Regression #14059
closed
Old states are still used after a route change
0%
Description
- The relevant child SA has not yet been installed.
- The tunnel is currently disabled.
- The IPsec service is stopped.
When this happens, states are created on the WAN. In the following example, lagg0.6 is the WAN. The top two states were created when there was an outage with the tunnel, and the bottom two states' age closely match the firewall's uptime.
all udp 10.102.242.3:5060 <- 10.103.242.116:5060 NO_TRAFFIC:SINGLE age 05:29:11, expires in 00:02:24, 2899:0 pkts, 1096243:0 bytes, rule 128 id: b198066400000000 creatorid: d0dc7a29 gateway: 0.0.0.0 origif: lagg0.10 all udp 198.51.100.234:21991 (10.103.242.116:5060) -> 10.102.242.3:5060 SINGLE:NO_TRAFFIC age 05:29:11, expires in 00:02:24, 2852:0 pkts, 1078191:0 bytes, rule 74 id: b298066400000000 creatorid: d0dc7a29 gateway: 198.51.100.233 origif: lagg0.6 all udp 10.102.242.3:5060 <- 10.103.242.117:5060 NO_TRAFFIC:SINGLE age 21:06:18, expires in 00:02:28, 16580:0 pkts, 3820099:0 bytes, rule 128 id: dd08f46300000000 creatorid: d0dc7a29 gateway: 0.0.0.0 origif: lagg0.10 all udp 198.51.100.234:57200 (10.103.242.117:5060) -> 10.102.242.3:5060 SINGLE:NO_TRAFFIC age 21:06:18, expires in 00:02:28, 16519:0 pkts, 3806685:0 bytes, rule 74 id: de08f46300000000 creatorid: d0dc7a29 gateway: 198.51.100.233 origif: lagg0.6
In 22.05 once the relevant child SA is installed, the firewall correctly replaces the states and sends the traffic through the tunnel.
In 23.01, the states remain active on WAN regardless of the child SA being installed afterwards. While the bad states exist and the child SA is installed, running a packet capture on enc0 shows the traffic makes it into the interface, but is not actually sent through the tunnel. To work around the issue, the following floating rule may be created:
Action: Block | Quick: Checked | Interface: <select all WAN interfaces> | Direction: out | Address Family: IPv4 | Protocol: Any | Source: Any | Destination: 10.0.0.0/8 (or the relevant subnet)
With this rule in place, states are prevented from being created on the WAN, and hence traffic flows correctly once the child SA is installed.
Updated by Chris Linstruth over 1 year ago
I believe the change that makes this a problem where it wasn't before is that states are now created on interface "all".
Updated by Jim Pingle over 1 year ago
That sort of leakage has always been a problem without rules to ensure it doesn't escape. The old states sort of masked it but it really only worked by luck (and not for all types of traffic).
IMO there may not be anything actionable here other than encouraging people to use proper rules to prevent leaks.
We could maybe come up with some automatic rules to disallow egress of traffic matching P2 selectors out WAN but that could get tricky fast.
Updated by Marcos M over 1 year ago
This state behavior can affect other VPN types, not only IPsec.
Edit: e.g. https://www.reddit.com/r/PFSENSE/comments/187hhkz/states_issue_with_3cx_and_yealink_phones_over/
Updated by Marcos M 12 months ago
- Has duplicate Feature #14630: FRR script hook for clearing states on routing changes added
Updated by Marcos M 12 months ago
- Status changed from New to Not a Bug
Related: https://redmine.pfsense.org/issues/14630, https://redmine.pfsense.org/issues/15140
It seems like some sort of route change event queue would need to be maintained (either in userspace or in pf directly) which would trigger the clearing of related states. Even so, there there may be other issues that present themselves (e.g. unintended behavior, race conditions); additionally some setups may want prefer one behavior over the other. Correcting the behavior with rules seems preferable to doing all of that.