ifconfig fails setting subnet-mask

[Horst Westenkirchner]

Hello everybody,

we are internally using a C-Class IP-address and have setup within this address
space 4 subnets. Our HP/UX-server ist located in the first subnet of those four
with IP-address 192.67.170.1.

After upgrading to HP-UX 10 (and 11) the ifconfig-call for setting this netmask
fails and tells me, that this is no valid netmask.

The call is e.g.:
ifconfig lan0 inet 192.67.170.1 netmask 255.255.255.192 broadcast 192.67.170.63

With HP-UX 9.04 this worked fine and also does with all Windows machines and
other UNIXes.

What's wrong ?

Regards
Horst Westenkirchner

[Paul Heffels_1]

Hi,

Has this really worked before? Did the machine use the same ip-address of
192.67.170.1? Is lan0 still correct?

I could be wrong but I think your ip address is invalid. Here's why:
your ip address is binary 00000001
subnet mask 192 is binary 11000000
========
after ANDing: 00000000

I think to remember that this is not possible. The first valid host-address
would be:
0
64
--- +
64 192.67.170.64 through 192.67.170.127 subnet1
64 ==
--- +
128 192.67.170.128 through 192.67.170.191 subnet2
64
--- +
192
64
--- +
256

But then again, if you say it worked before then you're probably right!

[Horst Westenkirchner]

Hello Paul,

you may be right after I had a look at the ifconfig manual page.

But why does this restriction exist in HP-UX 10 and 11 ?

By this restriction I loose 2 of our 4 possible subnets. The address range from
1 to 62 (63 would be the broadcast address for subnet 0) and 193 to 254 (also
255 would be the broadcast address for subnet 3) will be lost by this way - and
that does not make any sense to me.

Furthermore the other UNIXes like Linux or SunOS (and also HP-UX 9.x), that we
are running in our environment, and also Windows NT do not have this
restriction.

Is there any workaround without changing the whole address schema that we are
using ?


Thanks
Horst Westenkirchner

[Paul Heffels_1]

Hi Horst,

To be honest: I don't know any other way then loosing the first possible range.
That's the way I was taught. I was very surprised to see that you have been
able to use it without problems. Since I learned TCP/IP through the MS-TCP/IP
study guide, which clearly states this, I was also very surprised to see that
NT doesn't nag about using this subnet mask. I have seen situations though
where NT lets you enter an invalid subnet mask, but restores it to the default
mask for your address on startup without ever telling you it did so.

To get more info:
How many hosts and clients do you have?

[Horst Westenkirchner]

Hello Paul,

we have four offices sparsed over germany and every office is using one of the
subnets. They are connected via ISDN-routers (Bianca Brick) and even those
routers have no problems with our subnet mask.
The router at our office here in Munich has the internal address 192.67.170.62
and netmask 255.255.255.192 - and works fine.

At every office we have about 10 to 15 workstations and some servers (1 to 4).

Seems that I have to take some experiments with ip-addresses and netmasks with
our HP-UX ... :-((

Thanks for your advice
Horst

[Ralf Hildebrandt_1]

Try this:
/usr/contrib/bin/nettune -s ip_check_subnet_addr 0
You need the latest ARPA patches!

[Horst Westenkirchner]

Hello Ralf,

thank you for this hint. It was the right one.

As an addition:
The nettune-command has been replaced since HP-UX 10.30 by the ndd-command.

The syntax is:
ndd -set /dev/ip ip_check_subnet_addr 0

With this setting my ifconfig call works fine.

But the online-help of ndd says, that this parameter is not supported by HP and
they do not recommend to change it. Why this ?

Regards
Horst

[Doug Cook]

Paul,
I believe you are right, but maybe looking a little too deeply at the masking
issue. You're right about the ANDing....
00000001 .1 host address
AND
11000000 .192 mask address
yeilds
00000000 .0 network address zero

But you have to remeber that the gateway making the "routing" decision is only
looking at the leftmost 2 bits of the octet to make a determination which
interface to forward it on (network 0). The individual hosts on the target
subnet will look at the rightmost 6 bits of the address of the forwarded packet
to determine if it is destined for them.

I believe the originaly suggested nettune/ndd fix is the right answer, but
won't know for sure till I get back to my problem 10.2 box tomorrow.

HTH,

[Jim Kay]

Setting the value in /etc/rc.config.d/nddconf doesn't seem to work for me. My
/sbin/init.d/net script is trying to activate the lan card BEFORE it reads
nddconf, thus it fails anyway.

My kludge was to put an ndd command (as kindly provided by another poster) as
the first thing in the net script.

It fixes the immediate problem, but is not exactly 'elegant'.

As a very newcomer to Unix, this was sure an ugly headache.

I'm sure glad you guys found this first.