ipvs: fix crash if scheduler is changed
authorJulian Anastasov <ja@ssi.bg>
Mon, 29 Jun 2015 18:51:40 +0000 (21:51 +0300)
committerSimon Horman <horms@verge.net.au>
Tue, 14 Jul 2015 07:41:27 +0000 (16:41 +0900)
commit05f00505a89acd21f5d0d20f5797dfbc4cf85243
tree9e4948698b71fcec2dc3b233c22893efded4c16b
parent4754957f04f5f368792a0eb7dab0ae89fb93dcfd
ipvs: fix crash if scheduler is changed

I overlooked the svc->sched_data usage from schedulers
when the services were converted to RCU in 3.10. Now
the rare ipvsadm -E command can change the scheduler
but due to the reverse order of ip_vs_bind_scheduler
and ip_vs_unbind_scheduler we provide new sched_data
to the old scheduler resulting in a crash.

To fix it without changing the scheduler methods we
have to use synchronize_rcu() only for the editing case.
It means all svc->scheduler readers should expect a
NULL value. To avoid breakage for the service listing
and ipvsadm -R we can use the "none" name to indicate
that scheduler is not assigned, a state when we drop
new connections.

Reported-by: Alexander Vasiliev <a.vasylev@404-group.com>
Fixes: ceec4c381681 ("ipvs: convert services to rcu")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
net/netfilter/ipvs/ip_vs_core.c
net/netfilter/ipvs/ip_vs_ctl.c
net/netfilter/ipvs/ip_vs_sched.c