T8454: fix VRF-bind port availability check in service_https

When service https is configured with a listen-address on a VRF
interface, adding or changing the vrf option causes the commit to
fail with "TCP port 443 is used by another service!" even though
no conflict exists.

Root cause: check_port_availability() performs a socket.bind() in
the default network namespace. When the listen-address belongs to
a VRF interface the address is unreachable from the default
namespace, so the bind fails with OSError which is misinterpreted
as "port in use". The secondary check via is_listen_port_bind_service()
also fails because psutil cannot see sockets bound inside a VRF.

Fix: add an optional vrf parameter to check_port_availability().
When set, the test socket is bound to the VRF master device via
SO_BINDTODEVICE before the bind() call, so that the address is
resolved in the correct L3 domain. This is done in-process (no
subprocess) and works for both IPv4 and IPv6.

service_https verify() now passes the configured VRF (if any) to
check_port_availability(). Non-VRF configurations are unaffected.

Add smoke test for HTTPS with listen-address inside a VRF.

Co-authored-by: Christian Breunig <christian@breunig.cc>

1 files changed, 13 insertions, 6 deletions

diff --git a/python/vyos/utils/network.py b/python/vyos/utils/network.py
index 8985224f3..8544b2163 100644
--- a/python/vyos/utils/network.py
+++ b/python/vyos/utils/network.py
@@ -305,7 +305,9 @@ def mac2eui64(mac, prefix=None):
         except:  # pylint: disable=bare-except
             return
 
-def check_port_availability(address: str=None, port: int=0, protocol: str='tcp') -> bool:
+
+def check_port_availability(address: str = None, port: int = 0,
+                            protocol: str = 'tcp', vrf: str = None) -> bool:
     """
     Check if given port is available and not used by any service.
 
@@ -315,7 +317,9 @@ def check_port_availability(address: str=None, port: int=0, protocol: str='tcp')
     Args:
       address: IPv4 or IPv6 address - if None, checks on all interfaces
       port:  TCP/UDP port number.
-
+      vrf:   VRF name to test the bind in - when set, the socket is bound
+             to the VRF master device via SO_BINDTODEVICE so that the
+             port check runs in the correct L3 domain.
 
     Returns:
       False if a port is busy or IP address does not exists
@@ -348,12 +352,15 @@ def check_port_availability(address: str=None, port: int=0, protocol: str='tcp')
         try:
             with socket.socket(family, socktype, proto) as s:
                 s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+                # When a VRF is specified, bind the socket to the VRF master
+                # device so the address is resolved in that VRF's L3 domain.
+                if vrf:
+                    s.setsockopt(socket.SOL_SOCKET, socket.SO_BINDTODEVICE,
+                                 (vrf + '\0').encode())
                 s.bind(sockaddr)
-                # port is free to use
-                return True
+                return True # port is free to use
         except OSError:
-            # port is already in use
-            return False
+            return False # port is already in use
 
     # if we reach this point, no socket was tested and we assume the port is
     # already in use - better safe then sorry