Skip to contentSkip to conversation

protocol: pm callback must be done after msk state is set up

PATCH In review Florian Westphal 3 participants 6 messages Jan 9, 2020

Subsystems (2)
lore
Share & tools
lore

Florian Westphal

author
#0 lore raw

PATCH 1/2 protocol: pm callback must be done after msk state is set up

Squashto: mptcp: Add path manager interface

current location calls this when msk->token is still 0, so pm worker will never find the mptcp socket it should be using.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/protocol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
net/mptcp/protocol.c +2 -2
index 475ef4c90460..a33c4c58de78 100644
@@ -1371,8 +1371,6 @@ void mptcp_finish_connect(struct sock *ssk)
 
 	pr_debug("msk=%p, token=%u", sk, subflow->token);
 
-	mptcp_pm_new_connection(msk, 0);
-
 	mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
 	ack_seq++;
 	subflow->map_seq = ack_seq;
@@ -1389,6 +1387,8 @@ void mptcp_finish_connect(struct sock *ssk)
 	WRITE_ONCE(msk->ack_seq, ack_seq);
 	WRITE_ONCE(msk->can_ack, 1);
 	atomic64_set(&msk->snd_una, msk->write_seq);
+
+	mptcp_pm_new_connection(msk, 0);
 }
 
 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
Signature · 1 line
2.24.1
author
#1 lore raw

PATCH 2/2 protocol: fix list corruption

squashto: subflow: place further subflows on new 'join_list'

plugs:
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
kworker/u8:1/31 [HC0[0]:SC1[1]:HE1:SE0] takes:
ffff88810c43d520 (&(&msk->join_list_lock)->rlock){+.?.}, at: mptcp_finish_join+0x30f/0x4d0

and a list corruption, we need to re-init the join list head, else next list_empty() test will splice this a second time, resulting in a corrupted conn list.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/protocol.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
net/mptcp/protocol.c +3 -3
index a33c4c58de78..6df4eb20916c 100644
@@ -153,7 +153,7 @@ static void __mptcp_flush_join_list(struct mptcp_sock *msk)
 		return;
 
 	spin_lock_bh(&msk->join_list_lock);
-	list_splice_tail(&msk->join_list, &msk->conn_list);
+	list_splice_tail_init(&msk->join_list, &msk->conn_list);
 	spin_unlock_bh(&msk->join_list_lock);
 }
 
@@ -1420,9 +1420,9 @@ bool mptcp_finish_join(struct sock *sk)
 
 		/* active connections are already on conn_list */
 		if (list_empty(&subflow->node)) {
-			spin_lock(&msk->join_list_lock);
+			spin_lock_bh(&msk->join_list_lock);
 			list_add_tail(&subflow->node, &msk->join_list);
-			spin_unlock(&msk->join_list_lock);
+			spin_unlock_bh(&msk->join_list_lock);
 		}
 	}
 	return true;
Signature · 1 line
2.24.1
#2 lore raw

[MPTCP] Re: [PATCH 2/2] protocol: fix list corruption

Quoted 31 lines · Florian Westphal

squashto: subflow: place further subflows on new 'join_list'

plugs:
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
kworker/u8:1/31 [HC0[0]:SC1[1]:HE1:SE0] takes:
ffff88810c43d520 (&(&msk->join_list_lock)->rlock){+.?.}, at: mptcp_finish_join+0x30f/0x4d0

and a list corruption, we need to re-init the join list head, else next list_empty() test will splice this a second time, resulting in a corrupted conn list.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/protocol.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
net/mptcp/protocol.c +1 -1
index a33c4c58de78..6df4eb20916c 100644
@@ -153,7 +153,7 @@ static void __mptcp_flush_join_list(struct mptcp_sock *msk)
 		return;
 
 	spin_lock_bh(&msk->join_list_lock);
-	list_splice_tail(&msk->join_list, &msk->conn_list);
+	list_splice_tail_init(&msk->join_list, &msk->conn_list);
 	spin_unlock_bh(&msk->join_list_lock);
 }
 
@@ -1420,9 +1420,9 @@ bool mptcp_finish_join(struct sock *sk)
 

net/mptcp/protocol.c bool mptcp_finish_join(struct sock *sk) · line 1428

Quoted 10 lines
net/mptcp/protocol.c +2 -2
 		/* active connections are already on conn_list */
 		if (list_empty(&subflow->node)) {
-			spin_lock(&msk->join_list_lock);
+			spin_lock_bh(&msk->join_list_lock);
 			list_add_tail(&subflow->node, &msk->join_list);
-			spin_unlock(&msk->join_list_lock);
+			spin_unlock_bh(&msk->join_list_lock);
 		}
 	}
 	return true;

whoops, I thought this would run in BH only - I was wrong!

LGTM!

Paolo

#3 lore raw

[MPTCP] Re: [PATCH 2/2] protocol: fix list corruption

Hi Florian, Paolo,

Quoted 32 lines · Paolo Abeni
Quoted 31 lines · Florian Westphal

squashto: subflow: place further subflows on new 'join_list'

plugs:
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
kworker/u8:1/31 [HC0[0]:SC1[1]:HE1:SE0] takes:
ffff88810c43d520 (&(&msk->join_list_lock)->rlock){+.?.}, at: mptcp_finish_join+0x30f/0x4d0

and a list corruption, we need to re-init the join list head, else next list_empty() test will splice this a second time, resulting in a corrupted conn list.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
  net/mptcp/protocol.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
net/mptcp/protocol.c +1 -1
index a33c4c58de78..6df4eb20916c 100644
@@ -153,7 +153,7 @@ static void __mptcp_flush_join_list(struct mptcp_sock *msk)
  		return;
 
  	spin_lock_bh(&msk->join_list_lock);
-	list_splice_tail(&msk->join_list, &msk->conn_list);
+	list_splice_tail_init(&msk->join_list, &msk->conn_list);
  	spin_unlock_bh(&msk->join_list_lock);
  }
 
@@ -1420,9 +1420,9 @@ bool mptcp_finish_join(struct sock *sk)
 
Quoted 12 lines

net/mptcp/protocol.c bool mptcp_finish_join(struct sock *sk) · line 1428

Quoted 10 lines
net/mptcp/protocol.c +2 -2
  		/* active connections are already on conn_list */
  		if (list_empty(&subflow->node)) {
-			spin_lock(&msk->join_list_lock);
+			spin_lock_bh(&msk->join_list_lock);
  			list_add_tail(&subflow->node, &msk->join_list);
-			spin_unlock(&msk->join_list_lock);
+			spin_unlock_bh(&msk->join_list_lock);
  		}
  	}
  	return true;

whoops, I thought this would run in BH only - I was wrong!

Thank you for the patches and the review!

- f11b0171d4ee: "squashed" patch 1/2 in "mptcp: Add path manager interface" (with a small conflict)
- 8dc2fbf1951b: "Signed-off-by" + "Co-developed-by"
- bf09e56f4f2b: conflict in t/mptcp-update-per-unacked-sequence-on-pkt-reception
- 11cb91b0f172..0bc44ebe6199: result
- a4445faa25c7: "squashed" patch 2/2 in "subflow: place further subflows on new 'join_list'"
- 0bc44ebe6199..23c5e836256b: result

Tests are in progress.

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
#4 lore raw
In reply to Florian Westphal
Quoted 24 lines · Florian Westphal

Squashto: mptcp: Add path manager interface

current location calls this when msk->token is still 0, so pm worker will never find the mptcp socket it should be using.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/protocol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
net/mptcp/protocol.c +0 -2
index 475ef4c90460..a33c4c58de78 100644
@@ -1371,8 +1371,6 @@ void mptcp_finish_connect(struct sock *ssk)
 
 	pr_debug("msk=%p, token=%u", sk, subflow->token);
 
-	mptcp_pm_new_connection(msk, 0);
-
 	mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
 	ack_seq++;
 	subflow->map_seq = ack_seq;

net/mptcp/protocol.c void mptcp_finish_connect(struct sock *ssk) · line 1394

Quoted 8 lines
net/mptcp/protocol.c +2 -0
@@ -1389,6 +1387,8 @@ void mptcp_finish_connect(struct sock *ssk)
 	WRITE_ONCE(msk->ack_seq, ack_seq);
 	WRITE_ONCE(msk->can_ack, 1);
 	atomic64_set(&msk->snd_una, msk->write_seq);
+
+	mptcp_pm_new_connection(msk, 0);
 }
 
 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)

LGTM, thanks Florian! (This was fast!)

Does this fix the MP_JOIN handshake or is there any other pending known issue ? (beyond the list corruption fixed by the next patch...)

Paolo

author
#5 lore raw
Quoted 32 lines · Paolo Abeni
Quoted 31 lines · Florian Westphal

Squashto: mptcp: Add path manager interface

current location calls this when msk->token is still 0, so pm worker will never find the mptcp socket it should be using.

Signed-off-by Florian Westphal <fw(a)strlen.de>
---
 net/mptcp/protocol.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
net/mptcp/protocol.c +2 -2
index 475ef4c90460..a33c4c58de78 100644
@@ -1371,8 +1371,6 @@ void mptcp_finish_connect(struct sock *ssk)
 
 	pr_debug("msk=%p, token=%u", sk, subflow->token);
 
-	mptcp_pm_new_connection(msk, 0);
-
 	mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
 	ack_seq++;
 	subflow->map_seq = ack_seq;
@@ -1389,6 +1387,8 @@ void mptcp_finish_connect(struct sock *ssk)
 	WRITE_ONCE(msk->ack_seq, ack_seq);
 	WRITE_ONCE(msk->can_ack, 1);
 	atomic64_set(&msk->snd_una, msk->write_seq);
+
+	mptcp_pm_new_connection(msk, 0);
 }
 
Quoted 6 lines

net/mptcp/protocol.c void mptcp_finish_connect(struct sock *ssk) · line 1394

Quoted 1 line
net/mptcp/protocol.c +0 -0
 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)

LGTM, thanks Florian! (This was fast!)

Does this fix the MP_JOIN handshake or is there any other pending known issue ? (beyond the list corruption fixed by the next patch...)

Yes and yes. Things only work when slowing down the test prog to give the join time to complete.

Else, this hits the 'is parent not established anymore' check in mptcp_finish_join. This also gives a memory leak of the socket.