Re: kerneld/multicast bug (tickled by gated)
Sun, 15 Jun 1997 20:22:01 +0200

Brian Candler <> writes:
>> Disable kerneld or remove the request route rubbish. I've been trying to
>> stop people adding it since 1.2, and as of 2.1 I've finally got rid of the
>> buggy rubbish.
>Alan, do you suggest removing it in the kernel, or in kerneld?

Well, if we'd been a bit faster it would be gone in the 2.0.x kernels,
but as it stands it's a bit late to take it away from a stable series.
[Read this as: I don't want to deal with the screams of anguish if
we take it away.]

>I note from
>the kerneld source that chmod -x on /sbin/request-route should be enough to
>prevent kerneld from attempting to run it.

I think you can just do "rm /sbin/request-route" and get it over with :)
I don't have it installed on my machine and I run kerneld without problems.

>However, if the bug is to do with the communication between the kernel and
>kerneld, I guess it would be better to patch the kernel instead (and
>presumably this would be more in line with 2.1)... but then people running
>gated under Linux would need a patched kernel. It would be nice to have the
>ability to keep kerneld running.

There are more races then you can shake a stick at in running a
user space program in request to a kernel routing call. The kernel
networking and routing code was written under the assumption that the
code would be atomic. Going to sleep to wait for a user space program
to run breaks these assumptions badly. It's really quite surprising
that the thing works at all.

Eric Schenk                               www:
Dept. of Comp. Sci., Lund University          email:
Box 118, S-221 00 LUND, Sweden   fax: +46-46 13 10 21  ph: +46-46 222 96 38