Re: [PATCH v1.5 3/5] key: add tpm_send command

[Serge Hallyn <serge.hallyn@canonical.com>]

Quoting David Safford (safford@watson.ibm.com):
> On Tue, 2010-11-23 at 20:32 -0600, Serge Hallyn wrote:
> > Quoting Mimi Zohar (zohar@linux.vnet.ibm.com):
> > > Add internal kernel tpm_send() command used to seal/unseal keys.
> ... 
> > > +int tpm_send(u32 chip_num, void *cmd, size_t buflen)
> > 
> > Hate to nit-pick, but any particular reason you're not following the
> > rest of the file and using 'struct tpm_cmd_t *cmd' here?
> > 
> > Acked-by: Serge E. Hallyn <serge.hallyn@canonical.com>
> 
> We put some thought into this one. TPM command packets are
> binary blobs with lots of optional and variable length fields,
> and there are at least three common approaches to creating them:
> structures (as used in tpm.c), load/store (as used in trousers
> and trusted-keys), and an sprintf like format string (as used
> in the original libtpm.) Each has its advantages and disadvantages.
> Structures are nice for the simple TPM commands, but they become
> unwieldy for the complex commands like seal and unseal. Load/store
> is much more readable for the complex seal and unseal commands.
> Format strings are nice for creating the most complex commands
> in the fewest lines of code, but are way overkill for simple ones.
> 
> With the void *cmd, we are allowing the other modules to pick
> whichever method most suits their needs.

Jinkeys, that's complicated :)

But doesn't that mean that the transmit_cmd() parameters are lying?
Should the second argument for transmit_cmd() be a union?

(If only to help out the lamentable reader)

thanks,
-serge