Dear Maintainer,

*** Reporter, please consider answering these questions, where appropriate ***

   * What led up to the situation?
We tried to set up a bootpc testbed to develop a solution for a customer who
requires bootp.

   * What exactly did you do (or not do) that was effective (or
     ineffective)?

We found that replies from the bootpd server (from the debian bootp package)
only make it to the bootpc client's kernel, not the actual bootpc process.

We inspected the replies with tdpcump and dropwatch. We did this with and
without the --serverbcast option.

   * What was the outcome of this action?

We found that the replies' destination IP address is the IP address that is
being assigned to the client. This IP address has not been configured on the
client yet, so the kernel considers it non-local, and drops the packets.

With --serverbcast, the replies's destination IP address is the subnet
broadcast address (e.g. 10.0.1.255). Since the IP address is not configured on
the client yet, this is also consiered non-local by the kernerl, and dropped.

   * What outcome did you expect instead?

We expected replies to make it to the bootpc process. In order for these
replies to make it to the process as-is, bootpc would need to open a raw
socket. Since there is nothing in the RFC to prohibit sending replies this way,
bootpc should have the option to use a raw socket.

*** End of the template - remove these template lines ***