	csk->sk_state_change = psock_state_change;
+
 	write_unlock_bh(&csk->sk_callback_lock);
 
+	sock_hold(csk);
+
 	/* Finished initialization, now add the psock to the MUX. */
 	spin_lock_bh(&mux->lock);
 	head = &mux->psocks;
-- 
2.11.0

Re: [PATCH net-next 1/2] kcm: Only allow TCP sockets to be attached to a KCM mux

[Guillaume Nault]

On Tue, Jan 23, 2018 at 05:00:13PM -0800, Tom Herbert wrote:
> TCP sockets for IPv4 and IPv6 that are not listeners or in closed
> stated are allowed to be attached to a KCM mux.
> 
> Fixes: ab7ac4eb9832 ("kcm: Kernel Connection Multiplexor module")
> Reported-by: syzbot+8865eaff7f9acd593945@syzkaller.appspotmail.com
> Signed-off-by: Tom Herbert <tom@quantonium.net>
> ---
>  net/kcm/kcmsock.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/net/kcm/kcmsock.c b/net/kcm/kcmsock.c
> index d4e98f20fc2a..474ea2200592 100644
> --- a/net/kcm/kcmsock.c
> +++ b/net/kcm/kcmsock.c
> @@ -1391,6 +1391,15 @@ static int kcm_attach(struct socket *sock, struct socket *csock,
>  	if (csk->sk_family == PF_KCM)
>  		return -EOPNOTSUPP;
>  
> +	/* Only allow TCP sockets to be attached for now */
> +	if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
> +	    csk->sk_protocol != IPPROTO_TCP)
> +		return -EOPNOTSUPP;
> +
This renders the 'csk->sk_family == PF_KCM' test useless. So why not
removing it?