[manet] Re: Kernel work, Linux

Thanks Charlie. I’d like to see it of course, although I don’t expect to be able to usefully critique it. I hope those who have done related work can let us know their thoughts on whether and how it impacts on the implementability issue. (There are other things to discuss of course, but this is one thing.)

On 21 Jul 2025, at 21:15, Charlie Perkins <[email protected]> wrote:



Hello folks,

Reactive protocols such as DSR and AODVv2 and others typically
need certain kernel modifications. To my knowledge, the need for
a kernel modification has never been a reason to stop work on
charter items.

Kernel work does demand careful programming. The amount of
security exposure from reactive notifications for broken or
nonexistent routes is considerably less than from making a new
device driver, memory handler, or scheduler.

Finally, it seems to me that the [manet] WG can (and probably
should) be involved in creating ad hoc network IETF protocol
specifications not necessarily targeted at Linux implementations.
Is that no longer correct?

Regards,

Charlie P.

On 7/16/2025 11:06 PM, Henning Rogge
wrote:

On Thu, Jul 17, 2025 at 12:39 AM Charlie Perkins
<[email protected]> wrote:

. . .
I remember a few implementations of AODV (v1) when I tried to get a
running version a couple of years ago... all of them did not work
because they brought a custom kernel module for "kernel 2.6" (or
similar) with them... demanding a kernel module is not a viable
option.

. . .

Beauty is in the eye of the beholder. In AODV my favorite solutions fit perfectly with the design of ICMP and the meaning of the word "reactive". The same design would work for AODVv2. Perhaps the [manet] WG considers to disqualify any protocol development that depends on such real-time reaction. At least part of that reaction should properly reside in the kernel.

I am not aware of the huge security risk, but if you could provide more details I would be interested to learn. Maybe we could incorporate some discussion about it in Section 14 of the draft.

The issue is that (at least) 99% good C programmers (myself included)
don't know how to write a reasonable and safe kernel module acceptable
by the kernel.org group... and if your module is NOT accepted into the
kernel it will be outdated and unusable very quickly.

Inserting code into the kernel into networking is always a big security risk...

Henning Rogge

_______________________________________________
manet mailing list -- [email protected]
To unsubscribe send an email to [email protected]

_______________________________________________
manet mailing list -- [email protected]
To unsubscribe send an email to [email protected]

_______________________________________________
manet mailing list -- [email protected]
To unsubscribe send an email to [email protected]