Subject: [PATCH netfilter] ipvs: Fix crash when ipv6 route unreach

From: hujunwei
Date: Mon May 06 2019 - 22:36:09 EST

From: Junwei Hu <hujunwei4@xxxxxxxxxx>

When Tcp send RST packet in ipvs, crash occurs with the following
stack trace:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
PID: 0 COMMAND: "swapper/2"
TASK: ffff9ec83889bf40 (1 of 4) [THREAD_INFO: ffff9ec8388b0000]
[exception RIP: __ip_vs_get_out_rt_v6+1250]
RIP: ffffffffc0d566f2 RSP: ffff9ec83ed03c68 RFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff9ec835e85000 RCX: 000000000005e1f9
RDX: 000000000005e1f8 RSI: 0000000000000200 RDI: ffff9ec83e801b00
RBP: ffff9ec83ed03cd8 R8: 000000000001bb40 R9: ffffffffc0d5673f
R10: ffff9ec83ed1bb40 R11: ffffe2d384d4fdc0 R12: ffff9ec7b7ad5900
R13: 0000000000000000 R14: 0000000000000007 R15: ffff9ec8353f7580
ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
[ffff9ec83ed03ce0] ip_vs_fnat_xmit_v6 at ffffffffc0d5b42c [ip_vs]
[ffff9ec83ed03d70] tcp_send_rst_ipv6 at ffffffffc0d6542a [ip_vs]
[ffff9ec83ed03df8] tcp_conn_expire_handler at ffffffffc0d65823 [ip_vs]
[ffff9ec83ed03e20] ip_vs_conn_expire at ffffffffc0d42373 [ip_vs]
[ffff9ec83ed03e70] call_timer_fn at ffffffffae0a6b58
[ffff9ec83ed03ea8] run_timer_softirq at ffffffffae0a904d
[ffff9ec83ed03f20] __do_softirq at ffffffffae09fa85
[ffff9ec83ed03f90] call_softirq at ffffffffae739dac
[ffff9ec83ed03fa8] do_softirq at ffffffffae02e62b
[ffff9ec83ed03fc0] irq_exit at ffffffffae09fe25
[ffff9ec83ed03fd8] smp_apic_timer_interrupt at ffffffffae73b158
[ffff9ec83ed03ff0] apic_timer_interrupt at ffffffffae737872

TCP connection timeout and send a RST packet, the skb is alloc
by alloc_skb, the pointer skb->dev and skb_dst(skb) is NULL,
however, ipv6 route unreach at that time, so go into err_unreach.
In err_unreach, crash occurs when skb->dev and skb_dst(skb) is NULL.

The code is added by the following patch:
commit 326bf17ea5d4 ("ipvs: fix ipv6 route unreach panic")
because the ip6_link_failure function requires the skb->dev
in icmp6_send with that version.

This patch only fix the problem in specific scene, and icmp6_send in
current version is robust against null skb->dev by adding the
following patch.
commit 8d9336704521
("ipv6: make icmp6_send() robust against null skb->dev")

So I delete the code, make __ip_vs_get_out_rt_v6() robust, when
skb->dev and skb_dst(skb) is NULL.

Fixes: 326bf17ea5d4 ("ipvs: fix ipv6 route unreach panic")
Signed-off-by: Junwei Hu <hujunwei4@xxxxxxxxxx>
Reported-by: Wenhao Zhang <zhangwenhao8@xxxxxxxxxx>
net/netfilter/ipvs/ip_vs_xmit.c | 7 -------
1 file changed, 7 deletions(-)

diff --git a/net/netfilter/ipvs/ip_vs_xmit.c b/net/netfilter/ipvs/ip_vs_xmit.c
index 175349fcf91f..e2bb6c223396 100644
--- a/net/netfilter/ipvs/ip_vs_xmit.c
+++ b/net/netfilter/ipvs/ip_vs_xmit.c
@@ -561,13 +561,6 @@ __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
return -1;

- /* The ip6_link_failure function requires the dev field to be set
- * in order to get the net (further for the sake of fwmark
- * reflection).
- */
- if (!skb->dev)
- skb->dev = skb_dst(skb)->dev;
return -1;