Netgraph/mpd5 stability issues

[Mike Tancsa]

On 1/31/2011 12:10 PM, Julian Elischer wrote:
> On 1/31/11 8:58 AM, Julian Elischer wrote:
>> On 1/31/11 7:13 AM, Przemyslaw Frasunek wrote:
>>>> And in this one, can you please show *hook->hk_peer ?
>>> (kgdb) print *hook->hk_peer
>>> $2 = {
>>>    hk_name = "\b\000\000\000
>>> \000\000\000\004\000\000\000\001\000\000\000ŐRí\003\003ö\0248cmd4\000\000\000",
>>>
>>> hk_private = 0x0, hk_flags = 0, hk_refs = 0,
>>>    hk_type = 0, hk_peer = 0x0, hk_node = 0x0, hk_hooks = {le_next =
>>> 0x566226,
>>>      le_prev = 0x99e79c03}, hk_rcvmsg = 0x38ef45, hk_rcvdata =
>>> 0x28d6a8a1}
>>
>> that's not supposed to be able to happen.
>> It's supposed to point to SOMETHING, even if it's the "dead" hook.
>> does the dead hook point to itself? itprobably should if it doesn't.
>> (and it should have a name of 'dead' if it doesn't already).
> 
> Replying to self.. all these things are in fact true, (just looked at
> source) so this is not a pointer
> to the dead node.   So, how do we get a NULL peer?
> unless the hook is destroyed by another thread while we are accessing it,
> but I think from memory that that should set it to 'dead' not NULL.



If there is extra debugging code you would like me to add, it does not
take too long for this to happen on my one LNS... 4-5 days.

	---Mike