Fix service name generation to enforce RFC 6763 length limit

Truncate mDNS service instance names exceeding 63 bytes with a
hash suffix to ensure RFC 6763 compliance, since Avahi does not
automatically truncate like mDNSResponder. Also use underscore
rather than colon as the host/port separator, and declare
service_name in the header. Add unit tests for service_name.

Made-with: Cursor

Author: garethsb

Development/cmake/NmosCppTest.cmake

@@ -55,6 +55,7 @@ set(NMOS_CPP_TEST_NMOS_TEST_SOURCES
     nmos/test/json_validator_test.cpp
     nmos/test/jwt_generator_test.cpp
     nmos/test/jwt_validation_test.cpp
+    nmos/test/mdns_test.cpp
     nmos/test/paging_utils_test.cpp
     nmos/test/query_api_test.cpp
     nmos/test/sdp_test_utils.cpp

Development/nmos/mdns.cpp

@@ -1,5 +1,7 @@
 #include "nmos/mdns.h"
 
+#include <functional>
+#include <iomanip>
 #include <boost/algorithm/string/erase.hpp>
 #include <boost/algorithm/string/join.hpp>
 #include <boost/algorithm/string/predicate.hpp>
@@ -330,6 +332,13 @@ namespace nmos
                 return utility::us2s(nmos::fields::service_name_prefix(settings)) + "_" + service_api(service);
             }
 
+            inline std::string hash_string(const std::string& s)
+            {
+                std::ostringstream os;
+                os << std::hex << std::setfill('0') << std::setw(8) << (std::hash<std::string>{}(s) & 0xFFFFFFFF);
+                return os.str();
+            }
+
             inline std::set<nmos::api_version> service_versions(const nmos::service_type& service, const nmos::settings& settings)
             {
                 // the System API is defined by IS-09 (having been originally specified in JT-NM TR-1001-1:2018 Annex A)
@@ -341,11 +350,33 @@ namespace nmos
             }
         }
 
+        // generate a stable unique name for the specified service, based on the service type, host and port
         std::string service_name(const nmos::service_type& service, const nmos::settings& settings)
         {
-            // this just serves as an example of a possible service naming strategy
-            // replacing '.' with '-', since although '.' is legal in service names, some DNS-SD implementations just don't like it
-            return boost::algorithm::replace_all_copy(details::service_base_name(service, settings) + "_" + utility::us2s(nmos::get_host(settings)) + ":" + utility::us2s(utility::ostringstreamed(details::service_port(service, settings))), ".", "-");
+            // replace '.' with '-', since although '.' is legal in service names, some DNS-SD implementations just don't like it
+            const auto name = boost::algorithm::replace_all_copy(details::service_base_name(service, settings) + "_" + utility::us2s(nmos::get_host(settings)) + "_" + utility::us2s(utility::ostringstreamed(details::service_port(service, settings))), ".", "-");
+
+            // RFC 6763 Section 4.1.1 specifies that instance names must not exceed 63 bytes
+            // see https://tools.ietf.org/html/rfc6763#section-4.1.1
+            const size_t max_length = 63;
+            if (name.size() <= max_length) return name;
+
+            // truncate over-long names, because Avahi does not automatically truncate like mDNSResponder does,
+            // and include a unique suffix based on the full name, to reduce collisions
+
+            // RFC 6763 explains that DNS-SD names are intended to be user-visible, "short and descriptive",
+            // and avoid "incomprehensible hexadecimal strings"; because this is likely to result in collisions,
+            // RFC 6762 defines a conflict resolution mechanism designed to keep names human-readable,
+            // implemented by Avahi ("<name> #2") and mDNSResponder ("<name> (2)"), and recommends the newly
+            // chosen name is recorded in persistent storage so that the service will use the same name when it
+            // restarts...
+            // because NMOS service names are unlikely to be presented to the user, maximizing the likelihood
+            // of a unique name without the need for persistent storage is higher priority than no hex strings!
+
+            const auto suffix = details::hash_string(name);
+            const auto max_prefix_length = max_length - 1 - suffix.size();
+            const auto prefix_length = name.find_last_not_of("-_", max_prefix_length - 1) + 1;
+            return name.substr(0, prefix_length) + "-" + suffix;
         }
 
         // helper function for registering addresses when the host name is explicitly configured

Development/nmos/mdns.h

@@ -148,6 +148,9 @@ namespace nmos
 
     namespace experimental
     {
+        // generate a stable unique name for the specified service, based on the service type, host and port
+        std::string service_name(const nmos::service_type& service, const nmos::settings& settings);
+
         // helper function for registering addresses when the host name is explicitly configured
         void register_addresses(mdns::service_advertiser& advertiser, const nmos::settings& settings);
 

Development/nmos/test/mdns_test.cpp

@@ -0,0 +1,198 @@
+// The first "test" is of course whether the header compiles standalone
+#include "nmos/mdns.h"
+
+#include "bst/test/test.h"
+#include "cpprest/basic_utils.h"
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameFormat)
+{
+    using web::json::value_of;
+
+    // with default settings, service_name should be "<prefix>_<api>_<host>_<port>"
+    // with dots replaced by dashes
+    const nmos::settings settings;
+
+    BST_CHECK_EQUAL("nmos-cpp_node_127-0-0-1_3212", nmos::experimental::service_name(nmos::service_types::node, settings));
+    BST_CHECK_EQUAL("nmos-cpp_query_127-0-0-1_3211", nmos::experimental::service_name(nmos::service_types::query, settings));
+    BST_CHECK_EQUAL("nmos-cpp_registration_127-0-0-1_3210", nmos::experimental::service_name(nmos::service_types::registration, settings));
+    BST_CHECK_EQUAL("nmos-cpp_system_127-0-0-1_10641", nmos::experimental::service_name(nmos::service_types::system, settings));
+    BST_CHECK_EQUAL("nmos-cpp_auth_127-0-0-1_443", nmos::experimental::service_name(nmos::service_types::authorization, settings));
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameCustomPrefix)
+{
+    using web::json::value_of;
+
+    // dots in prefix should be replaced with dashes
+    const nmos::settings settings = value_of({
+        { nmos::fields::service_name_prefix, U("my.prefix") }
+    });
+
+    BST_CHECK_EQUAL("my-prefix_node_127-0-0-1_3212", nmos::experimental::service_name(nmos::service_types::node, settings));
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameDotReplacement)
+{
+    using web::json::value_of;
+
+    // dots in host name should be replaced with dashes
+    const nmos::settings settings = value_of({
+        { nmos::fields::host_name, U("my-server.example.com") },
+        { nmos::experimental::fields::href_mode, 1 }
+    });
+
+    BST_CHECK_EQUAL("nmos-cpp_node_my-server-example-com_3212", nmos::experimental::service_name(nmos::service_types::node, settings));
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameCustomPort)
+{
+    using web::json::value_of;
+
+    const nmos::settings settings = value_of({
+        { nmos::fields::node_port, 8080 }
+    });
+
+    BST_CHECK_EQUAL("nmos-cpp_node_127-0-0-1_8080", nmos::experimental::service_name(nmos::service_types::node, settings));
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameMaxLength)
+{
+    using web::json::value_of;
+
+    // RFC 6763 Section 4.1.1 specifies instance names must not exceed 63 bytes
+    const size_t max_length = 63;
+
+    // a name that fits within 63 bytes should pass through unchanged
+    {
+        const nmos::settings settings;
+        const auto name = nmos::experimental::service_name(nmos::service_types::node, settings);
+        BST_REQUIRE(name.size() <= max_length);
+        // no hash suffix
+        BST_CHECK_EQUAL("nmos-cpp_node_127-0-0-1_3212", name);
+    }
+
+    // a name that would exceed 63 bytes should be truncated with a hash suffix
+    {
+        const nmos::settings settings = value_of({
+            { nmos::fields::host_name, U("a-host-name-that-is-itself-valid-but-already-63-characters-long.example.com") },
+            { nmos::experimental::fields::href_mode, 1 }
+        });
+
+        const auto name = nmos::experimental::service_name(nmos::service_types::registration, settings);
+        BST_REQUIRE(name.size() <= max_length);
+    }
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameTruncationStable)
+{
+    using web::json::value_of;
+
+    // the truncated name should be stable (consistent) even across runs of the same program
+    // but that's hard to test (and actually, std::hash is only required to produce the same
+    // result for the same input within a single execution of a program; this allows salted
+    // hashes that prevent collision denial-of-service attacks, which would be unfortunate
+    // for this use case, but common standard library implementations don't do that...)
+
+    const nmos::settings settings = value_of({
+        { nmos::fields::host_name, U("a-host-name-that-is-itself-valid-but-already-63-characters-long.example.com") },
+        { nmos::experimental::fields::href_mode, 1 }
+    });
+
+    const auto name1 = nmos::experimental::service_name(nmos::service_types::registration, settings);
+    const auto name2 = nmos::experimental::service_name(nmos::service_types::registration, settings);
+    BST_CHECK_EQUAL(name1, name2);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameTruncationSuffix)
+{
+    using web::json::value_of;
+
+    const size_t max_length = 63;
+
+    // the truncated name should end with a dash followed by an 8-character hex hash
+    const nmos::settings settings = value_of({
+        { nmos::fields::host_name, U("a-host-name-that-is-itself-valid-but-already-63-characters-long.example.com") },
+        { nmos::experimental::fields::href_mode, 1 }
+    });
+
+    const auto name = nmos::experimental::service_name(nmos::service_types::registration, settings);
+    BST_REQUIRE(name.size() <= max_length);
+
+    const auto last_dash = name.rfind('-');
+    BST_REQUIRE(std::string::npos != last_dash);
+
+    // the suffix after the last dash should be exactly 8 hex characters
+    const auto suffix = name.substr(last_dash + 1);
+    BST_CHECK_EQUAL(8, suffix.size());
+    BST_CHECK(suffix.end() == std::find_if(suffix.begin(), suffix.end(), [](char c)
+    {
+        return !std::isxdigit(static_cast<unsigned char>(c));
+    }));
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameExactly63Bytes)
+{
+    using web::json::value_of;
+
+    // a name that is exactly 63 bytes should not be truncated
+    // "nmos-cpp_node_" = 14 chars, "_3212" = 5 chars, so host needs to be 63 - 14 - 5 = 44 chars
+    // after dot replacement: e.g. a host name with no dots that is 44 chars
+    const std::string host_44(44, 'x');
+    const nmos::settings settings = value_of({
+        { nmos::fields::host_name, utility::s2us(host_44) },
+        { nmos::experimental::fields::href_mode, 1 }
+    });
+
+    const auto name = nmos::experimental::service_name(nmos::service_types::node, settings);
+    BST_CHECK_EQUAL(63, name.size());
+    BST_CHECK_EQUAL("nmos-cpp_node_" + host_44 + "_3212", name);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameOneOver63Bytes)
+{
+    using web::json::value_of;
+
+    // a name that is 64 bytes should be truncated
+    const std::string host_45(45, 'x');
+    const nmos::settings settings = value_of({
+        { nmos::fields::host_name, utility::s2us(host_45) },
+        { nmos::experimental::fields::href_mode, 1 }
+    });
+
+    const auto name = nmos::experimental::service_name(nmos::service_types::node, settings);
+    BST_REQUIRE(name.size() <= 63);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////
+BST_TEST_CASE(testServiceNameTruncationDistinct)
+{
+    using web::json::value_of;
+
+    // even when truncated, names that differ only in the port (which is past the truncation point)
+    // should produce distinct names due to the hash suffix being based on the full untruncated name
+    const auto make_settings = [](int port)
+    {
+        return value_of({
+            { nmos::fields::host_name, U("a-host-name-that-is-itself-valid-but-already-63-characters-long.example.com") },
+            { nmos::experimental::fields::href_mode, 1 },
+            { nmos::fields::node_port, port }
+        });
+    };
+
+    const auto name1 = nmos::experimental::service_name(nmos::service_types::node, make_settings(3212));
+    const auto name2 = nmos::experimental::service_name(nmos::service_types::node, make_settings(3213));
+    const auto name3 = nmos::experimental::service_name(nmos::service_types::node, make_settings(3214));
+
+    BST_CHECK(name1 != name2);
+    BST_CHECK(name1 != name3);
+    BST_CHECK(name2 != name3);
+}