[srp-server] retry other ports when failing to prepareSocket (#9981)

Unicast SRP dataset uses a ephemeral UDP port which could be taken by
another process. Currently SRP server creates the socket at the port
after the server is added into netdata. However, at that moment the
port may not be available on the platform so it may fail to create the
socket and start the server.

This commit adds the logic to restart the enabling process with
another port candidate if SRP server fails to create the socket.

src/core/net/srp_server.cpp

@@ -91,7 +91,7 @@ Server::Server(Instance &aInstance)
     , mOutstandingUpdatesTimer(aInstance)
     , mCompletedUpdateTask(aInstance)
     , mServiceUpdateId(Random::NonCrypto::GetUint32())
-    , mPort(kUdpPortMin)
+    , mPort(kUninitializedPort)
     , mState(kStateDisabled)
     , mAddressMode(kDefaultAddressMode)
     , mAnycastSequenceNumber(0)
@@ -595,7 +595,7 @@ exit:
     }
 }
 
-void Server::SelectPort(void)
+void Server::InitPort(void)
 {
     mPort = kUdpPortMin;
 
@@ -605,24 +605,35 @@ void Server::SelectPort(void)
 
         if (Get<Settings>().Read(info) == kErrorNone)
         {
-            mPort = info.GetPort() + 1;
-            if (mPort < kUdpPortMin || mPort > kUdpPortMax)
-            {
-                mPort = kUdpPortMin;
-            }
+            mPort = info.GetPort();
         }
     }
 #endif
+}
+
+void Server::SelectPort(void)
+{
+    if (mPort == kUninitializedPort)
+    {
+        InitPort();
+    }
+    ++mPort;
+    if (mPort < kUdpPortMin || mPort > kUdpPortMax)
+    {
+        mPort = kUdpPortMin;
+    }
 
     LogInfo("Selected port %u", mPort);
 }
 
 void Server::Start(void)
 {
+    Error error = kErrorNone;
+
     VerifyOrExit(mState == kStateStopped);
 
     mState = kStateRunning;
-    PrepareSocket();
+    SuccessOrExit(error = PrepareSocket());
     LogInfo("Start listening on port %u", mPort);
 
 #if OPENTHREAD_CONFIG_SRP_SERVER_ADVERTISING_PROXY_ENABLE
@@ -630,10 +641,15 @@ void Server::Start(void)
 #endif
 
 exit:
-    return;
+    // Re-enable server to select a new port.
+    if (error != kErrorNone)
+    {
+        Disable();
+        Enable();
+    }
 }
 
-void Server::PrepareSocket(void)
+Error Server::PrepareSocket(void)
 {
     Error error = kErrorNone;
 
@@ -659,8 +675,11 @@ exit:
     if (error != kErrorNone)
     {
         LogCrit("Failed to prepare socket: %s", ErrorToString(error));
+        IgnoreError(mSocket.Close());
         Stop();
     }
+
+    return error;
 }
 
 Ip6::Udp::Socket &Server::GetSocket(void)
@@ -689,7 +708,7 @@ void Server::HandleDnssdServerStateChange(void)
 
     if (mState == kStateRunning)
     {
-        PrepareSocket();
+        IgnoreError(PrepareSocket());
     }
 }
 

src/core/net/srp_server.hpp

@@ -912,6 +912,7 @@ private:
     static constexpr AddressMode kDefaultAddressMode =
         static_cast<AddressMode>(OPENTHREAD_CONFIG_SRP_SERVER_DEFAULT_ADDRESS_MODE);
 
+    static constexpr uint16_t kUninitializedPort      = 0;
     static constexpr uint16_t kAnycastAddressModePort = 53;
 
     // Metadata for a received SRP Update message.
@@ -971,8 +972,9 @@ private:
     void              Disable(void);
     void              Start(void);
     void              Stop(void);
+    void              InitPort(void);
     void              SelectPort(void);
-    void              PrepareSocket(void);
+    Error             PrepareSocket(void);
     Ip6::Udp::Socket &GetSocket(void);
     LinkedList<Host> &GetHosts(void) { return mHosts; }
 

tests/scripts/thread-cert/border_router/test_advertising_proxy.py

@@ -27,7 +27,6 @@
 #  POSSIBILITY OF SUCH DAMAGE.
 #
 import ipaddress
-import logging
 import unittest
 
 import config
@@ -81,6 +80,12 @@ class SingleHostAndService(thread_cert.TestCase):
         host.start(start_radvd=False)
         self.simulator.go(5)
 
+        # Reserve UDP ports to verify that SRP server can skip the unavailable
+        # ports correctly
+        server.reserve_udp_port(53535)
+        server.reserve_udp_port(53536)
+        server.reserve_udp_port(53537)
+
         self.assertEqual(server.srp_server_get_state(), 'disabled')
         server.srp_server_set_enabled(True)
         server.srp_server_set_lease_range(LEASE, LEASE, KEY_LEASE, KEY_LEASE)
@@ -88,6 +93,7 @@ class SingleHostAndService(thread_cert.TestCase):
         self.simulator.go(config.BORDER_ROUTER_STARTUP_DELAY)
         self.assertEqual('leader', server.get_state())
         self.assertEqual(server.srp_server_get_state(), 'running')
+        self.assertNotIn(server.get_srp_server_port(), [53535, 53536, 53537])
 
         client.start()
         self.simulator.go(config.ROUTER_STARTUP_DELAY)

tests/scripts/thread-cert/node.py

@@ -197,6 +197,9 @@ class OtbrDocker:
         self.pexpect.wait()
         self.pexpect.proc.kill()
 
+    def reserve_udp_port(self, port):
+        self.bash(f'socat -u UDP6-LISTEN:{port},bindtodevice=wpan0 - &')
+
     def destroy(self):
         logging.info("Destroying %s", self)
         self._shutdown_docker()

tests/scripts/thread-cert/test_srp_server_reboot_port.py

@@ -94,14 +94,14 @@ class SrpServerRebootPort(thread_cert.TestCase):
 
         #
         # 2. Reboot the server without any service registered. The server should
-        # listen to the same port after the reboot.
+        # switch to a new port after re-enabling.
         #
         old_port = server.get_srp_server_port()
         server.srp_server_set_enabled(False)
         self.simulator.go(5)
         server.srp_server_set_enabled(True)
         self.simulator.go(5)
-        self.assertEqual(old_port, server.get_srp_server_port())
+        self.assertNotEqual(old_port, server.get_srp_server_port())
 
         #
         # 3. Register a service