On Fri, Jul 08, 2005 at 02:14:54PM -0700, Mitch Williams wrote:
> (Adding vger to Cc: list, so as to spread the joy around.)
> On Thu, 7 Jul 2005, Greg KH wrote:
> > >
> > > Moving the individual bond directories to a bonds/ directory
> > > is problematic. Because each bond shows up a just another network
> > > interface, they show up in /sys/class/net automatically. We'd have to
> > > make a bunch of changes to the device model to account for bonding, and
> > > we'd break the semantics of the /sys/class/net hierarchy.
> > Why not just put them in /sys/class/bond/ instead?
> As Aaron Brown noted in another reply, it's inappropriate for bonding to
> do this. Bonding is really just another network driver, and its
> interfaces show up in /sys/class/net courtesy of the device model. We
> don't need to replicate this capability; we just need to extend it a bit.
> Adding /sys/class/bond/ would cause a much larger protest that what we're
> dealing with now (i.e. pretty much only you).
Ok, as I said before, I don't care where you put it, just writting and
reading multiple values at once in sysfs files is not to be tolerated.
> > > reading of Documentation/filesystems/sysfs.txt indicates that such a usage
> > > is "socially acceptable". (Or at least it was to Patrick Mochel back in
> > > January of 2003.)
> > Pat was just trying to be nice. I'm not. :)
> Well, if "nice" isn't the policy any more, then the doc needs to change.
Ok, I will do so.
> However, I did a little more poking, and found a bunch of files that have
> more than one item in them. HW resource listings, block device scheduling
> stuff, plenty of examples. Are they socially acceptable?
The block device stuff I understand as they want a single snapshot.
They also can't create multiple files for every value or the 20,000 disk
system admins will complain :)
As for the other ones, have specific examples?
And none of them are "multiple value write", correct? That, and due to
the fact that your implementation will drop needed data is why I object
to your patch so much.
I'm sure you can understand that desiging in a limitation that could
cause data to be dropped that people really want is not a good thing.
> > > bonding_masters. This currently is a problem, since sysfs itself does not
> > > handle appends properly.
> > Because you are not supposed to do that.
> Sysfs will happily accept O_APPEND opens, and will happily report success
> on seek attempts, although neither operation is permitted. Whether or not
> you're supposed to, this behavior is just plain wrong.
Ok, care to resend the patches in different emails? From what I last
remember, they either were not able to be applied, or some other trival
problem with them.