Re: Suggested dual human/binary interface for proc/devfs

[H. Peter Anvin]

Linus Torvalds wrote:
> >
> > So if we removed the magic ID's (and sysctl(2)), Linus would be happy?
> 
> The actual internal implementation of sysctl() is horribly unclear. I'd
> rather have a dentry tree where the dentries contain the filenames
> (obviously) and also the values (in the private field or in the inode).
> That makes adding and removing entries quite trivial.
> 

This sounds like a job for Al Viro's detached filesystem tree
modifications.  Seriously.  The more things like this crop up, the more
I'm starting to appreciate how that simplifies life: without them, we
couldn't start build the dentry tree until the filesystem is mounted, at
which point we'll have had a bunch of registrations already.

	-hpa