[RFC] Proof-of-concept for netlink listener for mdev -i

Natanael Copa ncopa at alpinelinux.org
Mon Mar 16 18:49:47 UTC 2015


On Mon, 16 Mar 2015 18:13:17 +0100
Harald Becker <ralda at gmx.de> wrote:

> On 16.03.2015 08:16, Natanael Copa wrote:
> > This give me exactly what I am interested in: a hotplug handler that is
> > fast while keep memory consumption at a minimal during long periods
> > with no events.
> 
> This is the essential of your message, I would give *you* the expected 
> result, but not to everybody else.

That is correct. A netlink socket activator is completely useless for
those who don't want read events from netlink.

I am perfectly happy that not *everybody* else wants to read from
netlink and I am happy that most of those people simply ignore my code
in silence.

> What about the ability of logging and debugging purposes? And I don't 
> talk about debugging netlink / mdev / xdev code, I mean debugging kernel 
> device event messages. Your approach would need a separate new piece of 
> code (socket / netlink aware), to handle those, 

Correct. It does absolutely no reading. To actually read from a netlink
socket you need more code that is netlink aware. I just want avoid
having that code in memory when there are no events to read. That is
the only value this code brings to the table.

> whereas a slightly 
> different approach gives more modularity and compatibility with other 
> Unix tools, e.g.
> 
> netlink reader | tee /dev/ttyX | device operation handler

This looks good to me.

If you want avoid that this netlink reader in your example is in memory
at all times, then feel free to use my netlink socket activator to
activate it. Otherwise, please ignore it.

> (Yes this is a normal pipe, used in shell scripts, to display all 
> incoming event messages on a specific console).

-nc


More information about the busybox mailing list