1 files changed, 46 insertions, 3 deletions

diff --git a/src/btfrv.cpp b/src/btfrv.cpp
index 133eb89..8ab735c 100644
--- a/src/btfrv.cpp
+++ b/src/btfrv.cpp
@@ -40,6 +40,8 @@ typedef int SOCKET;
 #include <thread>
 #include <mutex>
 #include <atomic>
+#include <chrono>
+#include <system_error>
 
 namespace btf
 {
@@ -209,8 +211,28 @@ static void HandleIncoming(SOCKET s)
         bool a = WriteN(svc, &ok, 1);
         bool b = WriteN(s, &ok, 1);
         if (!a || !b) { CLOSESOCK(svc); CLOSESOCK(s); return; }
-        std::thread(Forward, s, svc).detach();
-        std::thread(Forward, svc, s).detach();
+        // One thread pumps s -> svc; this thread pumps svc -> s. Spawning a
+        // thread per direction left no clean way to fail the second spawn:
+        // std::thread throws std::system_error when pthread_create is refused
+        // (NPROC limit, container pids cgroup, memory pressure), the first
+        // pump was already detached, and closing its sockets from here would
+        // race it -- so nothing caught the throw and it reached
+        // std::terminate, killing the relay over one pair it could not serve.
+        // Running one direction on this thread leaves exactly one spawn to
+        // guard, and a failure here has nothing running yet to race.
+        try
+        {
+            std::thread(Forward, s, svc).detach();
+        }
+        catch (const std::system_error& e)
+        {
+            fprintf(stderr, "rendezvous relay: cannot spawn forward thread (%s), pair dropped\n",
+                    e.what());
+            CLOSESOCK(svc);
+            CLOSESOCK(s);
+            return;
+        }
+        Forward(svc, s);
     }
     else
     {
@@ -236,6 +258,7 @@ bool RvRelayRun(unsigned short port)
     if (bind(g_listen, (struct sockaddr*)&addr, sizeof(addr)) != 0) { CLOSESOCK(g_listen); return false; }
     if (listen(g_listen, 16) != 0) { CLOSESOCK(g_listen); return false; }
 
+    int nSpawnFails = 0;
     while (!g_stop)
     {
         struct sockaddr_in cli;
@@ -246,7 +269,27 @@ bool RvRelayRun(unsigned short port)
             if (g_stop) break;
             continue;
         }
-        std::thread(HandleIncoming, s).detach();
+        try
+        {
+            std::thread(HandleIncoming, s).detach();
+            nSpawnFails = 0;
+        }
+        catch (const std::system_error& e)
+        {
+            // A refused pthread_create is backpressure, not a reason to die:
+            // the condition behind it (NPROC, pids cgroup, memory) does not
+            // clear in milliseconds, and the connection count feeding it is
+            // attacker-influenced. Drop this one peer, stop accepting for a
+            // moment -- longer each time, capped -- and keep serving the
+            // connections already established.
+            fprintf(stderr, "rendezvous relay: cannot spawn connection thread (%s), connection dropped\n",
+                    e.what());
+            CLOSESOCK(s);
+            nSpawnFails++;
+            int nWaitMs = nSpawnFails < 5 ? nSpawnFails * 1000 : 5000;
+            for (int i = 0; i < nWaitMs && !g_stop; i += 100)
+                std::this_thread::sleep_for(std::chrono::milliseconds(100));
+        }
     }
     CLOSESOCK(g_listen);
     g_listen = INVALID_SOCKET;