Re: [PATCH] [RFC] increase MAX_ADDR_LEN

[Roland Dreier]

>>>>> "David" == David S Miller <[e-mail]> writes:

    David> So how are apps able to specify such larger hw addresses to
    David> configure a driver if IFHWADDRLEN is still 6?

In the InfiniBand case, the device's hardware address comes from a
combination of the port GID (which is set by the InfiniBand subnet
manager through an IB-specific mechanism) and the queue pair number
that the driver gets when it initializes.  There definitely still are 
problems to solve, such as specifying static ARP entries.

    David> I'm not going to increase MAX_ADDR_LEN if there is no user
    David> ABI capable of configuring such larger addresses properly.

What would you consider a palatable ABI?  (I'm happy to implement it)
Enlarging sa_data in struct sockaddr doesn't seem feasible.  I guess
we could add a new socket ioctl() or extend SIOCGIFHWADDR/SIOCSIFHWADDR
somehow....

Thanks,
  Roland <[e-mail]>