[Freeipmi-devel] FreeIPMI Issues

[Levi Pearson]

I've been looking at FreeIPMI for the past few days for use on clusters
of Dell 1950/2950/1435 nodes.  For the most part, it's perfectly suited
for the task, but I've run into a couple of problems and found some
features I'd like to see, and perhaps implement in-house.

I'd like to work with the core developers so that any fixes and new
features we make are acceptable to you, so we don't have to maintain a
separate branch.  With that in mind, here's what I've found so far.

With ipmiconsole, I discovered that the internal handling of the Kg key
is done with string manipulation functions, and that there is also an
off-by-one error.  I'd like the ability to enter the key in hexadecimal
and have it treated as a 20-byte binary field instead of a string, which
matches how the key is handled in the Dell utilities.  I've started
working on a patch for this, which adds some new command line options
for hex entry of the key, but I put it aside to look through the rest of
the FreeIPMI suite.

I'd also like to integrate libfreeipmi with conman, and I'd appreciate
hearing if anything has been done in that direction yet.

With the exception of ipmipower and ipmiconsole, the FreeIPMI utilities
have issues dealing with having Per-Message Authentication and
User-Level Authentication disabled.  This wouldn't be much of a problem,
but the Dell BMCs have Per-Message Authentication disabled by default
and the utilities can't communicate with them to change that option, nor
does there seem to be an option to change it in bmc-config yet.  There's
also no support for changing it in Dell's configuration utilities, so I
had to use ipmitool's raw command to set up a BMC so I could check
things out.

I've done some investigation into the best place to put checks for those
options.  Right now, it seems like the best way would be to have
ipmi_cmd_get_channel_authentication_capabilities set some flags or new
struct members in the ipmi_device_t that is passed to it.  Then,
ipmi_lan_open_session could change the authentication type in the
ipmi_device_t to NONE after it finishes authenticating the session (if
the appropriate flags are set, of course, and barring the need for the
workaround present in ipmipower).  Any thoughts on this?

We'd also like some more extensive PEF configuration options in
bmc-config, but I haven't looked into that with any detail yet.

Anyway, thanks for the excellent software, and let me know what you
think about my ideas above.

                --Levi