Re: [PATCH v3 net-next 00/23] net: add preliminary netdev refcount tracking

From: Eric Dumazet <edumazet@google.com>
To: Andrew Lunn <andrew@lunn.ch>
Cc: Eric Dumazet <eric.dumazet@gmail.com>,
	"David S . Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>, netdev <netdev@vger.kernel.org>,
	Dmitry Vyukov <dvyukov@google.com>
Subject: Re: [PATCH v3 net-next 00/23] net: add preliminary netdev refcount tracking
Date: Mon, 6 Dec 2021 15:44:57 -0800	[thread overview]
Message-ID: <CANn89iLPSianJ7TjzrpOw+a0PTgX_rpQmiNYbgxbn2K-PNouFg@mail.gmail.com> (raw)
In-Reply-To: <Ya6bj2nplJ57JPml@lunn.ch>

On Mon, Dec 6, 2021 at 3:24 PM Andrew Lunn <andrew@lunn.ch> wrote:
>
> On Sat, Dec 04, 2021 at 08:21:54PM -0800, Eric Dumazet wrote:
> > From: Eric Dumazet <edumazet@google.com>
> >
> > Two first patches add a generic infrastructure, that will be used
> > to get tracking of refcount increments/decrements.
>
> Hi Eric
>
> Using this i found:
>
> [  774.108901] unregister_netdevice: waiting for eth0 to become free. Usage count = 4
> [  774.110864] leaked reference.
> [  774.110874]  dst_alloc+0x7a/0x180
> [  774.110887]  ip6_dst_alloc+0x27/0x90
> [  774.110894]  ip6_pol_route+0x257/0x430
> [  774.110900]  ip6_pol_route_output+0x19/0x20
> [  774.110905]  fib6_rule_lookup+0x18b/0x270
> [  774.110914]  ip6_route_output_flags_noref+0xaa/0x110
> [  774.110918]  ip6_route_output_flags+0x32/0xa0
> [  774.110922]  ip6_dst_lookup_tail.constprop.0+0x181/0x240
> [  774.110929]  ip6_dst_lookup_flow+0x43/0xa0
> [  774.110934]  inet6_csk_route_socket+0x166/0x200
> [  774.110943]  inet6_csk_xmit+0x56/0x130
> [  774.110946]  __tcp_transmit_skb+0x53b/0xc30
> [  774.110953]  __tcp_send_ack.part.0+0xc6/0x1a0
> [  774.110958]  tcp_send_ack+0x1c/0x20
> [  774.110964]  __tcp_ack_snd_check+0x42/0x200
> [  774.110968]  tcp_rcv_established+0x27a/0x6f0
> [  774.110973] leaked reference.
> [  774.110975]  ipv6_add_dev+0x13e/0x4f0
> [  774.110982]  addrconf_notify+0x2ca/0x950
> [  774.110989]  raw_notifier_call_chain+0x49/0x60
> [  774.111000]  call_netdevice_notifiers_info+0x50/0x90
> [  774.111007]  __dev_change_net_namespace+0x30d/0x6c0
> [  774.111016]  do_setlink+0xdc/0x10b0
> [  774.111024]  __rtnl_newlink+0x608/0xa10
> [  774.111031]  rtnl_newlink+0x49/0x70
> [  774.111038]  rtnetlink_rcv_msg+0x14f/0x380
> [  774.111046]  netlink_rcv_skb+0x55/0x100
> [  774.111053]  rtnetlink_rcv+0x15/0x20
> [  774.111059]  netlink_unicast+0x230/0x340
> [  774.111064]  netlink_sendmsg+0x252/0x4b0
> [  774.111075]  sock_sendmsg+0x65/0x70
> [  774.111080]  ____sys_sendmsg+0x24e/0x290
> [  774.111084]  ___sys_sendmsg+0x81/0xc0
>
> I'm using GNS3 to simulate a network topology. So a collection of veth
> pairs, bridges and tap interfaces spread over a few namespaces. The
> network being simulated uses Segment Routing. And traceroute might also
> involved in this somehow. I have 3 patches applied, to make traceroute
> actually work when SRv6 is being used. You can find v3 here:
>
> https://lore.kernel.org/netdev/20211203162926.3680281-3-andrew@lunn.ch/T/
>
> I'm not sure if these patches are part of the problem or not. None of
> the traces i've seen are directly on the ICMP path. traceroute is
> using udp, and one of the traces above is for tcp, and the other looks
> like it is moving an interface into a different namespace?
>
> This is net-next from today.

I do not understand, net-next does not contain this stuff yet ?

I have other patches, this work is still in progress.