答复: [PATCH] ipv4: fix the problem of ping failure in some cases

["Guodeqing (A)" <geffrey.guo@huawei.com>]

> -----邮件原件-----
> 发件人: David Ahern [mailto:dsahern@gmail.com]
> 发送时间: Monday, August 24, 2020 10:34
> 收件人: Guodeqing (A) <geffrey.guo@huawei.com>; davem@davemloft.net
> 抄送: kuba@kernel.org; netdev@vger.kernel.org
> 主题: Re: [PATCH] ipv4: fix the problem of ping failure in some cases
> 
> On 8/23/20 8:27 PM, David Ahern wrote:
> > On 8/22/20 1:46 AM, guodeqing wrote:
> >> ie.,
> >> $ ifconfig eth0 9.9.9.9 netmask 255.255.255.0
> >>
> >> $ ping -I lo 9.9.9.9
> 
> If that ever worked it was wrong; the address is scoped to eth0, not lo.
> 
> >> ping: Warning: source address might be selected on device other than lo.
> >> PING 9.9.9.9 (9.9.9.9) from 9.9.9.9 lo: 56(84) bytes of data.
> >>
> >> 4 packets transmitted, 0 received, 100% packet loss, time 3068ms
> >>
> >> This is because the return value of __raw_v4_lookup in raw_v4_input
> >> is null, the packets cannot be sent to the ping application.
> >> The reason of the __raw_v4_lookup failure is that sk_bound_dev_if and
> >> dif/sdif are not equal in raw_sk_bound_dev_eq.
> >>
> >> Here I add a check of whether the sk_bound_dev_if is LOOPBACK_IFINDEX
> >> to solve this problem.
> >>
> >> Fixes: 19e4e768064a8 ("ipv4: Fix raw socket lookup for local
> >> traffic")
> >> Signed-off-by: guodeqing <geffrey.guo@huawei.com>
> >> ---
> >>  include/net/inet_sock.h | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/include/net/inet_sock.h b/include/net/inet_sock.h index
> >> a3702d1..7707b1d 100644
> >> --- a/include/net/inet_sock.h
> >> +++ b/include/net/inet_sock.h
> >> @@ -144,7 +144,7 @@ static inline bool inet_bound_dev_eq(bool
> >> l3mdev_accept, int bound_dev_if,  {
> >>  	if (!bound_dev_if)
> >>  		return !sdif || l3mdev_accept;
> >> -	return bound_dev_if == dif || bound_dev_if == sdif;
> >> +	return bound_dev_if == dif || bound_dev_if == sdif || bound_dev_if
> >> +== LOOPBACK_IFINDEX;
> >>  }
> >>
> >>  struct inet_cork {
> >>
> >
> > this is used by more than just raw socket lookups.
> >
> 
> And assuming it should work, this is definitely the wrong fix.

ok, I see now. I misunderstood the function of the loopback interface, thanks.