Re: mm snapshot broken-out-2007-01-26-00-36.tar.gz uploaded

[ebiederm@xmission.com (Eric W. Biederman)]

Valdis.Kletnieks@vt.edu writes:

> On Sun, 28 Jan 2007 23:08:17 MST, Eric W. Biederman said:
>> Valdis.Kletnieks@vt.edu writes:
>
>> Does it find sys?  If so perhaps I should do something even more significant.
>> I guess if I get many complaints about this I will figure out how to print
>> out an appropriate error message.
>
> It found sys, and then the second iteration in in xlate_proc_name it failed
> to find net because the de->subdir for sys/net wasn't set.

Thanks. 

>> > What's the intended semantics of create_proc_entry and xlate_proc_name in
>> > this new regime of no subdir pointers? Or am I just (yet again) one of the
>> > first to trip over a bug?
>> 
>> It is supposed to fail in this instance.  If you want something under
> /proc/sys
>> you are supposed to use register_sysctl like everyone else.  If it's not a
>> sysctl it should not show up under /proc/sys.
>
> Wasn't my code originally - I think the original author thought that since
> all the *other* config stuff for ipv4 was down under /proc/sys/net/ipv4, this
> one should be as well because that's where sysadmins would look for it, and
> wasn't thinking so much about whether it was a sysctl or not.
>
>> I'm glad to see my cleanup uncovering more bugs, I'm sorry you were the one
>> who had to find it.  I will you well fixing your out of tree ipfilter module.
>
> It's easy enough to move the entry under /proc/net or someplace instead.
>
> What's the current advice on what kernel interface to use for this scenario:
>
> In userspace, we do something like this:
>
> (while read foo; do echo $foo > /proc/my_file; done) < /etc/bunch_of_lines
>
> and we want to catch, parse, and save each line as it enters the kernel, and
> we end up with several dozen entries saved.
>
> If we do a 'cat /proc/my_file', we iterate across the list of saved lines
> and dump them all out.

Well it is a non-fixed number of entries so it does not sound appropriate for
a sysctl.  sysctl tends to work well for simple settings, not complicated things.

/proc/net would be the easy place, although I think traditionally /proc/net
is just reporting.

Given the context I will suggest setsockopt as most of iptables is configured
using that.

Generally going beyond one value per file is discouraged, in any context.

So I guess the general advice is don't have that scenario :)

In your situation I would find something in the kernel that does something
similar and see how it provides for a user interface.  If there is a way
that it is done for similar modules do that. 

If there is no precedent something under sysfs likely makes the most sense.
/proc is really supposed for things related to processes.

I usually spend my time keeping the old things going and useful in new and
interesting situations instead of figuring out the interface recommendation
of the week.

I would be really surprised if your configuration requirements were unique
and there weren't several somethings in the kernel that solved the
same problem in a different way.

Eric