Tag: nortel

Funny bugfixes in Nortel 6.2.4 ERS software release

There are few nice bugs in Nortel 6.2.4 ERS release (release notes). There are update surprises:

  • After Upgrading from 6.1.1 to 6.2.1, QoS configurations were lost (wi00838747)
  • Some links get disabled after upgrade from 5.1.x to 6.x (wi00731564)
  • Stack upgrade failure from 6.1.4.011s to 6.2.1.003s with a large config file (wi00882592)
  • After upgrading from 5.1.4 to 6.2.1 EDM routing/IGMP/SNOOPING table expanded indefinitely causing high CPU utilization (wi00886347)

Weird issues, which can be explained by lack of objects interoperability:

  • Unable to add static route under certain conditions, adding the route required a reboot (wi00937754)
  • With autosave enabled on 5632FD, if the power is recycled the fiber connectivity to 470-48T switch is lost (wi00941175)
  • Autonegotiation could not be disabled (wi00824799)
  • Ping or Telnet to any DNS hostname would sometimes cause loss of connectivity to the management VLAN requiring a reboot of the stack (wi00933202)

Very weird bugs, which cause many wtf-like questions on the source code:

  1. Using show running-configuration with 744 VLANs configured, spiked the CPU utilization to 100% for about 12-15 minutes (wi00907462)
  2. Switch does not learn MAC of format xx:59:xx:xx:xx:xx (wi00870510)
  3. Cannot give an IP address to the switch with the last octet as “0” (wi00872983)

I can explain the third bug by an error of checking network address in classless notation. I suppose, the code was checking an IP address in a classful way, i.e. if IP ends by zero, it’s a network address, hence can not be assigned.

How could they create bugs 1 and 2? They should’ve had hard coded numbers 744 and 0x59 (dec89). Why? what was the purpose? What was the algorithm? Can you guess an algorithm by knowing the information above?