nnpfs/xfs_message.h documentation

[Tomas Olsson]

Kristaps Johnson <kristaps at gradient-enterprises.com> writes:

> Hello everybody.  Does anybody know of API documentation for the 
> nnpfs_message/attr.h (xfs_*.h on some unices) files?
>
Not really.  In arla/doc/nnpfs.txt there's are slightly outdated
descriptions of the messages.  There's also a dot file describing the
various states that nodes can be in in Arla:
 http://cvsweb.stacken.kth.se/cvsweb.cgi/arla-doc/arla-programming/fcache-states.dot?rev=1.4;content-type=text%2Fplain
And 'The life of a file' in arla.info for an overview.

Often there are details in these docs that aren't valid any more, and some
omissions.  Please let me know what you find, nagging actually does help :)

> I'm implementing the xfs char-device with a non-arla superstructure.
>
Nice.  As always, I'm very curious about such things.  How much can you
tell us?

> Thusfar the arlad code itself has provided a clean resource but I'd like
> to be certain about some structures and valid values.
>
Well, you should trust the code more than docs...  Since nnpfs is meant to
be generic everything should be in the common nnpfs includes.  Any
omissions should be corrected.  

> does struct nnpfs_handle always correspond with cell, volume, vnode and
> unique?
>
In arlad, yes.  In nnpfs it can be anything your daemon wants it to be.
It's just a 128 bit node id.  In Keso (filesystem on top of a distributed
hash table) they used the file's hash (sort of).

I'm sure there are more things that need clarification; just ask.

/t (wants a lib that isolates daemon writers from nnpfs details)