Re: [RFC] [PATCH] sysctl for the latecomers

[Chase Venters]

On Tue, 1 Aug 2006, Amit Gud wrote:

> Chase Venters wrote:
> >  On Tue, 1 Aug 2006, Chase Venters wrote:
> >  Btw, wanted to add some comments on the specific approach:
> >
> >  1. A ring hard-coded to 32 elements is IMO unuseable. While it may not be
> >  a real limit for what use case you have in mind, if it's in the kernel
> >  sooner or later someone else is going to use it and get bitten. Imagine if
> >  they wrote in 33 entries, and the first one was some critical security
> >  setting that ended up getting silently ignored...
> >
> >  2. On the other hand, allowing it to grow unbounded is equally
> >  unacceptable without a mechanism to list and clear the current "pending"
> >  sysctl values. Unfortunately, at this point, you're starting to violate
> >  "KISS".
>
> You figured it right, theres no "correct" number of elements that I could 
> adhere to.
>
> >  Are the modules you refer to inserted during init at all? Because it seems
> >  like it would be a lot more appropriate to just move sysctl until after
> >  loading the modules, or perhaps running it again once they are loaded.
>
> I have a case where sunrpc module gets inserted and 
> sunrpc.tcp_slot_table_entries parameter is to be set _before_ nfs module is 
> inserted. I agree that for this particular case user-space works (either udev 
> rule, or modprobe.conf or sysctl after modprobe in initscripts), but am on a 
> lookout for a more generic way for handling such cases - be it user-space or 
> otherwise.

It just seems like something that belongs in user-space -- whether that be 
better init scripts, changes to modprobe, both, or something else 
entirely.

To make a kernel implementation of this concept that isn't 
fundamentally flawed from a usability and "least-surprise" 
perspective would mean enough code and behavior that the resulting 
implementation wouldn't be considered correct for the kernel anyway.

The kernel has, fundamentally, three kinds of configuration - CONFIG_*, 
the bootloader command-line, and 'live' configuration that gets set by 
user-space whenever appropriate.

> AG

Thanks,
Chase
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/