Recently a customer case came with following -

"It appears that the customer has hit the issue in the above [1] CR
where the  IP Instances introduced in the S10 Update 4 KU is preventing
the use of ndd to enable ip debug with ndd as done in the past. The
customer feels  that having to use mdb to do it on a live, running
system is risky and not user friendly. The previous ability to use ndd
should be restored/corrected in their opinion."

[1] 6621665
   IP Instances removed console printing when ndd -set /dev/ip ip_debug 1


Though I understand facility provided by ip_debug is being planned to
be provided by SDT probes, above quote points to the general feeling
towards modifying anything on a live system using mdb -kw. And there
are reasons for it - I do recall someone mentioned a system crash
while attempting to modify a variable using mdb. It was definitely
not a bug in mdb, I'd guess it was a typo. mdb -kw is unforgiving
to user errors.

There had been suggestions to pass a script to customers to modify
variables through mdb e.g.

   # echo "ip_debug/W 0xa"|mdb -kw

which will work fine in most of the cases but will fail in certain cases
where customer has to use it in both 32 bit and 64 bit kernels. e.g.
variables like ip_rput_pullups which are of type long differs in size -
4 bytes on 32 bit kernel vs 8 bytes on 64 bit kernel. It's easy
to miss this subtle point when writing a script.

Perhaps, Brussels can deliver a mdb dcmds, say, "::assign". This dcmd
can check the size of the variable and then writes to it. I'm
not sure if it will make customer feel completely comfortable,
it certainly will help.

-
prakash








_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to