Implement per-issuer lock to prevent thundering herd on new issuers

Co-authored-by: bbockelm <1093447+bbockelm@users.noreply.github.com>

Author: Copilot

src/scitokens_internal.cpp

@@ -33,6 +33,44 @@ CurlRaii myCurl;
 
 std::mutex key_refresh_mutex;
 
+// Per-issuer mutex map for preventing thundering herd on new issuers
+std::mutex issuer_mutex_map_lock;
+std::unordered_map<std::string, std::shared_ptr<std::mutex>> issuer_mutexes;
+constexpr size_t MAX_ISSUER_MUTEXES = 1000;
+
+// Get or create a mutex for a specific issuer
+std::shared_ptr<std::mutex> get_issuer_mutex(const std::string &issuer) {
+    std::lock_guard<std::mutex> guard(issuer_mutex_map_lock);
+    
+    auto it = issuer_mutexes.find(issuer);
+    if (it != issuer_mutexes.end()) {
+        return it->second;
+    }
+    
+    // Prevent resource exhaustion: limit the number of cached mutexes
+    if (issuer_mutexes.size() >= MAX_ISSUER_MUTEXES) {
+        // Remove mutexes that are no longer in use
+        for (auto iter = issuer_mutexes.begin(); iter != issuer_mutexes.end(); ) {
+            if (iter->second.use_count() == 1) {
+                // Only we hold a reference, safe to remove
+                iter = issuer_mutexes.erase(iter);
+            } else {
+                ++iter;
+            }
+        }
+        
+        // If still at capacity, clear the oldest entries
+        // Note: In a production system, we might use an LRU cache
+        if (issuer_mutexes.size() >= MAX_ISSUER_MUTEXES) {
+            issuer_mutexes.clear();
+        }
+    }
+    
+    auto mutex_ptr = std::make_shared<std::mutex>();
+    issuer_mutexes[issuer] = mutex_ptr;
+    return mutex_ptr;
+}
+
 } // namespace
 
 namespace scitokens {
@@ -829,8 +867,22 @@ Validator::get_public_key_pem(const std::string &issuer, const std::string &kid,
         }
     } else {
         // No keys in the DB, or they are expired, so get them from the web.
-        result = get_public_keys_from_web(
-            issuer, internal::SimpleCurlGet::default_timeout);
+        // Use per-issuer lock to prevent thundering herd for new issuers
+        auto issuer_mutex = get_issuer_mutex(issuer);
+        std::lock_guard<std::mutex> lock(*issuer_mutex);
+        
+        // Check again if keys are now in DB (another thread may have fetched them)
+        if (get_public_keys_from_db(issuer, now, result->m_keys,
+                                    result->m_next_update)) {
+            // Keys are now available, use them
+            result->m_continue_fetch = false;
+            result->m_do_store = false;
+            result->m_done = true;
+        } else {
+            // Still no keys, fetch them from the web
+            result = get_public_keys_from_web(
+                issuer, internal::SimpleCurlGet::default_timeout);
+        }
     }
     result->m_issuer = issuer;
     result->m_kid = kid;

test/main.cpp

@@ -941,6 +941,42 @@ TEST_F(IssuerSecurityTest, SpecialCharacterIssuer) {
     EXPECT_NE(error_message.find("\""), std::string::npos);
 }
 
+// Test for thundering herd prevention with per-issuer locks
+TEST_F(IssuerSecurityTest, ThunderingHerdPrevention) {
+    char *err_msg = nullptr;
+    
+    // Create tokens for a new issuer and pre-populate the cache
+    std::string test_issuer = "https://thundering-herd-test.example.org/gtest";
+    
+    auto rv = scitoken_set_claim_string(m_token.get(), "iss",
+                                        test_issuer.c_str(), &err_msg);
+    ASSERT_TRUE(rv == 0) << err_msg;
+    
+    // Store public key for this issuer in the cache
+    rv = scitoken_store_public_ec_key(test_issuer.c_str(), "1", ec_public,
+                                      &err_msg);
+    ASSERT_TRUE(rv == 0) << err_msg;
+    
+    char *token_value = nullptr;
+    rv = scitoken_serialize(m_token.get(), &token_value, &err_msg);
+    ASSERT_TRUE(rv == 0) << err_msg;
+    std::unique_ptr<char, decltype(&free)> token_value_ptr(token_value, free);
+    
+    // Successfully deserialize - the per-issuer lock should prevent thundering herd
+    // Since we pre-populated the cache, this should succeed without network access
+    rv = scitoken_deserialize_v2(token_value, m_read_token.get(), nullptr,
+                                 &err_msg);
+    ASSERT_TRUE(rv == 0) << err_msg;
+    
+    // Verify the issuer claim
+    char *value;
+    rv = scitoken_get_claim_string(m_read_token.get(), "iss", &value, &err_msg);
+    ASSERT_TRUE(rv == 0) << err_msg;
+    ASSERT_TRUE(value != nullptr);
+    std::unique_ptr<char, decltype(&free)> value_ptr(value, free);
+    EXPECT_STREQ(value, test_issuer.c_str());
+}
+
 int main(int argc, char **argv) {
     ::testing::InitGoogleTest(&argc, argv);
     return RUN_ALL_TESTS();