diff mbox series

Re: [MPTCP][PATCH v7 mptcp-next 1/7] mptcp: create the listening socket for new port

Message ID 8d83a6f809c0856b4488247da8d7d48bea1fa66f.camel@redhat.com
State Superseded, archived
Headers show
Series Re: [MPTCP][PATCH v7 mptcp-next 1/7] mptcp: create the listening socket for new port | expand

Commit Message

Paolo Abeni Dec. 11, 2020, 3:21 p.m. UTC 
On Thu, 2020-12-10 at 11:48 +0800, Geliang Tang wrote:
> Hi Paolo,
> 
> On Wed, Dec 09, 2020 at 04:25:33PM +0100, Paolo Abeni wrote:
> > On Wed, 2020-12-09 at 20:33 +0800, Geliang Tang wrote:
> > > The full log and the patch is attached. Apply this patch and run
> > > mptcp_join.sh can reproduce the warning.
> > 
> > I tried applying the v7 posted on the ML, plus some manging to let it
> > apply on top of current export. The end result is quite alike the patch
> > attached here.
> > 
> > Still I can't reproduce the issue.
> 
> Thanks for your help.
> 
> v7 should be added the following code in pm_netlink.c to invoke
> sock_release to reproduce the issue:
> 
> @@ -885,6 +968,8 @@ static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
>         spin_unlock_bh(&pernet->lock);
> 
>         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
> +       if (entry->lsk)
> +               sock_release(entry->lsk);
>         kfree_rcu(entry, rcu);
> 
>         return ret;
> 
> > I'm wondering if the kconfig matters here (beyond CONFIG_LOCKDEP=y, I
> > mean). Could you please share the config you are using ?

Ok, I can reproduce the issue with the above chunk. Looks like it's
just a neste lock issue. I'm unsure why lockdep did not splat before -
the issue has been there since sometime.

Anyhow the following fixes the issue here:
---
---
/P

Comments

Geliang Tang Dec. 14, 2020, 4:22 a.m. UTC | #1 
Hi Paolo,

Paolo Abeni <pabeni@redhat.com> 于2020年12月11日周五 下午11:21写道：
>
> On Thu, 2020-12-10 at 11:48 +0800, Geliang Tang wrote:
> > Hi Paolo,
> >
> > On Wed, Dec 09, 2020 at 04:25:33PM +0100, Paolo Abeni wrote:
> > > On Wed, 2020-12-09 at 20:33 +0800, Geliang Tang wrote:
> > > > The full log and the patch is attached. Apply this patch and run
> > > > mptcp_join.sh can reproduce the warning.
> > >
> > > I tried applying the v7 posted on the ML, plus some manging to let it
> > > apply on top of current export. The end result is quite alike the patch
> > > attached here.
> > >
> > > Still I can't reproduce the issue.
> >
> > Thanks for your help.
> >
> > v7 should be added the following code in pm_netlink.c to invoke
> > sock_release to reproduce the issue:
> >
> > @@ -885,6 +968,8 @@ static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
> >         spin_unlock_bh(&pernet->lock);
> >
> >         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
> > +       if (entry->lsk)
> > +               sock_release(entry->lsk);
> >         kfree_rcu(entry, rcu);
> >
> >         return ret;
> >
> > > I'm wondering if the kconfig matters here (beyond CONFIG_LOCKDEP=y, I
> > > mean). Could you please share the config you are using ?
>
> Ok, I can reproduce the issue with the above chunk. Looks like it's
> just a neste lock issue. I'm unsure why lockdep did not splat before -
> the issue has been there since sometime.
>
> Anyhow the following fixes the issue here:
> ---
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index 56bc1ed94ca7..63f3043c1c1d 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -2144,7 +2144,7 @@ void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
>
>         list_del(&subflow->node);
>
> -       lock_sock(ssk);
> +       lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
>
>         /* if we are invoked by the msk cleanup code, the subflow is
>          * already orphaned
> ---

Thanks for your help. I tested this patch, it works well. Please send this
patch to the ML and add my reported-and-tested tag in it:

Reported-and-tested-by: Geliang Tang <geliangtang@gmail.com>

Thanks very much.

-Geliang

> /P
>
>
diff mbox series

Patch

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 56bc1ed94ca7..63f3043c1c1d 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -2144,7 +2144,7 @@  void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
 
 	list_del(&subflow->node);
 
-	lock_sock(ssk);
+	lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
 
 	/* if we are invoked by the msk cleanup code, the subflow is
 	 * already orphaned