diff mbox

nf_conntrack_h323: Fix locking in process_urq

Message ID 1452894918.4622.3.camel@gmail.com
State Awaiting Upstream, archived
Delegated to: David Miller
Headers show

Commit Message

Sebastian =?ISO-8859-1?Q?P=F6hn?= Jan. 15, 2016, 9:55 p.m. UTC
nf_ct_remove_expectations has to be called under nf_conntrack_expect_lock
    
Signed-off-by: Sebastian Pöhn <sebastian.poehn@gmail.com>

---

Comments

Florian Westphal Jan. 15, 2016, 10:42 p.m. UTC | #1
Sebastian Pöhn <sebastian.poehn@gmail.com> wrote:

[ CC netfilter-devel ]

> nf_ct_remove_expectations has to be called under nf_conntrack_expect_lock

But nf_ct_remove_expectations grabs that lock?

Added in:

commit ca7433df3a672efc88e08222cfa4b3aa965ca324
Author: Jesper Dangaard Brouer <brouer@redhat.com>
netfilter: conntrack: seperate expect locking from nf_conntrack_lock

> diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
> index 9511af0..d477375 100644
> --- a/net/netfilter/nf_conntrack_h323_main.c
> +++ b/net/netfilter/nf_conntrack_h323_main.c
> @@ -1518,7 +1518,9 @@ static int process_urq(struct sk_buff *skb, struct nf_conn *ct,
>         }
>  
>         /* Clear old expect */
> +       spin_lock_bh(&nf_conntrack_expect_lock);
>         nf_ct_remove_expectations(ct);

... so I'd expect deadlock.
Sebastian =?ISO-8859-1?Q?P=F6hn?= Jan. 15, 2016, 10:57 p.m. UTC | #2
You are right. The problem is fixed since 3.15 with the expect_lock refactoring.
We found this in 3.12. Probably this is something for stable releases 3.14 and earlier.

Am 15.01.2016 11:42 nachm. schrieb "Florian Westphal" <fw@strlen.de>:
>
> Sebastian Pöhn <sebastian.poehn@gmail.com> wrote:
>
> [ CC netfilter-devel ]
>
> > nf_ct_remove_expectations has to be called under nf_conntrack_expect_lock
>
> But nf_ct_remove_expectations grabs that lock?
>
> Added in:
>
> commit ca7433df3a672efc88e08222cfa4b3aa965ca324
> Author: Jesper Dangaard Brouer <brouer@redhat.com>
> netfilter: conntrack: seperate expect locking from nf_conntrack_lock
>
> > diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
> > index 9511af0..d477375 100644
> > --- a/net/netfilter/nf_conntrack_h323_main.c
> > +++ b/net/netfilter/nf_conntrack_h323_main.c
> > @@ -1518,7 +1518,9 @@ static int process_urq(struct sk_buff *skb, struct nf_conn *ct,
> >         }
> >  
> >         /* Clear old expect */
> > +       spin_lock_bh(&nf_conntrack_expect_lock);
> >         nf_ct_remove_expectations(ct);
>
> ... so I'd expect deadlock.
>
diff mbox

Patch

diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
index 9511af0..d477375 100644
--- a/net/netfilter/nf_conntrack_h323_main.c
+++ b/net/netfilter/nf_conntrack_h323_main.c
@@ -1518,7 +1518,9 @@  static int process_urq(struct sk_buff *skb, struct nf_conn *ct,
        }
 
        /* Clear old expect */
+       spin_lock_bh(&nf_conntrack_expect_lock);
        nf_ct_remove_expectations(ct);
+       spin_unlock_bh(&nf_conntrack_expect_lock);
        info->sig_port[dir] = 0;
        info->sig_port[!dir] = 0;