Well, although I'm not a major kernel hacker, I would have tried to
use memset.  There's a good reason for open() and fprintf() not to
make sense inside the kernel.  I can't see a reason to have mem*() not
available in the kernel, and I think bzero() should be deprecated in
all contexts.  While there are obvious reasons for kernel code not to
be strictly conforming ANSI (even *I* see this...), I don't see much
advantage to gratuitous incompatabilities.

If bzero can be significantly faster than memset(x,y,0), then it
should be named memzero() for consistency.

Depending, we may be able to get a (very marginal) performance boost, because
ANSI allows the mem* functions to have a bit more information than the
bcopy() function can have.  (memcpy *may* be possible to make faster than
bcopy, for non-overlapping moves.)

For once, I'd be happy to look into contribiting the code.  I would say
I could write memmove, memcpy, memcmp, and memset in MI terms in, oh, say,
15 minutes.  Assuming we don't just want to import the libc ones.

-s