Thank you for your suggestion.  I'm sorry for my delayed response .

I understand "MUST NOT" isn't good for these objects.
Then, I'd like to discuss your suggested sentence in the WG meeting today.

Thank you.
Hirochika


On Jul 4, 2014, at 3:08 PM, "Black, David" <[email protected]> wrote:

>>> I also question the use of MUST NOT persist.
>>> How does an implementation allowing the enable setting to persist break the
>>> protocol or the managed entity?
>>> I think an appropriate RFC2119 usage would be “implementations MAY choose to
>>> not persist the values.”
>> 
>> Implementations that support SET for these objects can store the values
>> in transient storage such that a restart of the agent itself does not
>> lose the configuration. One example implementation would be
>> writing the state to ramdisk which would not persist across system reboot.
>> As usual sets can be complex to get right and its difficult to understand
>> what a mib module is telling you to do here.
>> Maybe some guidance on that point would be useful in the discussion.
> 
> I concur w/Dave Harrington's original concern - "MUST NOT" seems excessive.
> 
> Suggestion:
> 
> OLD
>      Changes to this  
>       object MUST NOT persist across re-initialization of     
>       the management system, e.g., SNMP agent.
> NEW
>      Changes to this  
>       object may be lost when the management system, e.g., SNMP agent,
>      is re-initialized.
> 
> I don't see the harm in persisting changes to notification enablement
> settings.  Mike's point seems to be that one cannot and should not rely
> on changes persisting, not that it's harmful or otherwise wrong to do so.
> 
> Thanks,
> --David
> 
>> -----Original Message-----
>> From: OPSAWG [mailto:[email protected]] On Behalf Of Michael MacFaden
>> Sent: Friday, July 04, 2014 1:43 PM
>> To: David Harrington
>> Cc: [email protected]
>> Subject: Re: [OPSAWG] I-D Action: draft-ietf-opsawg-vmm-mib-01.txt
>> 
>> dbh writes:
>>> I think there is an undocumented security vulnerability.
>>> Because we REQUIRE the read-write objects to not persist across reboots, an
>>> undetected reboot of the SNMP system shuts off notifications.
>>> If an operator turns on notifications via the enable objects in this module,
>>> they presumably want to be told of certain events that might occur.
>>> If notifications are used to report security issues, an attacker that can
>>> somehow cause the system to restart will disable notifications, presumably
>>> including those that warn an operator of an attack.
>>> (as well as those that warn an operator of an undesirable operational state,
>>> such as interface down).
>>> I think that’s a problem.
>>> 
>>> I also question the use of MUST NOT persist.
>>> How does an implementation allowing the enable setting to persist break the
>>> protocol or the managed entity?
>>> I think an appropriate RFC2119 usage would be “implementations MAY choose to
>>> not persist the values.”
>> 
>> Implementations that support SET for these objects can store the values
>> in transient storage such that a restart of the agent itself does not
>> lose the configuration. One example implementation would be
>> writing the state to ramdisk which would persist across system reboot.
>> As usual sets can be complex to get right and its difficult to understand
>> what a mib module is telling you to do here.
>> Maybe some guidance on that point would be useful in the discussion.
>> 
>> Regarding notification state on/of and it being security issue, this isn't
>> only an issue in this mib module I think. Its common place problem with
>> with event/push only designs. We  could add text to explain that
>> any design that is 100% reliant  on notifications as the sole means
>> of maintaining distributed state is a failed design to begin with.
>> notification only designs must be bounded either by poll or reverse
>> poll/periodic
>> event to detect a failure to receive notifications regardless of what caused
>> the failure.
>> 
>> Cheers,
>> Mike MacFaden
>> 
>> _______________________________________________
>> OPSAWG mailing list
>> [email protected]
>> https://www.ietf.org/mailman/listinfo/opsawg
> _______________________________________________
> OPSAWG mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/opsawg

-- 
Hirochika Asai <[email protected]>, The University of Tokyo

_______________________________________________
OPSAWG mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/opsawg

Reply via email to