What does ifup do when “interfaces” file has an invalid netmask?

nazekimi · 10-24-2013, 02:04 AM

The interfaces file looks like below:

Code:

auto lo
iface lo inet loopback

auto eth0
iface eth0 inet static
    address 192.168.0.130
    netmask 250.255.255.0
    gateway 192.168.0.254

As you can see the netmask address is sort of invalid. (250 instead of 255) The problem is when I use the following code to get the current IP address and netmask:

Code:

ioctl(iSock, SIOCGIFADDR, &stIFReq);
ioctl(iSock, SIOCGIFNETMASK, &stIFReq);

It always returns "72,0,xxx,xxx" (the last two octets seem to change depending on the device, but always constant for the same device) on both ioctl functions. I wonder if this is some sort of error code or bug during ifup. I would like to know what type of checking ifup does on netmasks/ip addresses and what does it do if the address is judged as invalid.

What steps can be taken to recover from a "bad" netmask address setting and maybe run a default "255.255.255.0" address?

Thanks.

Additional info:
Linux version 2.6.35.3, running on an ARM cpu embedded system
BusyBox v1.18.5
[ifup] is not a script but a C program on the system

nini09 · 10-25-2013, 04:04 PM

The 1 bits of correct netmask should be continue.

nazekimi · 10-28-2013, 01:36 AM

Quote:

Originally Posted by nini09

The 1 bits of correct netmask should be continue.

Is there anyway for ifup or interfaces to ignore this rule? What steps should I take on boot-up to recover from such a situation?

sag47 · 10-28-2013, 10:13 AM

You can run a script on startup that checks the current netmask and if it is not valid then set a default... See man 5 crontab and look at the @reboot alias.

Here's a python function for checking a single octet of a netmask... I don't have time to write the rest of the script to check a whole netmask but I'd recommend using it in a class with recursion to do it.

Code:

def check(partial_mask):
  x=0
  while True:
    if x==partial_mask:
      return partial_mask
    elif (x==255 and x!=partial_mask):
      return None
    x=x>>1|128

The core of your question *should be* is why does your netmask get arbitrarily set to something not proper? You should fix the root of the problem rather than create patch work around temporarily fixing it. Duct tape will only work for so long.

nini09 · 10-28-2013, 03:19 PM

No, you can't ignore the rule except your own kernel.

nazekimi · 10-30-2013, 04:57 AM

Quote:

Originally Posted by sag47

You can run a script on startup that checks the current netmask and if it is not valid then set a default... See man 5 crontab and look at the @reboot alias.

Here's a python function for checking a single octet of a netmask... I don't have time to write the rest of the script to check a whole netmask but I'd recommend using it in a class with recursion to do it.

Code:

def check(partial_mask):
  x=0
  while True:
    if x==partial_mask:
      return partial_mask
    elif (x==255 and x!=partial_mask):
      return None
    x=x>>1|128

The core of your question *should be* is why does your netmask get arbitrarily set to something not proper? You should fix the root of the problem rather than create patch work around temporarily fixing it. Duct tape will only work for so long.

Thank you for your suggestion. I guess configuring it manually is better than just feeding ifup the interfaces file.

As per the title of my post, I really was looking for reasons why ifup does what it does on such address settings. Perhaps my English is hard to understand.
Anyway, since there are really no clear information on that direction, as of now, I'll just go ahead and apply that duct-tape.

sree_ec · 10-30-2013, 08:03 AM

slightly offtopic:

why is this netmask invalid?

TIA

sag47 · 10-30-2013, 11:34 AM

Quote:

Originally Posted by sree_ec

why is this netmask invalid?

Quote:

Originally Posted by nazekimi

Code:

    netmask 250.255.255.0

Google search. The short answer is a netmask designates how many bits in an ip address are for network identification and how many bits in an ip address are for host identification.

Essentially the IP address is split into two halves... e.g. 255.255.0.0

Code:

11111111.11111111.00000000.00000000

Since there can only be two halves (network identification or host identification) then 250.255.255.0 breaks that standard because there can't be zeroes mixed into the network identification. The bits identifying network must be contiguous 1's.

Not valid: 250.255.255.0 (converted into binary below to emphasize point)

Code:

11111010.11111111.11111111.00000000

sree_ec · 10-30-2013, 01:59 PM

I never thought network Id is a part of netmask in the sense you have just explained. What my understanding was networkId=Ipaddress (logical AND) netmask.

I never thought it exists as a part of netamsk and should contain only 1s.

Thanks for explaining

sag47 · 10-30-2013, 02:01 PM

Quote:

Originally Posted by sree_ec

I never thought network Id is a part of netmask in the sense you have just explained. What my understanding was networkId=Ipaddress (logical AND) netmask.

You're correct. Network ID == IPADDRESS & NETMASK.

So if netmask doesn't contain all 1's from left to right contiguously then you'll get weird network ID's from some IP addresses but not others. That's why you can do the logical AND because of the format of the netmask.