svn commit: r227791 - head/sys/netinet

[Gleb Smirnoff]

On Wed, Nov 23, 2011 at 06:53:05AM -0800, Qing Li wrote:
Q> >  first I'd like to notice that we are speaking about obsoleted interfaces.
Q> 
Q> Yup, that's why you don't see me commenting on your other commits around
Q> ia_netmask stuff, do you ?

Oops, I didn't notice that one. Sorry! Can you please resend it?

Q> > Back to your comments:
Q> >
Q> > I have made a test case that proves, that usage of deleted address isn't
Q> > prevented when it is removed, but loopback route still exists.
Q> >
Q> > The test is the following run a race between this program:
Q> >
Q> >        struct ifreq ifr;
Q> >        int s;
Q> >
Q> >        bzero(&ifr, sizeof(struct ifreq));
Q> >
Q> >        strncpy(ifr.ifr_name, "igb1", sizeof ifr.ifr_name);
Q> >        ifr.ifr_addr.sa_family = AF_INET;
Q> >        ifr.ifr_addr.sa_len = sizeof(struct sockaddr_in);
Q> >        ((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr.s_addr = inet_addr("10.0.0.1");
Q> >
Q> >        s = socket(AF_INET, SOCK_DGRAM, 0);
Q> >
Q> >        for (;;)
Q> >                ioctl(s, SIOCSIFADDR, &ifr);
Q> >
Q> > And this script:
Q> >
Q> >        while (true); do nc -z 10.0.0.1 22 || echo Fail; done
Q> >
Q> 
Q> I am not sure if this test scenario is valid.
Q> 
Q> The loopback route is wiped at line #853 and then quickly inserted back at
Q> line #936 because you are SIOCSIFADDR the same address over and over again.

But during the procedure there is a state when we do not have this address,
but still we receive traffic to it. That is the thing I don't like.

If you really insist on this, let's shift in_ifscrub() down below as it was.
Since it affects only obsoleted SIOCSIFADDR, I'd better just ignore that
behavior rather than stand on its incorrectness.

-- 
Totus tuus, Glebius.