On 12 July 2014 00:55, Thomas Graf <tg...@noironetworks.com> wrote: > > Another sidenote: > > > Looking at the code, OVS does not handle NLM_F_DUMP_INTR in user space > yet > > > and the kernel dump does not call genl_dump_check_consistent() yet to > > > provide > > > the flag. So what can currently happen even without your patch is that > if > > > the dump requires more than one Netlink message, the RCU critical > section > > > is left and thus a list deletion or insertion of a flow may occur in > the > > > middle of the dump and thus provide a slightly incorrect set of flows. > > > > > > > Yes, I believe this is done deliberately for performance reasons. OVS > > userspace doesn't assume that it's getting a consistent snapshot of the > > flow table, it handles cases where flows are duplicated or missing in the > > dump. An idea has been floated to keep track of flow adds/deletes in the > > kernel to counteract this, but I got distracted by this idea first :-) > > Yes. Other code paths do the same thing. Which is why the > dump-interrupted flag has been introduced to ease detection on the > receiver side. >
I see, thanks for pointing this out. I guess that there are some cases that could benefit from this. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
