>>>> Why not, instead, have the kernel increase the cmsg_len values to
>>>> whatever boundary it thinks appropriate?
> cmsg is bidirectional.  when userland passes cmsg to kernel, we can
> not do such trick (userland does not know what kind of alignment
> kernel wants), so __cmsg_alignbytes is there for you.

Yessss...but should the kernel be counting on userland to do any
alignment?  IMO it shouldn't be; it should be obeying cmsg_len values
and, if the value isn't aligned the way it (the kernel) wants, copying.
Again, no need for CMSG_ALIGN.

I think that having bytes in the control data buffer that aren't
accounted for by some item's cmsg_len is a Very Bad Idea.  Aligning is
not unreasonable, but the alignment should be added into cmsg_len!  (I
also think that alignment should never be _required_, that if the
receiver doesn't like the alignment provided, it's up to the receiver
to copy the data.  But that is, strictly speaking, a separate issue.)

Is there any real spec for the control stuff?  The only doc I've ever
found on it was source code, and searching the source tree for SCM_, I
don't find anything but source.  A few manpages mention a few of the
possible things that can appear in the control data, but I couldn't
find any spec on things like CMSG_ALIGN.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B