Re: A common problem with SLAAC in "renumbering" scenarios

[Michael Richardson <[e-mail]>]

Mikael Abrahamsson <[e-mail]> wrote:
    >> That doesn't really make sense on a mobile platform. When you switch from
    >> wifi to cell data, you probably don't want to keep that connection alive
    >> until that specific wifi network comes back. Instead you want to give the
    >> app an error as soon as possible so it can retry the failed connection on
    >> the new network. On Android, TCP connections are closed the instant their
    >> IP address goes away. I don't know what iOS does.

    > I contacted linux-netdev about this, and I got no sympathy to close down the
    > sessions if the addresses were gone. "Oh, the addresses might come
    > back". Even when I said I'd like to be able to have this as a setting to
    > close down the sessions, that I knew it won't come back I kept getting ""how
    > do you know?"

Hmm. You know because you are writing the "DHCP" daemon :-)
This reminds me of the /dev/pty allocation problem, or the good-ole
/dev/tty00 background process.... some googling found:
        https://jdebp.eu/FGA/bernstein-on-ttys/cttys.html

There is a system call, which I can't remember, which kills all other access
to a tty... clearly we want to do exactly that when we realize that a device
has changed provisioning domains.

Well.. maybe each time we are in a new provisioning domain, we want to spawn
a "macvlan" device with the new randomized layer-2 address, and use that.
Just let the old one hang around until everything times out... or maybe, we
they say, "it could come back"

    > How do you close them in Android? Did you write your own userspace code to
    > find sockets bound to addresses that were no longer available and close them
    > down?

No, I don't think that this happens.




--
Michael Richardson <[e-mail]>, Sandelman Software Works
 -= IPv6 IoT consulting =-