Re: bridge/netfilter: regression in 2.6.39.1

Le lundi 06 juin 2011 à 09:13 -0400, Neil Horman a écrit :

Ok, this makes sense to me now, thanks.  The change to the dst initalization to
mark our fake routing table as read only means we need a cow_metrics method to
copy the metrics before we we can compete the dst_metric_set in br_change_mtu.
Thats fine, but given that its really a fake routing table with only one dst
entry which (I think) is only written under the rtnl lock, why not just modify
the dst_init_metrics call so that its not marked as read-only?

It _is_ read-only, its even a "const", so if you do that you'll trap if
kernel const pages are RO

Please check commit 0972ddb237 for a proper way to deal with this.