ipv6: do not set sk_destruct in IPV6_ADDRFORM sockopt
authorXin Long <lucien.xin@gmail.com>
Mon, 28 Aug 2017 02:45:01 +0000 (10:45 +0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 29 Aug 2017 17:54:40 +0000 (10:54 -0700)
ChunYu found a kernel warn_on during syzkaller fuzzing:

[40226.038539] WARNING: CPU: 5 PID: 23720 at net/ipv4/af_inet.c:152 inet_sock_destruct+0x78d/0x9a0
[40226.144849] Call Trace:
[40226.147590]  <IRQ>
[40226.149859]  dump_stack+0xe2/0x186
[40226.176546]  __warn+0x1a4/0x1e0
[40226.180066]  warn_slowpath_null+0x31/0x40
[40226.184555]  inet_sock_destruct+0x78d/0x9a0
[40226.246355]  __sk_destruct+0xfa/0x8c0
[40226.290612]  rcu_process_callbacks+0xaa0/0x18a0
[40226.336816]  __do_softirq+0x241/0x75e
[40226.367758]  irq_exit+0x1f6/0x220
[40226.371458]  smp_apic_timer_interrupt+0x7b/0xa0
[40226.376507]  apic_timer_interrupt+0x93/0xa0

The warn_on happned when sk->sk_rmem_alloc wasn't 0 in inet_sock_destruct.
As after commit f970bd9e3a06 ("udp: implement memory accounting helpers"),
udp has changed to use udp_destruct_sock as sk_destruct where it would
udp_rmem_release all rmem.

But IPV6_ADDRFORM sockopt sets sk_destruct with inet_sock_destruct after
changing family to PF_INET. If rmem is not 0 at that time, and there is
no place to release rmem before calling inet_sock_destruct, the warn_on
will be triggered.

This patch is to fix it by not setting sk_destruct in IPV6_ADDRFORM sockopt
any more. As IPV6_ADDRFORM sockopt only works for tcp and udp. TCP sock has
already set it's sk_destruct with inet_sock_destruct and UDP has set with
udp_destruct_sock since they're created.

Fixes: f970bd9e3a06 ("udp: implement memory accounting helpers")
Reported-by: ChunYu Wang <chunwang@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/ipv6_sockglue.c

index 02d795fe3d7f2c5e6e922a25dbbe69c8139919b6..a5e466d4e09310ed99c391fea4bb6126d83fbb56 100644 (file)
@@ -242,7 +242,6 @@ static int do_ipv6_setsockopt(struct sock *sk, int level, int optname,
                        pktopt = xchg(&np->pktoptions, NULL);
                        kfree_skb(pktopt);
 
-                       sk->sk_destruct = inet_sock_destruct;
                        /*
                         * ... and add it to the refcnt debug socks count
                         * in the new family. -acme