Re: Patch to add usbmon

[Marcel Holtmann <marcel@holtmann.org>]

Hi Pete,

> > While I am really thinking about starting usbdump, I may ask why you
> > have choosen to use debugfs as interface. This will not be available in
> > normal distribution kernels and I think a general USB monitoring ability
> > would be great. For example like we have it for Ethernet, Bluetooth and
> > IrDA. So my idea is to create some /dev/usbmonX (for each bus one) where
> > usbdump can read its information from. What do you think?
> 
> The debugfs will be available in distributions. And it's no different from
> having SOCK_PACKET enabled before tcpdump can work. There's no practical
> disadvantage, unless we consider a backport of usbmon into a legacy distribution
> such as RHEL 4 or SuSE 9.1.

I am not interested in a backport.

> Since usbmon rides raw file_operations, it can use a chardev interface with
> a minimal change. The advantage of debugfs is only not needing to allocate
> a major and dealing with minor partitioning. I also thought it would help
> to shut up the namespace pollution whiners (the debate of /dbg versus
> /sys/kernel/debug was rather mild by comparison).

Getting a major number should be no problem and the minor partitioning
is also easy, because the root hubs are already numbered by USB.

> It should make little difference for the tool anyway, the base path ought to
> be configurable. The biggest difference would be to scripts which launch the
> tool: in one case they need to mount debugfs if not mounted (if initscripts
> haven't), in other case they mknod if necessary (if hal hasn't done it).

The mknod reason is no reason for, because we have udev (not hal btw)
and this working perfect.

> It is too early to care about this anyway.

Since I am really thinking of writing usbdump, it is not to early. What
stuff is missing in your usbmon patch?

Regards

Marcel