On Thu, Aug 01, 2013 at 02:37:10AM +0200, Pablo Neira Ayuso wrote:
> On Wed, Jul 31, 2013 at 05:03:48PM -0700, David Miller wrote:
[...]
> > Therefore I don't see how NLM_F_REPLACE and NLM_F_EXCL can be used
> > at all, in those places, because the check is still "& NLM_F_DUMP"
> 
> The kind = type&3; is doing the magic there for rtnetlink. kind == 2
> means that this is a get command, and you can only set NLM_F_DUMP
> using the get command.
> 
> Since it doesn't make sense to use NLM_F_EXCL or NLM_F_REPLACE for get
> commands, there is no room for ambiguity and rtnetlink is fine.

I had re-read what I wrote to get your point. We can fix in a
different way by checking for: ops->flags & GENL_CMD_CAP_DUMP, which
means we have a .dumpit callback, so only in that case genetlink
should interpret the flags as NLM_F_DUMP.

Please, see patch attached.