[OpenAFS] Re: openafs does not put a [correct] value in fsinfo.f_type?

Derrick J Brashear shadow@dementia.org
Mon, 25 Dec 2006 20:16:55 -0500 (EST)

On Mon, 25 Dec 2006, Adam Megacz wrote:

> Derrick J Brashear <shadow@dementia.org> writes:
>>> This won't work for code which must build without dependencies on AFS,
>>> of which might be built on a non-AFS machine and executed on an
>>> AFS-client machine (ie for rpm/deb packaging).
>> Untrue. The kafs library has its on syscall prober thing which doesn't
>> depend on AFS and certainly manages to set tokens.
> Are you referring to this?
>  http://www.pdc.kth.se/kth-krb/
> The code in lib/kafs/afssys.c appears to catch the SIGSYS signal while
> doing some scary syscall()-guessing.


> This is the closest thing to an answer so far.  But that's a pretty
> huge amount of code just to figure out what filesystem a file sits on.

Ok, but, that's not "can't" and it's not "requires AFS"

> I'm also worried about using something like this to answer the
> question "does this machine have the AFS client" in a situation where
> a false negative means severe data corruption.  In the case of the
> krb4-kth library, a false negative in the syscall()-guessing just
> means that the user gets an error message.

AFS won't operate without either the syscall interface or for more modern 
linux the afs_ioctl file and the ioctl interface. you don't have it, afsd 
will never run.

No afsd, no afs. Move along.