mdev race conditions

[Bernard Blackham]

Mike Frysinger wrote:
> you'll also need to make sure you do this somewhere which has temporary 
> storage ... otherwise you'll wreck embedded flash systems

/dev is on tmpfs anyway.

> *shrug* anything not daemon/netlink based looks like a hack to me

And I'm inclined to agree.

 > which means i still dont understand why people dont go use udevd if
 > they need a real setup.

s/real/reliable/

What good is using mdev if it cannot be relied upon? (apart from using 
mdev -s at boot and an otherwise completely-static /dev).

[... snip ...]
 > better, just use udevd (which i honestly dont think is a big deal, it 
 > isnt exactly bloated ... it's 76k on x86_64 and it's built for speed,
 > not size).

It's 66K stripped on arm and built for size (the Makefile for udev-120 
does this by default). 66K is a lot for a critical daemon. It does a 
*lot* of stuff that nobody needs - it maintains a database in /dev/.udev 
of all devices and exports lots of information to userspace which 
probably does not matter to most embedded systems.

I'd like someone to answer my original proposal of whether or not a 
simple daemon running in a loop would suffice. I imagine the resulting 
binary would be a mere few kilobytes in size, as *all* it needs to do is 
serialize requests and pass them to mdev. Nothing more.

Bernard.