Skip to contentSkip to conversation

Forward-ports for mptcp-net-next

PATCH In review Christoph Paasch 2 participants 7 messages Nov 30, 2018

Subsystems (3)
lore
Share & tools
lore

Christoph Paasch

author
#0 lore raw

cover Forward-ports for mptcp-net-next

Hello,

here are some forward-ports for mptcp-net-next that come from mptcp_trunk.

Cheers,
Christoph

Christoph Paasch (3):
mptcp: Remove IPv6 duplicated address detection
mptcp: Always call sk_data_ready
mptcp: Correctly initialize IPv6 fields

Kostas Peletidis (2):
Export tcp_send_ack symbol to fix build issue with mptcp_fullmesh
module (mptcp_trunk). This commit should make it possible to revert 64a80850.
Revert "Fix building mptcp_fullmesh as module"

 net/ipv4/tcp_input.c       |  2 +-
 net/ipv4/tcp_output.c      |  1 +
 net/mptcp/mptcp_ctrl.c     | 30 ++++++++--------
 net/mptcp/mptcp_fullmesh.c | 85 ++++++----------------------------------------
 4 files changed, 29 insertions(+), 89 deletions(-)
Signature · 1 line
2.16.2
author
#1 lore raw

PATCH 1/5 Export tcp_send_ack symbol to fix build issue with mptcp_fullmesh module (mptcp_trunk). This commit should make it possible to revert 64a80850.

From: Kostas Peletidis <kostas(a)peletidis.net>

(cherry picked from commit 6f1b2a04fed74a7eb82cc2db6563c61bd0896eba)
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
---
 net/ipv4/tcp_output.c | 1 +
 1 file changed, 1 insertion(+)
net/ipv4/tcp_output.c +1 -0
index 63caedf168fd..344fed7b7a5d 100644
@@ -3709,6 +3709,7 @@ void tcp_send_ack(struct sock *sk)
 {
 	__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
 }
+EXPORT_SYMBOL_GPL(tcp_send_ack);
 
 /* This routine sends a packet with an out of date sequence
  * number. It assumes the other end will try to ack it.
Signature · 1 line
2.16.2
author
#2 lore raw

PATCH 2/5 Revert "Fix building mptcp_fullmesh as module"

In reply to Christoph Paasch

From: Kostas Peletidis <kostas(a)peletidis.net>

This reverts commit 64a808507dc425759bed331742fdbb62f0061870.

(cherry picked from commit 4b720160a051d82439314990c007065b0329c4d9)
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
---
 net/mptcp/mptcp_fullmesh.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)
net/mptcp/mptcp_fullmesh.c +10 -10
index 636642287541..e699c0e89224 100644
@@ -606,7 +606,7 @@ static void announce_remove_addr(u8 addr_id, struct sock *meta_sk)
 	mpcb->addr_signal = 1;
 
 	if (sk)
-		__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+		tcp_send_ack(sk);
 }
 
 static void update_addr_bitfields(struct sock *meta_sk,
@@ -897,7 +897,7 @@ static void mptcp_address_worker(struct work_struct *work)
 
 				sk = mptcp_select_ack_sock(meta_sk);
 				if (sk)
-					__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+					tcp_send_ack(sk);
 
 				full_mesh_create_subflows(meta_sk);
 			}
@@ -980,7 +980,7 @@ static void mptcp_address_worker(struct work_struct *work)
 							tp->mptcp->send_mp_prio = 1;
 							tp->mptcp->low_prio = event->low_prio;
 
-							__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+							tcp_send_ack(sk);
 						}
 					}
 
@@ -991,7 +991,7 @@ static void mptcp_address_worker(struct work_struct *work)
 							tp->mptcp->send_mp_prio = 1;
 							tp->mptcp->low_prio = event->low_prio;
 
-							__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+							tcp_send_ack(sk);
 						}
 					}
 				}
@@ -1399,10 +1399,10 @@ static void full_mesh_new_session(const struct sock *meta_sk)
 		fmp->announced_addrs_v6 |= (1 << index);
 
 	for (i = fmp->add_addr; i && fmp->add_addr; i--)
-		__tcp_send_ack(mpcb->master_sk, tcp_sk(mpcb->master_sk)->rcv_nxt);
+		tcp_send_ack(mpcb->master_sk);
 
 	if (master_tp->mptcp->send_mp_prio)
-		__tcp_send_ack(mpcb->master_sk, tcp_sk(mpcb->master_sk)->rcv_nxt);
+		tcp_send_ack(mpcb->master_sk);
 
 	return;
 
@@ -1492,7 +1492,7 @@ static void full_mesh_release_sock(struct sock *meta_sk)
 				tp->mptcp->send_mp_prio = 1;
 				tp->mptcp->low_prio = mptcp_local->locaddr4[i].low_prio;
 
-				__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+				tcp_send_ack(sk);
 			}
 		}
 
@@ -1504,7 +1504,7 @@ static void full_mesh_release_sock(struct sock *meta_sk)
 
 			sk = mptcp_select_ack_sock(meta_sk);
 			if (sk)
-				__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+				tcp_send_ack(sk);
 			full_mesh_create_subflows(meta_sk);
 		}
 	}
@@ -1536,7 +1536,7 @@ static void full_mesh_release_sock(struct sock *meta_sk)
 				tp->mptcp->send_mp_prio = 1;
 				tp->mptcp->low_prio = mptcp_local->locaddr6[i].low_prio;
 
-				__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+				tcp_send_ack(sk);
 			}
 		}
 
@@ -1548,7 +1548,7 @@ static void full_mesh_release_sock(struct sock *meta_sk)
 
 			sk = mptcp_select_ack_sock(meta_sk);
 			if (sk)
-				__tcp_send_ack(sk, tcp_sk(sk)->rcv_nxt);
+				tcp_send_ack(sk);
 			full_mesh_create_subflows(meta_sk);
 		}
 	}
Signature · 1 line
2.16.2
author
#3 lore raw

PATCH 3/5 mptcp: Remove IPv6 duplicated address detection

In reply to Christoph Paasch

It causes us to take a refcnt on the address and actually periodically schedule a timer until eventually the address gets either marked as confirmed or removed.

It causes regressions in the Android test-suite.

This patch simply removes the duplicate address detection. That way, when we get an IPv6-address we use it right away. Also, if the address turns out to be duplicate on the link (which is really a cornercase scenario), it will be removed and our subflow will be destroyed. I don't think that's a big deal.

I don't suggest to backport this patch as it's more of a design-change than a bugfix.

Github-issue: https://github.com/multipath-tcp/mptcp/issues/293

Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
Signed-off-by Matthieu Baerts <matthieu.baerts(a)tessares.net>
(cherry picked from commit 8e3743b348121ac66aba0df3074aba12c40591c2)
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
---
 net/mptcp/mptcp_fullmesh.c | 65 +---------------------------------------------
 1 file changed, 1 insertion(+), 64 deletions(-)
net/mptcp/mptcp_fullmesh.c +1 -64
index e699c0e89224..23f42a06cc70 100644
@@ -1122,69 +1122,9 @@ static struct notifier_block mptcp_pm_inetaddr_notifier = {
 
 #if IS_ENABLED(CONFIG_IPV6)
 
-/* IPV6-related address/interface watchers */
-struct mptcp_dad_data {
-	struct timer_list timer;
-	struct inet6_ifaddr *ifa;
-};
-
-static void dad_callback(struct timer_list *t);
 static int inet6_addr_event(struct notifier_block *this, unsigned long event,
 			    void *ptr);
 
-static bool ipv6_dad_finished(const struct inet6_ifaddr *ifa)
-{
-	return !(ifa->flags & IFA_F_TENTATIVE) ||
-	       ifa->state > INET6_IFADDR_STATE_DAD;
-}
-
-static void dad_init_timer(struct mptcp_dad_data *data,
-			   struct inet6_ifaddr *ifa)
-{
-	data->ifa = ifa;
-	if (ifa->idev->cnf.rtr_solicit_delay)
-		data->timer.expires = jiffies + ifa->idev->cnf.rtr_solicit_delay;
-	else
-		data->timer.expires = jiffies + (HZ/10);
-}
-
-static void dad_callback(struct timer_list *t)
-{
-	struct mptcp_dad_data *data = from_timer(data, t, timer);
-
-	/* DAD failed or IP brought down? */
-	if (data->ifa->state == INET6_IFADDR_STATE_ERRDAD ||
-	    data->ifa->state == INET6_IFADDR_STATE_DEAD)
-		goto exit;
-
-	if (!ipv6_dad_finished(data->ifa)) {
-		dad_init_timer(data, data->ifa);
-		add_timer(&data->timer);
-		return;
-	}
-
-	inet6_addr_event(NULL, NETDEV_UP, data->ifa);
-
-exit:
-	in6_ifa_put(data->ifa);
-	kfree(data);
-}
-
-static inline void dad_setup_timer(struct inet6_ifaddr *ifa)
-{
-	struct mptcp_dad_data *data;
-
-	data = kmalloc(sizeof(*data), GFP_ATOMIC);
-
-	if (!data)
-		return;
-
-	timer_setup(&data->timer, dad_callback, 0);
-	dad_init_timer(data, ifa);
-	add_timer(&data->timer);
-	in6_ifa_hold(ifa);
-}
-
 static void addr6_event_handler(const struct inet6_ifaddr *ifa, unsigned long event,
 				struct net *net)
 {
@@ -1233,10 +1173,7 @@ static int inet6_addr_event(struct notifier_block *this, unsigned long event,
 	      event == NETDEV_CHANGE))
 		return NOTIFY_DONE;
 
-	if (!ipv6_dad_finished(ifa6))
-		dad_setup_timer(ifa6);
-	else
-		addr6_event_handler(ifa6, event, net);
+	addr6_event_handler(ifa6, event, net);
 
 	return NOTIFY_DONE;
 }
Signature · 1 line
2.16.2
author
#4 lore raw

PATCH 4/5 mptcp: Always call sk_data_ready

In reply to Christoph Paasch

Since commit 03f45c883c6f ("tcp: avoid extra wakeups for SO_RCVLOWAT users"), tcp_data_queue checks for available bytes in the subflow's receive-queue before waking up the "app". However, with MPTCP we might have zero-length data-fins in the queue. Thus, we still need to wakeup the MPTCP-layer in case of a subflow.

Fixes dd64d1c40b48 ("Merge tag 'v4.18' into mptcp_trunk")
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
Signed-off-by Matthieu Baerts <matthieu.baerts(a)tessares.net>
(cherry picked from commit 8562bd5e62968e4bbdbf897decf933e90210a139)
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
---
 net/ipv4/tcp_input.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ipv4/tcp_input.c +1 -1
index c0efc3de8683..0f1ac709f15d 100644
@@ -4741,7 +4741,7 @@ void tcp_data_ready(struct sock *sk)
 	const struct tcp_sock *tp = tcp_sk(sk);
 	int avail = tp->rcv_nxt - tp->copied_seq;
 
-	if (avail < sk->sk_rcvlowat && !sock_flag(sk, SOCK_DONE))
+	if (avail < sk->sk_rcvlowat && !sock_flag(sk, SOCK_DONE) && !mptcp(tp))
 		return;
 
 	sk->sk_data_ready(sk);
Signature · 1 line
2.16.2
author
#5 lore raw

PATCH 5/5 mptcp: Correctly initialize IPv6 fields

In reply to Christoph Paasch

We are not correctly initializing all the IPv6 fields when creating an IPv6 socket in mptcp_alloc_mpcb.
This creates problems when IPv6 extensions are being used.

Github-issue: https://github.com/multipath-tcp/mptcp/issues/266

Fixes Zero-day bug
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
Signed-off-by Matthieu Baerts <matthieu.baerts(a)tessares.net>
(cherry picked from commit 708e2af4d23882c74b08bcbb67eb9c42bbeb5aee)
Signed-off-by Christoph Paasch <cpaasch(a)apple.com>
---
 net/mptcp/mptcp_ctrl.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)
net/mptcp/mptcp_ctrl.c +16 -14
index 626ccedd0878..916438e31b59 100644
@@ -1159,32 +1159,34 @@ static int mptcp_alloc_mpcb(struct sock *meta_sk, __u64 remote_key,
 
 #if IS_ENABLED(CONFIG_IPV6)
 	if (meta_icsk->icsk_af_ops == &mptcp_v6_mapped) {
+		struct tcp6_sock *master_tp6 = (struct tcp6_sock *)master_sk;
 		struct ipv6_pinfo *newnp, *np = inet6_sk(meta_sk);
 
-		inet_sk(master_sk)->pinet6 = &((struct tcp6_sock *)master_sk)->inet6;
+		inet_sk(master_sk)->pinet6 = &master_tp6->inet6;
 
 		newnp = inet6_sk(master_sk);
 		memcpy(newnp, np, sizeof(struct ipv6_pinfo));
-
-		newnp->ipv6_mc_list = NULL;
-		newnp->ipv6_ac_list = NULL;
-		newnp->ipv6_fl_list = NULL;
-		newnp->opt = NULL;
-		newnp->pktoptions = NULL;
-		(void)xchg(&newnp->rxpmtu, NULL);
 	} else if (meta_sk->sk_family == AF_INET6) {
+		struct tcp6_sock *master_tp6 = (struct tcp6_sock *)master_sk;
 		struct ipv6_pinfo *newnp, *np = inet6_sk(meta_sk);
+		struct ipv6_txoptions *opt;
 
-		inet_sk(master_sk)->pinet6 = &((struct tcp6_sock *)master_sk)->inet6;
+		inet_sk(master_sk)->pinet6 = &master_tp6->inet6;
 
+		/* The following heavily inspired from tcp_v6_syn_recv_sock() */
 		newnp = inet6_sk(master_sk);
 		memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 
-		newnp->hop_limit	= -1;
-		newnp->mcast_hops	= IPV6_DEFAULT_MCASTHOPS;
-		newnp->mc_loop	= 1;
-		newnp->pmtudisc	= IPV6_PMTUDISC_WANT;
-		master_sk->sk_ipv6only = sock_net(master_sk)->ipv6.sysctl.bindv6only;
+		opt = rcu_dereference(np->opt);
+		if (opt) {
+			opt = ipv6_dup_options(master_sk, opt);
+			RCU_INIT_POINTER(newnp->opt, opt);
+		}
+		inet_csk(master_sk)->icsk_ext_hdr_len = 0;
+		if (opt)
+			inet_csk(master_sk)->icsk_ext_hdr_len = opt->opt_nflen +
+								opt->opt_flen;
+
 	}
 #endif
 
Signature · 1 line
2.16.2
#6 lore raw
In reply to Christoph Paasch
Quoted 5 lines · Christoph Paasch

Hello,

here are some forward-ports for mptcp-net-next that come from mptcp_trunk.

Thanks Christoph. I merged net-next and applied your patches. mptcp_net-next has the updates.

Mat

Quoted 12 lines

Christoph Paasch (3):
mptcp: Remove IPv6 duplicated address detection
mptcp: Always call sk_data_ready
mptcp: Correctly initialize IPv6 fields

Kostas Peletidis (2):
Export tcp_send_ack symbol to fix build issue with mptcp_fullmesh
module (mptcp_trunk). This commit should make it possible to revert 64a80850.
Revert "Fix building mptcp_fullmesh as module"

Quoted 10 lines
net/ipv4/tcp_input.c       |  2 +-
net/ipv4/tcp_output.c      |  1 +
net/mptcp/mptcp_ctrl.c     | 30 ++++++++--------
net/mptcp/mptcp_fullmesh.c | 85 ++++++----------------------------------------
4 files changed, 29 insertions(+), 89 deletions(-)
Signature · 1 line
2.16.2

--
Mat Martineau
Intel OTC