Skip to content

Commit

Permalink
mptcp: don't overwrite sock_ops in mptcp_is_tcpsk()
Browse files Browse the repository at this point in the history 
Eric Dumazet suggests:

 > The fact that mptcp_is_tcpsk() was able to write over sock->ops was a
 > bit strange to me.
 > mptcp_is_tcpsk() should answer a question, with a read-only argument.

re-factor code to avoid overwriting sock_ops inside that function. Also,
change the helper name to reflect the semantics and to disambiguate from
its dual, sk_is_mptcp(). While at it, collapse mptcp_stream_accept() and
mptcp_accept() into a single function, where fallback / non-fallback are
separated into a single sk_is_mptcp() conditional.

Link: multipath-tcp/mptcp_net-next#432
Suggested-by: Eric Dumazet <[email protected]>
Signed-off-by: Davide Caratti <[email protected]>
  • Loading branch information
@dcaratti @intel-lab-lkp
dcaratti authored and intel-lab-lkp committed Dec 11, 2023
1 parent 8485436 commit 5eca689
Showing 1 changed file with 44 additions and 64 deletions.
108 changes: 44 additions & 64 deletions net/mptcp/protocol.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -55,28 +55,14 @@ u64 mptcp_wnd_end(const struct mptcp_sock *msk)
return READ_ONCE(msk->wnd_end);
}

static bool mptcp_is_tcpsk(struct sock *sk)
static const struct proto_ops *mptcp_fallback_tcp_ops(const struct sock *sk)
{
struct socket *sock = sk->sk_socket;

if (unlikely(sk->sk_prot == &tcp_prot)) {
/* we are being invoked after mptcp_accept() has
* accepted a non-mp-capable flow: sk is a tcp_sk,
* not an mptcp one.
*
* Hand the socket over to tcp so all further socket ops
* bypass mptcp.
*/
WRITE_ONCE(sock->ops, &inet_stream_ops);
return true;
#if IS_ENABLED(CONFIG_MPTCP_IPV6)
} else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
WRITE_ONCE(sock->ops, &inet6_stream_ops);
return true;
if (sk->sk_prot == &tcpv6_prot)
return &inet6_stream_ops;
#endif
}

return false;
WARN_ON_ONCE(sk->sk_prot != &tcp_prot);
return &inet_stream_ops;
}

static int __mptcp_socket_create(struct mptcp_sock *msk)
Expand Down Expand Up @@ -3258,44 +3244,6 @@ void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
}

static struct sock *mptcp_accept(struct sock *ssk, int flags, int *err,
bool kern)
{
struct sock *newsk;

pr_debug("ssk=%p, listener=%p", ssk, mptcp_subflow_ctx(ssk));
newsk = inet_csk_accept(ssk, flags, err, kern);
if (!newsk)
return NULL;

pr_debug("newsk=%p, subflow is mptcp=%d", newsk, sk_is_mptcp(newsk));
if (sk_is_mptcp(newsk)) {
struct mptcp_subflow_context *subflow;
struct sock *new_mptcp_sock;

subflow = mptcp_subflow_ctx(newsk);
new_mptcp_sock = subflow->conn;

/* is_mptcp should be false if subflow->conn is missing, see
* subflow_syn_recv_sock()
*/
if (WARN_ON_ONCE(!new_mptcp_sock)) {
tcp_sk(newsk)->is_mptcp = 0;
goto out;
}

newsk = new_mptcp_sock;
MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
} else {
MPTCP_INC_STATS(sock_net(ssk),
MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
}

out:
newsk->sk_kern_sock = kern;
return newsk;
}

void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
{
struct mptcp_subflow_context *subflow, *tmp;
Expand Down Expand Up @@ -3738,7 +3686,6 @@ static struct proto mptcp_prot = {
.connect = mptcp_connect,
.disconnect = mptcp_disconnect,
.close = mptcp_close,
.accept = mptcp_accept,
.setsockopt = mptcp_setsockopt,
.getsockopt = mptcp_getsockopt,
.shutdown = mptcp_shutdown,
Expand Down Expand Up @@ -3848,18 +3795,36 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
if (!ssk)
return -EINVAL;

newsk = mptcp_accept(ssk, flags, &err, kern);
pr_debug("ssk=%p, listener=%p", ssk, mptcp_subflow_ctx(ssk));
newsk = inet_csk_accept(ssk, flags, &err, kern);
if (!newsk)
return err;

lock_sock(newsk);

__inet_accept(sock, newsock, newsk);
if (!mptcp_is_tcpsk(newsock->sk)) {
struct mptcp_sock *msk = mptcp_sk(newsk);
pr_debug("newsk=%p, subflow is mptcp=%d", newsk, sk_is_mptcp(newsk));
if (sk_is_mptcp(newsk)) {
struct mptcp_subflow_context *subflow;
struct sock *new_mptcp_sock;

subflow = mptcp_subflow_ctx(newsk);
new_mptcp_sock = subflow->conn;

/* is_mptcp should be false if subflow->conn is missing, see
* subflow_syn_recv_sock()
*/
if (WARN_ON_ONCE(!new_mptcp_sock)) {
tcp_sk(newsk)->is_mptcp = 0;
goto tcpfallback;
}

newsk = new_mptcp_sock;
MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPCAPABLEPASSIVEACK);

newsk->sk_kern_sock = kern;
lock_sock(newsk);
__inet_accept(sock, newsock, newsk);

set_bit(SOCK_CUSTOM_SOCKOPT, &newsock->flags);
msk = mptcp_sk(newsk);
msk->in_accept_queue = 0;

/* set ssk->sk_socket of accept()ed flows to mptcp socket.
Expand All @@ -3881,6 +3846,21 @@ static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
if (unlikely(list_is_singular(&msk->conn_list)))
inet_sk_state_store(newsk, TCP_CLOSE);
}
} else {
MPTCP_INC_STATS(sock_net(ssk),
MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
tcpfallback:
newsk->sk_kern_sock = kern;
lock_sock(newsk);
__inet_accept(sock, newsock, newsk);
/* we are being invoked after accepting a non-mp-capable
* flow: sk is a tcp_sk, not an mptcp one.
*
* Hand the socket over to tcp so all further socket ops
* bypass mptcp.
*/
WRITE_ONCE(newsock->sk->sk_socket->ops,
mptcp_fallback_tcp_ops(newsock->sk));
}
release_sock(newsk);

Expand Down

0 comments on commit 5eca689

Please sign in to comment.