IPsec Peers
An IPsec peer is a site-to-site IPsec tunnel that terminates on one of your SecureLink edges and connects to a third-party firewall -- a FortiGate, a Cisco ASA or IOS router, a generic strongSwan host. The two ends negotiate the tunnel with IKE and authenticate with a pre-shared key, so the far end can be equipment SecureLink does not manage and cannot configure.
Use it when a site, a customer or a legacy datacentre has to reach your network but cannot run a SecureLink edge or connector.
How it differs from the other tunnel types
| IPsec Peers | Edge Peering | Connectors | |
|---|---|---|---|
| Far end | Third-party firewall | Another SecureLink edge | SecureLink connector |
| Protocol | IPsec, negotiated by IKE | WireGuard (or E2E IPsec), keyed by the orchestrator | Kernel WireGuard |
| Who configures the far end | The firewall's own administrator | The orchestrator configures both ends | The orchestrator configures both ends |
| Authentication | Pre-shared key plus IKE identity | Keys issued by the orchestrator | Keys issued by the orchestrator |
Because the far end negotiates, every crypto attribute is settable here -- and every attribute has to match what the far end offers. Edge peering has no such exposure: both ends are ours, so no negotiation takes place.
Roles: who dials whom
Most peers are responders. The edge listens, and the far side dials in. This is the normal arrangement when the far side has no fixed public address -- a FortiGate on a dynamic address behind NAT, for example -- and it is why a responder peer is matched by IKE identity rather than by source address: with no address to key on, the identity is the only thing that can select the right pre-shared key.
Set the role to initiator only when the far end has a fixed address or a DDNS name for the edge to dial. Both allows either side to start the negotiation.
A responder peer with no remote gateway address must have a remote identity (FQDN or key ID). Leave the remote ID type as "any" and the far side will dial in, offer a key the edge cannot match, and the tunnel will never authenticate.
Before you start
- The edge must have the IKEv2 App VPN enabled. The peer tunnel shares the App VPN's IKE ingress and its transit interface into VPP. An edge without it refuses the peer with a message naming the missing transit, rather than building a tunnel that passes nothing.
- The edge must not be in AF_PACKET mode. AF_PACKET edges have no transit TAP, and the peer is refused for the same reason.
- UDP 500 and UDP 4500 must reach the edge's public address. 4500 is what carries the traffic once either end is behind NAT.
- The far side must send an IKE identity when it dials in (see above).
- The proposals must match exactly. DES and MD5 are refused outright -- strongSwan ships them disabled, so a proposal naming them cannot negotiate at all.
- Know the far end's DH group. Vendors show a number; SecureLink stores the strongSwan name.
| strongSwan name | Vendor group number |
|---|---|
modp1024 | 2 |
modp1536 | 5 |
modp2048 | 14 |
modp3072 | 15 |
modp4096 | 16 |
ecp256 | 19 |
ecp384 | 20 |
Creating a peer
Go to Network > IPsec Peers, choose the terminating edge and add a peer. The fields are grouped:
Identity
A name for the peer (it also names the generated far-end objects, so keep it short), an optional description, and the enabled flag. A disabled peer is removed from the edge's configuration entirely rather than left loaded and idle.
Peer
The role, the remote gateway address or DDNS name (initiators only), the local identity the edge presents, and the remote identity it expects from the far end. The local identity defaults to the edge's public address; you can also present its WAN address, an FQDN, or a value you type.
Authentication
Pre-shared key. The key is write-only: once stored, the UI will tell you that a key is set but never shows it again. Keys must be at least 16 characters -- a dial-up responder accepts attempts from any source address, so the pre-shared key is the whole of the authentication. The form can generate a 32-byte random key for you.
Crypto
IKE version (2 recommended; IKEv1 main and aggressive mode are supported for older equipment), the IKE and ESP proposals, perfect forward secrecy and its DH group, and the SA lifetimes.
Rather than typing proposals, start from a vendor preset. The presets are served by the orchestrator, so they cannot drift from the validation that has to accept them. They describe what each platform ships by default -- the point is to match what the far end will already be offering, not to express a preference:
| Preset | IKE | ESP | PFS group | Notes |
|---|---|---|---|---|
| FortiGate 7.x | aes256-sha256-modp2048 | aes256-sha256 | modp2048 (14) | Verified against FortiOS 7.4.4 |
| FortiGate 6.x | aes256-sha256-modp1536,aes256-sha1-modp1536 | aes256-sha256 | modp1536 (5) | Older default is DH group 5; includes a SHA-1 fallback offer |
| Cisco ASA | aes256-sha256-modp2048 | aes256-sha256 | modp2048 (14) | ASA often rekeys on volume as well as time |
| Cisco IOS / IOS-XE | aes256-sha256-modp2048 | aes256-sha256 | modp2048 (14) | IOS defaults to a one-hour child SA lifetime |
| Generic strongSwan | aes256gcm16-sha256-modp2048 | aes256gcm16 | modp2048 (14) | AEAD throughout; no separate ESP integrity algorithm |
On an AEAD cipher such as aes256gcm16, a separate integrity algorithm is correct on the IKE proposal (it is the PRF) but meaningless on the ESP proposal. The form rejects it on ESP, because the far end would then report a different effective proposal from the one you configured.
Keepalive
Dead peer detection -- the action to take on a dead peer, the delay and the timeout -- and NAT traversal with its keepalive interval. Leave NAT traversal enabled unless you know neither end is behind NAT; force keeps UDP encapsulation on even when no NAT is detected.
Reachability
Tunnel mode, the local and remote subnets, whether to redistribute the remote subnets into the peering mesh, and whether to source-NAT overlay traffic to a tunnel address. These are explained below.
Then sync the edge
Creating, editing or deleting a peer changes the orchestrator's configuration only. Nothing reaches the edge until the edge is synced, and the UI offers the sync immediately after the change. If a peer is redistributed into the mesh, the other mesh members are marked for sync as well. See Syncing Configuration.
The reachability model
Getting a tunnel to establish is the easy half. Deciding what traffic may use it, and making sure the return path exists, is where tunnels to third-party firewalls usually fail.
Tunnel mode
Policy mode negotiates traffic selectors from the local and remote subnets you list -- the behaviour nearly every firewall expects, and the default. It needs at least one local and one remote subnet. Route mode negotiates a single any-to-any selector and relies on routing instead; use it only when the far end is configured the same way.
Remote subnets
The remote subnets are what the tunnel exists to reach. On the terminating edge they are installed as routes in both the Linux and the VPP forwarding planes, so traffic that arrives at that edge for those prefixes is sent into the tunnel.
Redistribute to mesh
With redistribute to mesh enabled, the edge advertises the peer's remote subnets into the edge peering mesh over BGP, and every other member's WireGuard allowed-IPs are updated to accept them. The effect is that every site in the mesh can reach the peer's subnets through the terminating edge, not just the edge the tunnel lands on.
Both halves have to be in place. BGP advertises the route as soon as the terminating edge is synced, but until the neighbours are synced their cryptokey routing still drops the packets -- the route is there and the traffic is not. This is why a redistributed peer marks the other mesh members for sync too; let those syncs complete.
The far side needs a return route
The far end must route your overlay prefixes back through the tunnel. Its selectors must also agree with yours: the far side's local subnets are your remote subnets and vice versa. A mismatch is accepted at phase 1 and then passes no traffic, which is the single most common symptom reported against this feature.
SNAT to tunnel address
Source-NAT translates all overlay traffic to one /32 -- the tunnel address -- before it enters the tunnel. The far end then needs exactly one return route and one traffic selector, no matter how many sites or clients are behind the mesh.
Recommended for first bring-up, and for legacy firewalls where every additional selector is a change request. The cost is source attribution: on the far side every packet appears to come from the tunnel address, so per-site or per-user identification is lost there.
If you enable source-NAT without choosing a tunnel address, one is allocated for you from the tenant's connector /24. Source-NAT without an address is refused by both the orchestrator and the edge rather than installed half-configured.
App VPN (road-warrior) clients do not yet reach IPsec peer subnets. Traffic from mesh sites does.
Far-end configuration
A peer's Far-end config action generates a paste-ready CLI block for the other device from the configuration you have already stored, for FortiOS, Cisco ASA or Cisco IOS. The local and remote selectors are inverted for the far side, the proposals are translated into that vendor's spelling, and the block can be copied or downloaded.
This exists because the two sides comparing notes by hand is what produces unnegotiable tunnels: the only symptom on the wire is NO_PROPOSAL_CHOSEN, which names no attribute and blames neither side. Generating the far end's half from ours removes the opportunity for a mismatch.
Alongside the block the generator lists:
- Warnings -- things it cannot know but that will break the tunnel if wrong, such as a missing remote identity, a placeholder WAN interface name, or more subnet pairs than the generated phase 2 entries cover.
- Manual steps -- what the block deliberately does not do: firewall policies, routing, and NAT exemption.
The pre-shared key is a <PRE-SHARED-KEY> placeholder unless you explicitly ask for it to be included. Including it requires update permission and is recorded in the audit log, so an ordinary read of the block cannot leak the key into a screenshot or a browser history entry. Supply the key out of band where you can.
SecureLink never pushes configuration to third-party equipment. These blocks are generated for a human to review and apply.
FortiGate walkthrough
The values below were verified against a FortiGate on FortiOS 7.4, with the edge as the responder and the FortiGate dialling in.
Phase 1:
config vpn ipsec phase1-interface
edit "vsnplus"
set interface "port1"
set ike-version 2
set proposal aes256-sha256
set dhgrp 14
set remote-gw <edge public IP>
set localid-type fqdn
set localid "<the identity the VSN+ side matches on>"
set peertype any
set psksecret <pre-shared key>
set nattraversal enable
set dpd on-idle
next
end
Phase 2:
config vpn ipsec phase2-interface
edit "vsnplus_p2"
set phase1name "vsnplus"
set proposal aes256-sha256
set pfs enable
set dhgrp 14
set auto-negotiate enable
next
end
set localid is the identity the VSN+ side matches the peer on -- it must equal the remote identity configured on the peer. set auto-negotiate enable keeps the child SA up instead of waiting for traffic, which is the difference between a tunnel that is there when you look at it and one that only appears under load.
Then, on the FortiGate:
- Firewall policies in both directions between the LAN interface and the tunnel interface, permitting the subnets concerned. Without them the tunnel establishes and passes nothing. This is the step most often missed.
- A static route for each overlay prefix via the tunnel interface.
- A blackhole route for the same prefixes at distance 254, so traffic is dropped rather than following the default route while the tunnel is down. This is the pattern the FortiOS VPN wizard itself uses.
Classic mistakes
| Mistake | What you see |
|---|---|
| Accepting the FortiOS wizard's default DES/MD5 proposals | NO_PROPOSAL_CHOSEN. DES and MD5 are refused by SecureLink and disabled in strongSwan |
| Leaving IKE version at the IKEv1 default | A dial-up peer authenticating with a pre-shared key cannot work in IKEv1 main mode. Use IKEv2 |
No set localid | The far end dials in with nothing to identify itself; the pre-shared key cannot be selected and the tunnel never authenticates |
| DH group 5 against a group 14 peer | NO_PROPOSAL_CHOSEN again -- the group is part of the proposal |
| Remote selector pointing at the management LAN | The tunnel establishes and passes no traffic. The remote selector must be the overlay prefix |
Peer status
Each peer reports its own state, refreshed on the edge's inform cycle:
| Field | Meaning |
|---|---|
| Status | Established, Connecting, Down, Failed, Not reported yet, or Disabled |
| IKE / child SA state | The raw states from the edge's IKE daemon, for example ESTABLISHED and INSTALLED |
| Last established | When the tunnel last came up |
| Remote address | The address the peer actually dialled in from -- useful when the far end is on a dynamic address |
| Bytes in / out | Counters from the child SA |
| Last report | When the edge last said anything about this peer |
Three states are easy to confuse:
- Not reported yet means the peer is enabled but the edge has not said anything about it. This is the normal state between creating a peer and syncing the edge. It is not a fault.
- Down means the edge reported the tunnel as down. That is a fault.
- A stale report -- shown separately, as the age of the last report -- means the edge has not reported this peer recently. The status badge keeps saying whatever the edge last claimed, because an established tunnel last seen twenty minutes ago is neither established nor down, and asserting either would be a guess. Check whether the edge itself is still online.
Redundant tunnels to one far end
A third-party device can terminate more than one tunnel, on more than one of your edges, so that traffic to it survives the loss of a single edge or a single uplink. The peer list presents that as it is: one card per far end, with every tunnel to it listed underneath. A device with a single tunnel is simply a site with one tunnel.
Each tunnel is still a peer in its own right -- it has its own edge, its own crypto, its own status -- and clicking it opens the usual detail panel, where Edit, Enable/Disable, Far-end config, Sync and Delete behave exactly as before.
How tunnels are matched to a site
Tunnels are grouped by whichever of these identifies the far end first:
- The peer ID (the IKE identity the far end presents), when one is configured and the ID type is not any. This is the most reliable, because it still groups the tunnels correctly when they leave your network through different ISPs.
- The remote address the edge observed, for a dial-up peer with no identity configured.
- The remote gateway you configured.
A peer with none of these -- no identity, nothing observed yet, no gateway -- is shown as its own site. Two such peers are never merged. The card says which field it matched on, so you can see why two tunnels were grouped, or why two you expected to be grouped were not. If a redundant pair is not grouping, give both tunnels the same peer ID.
Aggregate status
The heading on a multi-tunnel site reports the pair, not a tunnel:
| Status | Meaning |
|---|---|
| Redundant | Two or more tunnels are enabled and established. Losing one leaves the far end reachable |
| Degraded | Exactly one tunnel is up. The far end is reachable, but there is nothing left to fail over to |
| Down | No tunnel is up. The far end is unreachable |
A tunnel only counts towards these if it is enabled as well as established: a disabled tunnel is not a standby, it is a tunnel configured not to run.
A single-tunnel site makes no redundancy claim at all. It shows that tunnel's own status, because calling one working tunnel "degraded" would put a warning on every ordinary peer.
Degraded is the status worth alerting on. A degraded pair passes traffic and looks healthy from the far end, so without the aggregate it stays unnoticed until the surviving tunnel also drops -- which is the whole reason the far end was given two tunnels.
The redistribution warning
On a site with two or more tunnels, any tunnel with redistribute to mesh off is flagged: this tunnel gives no mesh-level redundancy.
Both tunnels establishing is not enough. With redistribution off, that tunnel's remote subnets are reachable only from the edge it terminates on, so when its sibling fails the mesh has no route to the far end through it. The tunnel is up, reports established, and carries nothing. Turn redistribution on for every tunnel in a redundant pair, and let the resulting neighbour syncs complete.
Troubleshooting
The tunnel is established but no traffic passes
Almost always selectors or routing, not crypto -- crypto failures prevent establishment. Check that your local subnets are the far end's remote subnets and vice versa, that the far end has a return route for your overlay prefixes, and that its firewall policies permit the traffic in both directions. If the far end NATs its internet traffic, it also needs a NAT exemption for the tunnelled subnets, or the traffic is translated before it reaches the tunnel and never matches the selectors.
Enabling SNAT to tunnel address resolves most of these at once, by reducing the far end's requirement to one route and one selector.
NO_PROPOSAL_CHOSEN
An attribute does not match, and the message does not say which. Compare the IKE proposal, the ESP proposal, the DH group and the PFS group against the far end, remembering that vendors show group numbers and SecureLink stores names. Start from the vendor preset for the far end's platform, and generate the far-end block rather than transcribing values.
The peer never dials in
Confirm the far end sends an IKE identity and that it matches the remote identity on the peer; that UDP 500 and 4500 reach the edge's public address; and that NAT traversal is enabled on both ends. Check that the edge has the IKEv2 App VPN enabled -- without it the peer configuration is refused.
Other mesh edges cannot reach the peer's subnets
Check that redistribute to mesh is enabled on the peer, then confirm the other mesh members have been synced. Until a neighbour is synced it has the BGP route but not the matching WireGuard allowed-IPs, and it drops the packets.