grpc 1.30.2

data/src/core/lib/iomgr/tcp_uv.cc

@@ -0,0 +1,419 @@
+/*
+ *
+ * Copyright 2016 gRPC authors.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ */
+
+#include <grpc/support/port_platform.h>
+
+#include "src/core/lib/iomgr/port.h"
+
+#ifdef GRPC_UV
+#include <limits.h>
+#include <string.h>
+
+#include <grpc/slice_buffer.h>
+
+#include <grpc/support/alloc.h>
+#include <grpc/support/log.h>
+#include <grpc/support/string_util.h>
+
+#include "src/core/lib/gpr/string.h"
+#include "src/core/lib/iomgr/error.h"
+#include "src/core/lib/iomgr/iomgr_custom.h"
+#include "src/core/lib/iomgr/resolve_address_custom.h"
+#include "src/core/lib/iomgr/resource_quota.h"
+#include "src/core/lib/iomgr/tcp_custom.h"
+#include "src/core/lib/slice/slice_internal.h"
+#include "src/core/lib/slice/slice_string_helpers.h"
+
+#include <uv.h>
+
+#define IGNORE_CONST(addr) ((grpc_sockaddr*)(uintptr_t)(addr))
+
+typedef struct uv_socket_t {
+  uv_connect_t connect_req;
+  uv_write_t write_req;
+  uv_shutdown_t shutdown_req;
+  uv_tcp_t* handle;
+  uv_buf_t* write_buffers;
+
+  char* read_buf;
+  size_t read_len;
+
+  int pending_connections;
+  grpc_custom_socket* accept_socket;
+  grpc_error* accept_error;
+
+  grpc_custom_connect_callback connect_cb;
+  grpc_custom_write_callback write_cb;
+  grpc_custom_read_callback read_cb;
+  grpc_custom_accept_callback accept_cb;
+  grpc_custom_close_callback close_cb;
+
+} uv_socket_t;
+
+static grpc_error* tcp_error_create(const char* desc, int status) {
+  if (status == 0) {
+    return GRPC_ERROR_NONE;
+  }
+  grpc_error* error = GRPC_ERROR_CREATE_FROM_STATIC_STRING(desc);
+  /* All tcp errors are marked with UNAVAILABLE so that application may
+   * choose to retry. */
+  error = grpc_error_set_int(error, GRPC_ERROR_INT_GRPC_STATUS,
+                             GRPC_STATUS_UNAVAILABLE);
+  return grpc_error_set_str(error, GRPC_ERROR_STR_OS_ERROR,
+                            grpc_slice_from_static_string(uv_strerror(status)));
+}
+
+static void uv_socket_destroy(grpc_custom_socket* socket) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  gpr_free(uv_socket->handle);
+  gpr_free(uv_socket);
+}
+
+static void alloc_uv_buf(uv_handle_t* handle, size_t suggested_size,
+                         uv_buf_t* buf) {
+  uv_socket_t* uv_socket =
+      (uv_socket_t*)((grpc_custom_socket*)handle->data)->impl;
+  (void)suggested_size;
+  buf->base = uv_socket->read_buf;
+  buf->len = uv_socket->read_len;
+}
+
+static void uv_read_callback(uv_stream_t* stream, ssize_t nread,
+                             const uv_buf_t* buf) {
+  grpc_error* error = GRPC_ERROR_NONE;
+  if (nread == 0) {
+    // Nothing happened. Wait for the next callback
+    return;
+  }
+  // TODO(murgatroid99): figure out what the return value here means
+  uv_read_stop(stream);
+  if (nread == UV_EOF) {
+    error = GRPC_ERROR_CREATE_FROM_STATIC_STRING("EOF");
+  } else if (nread < 0) {
+    error = tcp_error_create("TCP Read failed", nread);
+  }
+  grpc_custom_socket* socket = (grpc_custom_socket*)stream->data;
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  uv_socket->read_cb(socket, (size_t)nread, error);
+}
+
+static void uv_close_callback(uv_handle_t* handle) {
+  grpc_custom_socket* socket = (grpc_custom_socket*)handle->data;
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  if (uv_socket->accept_socket) {
+    uv_socket->accept_cb(socket, uv_socket->accept_socket,
+                         GRPC_ERROR_CREATE_FROM_STATIC_STRING("socket closed"));
+  }
+  uv_socket->close_cb(socket);
+}
+
+static void uv_socket_read(grpc_custom_socket* socket, char* buffer,
+                           size_t length, grpc_custom_read_callback read_cb) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  int status;
+  grpc_error* error;
+  uv_socket->read_cb = read_cb;
+  uv_socket->read_buf = buffer;
+  uv_socket->read_len = length;
+  // TODO(murgatroid99): figure out what the return value here means
+  status =
+      uv_read_start((uv_stream_t*)uv_socket->handle, (uv_alloc_cb)alloc_uv_buf,
+                    (uv_read_cb)uv_read_callback);
+  if (status != 0) {
+    error = tcp_error_create("TCP Read failed at start", status);
+    uv_socket->read_cb(socket, 0, error);
+  }
+}
+
+static void uv_write_callback(uv_write_t* req, int status) {
+  grpc_custom_socket* socket = (grpc_custom_socket*)req->data;
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  gpr_free(uv_socket->write_buffers);
+  uv_socket->write_cb(socket, tcp_error_create("TCP Write failed", status));
+}
+
+void uv_socket_write(grpc_custom_socket* socket,
+                     grpc_slice_buffer* write_slices,
+                     grpc_custom_write_callback write_cb) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  uv_socket->write_cb = write_cb;
+  uv_buf_t* uv_buffers;
+  uv_write_t* write_req;
+
+  uv_buffers = (uv_buf_t*)gpr_malloc(sizeof(uv_buf_t) * write_slices->count);
+  for (size_t i = 0; i < write_slices->count; i++) {
+    uv_buffers[i].base = (char*)GRPC_SLICE_START_PTR(write_slices->slices[i]);
+    uv_buffers[i].len = GRPC_SLICE_LENGTH(write_slices->slices[i]);
+  }
+
+  uv_socket->write_buffers = uv_buffers;
+  write_req = &uv_socket->write_req;
+  write_req->data = socket;
+  // TODO(murgatroid99): figure out what the return value here means
+  uv_write(write_req, (uv_stream_t*)uv_socket->handle, uv_buffers,
+           write_slices->count, uv_write_callback);
+}
+
+static void shutdown_callback(uv_shutdown_t* req, int status) {}
+
+static void uv_socket_shutdown(grpc_custom_socket* socket) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  uv_shutdown_t* req = &uv_socket->shutdown_req;
+  uv_shutdown(req, (uv_stream_t*)uv_socket->handle, shutdown_callback);
+}
+
+static void uv_socket_close(grpc_custom_socket* socket,
+                            grpc_custom_close_callback close_cb) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  uv_socket->close_cb = close_cb;
+  uv_close((uv_handle_t*)uv_socket->handle, uv_close_callback);
+}
+
+static grpc_error* uv_socket_init_helper(uv_socket_t* uv_socket, int domain) {
+  uv_tcp_t* tcp = (uv_tcp_t*)gpr_malloc(sizeof(uv_tcp_t));
+  uv_socket->handle = tcp;
+  int status = uv_tcp_init_ex(uv_default_loop(), tcp, (unsigned int)domain);
+  if (status != 0) {
+    return tcp_error_create("Failed to initialize UV tcp handle", status);
+  }
+#if defined(GPR_LINUX) && defined(SO_REUSEPORT)
+  if (domain == AF_INET || domain == AF_INET6) {
+    int enable = 1;
+    int fd;
+    uv_fileno((uv_handle_t*)tcp, &fd);
+    // TODO Handle error here.
+    setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &enable, sizeof(enable));
+  }
+#endif
+  uv_socket->write_buffers = nullptr;
+  uv_socket->read_len = 0;
+  uv_tcp_nodelay(uv_socket->handle, 1);
+  // Node uses a garbage collector to call destructors, so we don't
+  // want to hold the uv loop open with active gRPC objects.
+  uv_unref((uv_handle_t*)uv_socket->handle);
+  uv_socket->pending_connections = 0;
+  uv_socket->accept_socket = nullptr;
+  uv_socket->accept_error = GRPC_ERROR_NONE;
+  return GRPC_ERROR_NONE;
+}
+
+static grpc_error* uv_socket_init(grpc_custom_socket* socket, int domain) {
+  uv_socket_t* uv_socket = (uv_socket_t*)gpr_malloc(sizeof(uv_socket_t));
+  grpc_error* error = uv_socket_init_helper(uv_socket, domain);
+  if (error != GRPC_ERROR_NONE) {
+    return error;
+  }
+  uv_socket->handle->data = socket;
+  socket->impl = uv_socket;
+  return GRPC_ERROR_NONE;
+}
+
+static grpc_error* uv_socket_getpeername(grpc_custom_socket* socket,
+                                         const grpc_sockaddr* addr,
+                                         int* addr_len) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  int err = uv_tcp_getpeername(uv_socket->handle,
+                               (struct sockaddr*)IGNORE_CONST(addr), addr_len);
+  return tcp_error_create("getpeername failed", err);
+}
+
+static grpc_error* uv_socket_getsockname(grpc_custom_socket* socket,
+                                         const grpc_sockaddr* addr,
+                                         int* addr_len) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  int err = uv_tcp_getsockname(uv_socket->handle,
+                               (struct sockaddr*)IGNORE_CONST(addr), addr_len);
+  return tcp_error_create("getsockname failed", err);
+}
+
+static void accept_new_connection(grpc_custom_socket* socket) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  if (uv_socket->pending_connections == 0 || !uv_socket->accept_socket) {
+    return;
+  }
+  grpc_custom_socket* new_socket = uv_socket->accept_socket;
+  grpc_error* error = uv_socket->accept_error;
+  uv_socket->accept_socket = nullptr;
+  uv_socket->accept_error = GRPC_ERROR_NONE;
+  uv_socket->pending_connections -= 1;
+  if (uv_socket->accept_error != GRPC_ERROR_NONE) {
+    uv_stream_t dummy_handle;
+    uv_accept((uv_stream_t*)uv_socket->handle, &dummy_handle);
+    uv_socket->accept_cb(socket, new_socket, error);
+  } else {
+    uv_socket_t* uv_new_socket = (uv_socket_t*)gpr_malloc(sizeof(uv_socket_t));
+    uv_socket_init_helper(uv_new_socket, AF_UNSPEC);
+    // UV documentation says this is guaranteed to succeed
+    GPR_ASSERT(uv_accept((uv_stream_t*)uv_socket->handle,
+                         (uv_stream_t*)uv_new_socket->handle) == 0);
+    new_socket->impl = uv_new_socket;
+    uv_new_socket->handle->data = new_socket;
+    uv_socket->accept_cb(socket, new_socket, error);
+  }
+}
+
+static void uv_on_connect(uv_stream_t* server, int status) {
+  grpc_custom_socket* socket = (grpc_custom_socket*)server->data;
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  if (status < 0) {
+    switch (status) {
+      case UV_EINTR:
+      case UV_EAGAIN:
+        return;
+      default:
+        uv_socket->accept_error = tcp_error_create("accept failed", status);
+    }
+  }
+  uv_socket->pending_connections += 1;
+  accept_new_connection(socket);
+}
+
+void uv_socket_accept(grpc_custom_socket* socket,
+                      grpc_custom_socket* new_socket,
+                      grpc_custom_accept_callback accept_cb) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  uv_socket->accept_cb = accept_cb;
+  GPR_ASSERT(uv_socket->accept_socket == nullptr);
+  uv_socket->accept_socket = new_socket;
+  accept_new_connection(socket);
+}
+
+static grpc_error* uv_socket_bind(grpc_custom_socket* socket,
+                                  const grpc_sockaddr* addr, size_t len,
+                                  int flags) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  int status =
+      uv_tcp_bind((uv_tcp_t*)uv_socket->handle, (struct sockaddr*)addr, 0);
+  return tcp_error_create("Failed to bind to port", status);
+}
+
+static grpc_error* uv_socket_listen(grpc_custom_socket* socket) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  int status =
+      uv_listen((uv_stream_t*)uv_socket->handle, SOMAXCONN, uv_on_connect);
+  return tcp_error_create("Failed to listen to port", status);
+}
+
+static void uv_tc_on_connect(uv_connect_t* req, int status) {
+  grpc_custom_socket* socket = (grpc_custom_socket*)req->data;
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  grpc_error* error;
+  if (status == UV_ECANCELED) {
+    // This should only happen if the handle is already closed
+    error = GRPC_ERROR_CREATE_FROM_STATIC_STRING("Timeout occurred");
+  } else {
+    error = tcp_error_create("Failed to connect to remote host", status);
+  }
+  uv_socket->connect_cb(socket, error);
+}
+
+static void uv_socket_connect(grpc_custom_socket* socket,
+                              const grpc_sockaddr* addr, size_t len,
+                              grpc_custom_connect_callback connect_cb) {
+  uv_socket_t* uv_socket = (uv_socket_t*)socket->impl;
+  uv_socket->connect_cb = connect_cb;
+  uv_socket->connect_req.data = socket;
+  int status = uv_tcp_connect(&uv_socket->connect_req, uv_socket->handle,
+                              (struct sockaddr*)addr, uv_tc_on_connect);
+  if (status != 0) {
+    // The callback will not be called
+    uv_socket->connect_cb(socket, tcp_error_create("connect failed", status));
+  }
+}
+
+static grpc_resolved_addresses* handle_addrinfo_result(
+    struct addrinfo* result) {
+  struct addrinfo* resp;
+  size_t i;
+  grpc_resolved_addresses* addresses =
+      (grpc_resolved_addresses*)gpr_malloc(sizeof(grpc_resolved_addresses));
+  addresses->naddrs = 0;
+  for (resp = result; resp != nullptr; resp = resp->ai_next) {
+    addresses->naddrs++;
+  }
+  addresses->addrs = (grpc_resolved_address*)gpr_malloc(
+      sizeof(grpc_resolved_address) * addresses->naddrs);
+  for (resp = result, i = 0; resp != nullptr; resp = resp->ai_next, i++) {
+    memcpy(&addresses->addrs[i].addr, resp->ai_addr, resp->ai_addrlen);
+    addresses->addrs[i].len = resp->ai_addrlen;
+  }
+  // addrinfo objects are allocated by libuv (e.g. in uv_getaddrinfo)
+  // and not by gpr_malloc
+  uv_freeaddrinfo(result);
+  return addresses;
+}
+
+static void uv_resolve_callback(uv_getaddrinfo_t* req, int status,
+                                struct addrinfo* res) {
+  grpc_custom_resolver* r = (grpc_custom_resolver*)req->data;
+  gpr_free(req);
+  grpc_resolved_addresses* result = nullptr;
+  if (status == 0) {
+    result = handle_addrinfo_result(res);
+  }
+  grpc_custom_resolve_callback(r, result,
+                               tcp_error_create("getaddrinfo failed", status));
+}
+
+static grpc_error* uv_resolve(const char* host, const char* port,
+                              grpc_resolved_addresses** result) {
+  int status;
+  uv_getaddrinfo_t req;
+  struct addrinfo hints;
+  memset(&hints, 0, sizeof(struct addrinfo));
+  hints.ai_family = AF_UNSPEC;     /* ipv4 or ipv6 */
+  hints.ai_socktype = SOCK_STREAM; /* stream socket */
+  hints.ai_flags = AI_PASSIVE;     /* for wildcard IP address */
+  status = uv_getaddrinfo(uv_default_loop(), &req, NULL, host, port, &hints);
+  if (status != 0) {
+    *result = nullptr;
+  } else {
+    *result = handle_addrinfo_result(req.addrinfo);
+  }
+  return tcp_error_create("getaddrinfo failed", status);
+}
+
+static void uv_resolve_async(grpc_custom_resolver* r, const char* host,
+                             const char* port) {
+  int status;
+  uv_getaddrinfo_t* req =
+      (uv_getaddrinfo_t*)gpr_malloc(sizeof(uv_getaddrinfo_t));
+  req->data = r;
+  struct addrinfo hints;
+  memset(&hints, 0, sizeof(struct addrinfo));
+  hints.ai_family = GRPC_AF_UNSPEC;     /* ipv4 or ipv6 */
+  hints.ai_socktype = GRPC_SOCK_STREAM; /* stream socket */
+  hints.ai_flags = GRPC_AI_PASSIVE;     /* for wildcard IP address */
+  status = uv_getaddrinfo(uv_default_loop(), req, uv_resolve_callback, host,
+                          port, &hints);
+  if (status != 0) {
+    gpr_free(req);
+    grpc_error* error = tcp_error_create("getaddrinfo failed", status);
+    grpc_custom_resolve_callback(r, NULL, error);
+  }
+}
+
+grpc_custom_resolver_vtable uv_resolver_vtable = {uv_resolve, uv_resolve_async};
+
+grpc_socket_vtable grpc_uv_socket_vtable = {
+    uv_socket_init,     uv_socket_connect,     uv_socket_destroy,
+    uv_socket_shutdown, uv_socket_close,       uv_socket_write,
+    uv_socket_read,     uv_socket_getpeername, uv_socket_getsockname,
+    uv_socket_bind,     uv_socket_listen,      uv_socket_accept};
+
+#endif

data/src/core/lib/iomgr/tcp_windows.cc

@@ -0,0 +1,533 @@
+/*
+ *
+ * Copyright 2015 gRPC authors.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ */
+
+#include <grpc/support/port_platform.h>
+
+#include "src/core/lib/iomgr/port.h"
+
+#ifdef GRPC_WINSOCK_SOCKET
+
+#include <limits.h>
+
+#include "src/core/lib/iomgr/sockaddr_windows.h"
+
+#include <grpc/slice_buffer.h>
+#include <grpc/support/alloc.h>
+#include <grpc/support/log.h>
+#include <grpc/support/log_windows.h>
+#include <grpc/support/string_util.h>
+
+#include "src/core/lib/gpr/useful.h"
+#include "src/core/lib/iomgr/iocp_windows.h"
+#include "src/core/lib/iomgr/sockaddr.h"
+#include "src/core/lib/iomgr/sockaddr_utils.h"
+#include "src/core/lib/iomgr/socket_windows.h"
+#include "src/core/lib/iomgr/tcp_client.h"
+#include "src/core/lib/iomgr/tcp_windows.h"
+#include "src/core/lib/iomgr/timer.h"
+#include "src/core/lib/slice/slice_internal.h"
+#include "src/core/lib/slice/slice_string_helpers.h"
+
+#if defined(__MSYS__) && defined(GPR_ARCH_64)
+/* Nasty workaround for nasty bug when using the 64 bits msys compiler
+   in conjunction with Microsoft Windows headers. */
+#define GRPC_FIONBIO _IOW('f', 126, uint32_t)
+#else
+#define GRPC_FIONBIO FIONBIO
+#endif
+
+extern grpc_core::TraceFlag grpc_tcp_trace;
+
+grpc_error* grpc_tcp_set_non_block(SOCKET sock) {
+  int status;
+  uint32_t param = 1;
+  DWORD ret;
+  status = WSAIoctl(sock, GRPC_FIONBIO, &param, sizeof(param), NULL, 0, &ret,
+                    NULL, NULL);
+  return status == 0
+             ? GRPC_ERROR_NONE
+             : GRPC_WSA_ERROR(WSAGetLastError(), "WSAIoctl(GRPC_FIONBIO)");
+}
+
+static grpc_error* set_dualstack(SOCKET sock) {
+  int status;
+  unsigned long param = 0;
+  status = setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, (const char*)&param,
+                      sizeof(param));
+  return status == 0
+             ? GRPC_ERROR_NONE
+             : GRPC_WSA_ERROR(WSAGetLastError(), "setsockopt(IPV6_V6ONLY)");
+}
+
+static grpc_error* enable_socket_low_latency(SOCKET sock) {
+  int status;
+  BOOL param = TRUE;
+  status = ::setsockopt(sock, IPPROTO_TCP, TCP_NODELAY,
+                        reinterpret_cast<char*>(&param), sizeof(param));
+  if (status == SOCKET_ERROR) {
+    status = WSAGetLastError();
+  }
+  return status == 0 ? GRPC_ERROR_NONE
+                     : GRPC_WSA_ERROR(status, "setsockopt(TCP_NODELAY)");
+}
+
+grpc_error* grpc_tcp_prepare_socket(SOCKET sock) {
+  grpc_error* err;
+  err = grpc_tcp_set_non_block(sock);
+  if (err != GRPC_ERROR_NONE) return err;
+  err = set_dualstack(sock);
+  if (err != GRPC_ERROR_NONE) return err;
+  err = enable_socket_low_latency(sock);
+  if (err != GRPC_ERROR_NONE) return err;
+  return GRPC_ERROR_NONE;
+}
+
+typedef struct grpc_tcp {
+  /* This is our C++ class derivation emulation. */
+  grpc_endpoint base;
+  /* The one socket this endpoint is using. */
+  grpc_winsocket* socket;
+  /* Refcounting how many operations are in progress. */
+  gpr_refcount refcount;
+
+  grpc_closure on_read;
+  grpc_closure on_write;
+
+  grpc_closure* read_cb;
+  grpc_closure* write_cb;
+
+  /* garbage after the last read */
+  grpc_slice_buffer last_read_buffer;
+
+  grpc_slice_buffer* write_slices;
+  grpc_slice_buffer* read_slices;
+
+  grpc_resource_user* resource_user;
+
+  /* The IO Completion Port runs from another thread. We need some mechanism
+     to protect ourselves when requesting a shutdown. */
+  gpr_mu mu;
+  int shutting_down;
+  grpc_error* shutdown_error;
+
+  char* peer_string;
+} grpc_tcp;
+
+static void tcp_free(grpc_tcp* tcp) {
+  grpc_winsocket_destroy(tcp->socket);
+  gpr_mu_destroy(&tcp->mu);
+  gpr_free(tcp->peer_string);
+  grpc_slice_buffer_destroy_internal(&tcp->last_read_buffer);
+  grpc_resource_user_unref(tcp->resource_user);
+  if (tcp->shutting_down) GRPC_ERROR_UNREF(tcp->shutdown_error);
+  gpr_free(tcp);
+}
+
+#ifndef NDEBUG
+#define TCP_UNREF(tcp, reason) tcp_unref((tcp), (reason), __FILE__, __LINE__)
+#define TCP_REF(tcp, reason) tcp_ref((tcp), (reason), __FILE__, __LINE__)
+static void tcp_unref(grpc_tcp* tcp, const char* reason, const char* file,
+                      int line) {
+  if (grpc_tcp_trace.enabled()) {
+    gpr_atm val = gpr_atm_no_barrier_load(&tcp->refcount.count);
+    gpr_log(file, line, GPR_LOG_SEVERITY_DEBUG,
+            "TCP unref %p : %s %" PRIdPTR " -> %" PRIdPTR, tcp, reason, val,
+            val - 1);
+  }
+  if (gpr_unref(&tcp->refcount)) {
+    tcp_free(tcp);
+  }
+}
+
+static void tcp_ref(grpc_tcp* tcp, const char* reason, const char* file,
+                    int line) {
+  if (grpc_tcp_trace.enabled()) {
+    gpr_atm val = gpr_atm_no_barrier_load(&tcp->refcount.count);
+    gpr_log(file, line, GPR_LOG_SEVERITY_DEBUG,
+            "TCP   ref %p : %s %" PRIdPTR " -> %" PRIdPTR, tcp, reason, val,
+            val + 1);
+  }
+  gpr_ref(&tcp->refcount);
+}
+#else
+#define TCP_UNREF(tcp, reason) tcp_unref((tcp))
+#define TCP_REF(tcp, reason) tcp_ref((tcp))
+static void tcp_unref(grpc_tcp* tcp) {
+  if (gpr_unref(&tcp->refcount)) {
+    tcp_free(tcp);
+  }
+}
+
+static void tcp_ref(grpc_tcp* tcp) { gpr_ref(&tcp->refcount); }
+#endif
+
+/* Asynchronous callback from the IOCP, or the background thread. */
+static void on_read(void* tcpp, grpc_error* error) {
+  grpc_tcp* tcp = (grpc_tcp*)tcpp;
+  grpc_closure* cb = tcp->read_cb;
+  grpc_winsocket* socket = tcp->socket;
+  grpc_winsocket_callback_info* info = &socket->read_info;
+
+  if (grpc_tcp_trace.enabled()) {
+    gpr_log(GPR_INFO, "TCP:%p on_read", tcp);
+  }
+
+  GRPC_ERROR_REF(error);
+
+  if (error == GRPC_ERROR_NONE) {
+    if (info->wsa_error != 0 && !tcp->shutting_down) {
+      char* utf8_message = gpr_format_message(info->wsa_error);
+      error = GRPC_ERROR_CREATE_FROM_COPIED_STRING(utf8_message);
+      gpr_free(utf8_message);
+      grpc_slice_buffer_reset_and_unref_internal(tcp->read_slices);
+    } else {
+      if (info->bytes_transferred != 0 && !tcp->shutting_down) {
+        GPR_ASSERT((size_t)info->bytes_transferred <= tcp->read_slices->length);
+        if (static_cast<size_t>(info->bytes_transferred) !=
+            tcp->read_slices->length) {
+          grpc_slice_buffer_trim_end(
+              tcp->read_slices,
+              tcp->read_slices->length -
+                  static_cast<size_t>(info->bytes_transferred),
+              &tcp->last_read_buffer);
+        }
+        GPR_ASSERT((size_t)info->bytes_transferred == tcp->read_slices->length);
+
+        if (grpc_tcp_trace.enabled()) {
+          size_t i;
+          for (i = 0; i < tcp->read_slices->count; i++) {
+            char* dump = grpc_dump_slice(tcp->read_slices->slices[i],
+                                         GPR_DUMP_HEX | GPR_DUMP_ASCII);
+            gpr_log(GPR_INFO, "READ %p (peer=%s): %s", tcp, tcp->peer_string,
+                    dump);
+            gpr_free(dump);
+          }
+        }
+      } else {
+        if (grpc_tcp_trace.enabled()) {
+          gpr_log(GPR_INFO, "TCP:%p unref read_slice", tcp);
+        }
+        grpc_slice_buffer_reset_and_unref_internal(tcp->read_slices);
+        error = tcp->shutting_down
+                    ? GRPC_ERROR_CREATE_REFERENCING_FROM_STATIC_STRING(
+                          "TCP stream shutting down", &tcp->shutdown_error, 1)
+                    : GRPC_ERROR_CREATE_FROM_STATIC_STRING("End of TCP stream");
+      }
+    }
+  }
+
+  tcp->read_cb = NULL;
+  TCP_UNREF(tcp, "read");
+  grpc_core::ExecCtx::Run(DEBUG_LOCATION, cb, error);
+}
+
+#define DEFAULT_TARGET_READ_SIZE 8192
+#define MAX_WSABUF_COUNT 16
+static void win_read(grpc_endpoint* ep, grpc_slice_buffer* read_slices,
+                     grpc_closure* cb, bool urgent) {
+  grpc_tcp* tcp = (grpc_tcp*)ep;
+  grpc_winsocket* handle = tcp->socket;
+  grpc_winsocket_callback_info* info = &handle->read_info;
+  int status;
+  DWORD bytes_read = 0;
+  DWORD flags = 0;
+  WSABUF buffers[MAX_WSABUF_COUNT];
+  size_t i;
+
+  if (grpc_tcp_trace.enabled()) {
+    gpr_log(GPR_INFO, "TCP:%p win_read", tcp);
+  }
+
+  if (tcp->shutting_down) {
+    grpc_core::ExecCtx::Run(
+        DEBUG_LOCATION, cb,
+        GRPC_ERROR_CREATE_REFERENCING_FROM_STATIC_STRING(
+            "TCP socket is shutting down", &tcp->shutdown_error, 1));
+    return;
+  }
+
+  tcp->read_cb = cb;
+  tcp->read_slices = read_slices;
+  grpc_slice_buffer_reset_and_unref_internal(read_slices);
+  grpc_slice_buffer_swap(read_slices, &tcp->last_read_buffer);
+
+  if (tcp->read_slices->length < DEFAULT_TARGET_READ_SIZE / 2 &&
+      tcp->read_slices->count < MAX_WSABUF_COUNT) {
+    // TODO(jtattermusch): slice should be allocated using resource quota
+    grpc_slice_buffer_add(tcp->read_slices,
+                          GRPC_SLICE_MALLOC(DEFAULT_TARGET_READ_SIZE));
+  }
+
+  GPR_ASSERT(tcp->read_slices->count <= MAX_WSABUF_COUNT);
+  for (i = 0; i < tcp->read_slices->count; i++) {
+    buffers[i].len = (ULONG)GRPC_SLICE_LENGTH(
+        tcp->read_slices->slices[i]);  // we know slice size fits in 32bit.
+    buffers[i].buf = (char*)GRPC_SLICE_START_PTR(tcp->read_slices->slices[i]);
+  }
+
+  TCP_REF(tcp, "read");
+
+  /* First let's try a synchronous, non-blocking read. */
+  status = WSARecv(tcp->socket->socket, buffers, (DWORD)tcp->read_slices->count,
+                   &bytes_read, &flags, NULL, NULL);
+  info->wsa_error = status == 0 ? 0 : WSAGetLastError();
+
+  /* Did we get data immediately ? Yay. */
+  if (info->wsa_error != WSAEWOULDBLOCK) {
+    info->bytes_transferred = bytes_read;
+    grpc_core::ExecCtx::Run(DEBUG_LOCATION, &tcp->on_read, GRPC_ERROR_NONE);
+    return;
+  }
+
+  /* Otherwise, let's retry, by queuing a read. */
+  memset(&tcp->socket->read_info.overlapped, 0, sizeof(OVERLAPPED));
+  status = WSARecv(tcp->socket->socket, buffers, (DWORD)tcp->read_slices->count,
+                   &bytes_read, &flags, &info->overlapped, NULL);
+
+  if (status != 0) {
+    int wsa_error = WSAGetLastError();
+    if (wsa_error != WSA_IO_PENDING) {
+      info->wsa_error = wsa_error;
+      grpc_core::ExecCtx::Run(DEBUG_LOCATION, &tcp->on_read,
+                              GRPC_WSA_ERROR(info->wsa_error, "WSARecv"));
+      return;
+    }
+  }
+
+  grpc_socket_notify_on_read(tcp->socket, &tcp->on_read);
+}
+
+/* Asynchronous callback from the IOCP, or the background thread. */
+static void on_write(void* tcpp, grpc_error* error) {
+  grpc_tcp* tcp = (grpc_tcp*)tcpp;
+  grpc_winsocket* handle = tcp->socket;
+  grpc_winsocket_callback_info* info = &handle->write_info;
+  grpc_closure* cb;
+
+  if (grpc_tcp_trace.enabled()) {
+    gpr_log(GPR_INFO, "TCP:%p on_write", tcp);
+  }
+
+  GRPC_ERROR_REF(error);
+
+  gpr_mu_lock(&tcp->mu);
+  cb = tcp->write_cb;
+  tcp->write_cb = NULL;
+  gpr_mu_unlock(&tcp->mu);
+
+  if (error == GRPC_ERROR_NONE) {
+    if (info->wsa_error != 0) {
+      error = GRPC_WSA_ERROR(info->wsa_error, "WSASend");
+    } else {
+      GPR_ASSERT(info->bytes_transferred == tcp->write_slices->length);
+    }
+  }
+
+  TCP_UNREF(tcp, "write");
+  grpc_core::ExecCtx::Run(DEBUG_LOCATION, cb, error);
+}
+
+/* Initiates a write. */
+static void win_write(grpc_endpoint* ep, grpc_slice_buffer* slices,
+                      grpc_closure* cb, void* arg) {
+  grpc_tcp* tcp = (grpc_tcp*)ep;
+  grpc_winsocket* socket = tcp->socket;
+  grpc_winsocket_callback_info* info = &socket->write_info;
+  unsigned i;
+  DWORD bytes_sent;
+  int status;
+  WSABUF local_buffers[MAX_WSABUF_COUNT];
+  WSABUF* allocated = NULL;
+  WSABUF* buffers = local_buffers;
+  size_t len;
+
+  if (grpc_tcp_trace.enabled()) {
+    size_t i;
+    for (i = 0; i < slices->count; i++) {
+      char* data =
+          grpc_dump_slice(slices->slices[i], GPR_DUMP_HEX | GPR_DUMP_ASCII);
+      gpr_log(GPR_INFO, "WRITE %p (peer=%s): %s", tcp, tcp->peer_string, data);
+      gpr_free(data);
+    }
+  }
+
+  if (tcp->shutting_down) {
+    grpc_core::ExecCtx::Run(
+        DEBUG_LOCATION, cb,
+        GRPC_ERROR_CREATE_REFERENCING_FROM_STATIC_STRING(
+            "TCP socket is shutting down", &tcp->shutdown_error, 1));
+    return;
+  }
+
+  tcp->write_cb = cb;
+  tcp->write_slices = slices;
+  GPR_ASSERT(tcp->write_slices->count <= UINT_MAX);
+  if (tcp->write_slices->count > GPR_ARRAY_SIZE(local_buffers)) {
+    buffers = (WSABUF*)gpr_malloc(sizeof(WSABUF) * tcp->write_slices->count);
+    allocated = buffers;
+  }
+
+  for (i = 0; i < tcp->write_slices->count; i++) {
+    len = GRPC_SLICE_LENGTH(tcp->write_slices->slices[i]);
+    GPR_ASSERT(len <= ULONG_MAX);
+    buffers[i].len = (ULONG)len;
+    buffers[i].buf = (char*)GRPC_SLICE_START_PTR(tcp->write_slices->slices[i]);
+  }
+
+  /* First, let's try a synchronous, non-blocking write. */
+  status = WSASend(socket->socket, buffers, (DWORD)tcp->write_slices->count,
+                   &bytes_sent, 0, NULL, NULL);
+  info->wsa_error = status == 0 ? 0 : WSAGetLastError();
+
+  /* We would kind of expect to get a WSAEWOULDBLOCK here, especially on a busy
+     connection that has its send queue filled up. But if we don't, then we can
+     avoid doing an async write operation at all. */
+  if (info->wsa_error != WSAEWOULDBLOCK) {
+    grpc_error* error = status == 0
+                            ? GRPC_ERROR_NONE
+                            : GRPC_WSA_ERROR(info->wsa_error, "WSASend");
+    grpc_core::ExecCtx::Run(DEBUG_LOCATION, cb, error);
+    if (allocated) gpr_free(allocated);
+    return;
+  }
+
+  TCP_REF(tcp, "write");
+
+  /* If we got a WSAEWOULDBLOCK earlier, then we need to re-do the same
+     operation, this time asynchronously. */
+  memset(&socket->write_info.overlapped, 0, sizeof(OVERLAPPED));
+  status = WSASend(socket->socket, buffers, (DWORD)tcp->write_slices->count,
+                   &bytes_sent, 0, &socket->write_info.overlapped, NULL);
+  if (allocated) gpr_free(allocated);
+
+  if (status != 0) {
+    int wsa_error = WSAGetLastError();
+    if (wsa_error != WSA_IO_PENDING) {
+      TCP_UNREF(tcp, "write");
+      grpc_core::ExecCtx::Run(DEBUG_LOCATION, cb,
+                              GRPC_WSA_ERROR(wsa_error, "WSASend"));
+      return;
+    }
+  }
+
+  /* As all is now setup, we can now ask for the IOCP notification. It may
+     trigger the callback immediately however, but no matter. */
+  grpc_socket_notify_on_write(socket, &tcp->on_write);
+}
+
+static void win_add_to_pollset(grpc_endpoint* ep, grpc_pollset* ps) {
+  grpc_tcp* tcp;
+  (void)ps;
+  tcp = (grpc_tcp*)ep;
+  grpc_iocp_add_socket(tcp->socket);
+}
+
+static void win_add_to_pollset_set(grpc_endpoint* ep, grpc_pollset_set* pss) {
+  grpc_tcp* tcp;
+  (void)pss;
+  tcp = (grpc_tcp*)ep;
+  grpc_iocp_add_socket(tcp->socket);
+}
+
+static void win_delete_from_pollset_set(grpc_endpoint* ep,
+                                        grpc_pollset_set* pss) {}
+
+/* Initiates a shutdown of the TCP endpoint. This will queue abort callbacks
+   for the potential read and write operations. It is up to the caller to
+   guarantee this isn't called in parallel to a read or write request, so
+   we're not going to protect against these. However the IO Completion Port
+   callback will happen from another thread, so we need to protect against
+   concurrent access of the data structure in that regard. */
+static void win_shutdown(grpc_endpoint* ep, grpc_error* why) {
+  grpc_tcp* tcp = (grpc_tcp*)ep;
+  gpr_mu_lock(&tcp->mu);
+  /* At that point, what may happen is that we're already inside the IOCP
+     callback. See the comments in on_read and on_write. */
+  if (!tcp->shutting_down) {
+    tcp->shutting_down = 1;
+    tcp->shutdown_error = why;
+  } else {
+    GRPC_ERROR_UNREF(why);
+  }
+  grpc_winsocket_shutdown(tcp->socket);
+  gpr_mu_unlock(&tcp->mu);
+  grpc_resource_user_shutdown(tcp->resource_user);
+}
+
+static void win_destroy(grpc_endpoint* ep) {
+  grpc_tcp* tcp = (grpc_tcp*)ep;
+  grpc_slice_buffer_reset_and_unref_internal(&tcp->last_read_buffer);
+  TCP_UNREF(tcp, "destroy");
+}
+
+static char* win_get_peer(grpc_endpoint* ep) {
+  grpc_tcp* tcp = (grpc_tcp*)ep;
+  return gpr_strdup(tcp->peer_string);
+}
+
+static grpc_resource_user* win_get_resource_user(grpc_endpoint* ep) {
+  grpc_tcp* tcp = (grpc_tcp*)ep;
+  return tcp->resource_user;
+}
+
+static int win_get_fd(grpc_endpoint* ep) { return -1; }
+
+static bool win_can_track_err(grpc_endpoint* ep) { return false; }
+
+static grpc_endpoint_vtable vtable = {win_read,
+                                      win_write,
+                                      win_add_to_pollset,
+                                      win_add_to_pollset_set,
+                                      win_delete_from_pollset_set,
+                                      win_shutdown,
+                                      win_destroy,
+                                      win_get_resource_user,
+                                      win_get_peer,
+                                      win_get_fd,
+                                      win_can_track_err};
+
+grpc_endpoint* grpc_tcp_create(grpc_winsocket* socket,
+                               grpc_channel_args* channel_args,
+                               const char* peer_string) {
+  grpc_resource_quota* resource_quota = grpc_resource_quota_create(NULL);
+  if (channel_args != NULL) {
+    for (size_t i = 0; i < channel_args->num_args; i++) {
+      if (0 == strcmp(channel_args->args[i].key, GRPC_ARG_RESOURCE_QUOTA)) {
+        grpc_resource_quota_unref_internal(resource_quota);
+        resource_quota = grpc_resource_quota_ref_internal(
+            (grpc_resource_quota*)channel_args->args[i].value.pointer.p);
+      }
+    }
+  }
+  grpc_tcp* tcp = (grpc_tcp*)gpr_malloc(sizeof(grpc_tcp));
+  memset(tcp, 0, sizeof(grpc_tcp));
+  tcp->base.vtable = &vtable;
+  tcp->socket = socket;
+  gpr_mu_init(&tcp->mu);
+  gpr_ref_init(&tcp->refcount, 1);
+  GRPC_CLOSURE_INIT(&tcp->on_read, on_read, tcp, grpc_schedule_on_exec_ctx);
+  GRPC_CLOSURE_INIT(&tcp->on_write, on_write, tcp, grpc_schedule_on_exec_ctx);
+  tcp->peer_string = gpr_strdup(peer_string);
+  grpc_slice_buffer_init(&tcp->last_read_buffer);
+  tcp->resource_user = grpc_resource_user_create(resource_quota, peer_string);
+  grpc_resource_quota_unref_internal(resource_quota);
+
+  return &tcp->base;
+}
+
+#endif /* GRPC_WINSOCK_SOCKET */