Andrew Gillham wrote about Re: tftp on 1.3:
:> > Grrr, now I know why my NetBSD 1.3 machines are no longer usable as a tftp
:> > server for my cisco equipment!  Is there a good reason this was changed? 
:> > Particularly when '-s /tftpboot' is used? 
:> > If anyone already has a working patch, please mail it to me, otherwise I'll
:> > have to come up with one today at work.
:> > 
:> > I don't want to sound like I'm whining (which I am), but I use a NetBSD
:> > machine for a tftp server, and for rsh/rcp services with our cisco router
:> > network.  With 1.3, these were *all* broken for me.  The rsh/rcp problem
:> > is related to NIS, which I can disable to make it work ok. (though I have
:> > a patch that makes it work with NIS)
:> 	cd /tftpboot
:> 	ln -s tftpboot .
:lrwxr-xr-x  1 root  wheel     1 May 23  1996 tftpboot -> .
:Been there for some time.  I'm concerned about trying to write a file
:called "routername-confg" via tftp.  Used to work, doesn't now.  I am
:not about to retype the whole filename just to put a '/' on the front.

You can't do that, I don't think (the soft linking, I mean).  Look in
src/libexec/tftpd/tftpd.c in  validate_access() .  From memory, all
filenames had to either start with ``/'' or an approved directory path
handed to tftp via the command line (or maybe also some other method) to
allow you to write there.  For the ``-s'' flag this is annoying and I
didn't notice it till I was, funnily enough, doing stuff on my Cisco.

After a very quick look at it, I'd say you would only need to change:
	if (mode != RRQ || !strncmp(filename, "../", 3))
		return (EACCESS);
to
	if ((!secure && mode != RRQ) || !strncmp(filename, "../", 3))
		return (EACCESS);

but you better check that yourself, and maybe send-pr it if you reckon it's 
correct.  That should allow reads/writes to the default directory with the
``-s'' flag used and no other directories specified.
-- 
Geoff Wing   <gcw@pobox.com>            Mobile : 0412 162 441 
Work URL: http://www.primenet.com.au/   Ego URL: http://pobox.com/~gcw/