[PATCH mptcp-net] mptcp: avoid race on msk state changes

[PATCH mptcp-net] mptcp: avoid race on msk state changes

[Paolo Abeni]

The msk socket state is currently updated in a few spots without
owning the msk socket lock itself.

Some of such operations are safe, as they happens before exposing
the msk socket to user-space and can't race with other changes.

A couple of them, at connect time, can actually race with close()
or shutdown(), leaving breaking the socket state machine.

This change addresses the issue moving such update under the msk
socket lock with the usual:

<acquire spinlock>
<check sk lock onwers>
<ev defer to release_cb>

scheme.

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/56
Fixes: 8fd738049ac3 ("mptcp: fallback in case of simultaneous connect")
Fixes: c3c123d16c0e ("net: mptcp: don't hang in mptcp_sendmsg() after TCP fallbac")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
---
 net/mptcp/protocol.c |  2 ++
 net/mptcp/protocol.h |  2 ++
 net/mptcp/subflow.c  | 30 ++++++++++++++++++++++--------
 3 files changed, 26 insertions(+), 8 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 05c8382aafef..15c3b75516fb 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -2906,6 +2906,8 @@ static void mptcp_release_cb(struct sock *sk)
 		__mptcp_clean_una_wakeup(sk);
 	if (test_and_clear_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags))
 		__mptcp_error_report(sk);
+	if (test_and_clear_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags))
+		__mptcp_set_connected(sk);
 
 	/* push_pending may touch wmem_reserved, ensure we do the cleanup
 	 * later
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index 2480db50cbd2..515bb1e6acec 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -112,6 +112,7 @@
 #define MPTCP_ERROR_REPORT	8
 #define MPTCP_RETRANSMIT	9
 #define MPTCP_WORK_SYNC_SETSOCKOPT 10
+#define MPTCP_CONNECTED		11
 
 static inline bool before64(__u64 seq1, __u64 seq2)
 {
@@ -600,6 +601,7 @@ void mptcp_get_options(const struct sock *sk,
 		       struct mptcp_options_received *mp_opt);
 
 void mptcp_finish_connect(struct sock *sk);
+void __mptcp_set_connected(struct sock *sk);
 static inline bool mptcp_is_fully_established(struct sock *sk)
 {
 	return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
diff --git a/net/mptcp/subflow.c b/net/mptcp/subflow.c
index 6772802737e4..0b5d4a3eadcd 100644
--- a/net/mptcp/subflow.c
+++ b/net/mptcp/subflow.c
@@ -373,6 +373,24 @@ static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct soc
 	return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 }
 
+void __mptcp_set_connected(struct sock *sk)
+{
+	if (sk->sk_state == TCP_SYN_SENT) {
+		inet_sk_state_store(sk, TCP_ESTABLISHED);
+		sk->sk_state_change(sk);
+	}
+}
+
+static void mptcp_set_connected(struct sock *sk)
+{
+	mptcp_data_lock(sk);
+	if (!sock_owned_by_user(sk))
+		__mptcp_set_connected(sk);
+	else
+		set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags);
+	mptcp_data_unlock(sk);
+}
+
 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 {
 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
@@ -381,10 +399,6 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 
 	subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 
-	if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
-		inet_sk_state_store(parent, TCP_ESTABLISHED);
-		parent->sk_state_change(parent);
-	}
 
 	/* be sure no special action on any packet other than syn-ack */
 	if (subflow->conn_finished)
@@ -417,6 +431,7 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 			 subflow->remote_key);
 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 		mptcp_finish_connect(sk);
+		mptcp_set_connected(parent);
 	} else if (subflow->request_join) {
 		u8 hmac[SHA256_DIGEST_SIZE];
 
@@ -457,6 +472,7 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 	} else if (mptcp_check_fallback(sk)) {
 fallback:
 		mptcp_rcv_space_init(mptcp_sk(parent), sk);
+		mptcp_set_connected(parent);
 	}
 	return;
 
@@ -564,6 +580,7 @@ static void mptcp_sock_destruct(struct sock *sk)
 
 static void mptcp_force_close(struct sock *sk)
 {
+	/* the msk is not yet exposed to user-space */
 	inet_sk_state_store(sk, TCP_CLOSE);
 	sk_common_release(sk);
 }
@@ -1577,10 +1594,7 @@ static void subflow_state_change(struct sock *sk)
 		mptcp_rcv_space_init(mptcp_sk(parent), sk);
 		pr_fallback(mptcp_sk(parent));
 		subflow->conn_finished = 1;
-		if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
-			inet_sk_state_store(parent, TCP_ESTABLISHED);
-			parent->sk_state_change(parent);
-		}
+		mptcp_set_connected(parent);
 	}
 
 	/* as recvmsg() does not acquire the subflow socket for ssk selection
-- 
2.26.3

Re: [PATCH mptcp-net] mptcp: avoid race on msk state changes

[Mat Martineau]

On Wed, 16 Jun 2021, Paolo Abeni wrote:

> The msk socket state is currently updated in a few spots without
> owning the msk socket lock itself.
>
> Some of such operations are safe, as they happens before exposing
> the msk socket to user-space and can't race with other changes.
>
> A couple of them, at connect time, can actually race with close()
> or shutdown(), leaving breaking the socket state machine.
>
> This change addresses the issue moving such update under the msk
> socket lock with the usual:
>
> <acquire spinlock>
> <check sk lock onwers>
> <ev defer to release_cb>
>
> scheme.
>
> Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/56
> Fixes: 8fd738049ac3 ("mptcp: fallback in case of simultaneous connect")
> Fixes: c3c123d16c0e ("net: mptcp: don't hang in mptcp_sendmsg() after TCP fallbac")
> Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> ---
> net/mptcp/protocol.c |  2 ++
> net/mptcp/protocol.h |  2 ++
> net/mptcp/subflow.c  | 30 ++++++++++++++++++++++--------
> 3 files changed, 26 insertions(+), 8 deletions(-)
>
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index 05c8382aafef..15c3b75516fb 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -2906,6 +2906,8 @@ static void mptcp_release_cb(struct sock *sk)
> 		__mptcp_clean_una_wakeup(sk);
> 	if (test_and_clear_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags))
> 		__mptcp_error_report(sk);
> +	if (test_and_clear_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags))
> +		__mptcp_set_connected(sk);

Is it worth it to move the MPTCP_CONNECTED handling to be first in 
mptcp_release_cb()? Some of the other handlers would expect to have the 
msk connection state set first - it doesn't look like the other flags are 
likely to be set so early in the connection but I haven't (yet?) convinced 
myself it's impossible.

>
> 	/* push_pending may touch wmem_reserved, ensure we do the cleanup
> 	 * later
> diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
> index 2480db50cbd2..515bb1e6acec 100644
> --- a/net/mptcp/protocol.h
> +++ b/net/mptcp/protocol.h
> @@ -112,6 +112,7 @@
> #define MPTCP_ERROR_REPORT	8
> #define MPTCP_RETRANSMIT	9
> #define MPTCP_WORK_SYNC_SETSOCKOPT 10
> +#define MPTCP_CONNECTED		11
>
> static inline bool before64(__u64 seq1, __u64 seq2)
> {
> @@ -600,6 +601,7 @@ void mptcp_get_options(const struct sock *sk,
> 		       struct mptcp_options_received *mp_opt);
>
> void mptcp_finish_connect(struct sock *sk);
> +void __mptcp_set_connected(struct sock *sk);
> static inline bool mptcp_is_fully_established(struct sock *sk)
> {
> 	return inet_sk_state_load(sk) == TCP_ESTABLISHED &&
> diff --git a/net/mptcp/subflow.c b/net/mptcp/subflow.c
> index 6772802737e4..0b5d4a3eadcd 100644
> --- a/net/mptcp/subflow.c
> +++ b/net/mptcp/subflow.c
> @@ -373,6 +373,24 @@ static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct soc
> 	return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
> }
>
> +void __mptcp_set_connected(struct sock *sk)
> +{
> +	if (sk->sk_state == TCP_SYN_SENT) {
> +		inet_sk_state_store(sk, TCP_ESTABLISHED);
> +		sk->sk_state_change(sk);
> +	}
> +}
> +
> +static void mptcp_set_connected(struct sock *sk)
> +{
> +	mptcp_data_lock(sk);
> +	if (!sock_owned_by_user(sk))
> +		__mptcp_set_connected(sk);
> +	else
> +		set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags);
> +	mptcp_data_unlock(sk);
> +}
> +
> static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
> {
> 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
> @@ -381,10 +399,6 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
>
> 	subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
>
> -	if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
> -		inet_sk_state_store(parent, TCP_ESTABLISHED);
> -		parent->sk_state_change(parent);
> -	}
>

Minor: this leaves a double-newline in the code. Maybe Matthieu can touch 
it up, or if a v2 is needed for the mptcp_release_cb() change above you 
could fold this in.

> 	/* be sure no special action on any packet other than syn-ack */
> 	if (subflow->conn_finished)
> @@ -417,6 +431,7 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
> 			 subflow->remote_key);
> 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
> 		mptcp_finish_connect(sk);
> +		mptcp_set_connected(parent);
> 	} else if (subflow->request_join) {
> 		u8 hmac[SHA256_DIGEST_SIZE];
>
> @@ -457,6 +472,7 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
> 	} else if (mptcp_check_fallback(sk)) {
> fallback:
> 		mptcp_rcv_space_init(mptcp_sk(parent), sk);
> +		mptcp_set_connected(parent);
> 	}
> 	return;
>
> @@ -564,6 +580,7 @@ static void mptcp_sock_destruct(struct sock *sk)
>
> static void mptcp_force_close(struct sock *sk)
> {
> +	/* the msk is not yet exposed to user-space */
> 	inet_sk_state_store(sk, TCP_CLOSE);
> 	sk_common_release(sk);
> }
> @@ -1577,10 +1594,7 @@ static void subflow_state_change(struct sock *sk)
> 		mptcp_rcv_space_init(mptcp_sk(parent), sk);
> 		pr_fallback(mptcp_sk(parent));
> 		subflow->conn_finished = 1;
> -		if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
> -			inet_sk_state_store(parent, TCP_ESTABLISHED);
> -			parent->sk_state_change(parent);
> -		}
> +		mptcp_set_connected(parent);
> 	}
>
> 	/* as recvmsg() does not acquire the subflow socket for ssk selection
> -- 
> 2.26.3

--
Mat Martineau
Intel

Re: [PATCH mptcp-net] mptcp: avoid race on msk state changes

[Paolo Abeni]

On Wed, 2021-06-16 at 17:38 -0700, Mat Martineau wrote:
> On Wed, 16 Jun 2021, Paolo Abeni wrote:
> 
> > The msk socket state is currently updated in a few spots without
> > owning the msk socket lock itself.
> > 
> > Some of such operations are safe, as they happens before exposing
> > the msk socket to user-space and can't race with other changes.
> > 
> > A couple of them, at connect time, can actually race with close()
> > or shutdown(), leaving breaking the socket state machine.
> > 
> > This change addresses the issue moving such update under the msk
> > socket lock with the usual:
> > 
> > <acquire spinlock>
> > <check sk lock onwers>
> > <ev defer to release_cb>
> > 
> > scheme.
> > 
> > Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/56
> > Fixes: 8fd738049ac3 ("mptcp: fallback in case of simultaneous connect")
> > Fixes: c3c123d16c0e ("net: mptcp: don't hang in mptcp_sendmsg() after TCP fallbac")
> > Signed-off-by: Paolo Abeni <pabeni@redhat.com>
> > ---
> > net/mptcp/protocol.c |  2 ++
> > net/mptcp/protocol.h |  2 ++
> > net/mptcp/subflow.c  | 30 ++++++++++++++++++++++--------
> > 3 files changed, 26 insertions(+), 8 deletions(-)
> > 
> > diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> > index 05c8382aafef..15c3b75516fb 100644
> > --- a/net/mptcp/protocol.c
> > +++ b/net/mptcp/protocol.c
> > @@ -2906,6 +2906,8 @@ static void mptcp_release_cb(struct sock *sk)
> > 		__mptcp_clean_una_wakeup(sk);
> > 	if (test_and_clear_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags))
> > 		__mptcp_error_report(sk);
> > +	if (test_and_clear_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags))
> > +		__mptcp_set_connected(sk);
> 
> Is it worth it to move the MPTCP_CONNECTED handling to be first in 
> mptcp_release_cb()? Some of the other handlers would expect to have the 
> msk connection state set first

Good point! I placed it towards the bottom mainly to avoid processing
such bit before the MPTCP_PUSH_PENDING/MPTCP_RETRANSMIT loop, otherwise
we should check CONNECTED twice.

Double checking the existing code, it looks like
only MPTCP_ERROR_REPORT and MPTCP_CLEAN_UNA depends on it - and the
latter should never be set before MPTCP_CONNECTED, I think.

Moving the MPTCP_CONNECTED before ERROR_REPORT/CLEAN_UNA and
after PUSH_PENDING/RETRANSMIT should fit.

Thanks!

Paolo