About ioctl call

I have posted this recently on lkml, but I've got no good answer. I'm
hoping you can explain this to me.

I was studying ioctl recently and I found out that, in order to change
the IP of a certain interface, you have to create a dummy socket with
socket(2) and use the returned file descriptor in the ioctl call. That
seems strange to me because the IP is not changed only for that
particular descriptor, but globally. Why is that configuration handled
by ioctl? Maybe via /proc/sys?

I know that there's a long time (even before linux) that things were
this way. But why was things done like this in the first place? Do you
consider it reasonable? Why?

> I was studying ioctl recently and I found out that, in order to change
> the IP of a certain interface, you have to create a dummy socket with
> socket(2) and use the returned file descriptor in the ioctl call. That
> seems strange to me because the IP is not changed only for that
> particular descriptor, but globally. Why is that configuration handled
> by ioctl? Maybe via /proc/sys?

Because ioctl had an available schema (prototype for function arguments)
that was easy to use to communicate the required information between
user space and the kernel: an integer (usually a file descriptor),
a verb (command, request, etc.), and a pointer to a variable-sized struct
that can be either input or output or both. Use an interface that already
exists, instead of inventing an argument list that is totally new.
ioctl traditionally has been used as a "catch all" for interfaces
that are experimental, less rigorously defined, change often, difficult
to get correct on the first try, etc.

--