Dogface2k opened a new pull request, #13766: URL: https://github.com/apache/cloudstack/pull/13766
### Description This PR implements native, route-based Site-to-Site VPN for NSX NAT-mode VPCs. The VPN terminates on the VPC's NSX Tier-1 gateway; the CloudStack virtual router remains outside the IPsec data path. The implementation makes the existing built-in `Vpn/Nsx` service mapping functional and supersedes the capability-removal approach in the closed #13765. It does not change seeded offerings or the database schema. The main lifecycle is provider-dispatched through `Site2SiteVpnServiceProvider` instead of adding another provider-specific branch to `Site2SiteVpnManagerImpl`: - gateway creation validates or acquires a dedicated public IP, records whether NSX owns that allocation, and creates a deterministic Tier-1 `IPSecVpnService` and local endpoint; - connection start validates CloudStack crypto policy, creates NSX IKE/tunnel/DPD profiles, a `RouteBasedIPSecVpnSession`, and a VTI named `default-tunnel-interface`; - each remote CIDR receives a Tier-1 static route and a `NO_SNAT` rule so traffic is not rewritten by the VPC's catch-all source NAT before entering the tunnel; - stop disables the session and removes connection routes/NAT exemptions; restart recreates them; permanent deletion also removes the session and profiles; - gateway and VPC teardown remove VPN objects before the Tier-1 gateway is deleted; - a scheduled status poll maps NSX detailed session status back to CloudStack connection state without allowing a transient query failure to corrupt the last known state. The public endpoint is intentionally separate from the VPC source-NAT address. NSX requires the IPsec local endpoint to differ from the Tier-1 uplink address. Auto-acquired endpoint addresses are released on gateway deletion; operator-supplied addresses are detached but not released. Compatibility and failure handling are explicit: - the new provider SPI methods have defaults, so existing virtual-router and Netris implementations retain their behavior; - legacy gateway rows without an ownership marker retain the historical virtual-router fallback; - persisted provider ownership prevents an offering change from redirecting an existing gateway to a different implementation; - gateway creation refuses to adopt a pre-existing deterministic NSX service and reports ambiguous rollback state instead of silently releasing an address that may already be active; - Tier-1 VPN mutations share a per-gateway lock with Tier-1 teardown; - PSKs are excluded from CloudStack command logging while remaining present on the agent wire payload; - unsupported algorithms, invalid lifetimes, oversized PSKs, missing VPC/gateway rows, and deterministic VTI collisions fail with specific errors. The Tier-0 must redistribute `TIER1_IPSEC_LOCAL_ENDPOINT` routes for the dedicated gateway address to be externally reachable. Related context: #13764 and the superseded closed PR #13765. ### Types of changes - [ ] Breaking change (fix or feature that would cause existing functionality to change) - [x] New feature (non-breaking change which adds functionality) - [ ] Bug fix (non-breaking change which fixes an issue) - [ ] Enhancement (improves an existing feature and functionality) - [ ] Cleanup (Code refactoring and cleanup, that may add test cases) - [ ] Build/CI - [ ] Test (unit or integration test code) ### Feature/Enhancement Scale or Bug Severity #### Feature/Enhancement Scale - [x] Major - [ ] Minor #### Bug Severity - [ ] BLOCKER - [ ] Critical - [ ] Major - [ ] Minor - [ ] Trivial ### Screenshots (if appropriate): Not applicable. ### How Has This Been Tested? The exact upstream-clean commit was built and tested from current `main` at `4f117071c9397b1e4714c8fb61c384883c872842` with Java 17: ```text NSX plugin suite: Tests run: 179, Failures: 0, Errors: 0, Skipped: 0 Full server suite: Tests run: 3625, Failures: 0, Errors: 0, Skipped: 9 Site2SiteVpnManagerImplTest (included in the server total): Tests run: 51, Failures: 0, Errors: 0, Skipped: 0 ``` The changed `api`, `server`, and NSX plugin modules also passed: ```text Checkstyle: 0 violations in all three modules Apache RAT: BUILD SUCCESS for all three modules Packaging: BUILD SUCCESS for all three modules git diff --check: clean ``` The NSX suite covers all six management-server/agent command paths, including a real CloudStack `Request`/Gson round trip proving that the PSK remains on the wire while `@LogLevel(Off)` excludes it from log serialization. An earlier 4.22.1 backport/prototype was exercised against NSX 4.2.4. Live validation exposed five integration defects that are addressed in this branch: virtual-router command leakage, the virtual-router monitor overwriting NSX state, NSX marked-for-deletion races, idle tunnels being removed by periodic DPD, and catch-all SNAT rewriting VPN traffic. The PR is intentionally draft until this exact upstream-clean commit is rerun through the complete live lifecycle: gateway creation, tunnel establishment and bidirectional traffic, stop/start/reset, connection deletion, gateway deletion, VPC teardown, and the Tier-0 endpoint-route redistribution prerequisite. #### How did you try to break this feature and the system with this change? The negative and compatibility coverage exercises the failure boundaries rather than only the happy path: - requested endpoint IPs are rejected if they are outside the VPC, inactive, source-NAT/system addresses, or already used by static NAT, firewall, port-forwarding, or load-balancing rules; - auto-acquired and operator-supplied endpoint addresses have different, verified release behavior; - partial and ambiguous gateway creation failures preserve enough ownership state for safe operator recovery and never adopt an unrelated deterministic NSX service; - connection creation rolls back partial sessions, routes, and NAT exemptions; - changed peer CIDR lists do not strand old routes or `NO_SNAT` rules during deletion; - deterministic VTI address collisions are detected against live Tier-1 sessions and fail closed; - invalid/unsupported cipher, hash, DH group, IKE version, lifetime, and PSK values are rejected before an NSX mutation; - stop, restart, reset, permanent connection deletion, gateway deletion, missing-row cleanup, and full VPC teardown are separately covered; - source-NAT rule demotion preserves the complete original rule and sequence, and the exact sequence is restored when the final VPN exemption is removed; - operator-created VPN services on the Tier-1 are not swept during CloudStack teardown; - provider ownership remains stable if an offering mapping changes after gateway creation; - legacy virtual-router VPN selection and command generation remain covered, while NSX-owned connections are excluded from virtual-router commands and monitoring; - repeated status-query failures alert without forcing a false state transition, and successful `UP`, `DOWN`, degraded, and not-found results map to the intended CloudStack states; - missing VPC or gateway state fails start/stop explicitly instead of reporting a false success, while permanent deletion remains idempotent for cleanup. No generic CloudStack command proxy, broad exception suppression, schema migration, retry of an ambiguous CloudStack mutation, or unrelated UI/CKS/network-offering change is included. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
