[OpenAFS-devel] AFS with DNS using AFSDB records

[Jeff Riegel]

Derrick J Brashear wrote:
> On Sat, 28 Apr 2001, Shyh-Wei Luan wrote:
> 
> > Jeff Riegel also has an implementation of the AFSDB support that he is
> > currently testing.   Maybe we should compare the two independent
> > designs/implementations.   Since this is going to be a very important
> > enhancement to AFS and would affect all platforms, it would not hurt to
> be
> > a little more careful and to leverage this accidental dual
> implementation.
> 
> From what of it was leaked into the 95/98 port (it appears to be
> incomplete) this would appear to use a fixed list of cells but do a lookup
> at initialization time, which is different from kolya's patch, which does
> it on the fly while the client is running. Is that an accurate summary of
> the situation?
> 
> -D

The code I'm currently testing does a DNS lookup on the fly.  The DNS
is checked on calls to afs_GetCellByName().  This function also calls
afs_NewCell() to add the cell to the kernel's database so calls that find
the cell by cell number or index number will still work.  A DNS lookup
is also done in afsconf_GetCellInfo() which is called by klog and other
programs.  The CellServDB is still read and is used if the DNS lookups fail.

The code that talks to the name server assembles its own request packets
and parses the response packets from the server by hand.  In the kernel,
I used the network functions from rx_knet.c to send and receive network
data: rxk_NewSocket(), osi_NetSend(), etc.

I've tested the code in the Linux client and libuafs and it works so far.
I'd be interested in hearing more about kolya's approach if it differs
extensively from the above.

Derrick: I didn't intend for any DNS code to end up in the 9x patch
I submitted, since I haven't finished the DNS support for the NT/9x
client yet.  Could you let me know which files contain the code?

Jeff Riegel