Skip to contentSkip to conversation

Fix issues in prerequisite patches

PATCH Superseded Peter Krystad 3 participants 10 messages Oct 21, 2019 – Oct 23, 2019

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Peter Krystad

author
#0 lore raw

cover Fix issues in prerequisite patches

Fix issues found by Mat in the "IPv6 prerequisite" patch series when running against the multipath.org implementation.

Peter Krystad (3):
mptcp: fix mistake in path manager interface
mptcp: Save address family when creating socket
mptcp: Use correct address family when creating subflows

 net/mptcp/basic.c    | 28 ++++++++++++++++------------
 net/mptcp/pm.c       |  4 ++--
 net/mptcp/protocol.c |  1 +
 net/mptcp/protocol.h |  1 +
 4 files changed, 20 insertions(+), 14 deletions(-)
Signature · 1 line
2.17.2
author
#1 lore raw

PATCH 1/3 mptcp: fix mistake in path manager interface

Do not de-reference NULL pointer immediately after confirming it is NULL.

squashto: Implement path manager interface commands

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/pm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
net/mptcp/pm.c +2 -2
index f521bf0c5f69..d45406bc3f6c 100644
@@ -95,7 +95,7 @@ int mptcp_pm_create_subflow(u32 token, u8 remote_id, struct in_addr *addr)
 
 	local.sin_family = AF_INET;
 	local.sin_port = 0;
-	if (!addr)
+	if (addr)
 		local.sin_addr = *addr;
 	else
 		local.sin_addr.s_addr = htonl(INADDR_ANY);
@@ -133,7 +133,7 @@ int mptcp_pm_create_subflow6(u32 token, u8 remote_id, struct in6_addr *addr)
 
 	local.sin6_family = AF_INET6;
 	local.sin6_port = 0;
-	if (!addr)
+	if (addr)
 		local.sin6_addr = *addr;
 	else
 		local.sin6_addr = in6addr_any;
Signature · 1 line
2.17.2
author
#3 lore raw

PATCH 2/3 mptcp: Save address family when creating socket

In reply to Peter Krystad

For IPv6 the family will be needed when creating additional subflow sockets.

squashto: mptcp: Associate MPTCP context with TCP socket

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/protocol.c | 1 +
 net/mptcp/protocol.h | 1 +
 2 files changed, 2 insertions(+)
net/mptcp/protocol.c +1 -0
index c7755cde9810..e7b40987d1b6 100644
@@ -1091,6 +1091,7 @@ static struct socket *mptcp_socket_create_get(struct mptcp_sock *msk)
 	}
 
 	msk->subflow = ssock;
+	msk->family = ssock->sk->sk_family;
 	subflow = mptcp_subflow_ctx(msk->subflow->sk);
 	subflow->request_mptcp = 1; /* @@ if MPTCP enabled */
 	subflow->request_version = 0; /* currently only v0 supported */
net/mptcp/protocol.h +1 -0
index 791f2c19cfb8..8fe5f9383d38 100644
@@ -133,6 +133,7 @@ struct mptcp_sock {
 	u32		token;
 	unsigned long	flags;
 	u16		dport;
+	sa_family_t	family;
 	struct work_struct rtx_work;
 	struct list_head conn_list;
 	struct list_head rtx_queue;
Signature · 1 line
2.17.2
author
#4 lore raw

PATCH 3/3 mptcp: Use correct address family when creating subflows

In reply to Peter Krystad

Use the address family used when the initial MPTCP socket was created when making additional subflow sockets.

squashto: Implement basic path manager

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/basic.c | 28 ++++++++++++++++------------
 1 file changed, 16 insertions(+), 12 deletions(-)
net/mptcp/basic.c +16 -12
index d2fd0560c136..2557cb63fd61 100644
@@ -209,19 +209,23 @@ static void create_subflow_worker(struct work_struct *work)
 
 	pernet = net_generic(sock_net((struct sock *)msk), basic_pernet_id);
 
-	if (pernet->has_announce_v4)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id,
-					&pernet->announce_v4_addr);
-#if IS_ENABLED(CONFIG_IPV6)
-	else if (pernet->has_announce_v6)
-		mptcp_pm_create_subflow6(pm->token, pm->remote_id,
-					 &pernet->announce_v6_addr);
-#endif
-	else if (pm->local_family == AF_INET)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id, NULL);
+	if (msk->family == AF_INET) {
+		if (pernet->has_announce_v4)
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						&pernet->announce_v4_addr);
+		else
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						NULL);
+	}
 #if IS_ENABLED(CONFIG_IPV6)
-	else
-		mptcp_pm_create_subflow6(pm->token, pm->remote_id, NULL);
+	else if (msk->family == AF_INET6) {
+		if (pernet->has_announce_v6)
+			mptcp_pm_create_subflow6(pm->token, pm->remote_id,
+						 &pernet->announce_v6_addr);
+		else
+			mptcp_pm_create_subflow6(pm->token, pm->remote_id,
+						 NULL);
+	}
 #endif
 
 	sock_put((struct sock *)msk);
Signature · 1 line
2.17.2
#5 lore raw

[MPTCP] Re: [PATCH 3/3] mptcp: Use correct address family when creating subflows

Hi,

Quoted 26 lines · Peter Krystad

Use the address family used when the initial MPTCP socket was created when making additional subflow sockets.

squashto: Implement basic path manager

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/basic.c | 28 ++++++++++++++++------------
 1 file changed, 16 insertions(+), 12 deletions(-)
net/mptcp/basic.c +0 -8
index d2fd0560c136..2557cb63fd61 100644
@@ -209,19 +209,23 @@ static void create_subflow_worker(struct work_struct *work)
 
 	pernet = net_generic(sock_net((struct sock *)msk), basic_pernet_id);
 
-	if (pernet->has_announce_v4)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id,
-					&pernet->announce_v4_addr);
-#if IS_ENABLED(CONFIG_IPV6)
-	else if (pernet->has_announce_v6)
-		mptcp_pm_create_subflow6(pm->token, pm->remote_id,
-					 &pernet->announce_v6_addr);
-#endif

net/mptcp/basic.c static void create_subflow_worker(struct work_struct *work) · line 219

Quoted 10 lines
net/mptcp/basic.c +8 -2
-	else if (pm->local_family == AF_INET)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id, NULL);
+	if (msk->family == AF_INET) {
+		if (pernet->has_announce_v4)
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						&pernet->announce_v4_addr);
+		else
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						NULL);
+	}

Here I'm likely a bit lost, but what if the peer sends an ADD_ADDR v4, and the local msk socket is a v6 (or vice versa)?

I think checking vs the msk family is possibly not required here (IIRC the RFC code, it looks like an MPTCP client could manage correctly subflows with different inet families - but we need to check/match the local address family (has_announce_v4/has_announce_v6) vs the remote address/ADD_ADDR family. Otherwise we could try to connect to a corrupted/truncated/uncomplete address.

Additionally, I think we need to check/match the local additional address family (has_announce_v4/has_announce_v6) vs the msk family in announce_addr_worker(), so that the sever always announce addresses that is able to accept[1].

Thanks,

Paolo

[1] for IPv6 listener such check could be fairly more complex - unbound listener could accept any ipv4 address, bound ones could check for ipv4-mapped ipv6 address, etc... I think is better use to the simpler approach here.

#6 lore raw

[MPTCP] Re: [PATCH 3/3] mptcp: Use correct address family when creating subflows

Hi Paolo, Peter,

Quoted 34 lines · Paolo Abeni

Hi,

Quoted 31 lines · Peter Krystad

Use the address family used when the initial MPTCP socket was created when making additional subflow sockets.

squashto: Implement basic path manager

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
  net/mptcp/basic.c | 28 ++++++++++++++++------------
  1 file changed, 16 insertions(+), 12 deletions(-)
net/mptcp/basic.c +3 -10
index d2fd0560c136..2557cb63fd61 100644
@@ -209,19 +209,23 @@ static void create_subflow_worker(struct work_struct *work)
 
  	pernet = net_generic(sock_net((struct sock *)msk), basic_pernet_id);
 
-	if (pernet->has_announce_v4)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id,
-					&pernet->announce_v4_addr);
-#if IS_ENABLED(CONFIG_IPV6)
-	else if (pernet->has_announce_v6)
-		mptcp_pm_create_subflow6(pm->token, pm->remote_id,
-					 &pernet->announce_v6_addr);
-#endif
-	else if (pm->local_family == AF_INET)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id, NULL);
+	if (msk->family == AF_INET) {
+		if (pernet->has_announce_v4)
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
Quoted 8 lines

net/mptcp/basic.c static void create_subflow_worker(struct work_struct *work) · line 219

Quoted 5 lines
net/mptcp/basic.c +5 -0
+						&pernet->announce_v4_addr);
+		else
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						NULL);
+	}

Here I'm likely a bit lost, but what if the peer sends an ADD_ADDR v4, and the local msk socket is a v6 (or vice versa)?

On one hand, if the socket is v6 only, it would make sense not to establish a v4 subflow or announce a v4 IP. Same if the socket is v4 only.

But on the other hand, when an app creates both v4-only and v6-only sockets (like many webservers do), it is annoying because you cannot combine v4 and v6. In this case the PM doesn't announce the addresses for the other family because it will not accept addresses from this other family. We respect what the app asked but I don't know if we want to "ignore" v4-only/v6-only for other subflows.

Quoted 6 lines

I think checking vs the msk family is possibly not required here (IIRC the RFC code, it looks like an MPTCP client could manage correctly subflows with different inet families - but we need to check/match the local address family (has_announce_v4/has_announce_v6) vs the remote address/ADD_ADDR family. Otherwise we could try to connect to a corrupted/truncated/uncomplete address.

Regarding the code here, I guess we should check for sockets allowing both v4 and v6. And possibly use the v4 addressed mapped in v6.

Quoted 4 lines

Additionally, I think we need to check/match the local additional address family (has_announce_v4/has_announce_v6) vs the msk family in announce_addr_worker(), so that the sever always announce addresses that is able to accept[1].

Yes but only if the socket is v4-only or v6-only.

Quoted 8 lines

Thanks,

Paolo

[1] for IPv6 listener such check could be fairly more complex - unbound listener could accept any ipv4 address, bound ones could check for ipv4-mapped ipv6 address, etc... I think is better use to the simpler approach here.

So accept a mix of v4 and v6 later? It would make sense but it will need to be documented somewhere with a TODO I guess :)

Cheers,
Matt

PS: Paolo: is it OK for you if I apply the first patch of the series?

Signature · 5 lines
Matthieu Baerts | R&D Engineer
matthieu.baerts(a)tessares.net
Tessares SA | Hybrid Access Solutions
www.tessares.net
1 Avenue Jean Monnet, 1348 Louvain-la-Neuve, Belgium
#2 lore raw

[MPTCP] Re: [PATCH 1/3] mptcp: fix mistake in path manager interface

In reply to Peter Krystad

Hi Peter,

Quoted 4 lines · Peter Krystad

Do not de-reference NULL pointer immediately after confirming it is NULL.

squashto: Implement path manager interface commands

Thank you for the patch! I just applied this one but not the other ones due to the other discussions started on the last one.

- 166607d720d7: "squashed" patch 1/3 in "mptcp: Implement path manager interface commands"
- 088eb1a59555..ecb276617e3b: results

Tests have the same issue as what has been saw this night:

   00:06:43.648 # ns2  MPTCP -> ns3  (10.0.2.2:10027) MPTCP	[ OK ]
   00:06:44.698 # ns2  MPTCP -> ns3  (10.0.2.2:10028) TCP  	[ OK ]
   00:06:45.940 # ns2  TCP   -> ns3  (10.0.2.2:10029) MPTCP	[ OK ]
   00:06:46.992 # ns2  MPTCP -> ns3  (10.0.3.2:10030) MPTCP 

./mptcp_connect.sh: line 16: 797 Terminated ip netns exec
${listener_ns} ./mptcp_connect -t $timeout -l -p $port -s ${srv_proto} 0.0.0.0 < "$sin" > "$sout"
00:06:47.076 # ./mptcp_connect.sh: line 16: 798 Terminated
sleep 1
00:06:47.108 not ok 1 selftests: mptcp: mptcp_connect.sh # TIMEOUT

Maybe due to a modification in net-next.

Cheers,
Matt

Signature · 5 lines
Matthieu Baerts | R&D Engineer
matthieu.baerts(a)tessares.net
Tessares SA | Hybrid Access Solutions
www.tessares.net
1 Avenue Jean Monnet, 1348 Louvain-la-Neuve, Belgium
#7 lore raw

[MPTCP] Re: [PATCH 3/3] mptcp: Use correct address family when creating subflows

In reply to Matthieu Baerts
Quoted 3 lines · Matthieu Baerts
Quoted 2 lines · Paolo Abeni

but what if the peer sends an ADD_ADDR v4,
and the local msk socket is a v6 (or vice versa)?

Sorry, I was not clear enough. With the posted code, when msk is v6, pernet->has_announce_v6 is 1 and ADD_ADDR is ipv4 (pm->remote_family == AF_INET), I think mptcp_pm_create_subflow6() will end-up using uninitialized bytes for mptcp_subflow_connect().

To avoid that, we should check that the local and the remote address used to establish the connection belong to the same family.

Quoted 1 line

So accept a mix of v4 and v6 later?

The idea would be to keep the code as simple as possible. The listening socket in the current status can't accept easily a mix of ipv4 and ipv6 subflow, I think.

On the flip side, should be quite straigh-forward support mixed family for client sockets; that was my overall proposal here.

Quoted 2 lines

It would make sense but it will need
to be documented somewhere with a TODO I guess :)

Agreed.

Quoted 1 line

PS: Paolo: is it OK for you if I apply the first patch of the series?

too late, but... "yes"

Cheers,

Paolo

#8 lore raw

[MPTCP] Re: [PATCH 3/3] mptcp: Use correct address family when creating subflows

Hi Paolo,

Quoted 5 lines · Paolo Abeni
Quoted 3 lines · Matthieu Baerts
Quoted 2 lines · Paolo Abeni

but what if the peer sends an ADD_ADDR v4,
and the local msk socket is a v6 (or vice versa)?

Quoted 7 lines

Sorry, I was not clear enough. With the posted code, when msk is v6, pernet->has_announce_v6 is 1 and ADD_ADDR is ipv4 (pm->remote_family == AF_INET), I think mptcp_pm_create_subflow6() will end-up using uninitialized bytes for mptcp_subflow_connect().

To avoid that, we should check that the local and the remote address used to establish the connection belong to the same family.

I see, thank you for the clarification! Indeed, there is no check when doing this, good catch!

remote.sin_family = msk->pm.remote_family
remote.sin6_family = msk->pm.remote_family

Quoted 8 lines
Quoted 1 line

So accept a mix of v4 and v6 later?

The idea would be to keep the code as simple as possible. The listening socket in the current status can't accept easily a mix of ipv4 and ipv6 subflow, I think.

On the flip side, should be quite straigh-forward support mixed family for client sockets; that was my overall proposal here.

Sounds like a good idea!

Quoted 8 lines
Quoted 2 lines

It would make sense but it will need
to be documented somewhere with a TODO I guess :)

Agreed.

Quoted 1 line

PS: Paolo: is it OK for you if I apply the first patch of the series?

too late, but... "yes"

Sorry, the fix was straightforward, I went ahead without waiting :)

Cheers,
Matt

Signature · 5 lines
Matthieu Baerts | R&D Engineer
matthieu.baerts(a)tessares.net
Tessares SA | Hybrid Access Solutions
www.tessares.net
1 Avenue Jean Monnet, 1348 Louvain-la-Neuve, Belgium
author
#9 lore raw

[MPTCP] Re: [PATCH 3/3] mptcp: Use correct address family when creating subflows

In reply to Paolo Abeni

Thank-you for the review Paolo.

Quoted 43 lines · Paolo Abeni

Hi,

Quoted 26 lines · Peter Krystad

Use the address family used when the initial MPTCP socket was created when making additional subflow sockets.

squashto: Implement basic path manager

Signed-off-by Peter Krystad <peter.krystad(a)linux.intel.com>
---
 net/mptcp/basic.c | 28 ++++++++++++++++------------
 1 file changed, 16 insertions(+), 12 deletions(-)
net/mptcp/basic.c +0 -8
index d2fd0560c136..2557cb63fd61 100644
@@ -209,19 +209,23 @@ static void create_subflow_worker(struct work_struct *work)
 
 	pernet = net_generic(sock_net((struct sock *)msk), basic_pernet_id);
 
-	if (pernet->has_announce_v4)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id,
-					&pernet->announce_v4_addr);
-#if IS_ENABLED(CONFIG_IPV6)
-	else if (pernet->has_announce_v6)
-		mptcp_pm_create_subflow6(pm->token, pm->remote_id,
-					 &pernet->announce_v6_addr);
-#endif

net/mptcp/basic.c static void create_subflow_worker(struct work_struct *work) · line 219

Quoted 10 lines
net/mptcp/basic.c +8 -2
-	else if (pm->local_family == AF_INET)
-		mptcp_pm_create_subflow(pm->token, pm->remote_id, NULL);
+	if (msk->family == AF_INET) {
+		if (pernet->has_announce_v4)
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						&pernet->announce_v4_addr);
+		else
+			mptcp_pm_create_subflow(pm->token, pm->remote_id,
+						NULL);
+	}

Here I'm likely a bit lost, but what if the peer sends an ADD_ADDR v4, and the local msk socket is a v6 (or vice versa)?

Quoted 6 lines

I think checking vs the msk family is possibly not required here (IIRC the RFC code, it looks like an MPTCP client could manage correctly subflows with different inet families - but we need to check/match the local address family (has_announce_v4/has_announce_v6) vs the remote address/ADD_ADDR family. Otherwise we could try to connect to a corrupted/truncated/uncomplete address.

You're right, checking the msk->family is not required, but checking against the remote address family is. So client may start both types, but we won't be able to test that against our own implementation, for now.

Quoted 4 lines

Additionally, I think we need to check/match the local additional address family (has_announce_v4/has_announce_v6) vs the msk family in announce_addr_worker(), so that the sever always announce addresses that is able to accept[1].

Agreed, I'll add this.

Peter.

Quoted 9 lines

Thanks,

Paolo

[1] for IPv6 listener such check could be fairly more complex - unbound listener could accept any ipv4 address, bound ones could check for ipv4-mapped ipv6 address, etc... I think is better use to the simpler approach here.