This should be under .if ${MK_INET6_SUPPORT} != "no".

ifconfig(8) generally resolves a host name wherever an address can be specified. I am not sure if it must be supported here because a 6rd configuration is address-family specific, but this should be noted in the manual page at least.

This can be removed safely because strtonum() should also check whether it is a number or not.

I think either 0 or 32 must be explicity rejected because 0 is silently converted to 32 in the kernel.

"prefix length" is better than "width", I think.

memcpy()?

I think there is no need to define ia6 after fixing the following two lines.

• LHS should be simply addr6.
• RHS can be simplified by using IFA_IN6(is).
• memcpy() or bcopy() might be preferred. The line 438 uses it even for a 32-bit long address, for example. And by the way, L.392 rewrote bcopy() with memcpy(). Consistency in this file might be good.

• IA6_MASKIN6(ia)

in_nullhost()?

This should be inside the next if-conditional.

These two lines are confusing. How about v4prefix = ntohl(sc->srcv4_addr & (0xffffffffU << v4suffixlen))?

(sc->v4prefixlen > 32) must be rejected and return with NULL. It never happens if SIOCSDRVSPEC guarantees it is between 1 to 32, but having a sanity check is useful.

(plen > 64) must be rejected and return with NULL before this. It is still possible to happen.

sizeof(args)

bcopy or memcpy.

The value must be checked to guarantee that it is within 1 to 32.

Consistency. SIOCSDRVSPEC checks ifd_cmd first, and this checks ifd_len first.

All examples of 6rd configuration I can find uses the IP address for the BR. The pfSense implementation also only supports using an IP address.

All together I'd be inclined to keep this the way it is. It'll be easy to extend later if we do encounter users who want to use a DNS name.

The if_stf.4 changes for 6rd describe the configuration command as "The border router IPv4 address is configured with the
.Xr ifconfig 8
.Cm stfv4br
command."

That expects a struct in_addr, and we use in_addr_t here.