From: Joerg Reuter (rwybwk@relay.tunkki.fi)
Date: Wed Mar 15 2000 - 20:05:53 EET
Hi Jens,
sorry, took a while to respond. I haven't found the time to actually
test it, but I've skimmed over the patch. As far as I can see it
looks fine, but I've some remarks.
- Have you seen the solution for ax25_info in 2.3.x? That way it's
possible to easily remove support for the old structure after
we've given it enough time to phase out. Sure, it assigns some few
variables that won't get passed down to the application in case of
SIOCAX25GETINFOOLD, but it's IMO much easier to maintain that way.
- It doesn't matter for the experimental patch for kernel 2.2.x, but
you should really at least printk() warnings for features you've
removed.
- Regretfully, support for sockaddr_ax25 in ax25_bind(), ax25_connect()
and ax25_sendmsg() must stay for a while, in return we shouldn't be
too forgiving if someone calls these functions with a full_sockaddr_ax25
length that reserves room for digipeaters other than 6 or 8. (See
my recent patches)
- We'll have SO_BINDTODEVICE support in 2.4.x, it will supersede the
"bind to device via digipeater[0]" hack. We don't ever need to phase out
that hack, but we shouldn't encourage folks to use it, either.
[Identifying an AX.25 interface by its callsign is bogus. Unlike
an ethernet MAC it isn't unique at all.]
- calling "connect()" without bind()ing first is bogus. Hopefully we
can remove this mis-feature (it doesn't even work correctly in the
current implementation due to some left-overs from previous versions).
I'll know as soon as Linus applies my latest patch...
- Passing down additional information to the driver *that* way... Hmm.
I had proposed a similar interface before and it got rejected. And I
even agree on that. BTW, I can see exactly one valid reason to pass
down slot time and p-persistence now -- KISS TNCs.
- Neither most of the drivers, nor the kernel AX.25 is SMP safe yet.
I think we can wait until 2.5 for it, but I've heard people talking about
using an SMP box for AX.25. It mostly works as long as only the KISS
driver is used, but it would sure be nice to start thinking about where
a spinlock() is needed.
> - DAMA slave code
Yay! That's certainly one very (IMHO most) important feature. But I haven't
seen a way to turn the DAMA slave off. That switch _is_ important because
there are several BBS in Spain (I believe) with some weird TNC firmware and
some satellites in use that set the "reserved" bits to zero. You'll get
_very_ loud complaints if you do not take this into account.
> - Nice /proc/sys/ax25/.../media_* - Interface
Haven't looked into that yet.
Anyway, the patch mostly looks fine as far as I can see. Good work!
I don't know if Linus has accepted my last patch for 2.3.x, as soon
as that is in the porting work for 2.4.x kernels can start. It's probably
not trivial as several details have changed.
-- vy 73 de Joerg dl1bke/a ampr-net: jtlso.psclp@ihugdsl.com.au www : http://www.qsl.net/dl1bke/ AX-25 : DL1BKE @ DB0ACH.#NRW.DEU.EU Internet: terhi.victor@logonet.com
This archive was generated by hypermail 2b29 : Thu Mar 16 2000 - 22:17:41 EET