I'm sorry for late response.

> 	Here are two fixes to ancillary data alignment issue.  I would like
> 	to pick one of them, however, I'm still wondering which is more
> 	correct.
> 
> 	I personally vote for the latter, since the former will break if we
> 	try to run 32bit sparc binary on 64bit sparc kernel.
> 	However, I'm still not sure if it is okay to make CMSG_xx non-constant.

Hmm, I have a source which doesn't work with latter one.

The source does this:
	union {
		struct cmsghdr hdr; /* to make alignment right */
		char all[CMSG_SPACE(sizeof(passed_descriptor))];
	} cmsg;

It seems these macros are standardized in IEEE 1003.1g.
Does anyone have 1003.1g handy?
--
soda