Re: [Etherboot-developers] Re: Patch for TFTP-multicast released

[<ebi...@ln...>]

"Timothy Legge" <tl...@ro...> writes:

> > > Is it possible that a proto_atftp.c needs to be written for atftp?
> > 
> > It is more likely that proto_tftm.c or possibly atftp needs to be bug
> > fixed.  I don't know that either of them has had any interoperability
> > testing.
> > 
> 
> Yes, you are correct.  It looks like the server Anselm used was fairly
> simple.  There were a number of issues that had to be resolved related
> to the way that server was responding.  It looks like the server kept
> the port that it was initially contacted on so the code did not extract
> the src port from the udp).  
> 
> I am now able to download the file, but I think there are some bugs in
> the code to allocate and reassemble the file since I get an invalid
> file.

Are you getting a bad ELF checksum or is it some other problem?

> I should sort it out fairly soon.  Since my code is a hack of the
> current code, I would recommend either a rewrite of the current
> implementation or IMO a better solution: adding the multicast option to
> the standard tftp code.  If I understand what I have seen so far, it
> should be a fairly minor change.
> 
> Comments?

There is one large difference between a multicast download and a normal download.

When doing multicast the data packets potentially arrive in a random order so there
is not enough information to place them where they need to go immediately.

Because of this for a multicast transmission we need to allocate memory for the file
we are downloading and them move it to the correct location in memory as a second pass.
For low memory machines that can be a problem, so I would not recommend changing the
tftp code to use two passes to load a file.  So there is a reasonable justification for
keeping two separate code paths.  This does not mean that subroutines cannot be reused
just that they must be reused carefully.

Eric