[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20240307224423.53315-1-kuniyu@amazon.com>
Date: Thu, 7 Mar 2024 14:44:23 -0800
From: Kuniyuki Iwashima <kuniyu@...zon.com>
To: <edumazet@...gle.com>
CC: <allison.henderson@...cle.com>, <davem@...emloft.net>, <kuba@...nel.org>,
<kuni1840@...il.com>, <kuniyu@...zon.com>, <linux-rdma@...r.kernel.org>,
<netdev@...r.kernel.org>, <pabeni@...hat.com>, <rds-devel@....oracle.com>
Subject: Re: [PATCH v3 net 1/2] tcp: Restart iteration after removing reqsk in inet_twsk_purge().
From: Eric Dumazet <edumazet@...gle.com>
Date: Thu, 7 Mar 2024 10:51:15 +0100
> On Thu, Mar 7, 2024 at 12:05 AM Kuniyuki Iwashima <kuniyu@...zon.com> wrote:
> >
> > Commit 740ea3c4a0b2 ("tcp: Clean up kernel listener's reqsk in
> > inet_twsk_purge()") added changes in inet_twsk_purge() to purge
> > reqsk in per-netns ehash during netns dismantle.
> >
> > inet_csk_reqsk_queue_drop_and_put() will remove reqsk from per-netns
> > ehash, but the iteration uses sk_nulls_for_each_rcu(), which is not
> > safe.
> >
> > After removing reqsk, we need to restart iteration.
> >
> > Note that we need not check net->ns.count here because per-netns
> > ehash does not have reqsk in other live netns. We will check
> > net->ns.count in the following patch.
> >
> > Fixes: 740ea3c4a0b2 ("tcp: Clean up kernel listener's reqsk in inet_twsk_purge()")
> > Reported-by: Eric Dumazet <edumazet@...gle.com>
> > Signed-off-by: Kuniyuki Iwashima <kuniyu@...zon.com>
> > ---
> > net/ipv4/inet_timewait_sock.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/net/ipv4/inet_timewait_sock.c b/net/ipv4/inet_timewait_sock.c
> > index 5befa4de5b24..00cbebaa2c68 100644
> > --- a/net/ipv4/inet_timewait_sock.c
> > +++ b/net/ipv4/inet_timewait_sock.c
> > @@ -287,6 +287,8 @@ void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
> > struct request_sock *req = inet_reqsk(sk);
> >
> > inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
> > +
> > + goto restart;
> > }
> >
> > continue;
>
> Note how the RCU rules that I followed for TCP_TIME_WAIT made
> me to grab a reference on tw->tw_refcnt, using refcount_inc_not_zero()
>
> I think your code had multiple bugs, because
> inet_csk_reqsk_queue_drop_and_put() could cause UAF
> if the timer already fired and refcount went to zero already.
Ugh.. exactly.
I'll post v4 following the TIME_WAIT path.
---8<---
diff --git a/net/ipv4/inet_timewait_sock.c b/net/ipv4/inet_timewait_sock.c
index 961b1917c3eb..c81f83893fc7 100644
--- a/net/ipv4/inet_timewait_sock.c
+++ b/net/ipv4/inet_timewait_sock.c
@@ -278,20 +278,32 @@ void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
restart:
sk_nulls_for_each_rcu(sk, node, &head->chain) {
if (sk->sk_state != TCP_TIME_WAIT) {
+ struct request_sock *req;
+
+ if (likely(sk->sk_state != TCP_NEW_SYN_RECV))
+ continue;
+
/* A kernel listener socket might not hold refcnt for net,
* so reqsk_timer_handler() could be fired after net is
* freed. Userspace listener and reqsk never exist here.
*/
- if (unlikely(sk->sk_state == TCP_NEW_SYN_RECV &&
- !refcount_read(&sock_net(sk)->ns.count))) {
- struct request_sock *req = inet_reqsk(sk);
- inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
+ if (sk->sk_family != family ||
+ refcount_read(&sock_net(sk)->ns.count))
+ continue;
+
+ req = inet_reqsk(sk);
+ if (unlikely(!refcount_inc_not_zero(&req->rsk_refcnt)))
+ continue;
- goto restart;
+ if (unlikely(sk->sk_family != family ||
+ refcount_read(&sock_net(sk)->ns.count))) {
+ reqsk_put(req);
+ continue;
}
- continue;
+ inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
+ goto restart;
}
tw = inet_twsk(sk);
---8<---
>
> We also could add sk_nulls_for_each_rcu_safe() to avoid these pesky
> "goto restart;"
I'll post this followup for net-next in the next release cycle.
Thanks!
Powered by blists - more mailing lists