[9fans] silly dhcpd question

[9fans] silly dhcpd question

[Ronald G. Minnich]

I have this in /lib/ndb/local

sys=g1 ip=10.128.150.217  dom=vm ipgw=10.128.150.232
                ether=00D0C9670733 auth=v9 fs=v9 cpu=v9 
		authdom=vm bootf=/lib/ndb/local proto=il


I get this:
bootp from unknown hwa01_00d0c9670733 via 10.128.150.232
bootp  0.0.0.0->10.128.150.232 from hwa01_00d0c9670733 via 10.128.150.232, file 
bootp from unknown hwa01_00d0c9670733 via 10.128.150.232



any idea what to look for? 

ron

Re: [9fans] silly dhcpd question

["Nils O. Selåsdal"]

Ronald G. Minnich wrote:
> I have this in /lib/ndb/local
> 
> sys=g1 ip=10.128.150.217  dom=vm ipgw=10.128.150.232
>                 ether=00D0C9670733 auth=v9 fs=v9 cpu=v9 
> 		authdom=vm bootf=/lib/ndb/local proto=il
> 
> 
> I get this:
> bootp from unknown hwa01_00d0c9670733 via 10.128.150.232
> bootp  0.0.0.0->10.128.150.232 from hwa01_00d0c9670733 via 10.128.150.232, file 
> bootp from unknown hwa01_00d0c9670733 via 10.128.150.232
> 
> 
> 
> any idea what to look for? 
Use lowercase in /lib/ndb/local for the ether.
ether=00d0c9670733

Re: [9fans] silly dhcpd question

[Ronald G. Minnich]

On Sat, 14 May 2005, "Nils O. Selåsdal" wrote:

> Use lowercase in /lib/ndb/local for the ether.
> ether=00d0c9670733


yes, that was it. Weird, cause i tried lowercase first, then went upper 
... something else must have happened. 

Thanks

ron

Re: [9fans] silly dhcpd question

[boyd, rounin]

>> Use lowercase in /lib/ndb/local for the ether.
>> ether=00d0c9670733

> yes, that was it. Weird, cause i tried lowercase first, then went upper

yup, the patch i sent in to fix the doc to warn you of it got
most of 'em i think.

i had thought of changing the ndb lookup code,
but decided against it.  in all of the other cases
case is insignificant so a hack for 'ether' just
didn't seem to be justifiable.
--
MGRS 31U DQ 52572 12604

Re: [9fans] silly dhcpd question

[lucio]

> any idea what to look for? 

It's possible that NDB is still case sensitive.  Use a lower case "d"
in the ethernet address (or fix NDB).

++L

Re: [9fans] silly dhcpd question

[lucio]

>> any idea what to look for? 
> 
> It's possible that NDB is still case sensitive.  Use a lower case "d"
> in the ethernet address (or fix NDB).

I've submitted a patch for /sys/src/libip/parseether.c which relies on
strtoul().  Is strtoul() (I haven't heard from Doug Gwyn for a long
time) meant to be case sensitive?

++L

PS: In passing, the patch is untested :-(  And broken :-( :-( :-(

Re: [9fans] silly dhcpd question

[andrey mirtchovski]

> (I haven't heard from Doug Gwyn for a long
> time)

doug relied on the usenet->9fans part of the gateway to work. he
hasn't posted since that stopped working.

i know this because i spotted his mails in comp.os.plan9 and notified
him that they don't make it to the list. i asked in 9fans who the
owner of the gateway is in order to get at least a couple of people's
emails to make it back here, but i got no reply.

doug does post, his last one in comp.os.plan9 is from the 4th of may.
i'll take the liberty of quoting it here:

----
p...@post.cz wrote:
> i suppose it is a bad manner to #include foo1.h from within foo.h
> is it?

Not as such.  If foo1.h defines the interface to some
resources that foo.h needs, then it *ought* to be
#included, so that the user of foo.h doesn't need to
know implementation details.  For example, if foo.h
declares a structure maintained by functions in the
"foo" package (also declared by foo.h) and that
structure has a private member that is a Boolean
flag, foo.h ought to be #including <stdbool.h> so
that it can use the proper type for "bool".  This
is generic advice, and the Plan9 developers seem to
disagree (for no good reason so far as I have heard).
----

to see all other posts authored by him:

http://groups-beta.google.com/groups?q=group:comp.os.plan9+author:gwyn&start=0&scoring=d&num=10&hl=en&lr=&as_drrb=q&as_mind=1&as_minm=1&as_miny=1981&as_maxd=14&as_maxm=5&as_maxy=2005&safe=off&

andrey

ps: sorry about the third person, douglas.

Re: [9fans] silly dhcpd question

[andrey mirtchovski]

by "hasn't posted" i meant "his posts stopped making it to 9fans"

Re: [9fans] silly dhcpd question

[Russ Cox]

You guys don't appear to be running the Plan 9 that I am.

1. As Boyd pointed out, the correct fix is to use lower case
in ndb.  Ndb is treating its entries as plain old case-sensitive
strings.  For the most part, the lookup routines have no
idea what an IP address or an ethernet address is, so when
dhcpd asks for an entry with ether=00d0c9670733 it doesn't
match against ether=00D0C9670733.  

The correct fix, if there is one, is to change the ndb reading
routines to recognize certain attributes and canonicalize
them: ip, ether, dom, maybe sysname.  But then it will be
weird that dom is case-insensitive but bootf is not.  Should
fs be case-sensitive?  Etc.  I'm much happier requiring that
the ndb entries be stored in canonical form than doing magical
rewriting under the hood. 

2. Parseether (which ndb isn't using) correctly handles upper
case ethernet addresses already.  

3. Strtoul correctly handles upper case hex already, as required
by the C standard.  

The program below prints

    abcdef
    abcdefabcdef

for me.

Russ

#include <u.h>
#include <libc.h>
#include <ip.h>

void
main(void)
{
	uchar ea[6];

	fmtinstall('H', encodefmt);
	print("%x\n", strtoul("ABCDEF", 0, 16));
	parseether(ea, "ABCDEFABCDEF");
	print("%.6lH\n", ea);
}

Re: [9fans] silly dhcpd question

[boyd, rounin]

> I'm much happier requiring that the ndb entries be stored
> in canonical form than doing magical rewriting under the hood.

this was my decision too, after a bit of squiz at the problem
and not wanting to break canonical interfaces or anything
else for that matter.

> But then it will be weird that dom is case-insensitive but bootf
> is not.  Should fs be case-sensitive?  Etc.

and lots of questions like the above.  i didn't feel i was in a
position to decide.
--
MGRS 31U DQ 52572 12604

Re: [9fans] silly dhcpd question

[Axel Belinfante]

> > I'm much happier requiring that the ndb entries be stored
> > in canonical form than doing magical rewriting under the hood.
> 
> this was my decision too, after a bit of squiz at the problem
> and not wanting to break canonical interfaces or anything
> else for that matter.

Maybe some confusion could be avoided if anything that reports
ethernet addresses to the user would use this same canonical format?

Right now when I boot 9load reports the ether addres in lower case
(via %E where it installed eipfmt from ip(2)
 which uses %.2lux%.2lux%.2lux%.2lux%.2lux%.2lux)

  ether#0: elnk3: port 0xDC00 irq 11: 00c0df15fb45

and a little later the kernel reports it in upper case
(directly via %2.2uX%2.2uX%2.2uX%2.2uX%2.2uX%2.2uX)

  #l0: elnk3: 100Mbps port 0xDC00 irq 11: 00C04F15FB45


just my 0.02 whatever.
Axel.

Re: [9fans] silly dhcpd question

[Russ Cox]

This is a good idea.  I think I got all the places in the
kernel that print ethernet addresses.  Really we should
use %E but I'm not sure why we don't.

Done.

Russ

Re: [9fans] silly dhcpd question

[Ronald G. Minnich]

On Sat, 14 May 2005, Axel Belinfante wrote:
> Maybe some confusion could be avoided if anything that reports
> ethernet addresses to the user would use this same canonical format?

Or maybe just this at the top of the distributed /lib/ndb/local:

# Make the ethernet addresses all lower case or you'll regret it.

Then you don't have to do all that work :-)

ron