On Thu, Sep 06, 2007 at 02:50:19PM -0400, Mark Hahn wrote: > >But, I assume network configuration is viewed as a userland activity, > >and not something the kernel per-se should be involved in--your "simple > >convenience" argument of earlier. > > I'm actually persuaded by the concept of doing all IP config in user-space.
Agreed. That's the reason I heed HPA's advice. ;) > but since tftpd is user-space, ipappend seems like a great way for one > piece of user-space (tftpd) to communicate with another (initrd ip config) > without involving another fairly complicated tool (dhclient). > > in other words (and afaik), the kernel has never been tweaked > to recognize the ipappend string... net/ipv4/ipconfig.c handles it. Having said that, it would be trivial to handle it strictly in userland in an initramfs. In that vein, I have a shell script that emulates pxelinux, so that I can provide a superset of PXE booting in userland with only kexec and either a tftp client or wget to replace tftp with http or ftp. Clearly, supporting https, scp, multicast, et al. are also possible. -- David N. Lombard, Intel, Irvine, CA I do not speak for Intel Corporation; all comments are strictly my own. _______________________________________________ Beowulf mailing list, [email protected] To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
