Paul -

I'm sure it's the commit that is causing the issue, although I agree that
it might be a side effect of some other change( and I just read your new
email with analysis and I agree with it).  When I have the commit in my
tree and run my ospf smoke tests against it, ospf becomes a crash fest and
none of the tests pass.  In fact it crashes so much that it causes the
switch to run out of space with 1000's of core files.  When I revert the
change the tests now run and I don't see any core files at all.

donald

On Mon, Dec 21, 2015 at 6:04 AM, Paul Jakma <[email protected]> wrote:

> On Thu, 17 Dec 2015, Donald Sharp wrote:
>
> Joakim :
>>
>> I am experiencing issues with an ospf crash associated with this bug:
>>
>> commit 5c1791f28e2e831e4e9b92c3c2f7d8ed832cb968
>> Author: Joakim Tjernlund <[email protected]>
>> Date:   Fri Apr 25 14:36:16 2014 +0200
>>
>>    ospfd: PointToPoint neighbors are identified by router ID
>>
>>    According to RFC 2328, section 10.5  PointToPoint neighbors
>>    should be identified by router ID instead of source IP address.
>>
>>    Signed-off-by: Joakim Tjernlund <[email protected]>
>>
>>
>> My reproduction:
>>
>> Starting Config:
>> interface swp1
>> ip address 192.168.39.1/24
>> ipv6 address fc00:192:168:1::101/64
>> ipv6 nd suppress-ra
>> link-detect
>> !
>> !
>> router ospf
>> network 192.168.39.0/24 area 0.0.0.0
>> !
>>
>> Sequence of events to make it happen:
>> root@cumulus:~# vtysh
>>
>> Hello, this is Quagga (version 0.99.25-dev).
>> Copyright 1996-2005 Kunihiro Ishiguro, et al.
>>
>> cumulus# conf t
>> cumulus(config)# int swp1
>> cumulus(config-if)# ip ospf network point-to-point
>> Warning: closing connection to ospfd because of an I/O error!
>>
>
> I can only provoke this crash very sporadically, with master or with
> commit 5c1791f. Which makes me wonder, how can we be sure it is that commit?
>
> I havn't yet managed to get it to crash on earlier commit, but I'm having
> an equally hard time getting it to crash on later ones. So...
>
> The commit concerned is self-contained and looks good. The crash seems
> related to flushing LSAs, and probably a use-after-free bug of some
> data-struct pointed to by oi or nbr. My guess is the bug is extant and not
> in that commit, even if that one might help to tickle the bug in some
> cases. (Also, we don't support PtP on ethernet yet do we?).
>
> regards,
> --
> Paul Jakma      [email protected]  @pjakma Key ID: 64A2FF6A
> Fortune:
> People are like onions -- you cut them up, and they make you cry.
>
_______________________________________________
Quagga-dev mailing list
[email protected]
https://lists.quagga.net/mailman/listinfo/quagga-dev

Reply via email to