Hi! > From: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@xxxxxxxxxxxxx> > > [ Upstream commit cb07d915bf278a7a3938b983bbcb4921366b5eff ] > > Add rcu locks when accessing netdev when processing route request > and tunnel keep alive messages received from hardware. > /* Get the neighbour entry for the lookup */ > n = dst_neigh_lookup(&rt->dst, &flow.daddr); > ip_rt_put(rt); > if (!n) > - goto route_fail_warning; > - nfp_tun_write_neigh(n->dev, app, &flow, n, GFP_KERNEL); > + goto fail_rcu_unlock; > + nfp_tun_write_neigh(n->dev, app, &flow, n, GFP_ATOMIC); > neigh_release(n); > + rcu_read_unlock(); > return; This will make allocations more likely to fail. I thought it is okay to sleep in rcu lock critical sections...? -- (english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
