diff --git a/LICENSE b/LICENSE
index 17889aaec..e0dd48d4c 100644
--- a/LICENSE
+++ b/LICENSE
@@ -1,5 +1,5 @@
 Copyright (c) 2018 Jeff Becker
-Win32 port and portions copyright �2018 Rick V.
+Windows NT port and portions copyright �2018 Rick V.
 
 This software is provided 'as-is', without any express or implied
 warranty. In no event will the authors be held liable for any damages
diff --git a/crypto/libntrup/src/avx/mult.c b/crypto/libntrup/src/avx/mult.c
index e01e1cd33..cb2ef059a 100644
--- a/crypto/libntrup/src/avx/mult.c
+++ b/crypto/libntrup/src/avx/mult.c
@@ -5,17 +5,48 @@
 #include "rq.h"
 #include "r3.h"
 
+// 32-bit hosts: use compiler builtins and let compiler
+// perform register allocation and/or spilling to core
+//
+// Confirmed working 3/10/18 -despair
+// Less than 5% performance hit,
+// all in register loads/stores to/from core
+#ifndef __amd64__
 #define MULSTEP_gcc(j,h0,h1,h2,h3,h4) \
   gj = g[j]; \
   h0 += f0 * gj; \
-  _mm256_storeu_ps(&h[i + j],h0); \
+  _mm256_storeu_ps((float*)&h[i + j],h0); \
   h1 += f1 * gj; \
   h2 += f2 * gj; \
   h3 += f3 * gj; \
   h4 += f4 * gj; \
-  h0 = _mm256_loadu_ps(&h[i + j + 5]); \
+  h0 = _mm256_loadu_ps((float*)&h[i + j + 5]); \
   h0 += f5 * gj;
 
+#define MULSTEP MULSTEP_gcc
+
+#define MULSTEP_noload(j,h0,h1,h2,h3,h4) \
+  gj = g[j]; \
+  h0 += gj*f0; \
+  _mm256_storeu_ps((float*)&h[i+j], h0); \
+  h1 += gj*f1; \
+  h2 += gj*f2; \
+  h3 += gj*f3; \
+  h4 += gj*f4; \
+  h0 = gj* f5;
+
+#define MULSTEP_fromzero(j,h0,h1,h2,h3,h4) \
+  gj = g[j]; \
+  h0 = gj*f0; \
+  _mm256_storeu_ps((float*)&h[i+j], h0); \
+  h1 = gj*f1; \
+  h2 = gj*f2; \
+  h3 = gj*f3; \
+  h4 = gj*f4; \
+  h0 = gj*f5;
+#else
+// 64-bit hosts: use inline asm as before
+#define MULSTEP MULSTEP_asm
 #define MULSTEP_asm(j,h0,h1,h2,h3,h4) \
   gj = g[j]; \
   __asm__( \
@@ -30,8 +61,6 @@
     : "+x"(h0),"+x"(h1),"+x"(h2),"+x"(h3),"+x"(h4) \
     : "x"(gj),"x"(f0),"x"(f1),"x"(f2),"x"(f3),"x"(f4),"x"(f5),"m"(h[i+j]),"m"(h[i+j+5]));
 
-#define MULSTEP MULSTEP_asm
-
 #define MULSTEP_noload(j,h0,h1,h2,h3,h4) \
   gj = g[j]; \
   __asm__( \
@@ -57,6 +86,7 @@
     "vmulps %5,%11,%0 \n\t" \
     : "=&x"(h0),"=&x"(h1),"=&x"(h2),"=&x"(h3),"=&x"(h4) \
     : "x"(gj),"x"(f0),"x"(f1),"x"(f2),"x"(f3),"x"(f4),"x"(f5),"m"(h[i+j]));
+#endif
 
 static inline __m128i _mm_load_cvtepi8_epi16(const long long *x)
 {
diff --git a/crypto/libntrup/src/avx/rq.c b/crypto/libntrup/src/avx/rq.c
index b6dc8ab6c..9b3d4f365 100644
--- a/crypto/libntrup/src/avx/rq.c
+++ b/crypto/libntrup/src/avx/rq.c
@@ -21,6 +21,12 @@
 #define broadcast(r) _mm256_set1_pd(r)
 #define floor(x) _mm256_floor_pd(x)
 
+// 32-bit hosts only
+#ifndef __amd64__
+#define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
+                                                  __a[N];}))
+#endif
+
 void rq_encode(unsigned char *c,const modq *f)
 {
   crypto_int32 f0, f1, f2, f3, f4;
diff --git a/crypto/libntrup/src/avx/rq_mod3.c b/crypto/libntrup/src/avx/rq_mod3.c
index c2b7059b5..6f631d938 100644
--- a/crypto/libntrup/src/avx/rq_mod3.c
+++ b/crypto/libntrup/src/avx/rq_mod3.c
@@ -1,5 +1,6 @@
 #if __AVX2__
 #include <immintrin.h>
+#include <smmintrin.h>
 #include "mod3.h"
 #include "rq.h"
 
@@ -9,6 +10,12 @@
 #define v4591_16 _mm256_set1_epi16(4591)
 #define v10923_16 _mm256_set1_epi16(10923)
 
+// 32-bit hosts only
+#ifndef __amd64__
+#define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
+                                                  __a[N];}))
+#endif
+
 static inline __m256i squeeze(__m256i x)
 {
   __m256i q = _mm256_mulhrs_epi16(x,v7);
diff --git a/daemon/main.cpp b/daemon/main.cpp
index 47f0c8b4f..ba62731a0 100644
--- a/daemon/main.cpp
+++ b/daemon/main.cpp
@@ -100,7 +100,8 @@ main(int argc, char *argv[])
     conffname        = fname.string();
     if(basedir.string().empty())
     {
-      if(!llarp_ensure_config(fname.string().c_str(), nullptr, overWrite, asRouter))
+      if(!llarp_ensure_config(fname.string().c_str(), nullptr, overWrite,
+                              asRouter))
         return 1;
     }
     else
diff --git a/include/llarp/pathset.hpp b/include/llarp/pathset.hpp
index 68439337f..0a293cf3b 100644
--- a/include/llarp/pathset.hpp
+++ b/include/llarp/pathset.hpp
@@ -137,18 +137,18 @@ namespace llarp
       size_t m_NumPaths;
 
      private:
-
       typedef std::pair< RouterID, PathID_t > PathInfo_t;
 
       struct PathInfoHash
       {
-        size_t operator()(const PathInfo_t & i) const
+        size_t
+        operator()(const PathInfo_t& i) const
         {
           return *i.first.data_l() ^ *i.second.data_l();
         }
       };
 
-      typedef std::unordered_map< PathInfo_t, Path* , PathInfoHash> PathMap_t;
+      typedef std::unordered_map< PathInfo_t, Path*, PathInfoHash > PathMap_t;
       PathMap_t m_Paths;
     };
 
diff --git a/include/tuntap.h b/include/tuntap.h
index 2f0447309..28da2e51b 100644
--- a/include/tuntap.h
+++ b/include/tuntap.h
@@ -143,6 +143,9 @@ extern "C"
     char if_name[IF_NAMESIZE];
 #if defined(FreeBSD)
     int mode;
+#endif
+#if defined(Windows)
+    OVERLAPPED ovl;
 #endif
   };
 
diff --git a/llarp/dnsd.cpp b/llarp/dnsd.cpp
index 1e476b673..acc3dca29 100644
--- a/llarp/dnsd.cpp
+++ b/llarp/dnsd.cpp
@@ -82,7 +82,7 @@ writecname_dnss_response(std::string cname, const struct sockaddr *from,
                          dnsd_question_request *request)
 {
   const size_t BUFFER_SIZE = 1024 + (request->question.name.size() * 2);
-  char buf[BUFFER_SIZE];
+  char buf[BUFFER_SIZE];  // heh, another UNIX compiler extension: VLAs in C++
   memset(buf, 0, BUFFER_SIZE);
   char *write_buffer = buf;
   char *bufferBegin  = buf;
diff --git a/llarp/ev.cpp b/llarp/ev.cpp
index a5e5f4c88..07bf7d55b 100644
--- a/llarp/ev.cpp
+++ b/llarp/ev.cpp
@@ -96,7 +96,7 @@ llarp_ev_udp_sendto(struct llarp_udp_io *udp, const sockaddr *to,
                     const void *buf, size_t sz)
 {
   auto ret = static_cast< llarp::ev_io * >(udp->impl)->sendto(to, buf, sz);
-  if(ret == -1)
+  if(ret == -1 && errno)
   {
     llarp::LogWarn("sendto failed ", strerror(errno));
     errno = 0;
diff --git a/llarp/ev.hpp b/llarp/ev.hpp
index f60853113..d91fe37e2 100644
--- a/llarp/ev.hpp
+++ b/llarp/ev.hpp
@@ -5,14 +5,16 @@
 #ifndef _WIN32
 #include <sys/uio.h>
 #endif
-
-#ifndef _MSC_VER
 #include <unistd.h>
-#endif
 #include <llarp/buffer.h>
 #include <list>
 #include <llarp/codel.hpp>
 #include <vector>
+
+#ifdef _WIN32
+#include <variant>
+#endif
+
 #ifndef MAX_WRITE_QUEUE_SIZE
 #define MAX_WRITE_QUEUE_SIZE 1024
 #endif
@@ -29,13 +31,18 @@ namespace llarp
     int fd;
     ev_io(int f) : fd(f), m_writeq("writequeue"){};
 #else
-    SOCKET fd;
+    // on windows, udp event loops are socket fds
+    // and TUN device is a plain old fd
+    std::variant< SOCKET, HANDLE > fd;
     // the unique completion key that helps us to
     // identify the object instance for which we receive data
-    // Here, we'll use the address of the udp_listener instance, converted to
-    // its literal int/int64 representation.
+    // Here, we'll use the address of the udp_listener instance, converted
+    // to its literal int/int64 representation.
     ULONG_PTR listener_id = 0;
     ev_io(SOCKET f) : fd(f), m_writeq("writequeue"){};
+    ev_io(HANDLE t)
+        : fd(t), m_writeq("writequeue"){};  // overload for TUN device, which
+                                            // _is_ a regular file descriptor
 #endif
     virtual int
     read(void* buf, size_t sz) = 0;
@@ -50,7 +57,8 @@ namespace llarp
 #ifndef _WIN32
       return write(fd, data, sz) != -1;
 #else
-      return WriteFile((void*)fd, data, sz, nullptr, nullptr);
+      DWORD w;
+      return WriteFile(std::get< HANDLE >(fd), data, sz, &w, nullptr);
 #endif
     }
 
@@ -129,7 +137,7 @@ namespace llarp
 #ifndef _WIN32
       ::close(fd);
 #else
-      closesocket(fd);
+      closesocket(std::get< SOCKET >(fd));
 #endif
     };
   };
@@ -156,7 +164,11 @@ struct llarp_ev_loop
     auto ev = create_udp(l, src);
     if(ev)
     {
+#ifdef _WIN32
+      l->fd = std::get< SOCKET >(ev->fd);
+#else
       l->fd = ev->fd;
+#endif
     }
     return ev && add_ev(ev, false);
   }
diff --git a/llarp/ev_kqueue.hpp b/llarp/ev_kqueue.hpp
index 469986253..43dd1390c 100644
--- a/llarp/ev_kqueue.hpp
+++ b/llarp/ev_kqueue.hpp
@@ -206,13 +206,13 @@ struct llarp_kqueue_loop : public llarp_ev_loop
       while(idx < result)
       {
         llarp::ev_io* ev = static_cast< llarp::ev_io* >(events[idx].udata);
-        if (ev)
+        if(ev)
         {
           ev->read(readbuf, sizeof(readbuf));
         }
         else
         {
-          llarp::LogWarn("event[", idx,"] udata is not an ev_io");
+          llarp::LogWarn("event[", idx, "] udata is not an ev_io");
         }
         ++idx;
       }
@@ -313,7 +313,6 @@ struct llarp_kqueue_loop : public llarp_ev_loop
       close(fd);
       return -1;
     }
-
     return fd;
   }
 
diff --git a/llarp/ev_win32.hpp b/llarp/ev_win32.hpp
index a7fb93bd2..442ff49db 100644
--- a/llarp/ev_win32.hpp
+++ b/llarp/ev_win32.hpp
@@ -17,11 +17,13 @@ namespace llarp
     // we receive queued data in the OVERLAPPED data field,
     // much like the pipefds in the UNIX kqueue and loonix
     // epoll handles
-    // 0 is the read port, 1 is the write port
-    WSAOVERLAPPED portfds[2] = {0};
+    WSAOVERLAPPED portfd;
     size_t iosz;
 
-    udp_listener(SOCKET fd, llarp_udp_io* u) : ev_io(fd), udp(u){};
+    udp_listener(SOCKET fd, llarp_udp_io* u) : ev_io(fd), udp(u)
+    {
+      memset((void*)&portfd, 0, sizeof(WSAOVERLAPPED));
+    };
 
     ~udp_listener()
     {
@@ -43,8 +45,9 @@ namespace llarp
       unsigned long flags = 0;
       WSABUF wbuf         = {sz, static_cast< char* >(buf)};
       // WSARecvFrom
-      int ret = ::WSARecvFrom(fd, &wbuf, 1, nullptr, &flags, addr, &slen,
-                              &portfds[0], nullptr);
+      llarp::LogDebug("read ", sz, " bytes into socket");
+      int ret = ::WSARecvFrom(std::get< SOCKET >(fd), &wbuf, 1, nullptr, &flags,
+                              addr, &slen, &portfd, nullptr);
       // 997 is the error code for queued ops
       int s_errno = ::WSAGetLastError();
       if(ret && s_errno != 997)
@@ -74,9 +77,10 @@ namespace llarp
           return -1;
       }
       // WSASendTo
-      ssize_t sent =
-          ::WSASendTo(fd, &wbuf, 1, nullptr, 0, to, slen, &portfds[1], nullptr);
-      int s_errno = ::WSAGetLastError();
+      llarp::LogDebug("write ", sz, " bytes into socket");
+      ssize_t sent = ::WSASendTo(std::get< SOCKET >(fd), &wbuf, 1, nullptr, 0,
+                                 to, slen, &portfd, nullptr);
+      int s_errno  = ::WSAGetLastError();
       if(sent && s_errno != 997)
       {
         llarp::LogWarn("send socket error ", s_errno);
@@ -90,8 +94,9 @@ namespace llarp
   {
     llarp_tun_io* t;
     device* tunif;
+    OVERLAPPED* tun_async;
     tun(llarp_tun_io* tio)
-        : ev_io(-1)
+        : ev_io(INVALID_HANDLE_VALUE)
         , t(tio)
         , tunif(tuntap_init())
 
@@ -115,6 +120,12 @@ namespace llarp
       ev_io::flush_write();
     }
 
+    bool
+    do_write(void* data, size_t sz)
+    {
+      return WriteFile(std::get< HANDLE >(fd), data, sz, nullptr, tun_async);
+    }
+
     int
     read(void* buf, size_t sz)
     {
@@ -136,23 +147,24 @@ namespace llarp
         llarp::LogWarn("failed to start interface");
         return false;
       }
-      if(tuntap_up(tunif) == -1)
+      if(tuntap_set_ip(tunif, t->ifaddr, t->ifaddr, t->netmask) == -1)
       {
-        llarp::LogWarn("failed to put interface up: ", strerror(errno));
+        llarp::LogWarn("failed to set ip");
         return false;
       }
-      if(tuntap_set_ip(tunif, t->ifaddr, t->ifaddr, t->netmask) == -1)
+      if(tuntap_up(tunif) == -1)
       {
-        llarp::LogWarn("failed to set ip");
+        llarp::LogWarn("failed to put interface up: ", strerror(errno));
         return false;
       }
-      fd = (SOCKET)tunif->tun_fd;
-      if(fd == -1)
+
+      fd        = tunif->tun_fd;
+      tun_async = &tunif->ovl;
+      if(std::get< HANDLE >(fd) == INVALID_HANDLE_VALUE)
         return false;
 
-      // set non blocking
-      int on = 1;
-      return ioctlsocket(fd, FIONBIO, (u_long*)&on) != -1;
+      // we're already non-blocking
+      return true;
     }
 
     ~tun()
@@ -199,35 +211,29 @@ struct llarp_win32_loop : public llarp_ev_loop
     // as an arch-specific pointer value
     ULONG_PTR ev_id      = 0;
     WSAOVERLAPPED* qdata = nullptr;
-    int result           = 0;
     int idx              = 0;
+    BOOL result =
+        ::GetQueuedCompletionStatus(iocpfd, &iolen, &ev_id, &qdata, ms);
 
-    do
+    if(result && qdata)
     {
-      if(ev_id && qdata && iolen)
+      llarp::udp_listener* ev = reinterpret_cast< llarp::udp_listener* >(ev_id);
+      if(ev && !ev->fd.valueless_by_exception())
       {
-        llarp::udp_listener* ev =
-            reinterpret_cast< llarp::udp_listener* >(ev_id);
-        if(ev && ev->fd)
-        {
-          ev->getData(readbuf, sizeof(readbuf), iolen);
-        }
+        llarp::LogDebug("size: ", iolen, "\tev_id: ", ev_id,
+                        "\tqdata: ", qdata);
+        ev->getData(readbuf, sizeof(readbuf), iolen);
       }
       ++idx;
-    } while(::GetQueuedCompletionStatus(iocpfd, &iolen, &ev_id, &qdata, ms));
-
-    // tick_listeners inlined since win32 does not
-    // implement ev_tun
-    for(auto& l : udp_listeners)
-    {
-      if(l->tick)
-        l->tick(l);
     }
 
     if(!idx)
       return -1;
     else
+    {
+      tick_listeners();
       result = idx;
+    }
 
     return result;
   }
@@ -244,35 +250,30 @@ struct llarp_win32_loop : public llarp_ev_loop
     // as an arch-specific pointer value
     ULONG_PTR ev_id      = 0;
     WSAOVERLAPPED* qdata = nullptr;
-    int result           = 0;
     int idx              = 0;
+    int result =
+        ::GetQueuedCompletionStatus(iocpfd, &iolen, &ev_id, &qdata, 10);
 
     // unlike epoll and kqueue, we only need to run so long as the
     // system call returns TRUE
-    do
+    if(result)
     {
-      if(ev_id && qdata && iolen)
+      llarp::udp_listener* ev = reinterpret_cast< llarp::udp_listener* >(ev_id);
+      if(ev && !ev->fd.valueless_by_exception())
       {
-        llarp::udp_listener* ev =
-            reinterpret_cast< llarp::udp_listener* >(ev_id);
-        if(ev && ev->fd)
-        {
-          ev->getData(readbuf, sizeof(readbuf), iolen);
-        }
+        llarp::LogInfo("size: ", iolen, "\tev_id: ", ev_id, "\tqdata: ", qdata);
+        ev->getData(readbuf, sizeof(readbuf), iolen);
       }
       ++idx;
-    } while(::GetQueuedCompletionStatus(iocpfd, &iolen, &ev_id, &qdata, 10));
-
-    for(auto& l : udp_listeners)
-    {
-      if(l->tick)
-        l->tick(l);
     }
 
     if(!idx)
       return -1;
     else
+    {
+      tick_listeners();
       result = idx;
+    }
 
     return result;
   }
@@ -292,8 +293,8 @@ struct llarp_win32_loop : public llarp_ev_loop
       default:
         return INVALID_SOCKET;
     }
-    SOCKET fd = ::WSASocket(addr->sa_family, SOCK_DGRAM, 0, nullptr, 0,
-                            WSA_FLAG_OVERLAPPED);
+    DWORD on  = 1;
+    SOCKET fd = ::socket(addr->sa_family, SOCK_DGRAM, 0);
     if(fd == INVALID_SOCKET)
     {
       perror("WSASocket()");
@@ -322,7 +323,8 @@ struct llarp_win32_loop : public llarp_ev_loop
       closesocket(fd);
       return INVALID_SOCKET;
     }
-    llarp::LogInfo("socket fd is ", fd);
+    llarp::LogDebug("socket fd is ", fd);
+    ioctlsocket(fd, FIONBIO, &on);
     return fd;
   }
 
@@ -331,8 +333,9 @@ struct llarp_win32_loop : public llarp_ev_loop
   {
     // On Windows, just close the socket to decrease the iocp refcount
     // and stop any pending I/O
-    BOOL stopped = ::CancelIo(reinterpret_cast< HANDLE >(ev->fd));
-    return closesocket(ev->fd) == 0 && stopped == TRUE;
+    BOOL stopped =
+        ::CancelIo(reinterpret_cast< HANDLE >(std::get< SOCKET >(ev->fd)));
+    return closesocket(std::get< SOCKET >(ev->fd)) == 0 && stopped == TRUE;
   }
 
   llarp::ev_io*
@@ -361,12 +364,37 @@ struct llarp_win32_loop : public llarp_ev_loop
   bool
   add_ev(llarp::ev_io* ev, bool write)
   {
-    ev->listener_id = reinterpret_cast< ULONG_PTR >(ev);
-    if(!::CreateIoCompletionPort(reinterpret_cast< HANDLE >(ev->fd), iocpfd,
-                                 ev->listener_id, 0))
+    uint8_t buf[1024];
+    llarp::udp_listener* udp = nullptr;
+    llarp::tun* t            = nullptr;
+    ev->listener_id          = reinterpret_cast< ULONG_PTR >(ev);
+    memset(&buf, 0, 1024);
+
+    switch(ev->fd.index())
     {
-      delete ev;
-      return false;
+      case 0:
+        udp = dynamic_cast< llarp::udp_listener* >(ev);
+        if(!::CreateIoCompletionPort((HANDLE)std::get< 0 >(ev->fd), iocpfd,
+                                     ev->listener_id, 0))
+        {
+          delete ev;
+          return false;
+        }
+        ::ReadFile((HANDLE)std::get< 0 >(ev->fd), &buf, 1024, nullptr,
+                   &udp->portfd);
+        break;
+      case 1:
+        t = dynamic_cast< llarp::tun* >(ev);
+        if(!::CreateIoCompletionPort(std::get< 1 >(ev->fd), iocpfd,
+                                     ev->listener_id, 0))
+        {
+          delete ev;
+          return false;
+        }
+        ::ReadFile(std::get< 1 >(ev->fd), &buf, 1024, nullptr, t->tun_async);
+        break;
+      default:
+        return false;
     }
     return true;
   }
diff --git a/llarp/handlers/tun.cpp b/llarp/handlers/tun.cpp
index 90efd5adb..089a20fd8 100644
--- a/llarp/handlers/tun.cpp
+++ b/llarp/handlers/tun.cpp
@@ -110,7 +110,8 @@ namespace llarp
 
         llarp::Addr tunIp(source_addr);
         // related to dns_iptracker_setup_dotLokiLookup(&this->dll, tunIp);
-        dns_iptracker_setup(this->dll.ip_tracker, tunIp);  // claim GW IP to make sure it's not inuse
+        dns_iptracker_setup(this->dll.ip_tracker,
+                            tunIp);  // claim GW IP to make sure it's not inuse
         return true;
       }
       return Endpoint::SetOption(k, v);
@@ -150,7 +151,8 @@ namespace llarp
         llarp::Addr tunIp(tunif.ifaddr);
         dns_iptracker_setup_dotLokiLookup(
             &this->dll, tunIp);  // just set ups dll to use global iptracker
-        dns_iptracker_setup(this->dll.ip_tracker, tunIp);  // claim GW IP to make sure it's not inuse
+        dns_iptracker_setup(this->dll.ip_tracker,
+                            tunIp);  // claim GW IP to make sure it's not inuse
 
         // set up networking in currrent thread if we are not isolated
         if(!SetupNetworking())
@@ -162,8 +164,9 @@ namespace llarp
         llarp::Addr tunIp(tunif.ifaddr);
         this->dll.ip_tracker = new dns_iptracker;
         dns_iptracker_setup_dotLokiLookup(
-                                          &this->dll, tunIp);  // just set ups dll to use global iptracker
-        dns_iptracker_setup(this->dll.ip_tracker, tunIp);  // claim GW IP to make sure it's not inuse
+            &this->dll, tunIp);  // just set ups dll to use global iptracker
+        dns_iptracker_setup(this->dll.ip_tracker,
+                            tunIp);  // claim GW IP to make sure it's not inuse
       }
       // wait for result for network setup
       llarp::LogInfo("waiting for tun interface...");
@@ -254,9 +257,8 @@ namespace llarp
     {
       llarp::LogInfo("Set Up networking for ", Name());
       bool result = SetupTun();
-#ifndef _WIN32
-      m_TunSetupResult.set_value(result);
-#endif
+      m_TunSetupResult.set_value(
+          result);  // now that NT has tun, we don't need the CPP guard
       if(!NetworkIsIsolated())
       {
         // need to check to see if we have more than one hidden service
diff --git a/llarp/link/utp.cpp b/llarp/link/utp.cpp
index a1cc7124c..67ec4d119 100644
--- a/llarp/link/utp.cpp
+++ b/llarp/link/utp.cpp
@@ -350,7 +350,8 @@ namespace llarp
                         " bytes");
         if(::sendto(l->m_udp.fd, (char*)arg->buf, arg->len, arg->flags,
                     arg->address, arg->address_len)
-           == -1)
+               == -1
+           && errno)
         {
           llarp::LogError("sendto failed: ", strerror(errno));
         }
diff --git a/llarp/time.cpp b/llarp/time.cpp
index 6ec997cec..bef56e580 100644
--- a/llarp/time.cpp
+++ b/llarp/time.cpp
@@ -15,6 +15,9 @@ namespace llarp
 }  // namespace llarp
 
 // use std::chrono because otherwise the network breaks with Daylight Savings
+// this time, it doesn't get truncated -despair
+// that concern is what drove me back to the POSIX C time functions
+// in the first place
 llarp_time_t
 llarp_time_now_ms()
 {
diff --git a/readme.md b/readme.md
index f730d9b19..fc76bf82b 100644
--- a/readme.md
+++ b/readme.md
@@ -49,10 +49,4 @@ for a development environment:
 
 ## Usage
 
-### Windows
-
-Windows only supports client mode so you run `lokinet.exe` and that's it.
-
-### Linux
-
 see the [lokinet-builder](https://github.com/loki-project/lokinet-builder)
diff --git a/vendor/libtuntap-master/tuntap-windows.c b/vendor/libtuntap-master/tuntap-windows.c
index 625a065fc..83b7e880a 100644
--- a/vendor/libtuntap-master/tuntap-windows.c
+++ b/vendor/libtuntap-master/tuntap-windows.c
@@ -63,7 +63,7 @@ formated_error(LPWSTR pMessage, DWORD m, ...)
   LPWSTR pBuffer = NULL;
 
   va_list args = NULL;
-  va_start(args, pMessage);
+  va_start(args, m);
 
   FormatMessage(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_ALLOCATE_BUFFER,
                 pMessage, m, 0, (LPSTR)&pBuffer, 0, &args);
@@ -189,8 +189,11 @@ tuntap_start(struct device *dev, int mode, int tun)
 
   if(mode == TUNTAP_MODE_TUNNEL)
   {
-    tuntap_log(TUNTAP_LOG_NOTICE, "Layer 3 tunneling is not implemented");
-    return -1;
+    deviceid = reg_query(NETWORK_ADAPTERS);
+    snprintf(buf, sizeof buf, "\\\\.\\Global\\%s.tap", deviceid);
+    tun_fd = CreateFile(buf, GENERIC_WRITE | GENERIC_READ,
+                        FILE_SHARE_READ | FILE_SHARE_WRITE, 0, OPEN_EXISTING,
+                        FILE_ATTRIBUTE_SYSTEM | FILE_FLAG_OVERLAPPED, 0);
   }
   else if(mode != TUNTAP_MODE_ETHERNET)
   {
@@ -198,14 +201,9 @@ tuntap_start(struct device *dev, int mode, int tun)
     return -1;
   }
 
-  deviceid = reg_query(NETWORK_ADAPTERS);
-  snprintf(buf, sizeof buf, "\\\\.\\Global\\%s.tap", deviceid);
-  tun_fd = CreateFile(buf, GENERIC_WRITE | GENERIC_READ, 0, 0, OPEN_EXISTING,
-                      FILE_ATTRIBUTE_SYSTEM | FILE_FLAG_OVERLAPPED, 0);
   if(tun_fd == TUNFD_INVALID_VALUE)
   {
     int errcode = GetLastError();
-
     tuntap_log(TUNTAP_LOG_ERR, (const char *)formated_error(L"%1%0", errcode));
     return -1;
   }
@@ -241,8 +239,8 @@ tuntap_get_hwaddr(struct device *dev)
     char buf[128];
 
     (void)_snprintf(buf, sizeof buf,
-                      "MAC address: %.2x:%.2x:%.2x:%.2x:%.2x:%.2x", hwaddr[0],
-                      hwaddr[1], hwaddr[2], hwaddr[3], hwaddr[4], hwaddr[5]);
+                    "MAC address: %.2x:%.2x:%.2x:%.2x:%.2x:%.2x", hwaddr[0],
+                    hwaddr[1], hwaddr[2], hwaddr[3], hwaddr[4], hwaddr[5]);
     tuntap_log(TUNTAP_LOG_DEBUG, buf);
   }
   return (char *)hwaddr;
@@ -274,8 +272,7 @@ tuntap_sys_set_updown(struct device *dev, ULONG flag)
   {
     char buf[32];
 
-    (void)_snprintf(buf, sizeof buf, "Status: %s",
-                      flag ? "Up" : "Down");
+    (void)_snprintf(buf, sizeof buf, "Status: %s", flag ? "Up" : "Down");
     tuntap_log(TUNTAP_LOG_DEBUG, buf);
     return 0;
   }
@@ -330,26 +327,54 @@ tuntap_set_mtu(struct device *dev, int mtu)
 int
 tuntap_sys_set_ipv4(struct device *dev, t_tun_in_addr *s, uint32_t mask)
 {
-  IPADDR psock[4];
-  DWORD len;
-
-  /* Address + Netmask */
-  psock[0] = s->S_un.S_addr;
-  psock[1] = mask;
-  /* DHCP server address (We don't want it) */
-  psock[2] = 0;
-  /* DHCP lease time */
-  psock[3] = 0;
-
-  if(DeviceIoControl(dev->tun_fd, TAP_IOCTL_CONFIG_DHCP_MASQ, &psock,
-                     sizeof(psock), &psock, sizeof(psock), &len, NULL)
-     == 0)
+  IPADDR sock[3];
+  DWORD len, ret;
+  IPADDR ep[4];
+#pragma pack(push)
+#pragma pack(1)
+  struct
+  {
+    uint8_t dhcp_opt;
+    uint8_t length;
+    uint32_t value[2];
+  } dns;
+#pragma pack(pop)
+
+  sock[0] = s->S_un.S_addr;
+  sock[2] = mask;
+  sock[1] = sock[0] & sock[2];
+  ret = DeviceIoControl(dev->tun_fd, TAP_IOCTL_CONFIG_TUN, &sock, sizeof(sock),
+                        &sock, sizeof(sock), &len, NULL);
+  ep[0] = s->S_un.S_addr;
+  ep[1] = mask;
+  ep[2] = (s->S_un.S_addr | ~mask)
+      - (mask + 1); /* For the 10.x.0.y subnet (in a class C config), _should_
+                       be 10.x.0.254 i think */
+  ep[3] = 3153600;    /* one year */
+
+  ret = DeviceIoControl(dev->tun_fd, TAP_IOCTL_CONFIG_DHCP_MASQ, ep, sizeof(ep),
+                        ep, sizeof(ep), &len, NULL);
+
+  /* set DNS address to 127.0.0.1 as lokinet-client runs its own DNS resolver
+   * inline */
+  dns.dhcp_opt = 6;
+  dns.length   = 4;
+  dns.value[0] =
+      htonl(0x7F000001); /* apparently this doesn't show in network properties,
+                            but it works 🤷🏻‍♂️ */
+  dns.value[1] = 0;
+
+  ret = DeviceIoControl(dev->tun_fd, TAP_IOCTL_CONFIG_DHCP_SET_OPT, &dns,
+                        sizeof(dns), &dns, sizeof(dns), &len, NULL);
+
+  if(!ret)
   {
     int errcode = GetLastError();
 
     tuntap_log(TUNTAP_LOG_ERR, (const char *)formated_error(L"%1%0", errcode));
     return -1;
   }
+
   return 0;
 }
 
@@ -369,12 +394,17 @@ tuntap_read(struct device *dev, void *buf, size_t size)
 {
   DWORD len;
 
-  if(ReadFile(dev->tun_fd, buf, (DWORD)size, &len, NULL) == 0)
+  if(ReadFile(dev->tun_fd, buf, (DWORD)size, &len, &dev->ovl) == 0)
   {
     int errcode = GetLastError();
 
-    tuntap_log(TUNTAP_LOG_ERR, (const char *)formated_error(L"%1%0", errcode));
-    return -1;
+    if (errcode != 997)
+	{
+		tuntap_log(TUNTAP_LOG_ERR, (const char *)formated_error(L"%1%0", errcode));
+		return -1;
+	}
+	else
+		return 0;
   }
 
   return 0;
@@ -385,12 +415,17 @@ tuntap_write(struct device *dev, void *buf, size_t size)
 {
   DWORD len;
 
-  if(WriteFile(dev->tun_fd, buf, (DWORD)size, &len, NULL) == 0)
+  if(WriteFile(dev->tun_fd, buf, (DWORD)size, &len, &dev->ovl) == 0)
   {
-    int errcode = GetLastError();
-
-    tuntap_log(TUNTAP_LOG_ERR, (const char *)formated_error(L"%1%0", errcode));
-    return -1;
+  	int errcode = GetLastError();
+
+    if (errcode != 997)
+	{
+		tuntap_log(TUNTAP_LOG_ERR, (const char *)formated_error(L"%1%0", errcode));
+		return -1;
+	}
+	else
+		return 0;
   }
 
   return 0;
@@ -411,7 +446,7 @@ tuntap_set_nonblocking(struct device *dev, int set)
   (void)dev;
   (void)set;
   tuntap_log(TUNTAP_LOG_NOTICE,
-             "Your system does not support tuntap_set_nonblocking()");
+             "TUN/TAP devices on Windows are non-blocking by default using either overlapped I/O or IOCPs");
   return -1;
 }
 
@@ -438,7 +473,6 @@ tuntap_set_descr(struct device *dev, const char *descr)
 int
 tuntap_set_ifname(struct device *dev, const char *name)
 {
-  /* TODO: Check Windows API to know how to rename an interface */
   (void)dev;
   (void)name;
   tuntap_log(TUNTAP_LOG_NOTICE,
diff --git a/vendor/libtuntap-master/tuntap.cpp b/vendor/libtuntap-master/tuntap.cpp
index e821714a1..630107b23 100644
--- a/vendor/libtuntap-master/tuntap.cpp
+++ b/vendor/libtuntap-master/tuntap.cpp
@@ -68,6 +68,9 @@ extern "C"
     dev->tun_fd    = TUNFD_INVALID_VALUE;
     dev->ctrl_sock = -1;
     dev->flags     = 0;
+	#if defined(Windows)
+	memset(&dev->ovl, 0, sizeof(OVERLAPPED));
+	#endif
 
     __tuntap_log = &tuntap_log_default;
     return dev;