Re: RTNL: assertion failed at net/ipv6/addrconf.c (1699)

[Hannes Frederic Sowa <hannes@stressinduktion.org>]

On Di, 2014-09-02 at 11:40 -0700, Cong Wang wrote:
> On Tue, Sep 2, 2014 at 11:18 AM, Hannes Frederic Sowa
> <hannes@stressinduktion.org> wrote:
> > Those ASSERT_RTNLs were misplaced and only caught the callers mostly
> > from addrconf.c. I don't mind getting reports from stable kernel users
> > and fixing those, too (or help fixing those). ASSERT_RTNL is not
> > dangerous.
> >
> > We had a long history in not correctly using rtnl lock in ipv6/multicast
> > code and those wrongfully placed ASSERT_RTNLs were my bad when I fixed
> > the duplicate address detection handling.
> >
> > If enough multicast addresses are subscribed to an interface we might
> > again get those splats because enabling promisc mode on an interface
> > will also check for rtnl lock.
> >
> 
> Sure, I never doubt adding ASSERT_RTNL() is helpful, I just still think
> this should be for net-next, or at least a separated patch. I don't want
> my patch to be blamed in others' "Fixes:". :)

Come on, that's why we have community review. Nobody blames anyone
because of added regressions. It's more a fault of the community then,
and it works out fairly good I think! Even others are keen on fixing
your bugs sometimes. ;)

If fixes tag is well researched, it won't point to the addition of
ASSERT_RTNL() but your patch would help to discover a bug somewhere else
in the stack.

I think for this patch a fixes-tag is hard to find because it is hard to
find because it dates back to the beginning of the git history IMHO.

Bye,
Hannes