Use os_sys_calls in all code paths to replace raw syscalls

include/envoy/common/platform.h

@@ -29,6 +29,20 @@
 #include <io.h>
 #include <stdint.h>
 
+#define htole16(x) (x)
+#define htole32(x) (x)
+#define htole64(x) (x)
+#define le16toh(x) (x)
+#define le32toh(x) (x)
+#define le64toh(x) (x)
+
+#define htobe16(x) htons((x))
+#define htobe32(x) htonl((x))
+#define htobe64(x) htonll((x))
+#define be16toh(x) ntohs((x))
+#define be32toh(x) ntohl((x))
+#define be64toh(x) ntohll((x))
+
 #define PACKED_STRUCT(definition, ...)                                                             \
   __pragma(pack(push, 1)) definition, ##__VA_ARGS__;                                               \
   __pragma(pack(pop))
@@ -122,6 +136,24 @@ struct msghdr {
 #include <sys/wait.h>
 #include <unistd.h>
 
+#ifdef __APPLE__
+#include <libkern/OSByteOrder.h>
+#define htole16(x) OSSwapHostToLittleInt16((x))
+#define htole32(x) OSSwapHostToLittleInt32((x))
+#define htole64(x) OSSwapHostToLittleInt64((x))
+#define le16toh(x) OSSwapLittleToHostInt16((x))
+#define le32toh(x) OSSwapLittleToHostInt32((x))
+#define le64toh(x) OSSwapLittleToHostInt64((x))
+#define htobe16(x) OSSwapHostToBigInt16((x))
+#define htobe32(x) OSSwapHostToBigInt32((x))
+#define htobe64(x) OSSwapHostToBigInt64((x))
+#define be16toh(x) OSSwapBigToHostInt16((x))
+#define be32toh(x) OSSwapBigToHostInt32((x))
+#define be64toh(x) OSSwapBigToHostInt64((x))
+#else
+#include <endian.h>
+#endif
+
 #if defined(__linux__)
 #include <linux/netfilter_ipv4.h>
 #endif

source/common/common/byte_order.h

@@ -3,47 +3,9 @@
 #include <cstddef>
 #include <cstdint>
 
-// NOLINT(namespace-envoy)
-
-#ifdef __APPLE__
-
-#include <libkern/OSByteOrder.h>
-
-#define htole16(x) OSSwapHostToLittleInt16((x))
-#define htole32(x) OSSwapHostToLittleInt32((x))
-#define htole64(x) OSSwapHostToLittleInt64((x))
-#define le16toh(x) OSSwapLittleToHostInt16((x))
-#define le32toh(x) OSSwapLittleToHostInt32((x))
-#define le64toh(x) OSSwapLittleToHostInt64((x))
-
-#define htobe16(x) OSSwapHostToBigInt16((x))
-#define htobe32(x) OSSwapHostToBigInt32((x))
-#define htobe64(x) OSSwapHostToBigInt64((x))
-#define be16toh(x) OSSwapBigToHostInt16((x))
-#define be32toh(x) OSSwapBigToHostInt32((x))
-#define be64toh(x) OSSwapBigToHostInt64((x))
-
-#elif WIN32
-
 #include "envoy/common/platform.h"
 
-#define htole16(x) (x)
-#define htole32(x) (x)
-#define htole64(x) (x)
-#define le16toh(x) (x)
-#define le32toh(x) (x)
-#define le64toh(x) (x)
-
-#define htobe16(x) htons((x))
-#define htobe32(x) htonl((x))
-#define htobe64(x) htonll((x))
-#define be16toh(x) ntohs((x))
-#define be32toh(x) ntohl((x))
-#define be64toh(x) ntohll((x))
-
-#else
-#include <endian.h>
-#endif
+// NOLINT(namespace-envoy)
 
 enum class ByteOrder { Host, LittleEndian, BigEndian };
 

source/common/network/address_impl.cc

@@ -77,6 +77,10 @@ Address::InstanceConstSharedPtr addressFromSockAddr(const sockaddr_storage& ss,
 #if defined(__APPLE__)
       struct sockaddr_in sin = {
           {}, AF_INET, sin6->sin6_port, {sin6->sin6_addr.__u6_addr.__u6_addr32[3]}, {}};
+#elif defined(WIN32)
+      struct in_addr in_v4 = {};
+      in_v4.S_un.S_addr = reinterpret_cast<const uint32_t*>(sin6->sin6_addr.u.Byte)[3];
+      struct sockaddr_in sin = {AF_INET, sin6->sin6_port, in_v4, {}};
 #else
       struct sockaddr_in sin = {AF_INET, sin6->sin6_port, {sin6->sin6_addr.s6_addr32[3]}, {}};
 #endif
@@ -102,38 +106,57 @@ Address::InstanceConstSharedPtr addressFromSockAddr(const sockaddr_storage& ss,
 InstanceConstSharedPtr addressFromFd(os_fd_t fd) {
   sockaddr_storage ss;
   socklen_t ss_len = sizeof ss;
-  int rc = ::getsockname(fd, reinterpret_cast<sockaddr*>(&ss), &ss_len);
-  if (rc != 0) {
-    throw EnvoyException(
-        fmt::format("getsockname failed for '{}': ({}) {}", fd, errno, strerror(errno)));
+  auto& os_sys_calls = Api::OsSysCallsSingleton::get();
+  Api::SysCallIntResult result =
+      os_sys_calls.getsockname(fd, reinterpret_cast<sockaddr*>(&ss), &ss_len);
+  if (result.rc_ != 0) {
+    throw EnvoyException(fmt::format("getsockname failed for '{}': ({}) {}", fd, result.errno_,
+                                     strerror(result.errno_)));
   }
   int socket_v6only = 0;
   if (ss.ss_family == AF_INET6) {
     socklen_t size_int = sizeof(socket_v6only);
-    RELEASE_ASSERT(::getsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &socket_v6only, &size_int) == 0, "");
+    result = os_sys_calls.getsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &socket_v6only, &size_int);
+#ifdef WIN32
+    // On Windows, it is possible for this getsockopt() call to fail.
+    // This can happen if the address we are trying to connect to has nothing
+    // listening. So we can't use RELEASE_ASSERT and instead must throw an
+    // exception
+    if (SOCKET_FAILURE(result.rc_)) {
+      throw EnvoyException(fmt::format("getsockopt failed for '{}': ({}) {}", fd, result.errno_,
+                                       strerror(result.errno_)));
+    }
+#else
+    RELEASE_ASSERT(result.rc_ == 0, "");
+#endif
   }
-  return addressFromSockAddr(ss, ss_len, rc == 0 && socket_v6only);
+  return addressFromSockAddr(ss, ss_len, socket_v6only);
 }
 
 InstanceConstSharedPtr peerAddressFromFd(os_fd_t fd) {
   sockaddr_storage ss;
   socklen_t ss_len = sizeof ss;
-  const int rc = ::getpeername(fd, reinterpret_cast<sockaddr*>(&ss), &ss_len);
-  if (rc != 0) {
-    throw EnvoyException(fmt::format("getpeername failed for '{}': {}", fd, strerror(errno)));
+  auto& os_sys_calls = Api::OsSysCallsSingleton::get();
+  Api::SysCallIntResult result =
+      os_sys_calls.getpeername(fd, reinterpret_cast<sockaddr*>(&ss), &ss_len);
+  if (result.rc_ != 0) {
+    throw EnvoyException(
+        fmt::format("getpeername failed for '{}': {}", fd, strerror(result.errno_)));
   }
 #ifdef __APPLE__
-  if (ss_len == sizeof(sockaddr) && ss.ss_family == AF_UNIX) {
+  if (ss_len == sizeof(sockaddr) && ss.ss_family == AF_UNIX)
 #else
-  if (ss_len == sizeof(sa_family_t) && ss.ss_family == AF_UNIX) {
+  if (ss_len == sizeof(sa_family_t) && ss.ss_family == AF_UNIX)
 #endif
+  {
     // For Unix domain sockets, can't find out the peer name, but it should match our own
     // name for the socket (i.e. the path should match, barring any namespace or other
     // mechanisms to hide things, of which there are many).
     ss_len = sizeof ss;
-    const int rc = ::getsockname(fd, reinterpret_cast<sockaddr*>(&ss), &ss_len);
-    if (rc != 0) {
-      throw EnvoyException(fmt::format("getsockname failed for '{}': {}", fd, strerror(errno)));
+    result = os_sys_calls.getsockname(fd, reinterpret_cast<sockaddr*>(&ss), &ss_len);
+    if (result.rc_ != 0) {
+      throw EnvoyException(
+          fmt::format("getsockname failed for '{}': {}", fd, strerror(result.errno_)));
     }
   }
   return addressFromSockAddr(ss, ss_len);
@@ -166,14 +189,16 @@ IoHandlePtr InstanceBase::socketFromSocketType(SocketType socket_type) const {
     domain = AF_UNIX;
   }
 
-  const Api::SysCallSocketResult result = Api::OsSysCallsSingleton::get().socket(domain, flags, 0);
+  auto os_sys_calls = Api::OsSysCallsSingleton::get();
+  const Api::SysCallSocketResult result = os_sys_calls.socket(domain, flags, 0);
   RELEASE_ASSERT(SOCKET_VALID(result.rc_),
                  fmt::format("socket(2) failed, got error: {}", strerror(result.errno_)));
   IoHandlePtr io_handle = std::make_unique<IoSocketHandleImpl>(result.rc_);
 
 #ifdef __APPLE__
   // Cannot set SOCK_NONBLOCK as a ::socket flag.
-  RELEASE_ASSERT(fcntl(io_handle->fd(), F_SETFL, O_NONBLOCK) != -1, "");
+  const int rc = os_sys_calls.setsocketblocking(io_handle->fd(), false).rc_;
+  RELEASE_ASSERT(!SOCKET_FAILURE(rc), "");
 #endif
 
   return io_handle;
@@ -230,8 +255,7 @@ Api::SysCallIntResult Ipv4Instance::bind(os_fd_t fd) const {
 }
 
 Api::SysCallIntResult Ipv4Instance::connect(os_fd_t fd) const {
-  const int rc = ::connect(fd, sockAddr(), sockAddrLen());
-  return {rc, errno};
+  return Api::OsSysCallsSingleton::get().connect(fd, sockAddr(), sockAddrLen());
 }
 
 IoHandlePtr Ipv4Instance::socket(SocketType type) const { return socketFromSocketType(type); }
@@ -319,16 +343,17 @@ Api::SysCallIntResult Ipv6Instance::bind(os_fd_t fd) const {
 }
 
 Api::SysCallIntResult Ipv6Instance::connect(os_fd_t fd) const {
-  const int rc = ::connect(fd, sockAddr(), sockAddrLen());
-  return {rc, errno};
+  return Api::OsSysCallsSingleton::get().connect(fd, sockAddr(), sockAddrLen());
 }
 
 IoHandlePtr Ipv6Instance::socket(SocketType type) const {
   IoHandlePtr io_handle = socketFromSocketType(type);
   // Setting IPV6_V6ONLY restricts the IPv6 socket to IPv6 connections only.
   const int v6only = ip_.v6only_;
-  RELEASE_ASSERT(
-      ::setsockopt(io_handle->fd(), IPPROTO_IPV6, IPV6_V6ONLY, &v6only, sizeof(v6only)) != -1, "");
+  const Api::SysCallIntResult result = Api::OsSysCallsSingleton::get().setsockopt(
+      io_handle->fd(), IPPROTO_IPV6, IPV6_V6ONLY, reinterpret_cast<const char*>(&v6only),
+      sizeof(v6only));
+  RELEASE_ASSERT(!SOCKET_FAILURE(result.rc_), "");
   return io_handle;
 }
 
@@ -416,8 +441,7 @@ Api::SysCallIntResult PipeInstance::bind(os_fd_t fd) const {
 }
 
 Api::SysCallIntResult PipeInstance::connect(os_fd_t fd) const {
-  const int rc = ::connect(fd, sockAddr(), sockAddrLen());
-  return {rc, errno};
+  return Api::OsSysCallsSingleton::get().connect(fd, sockAddr(), sockAddrLen());
 }
 
 IoHandlePtr PipeInstance::socket(SocketType type) const { return socketFromSocketType(type); }

source/common/network/connection_impl.cc

@@ -10,6 +10,7 @@
 #include "envoy/event/timer.h"
 #include "envoy/network/filter.h"
 
+#include "common/api/os_sys_calls_impl.h"
 #include "common/common/assert.h"
 #include "common/common/empty_string.h"
 #include "common/common/enum_to_int.h"
@@ -54,17 +55,23 @@ ConnectionImpl::ConnectionImpl(Event::Dispatcher& dispatcher, ConnectionSocketPt
       write_end_stream_(false), current_write_end_stream_(false), dispatch_buffered_data_(false) {
   // Treat the lack of a valid fd (which in practice only happens if we run out of FDs) as an OOM
   // condition and just crash.
-  RELEASE_ASSERT(ioHandle().fd() != -1, "");
+  RELEASE_ASSERT(SOCKET_VALID(ioHandle().fd()), "");
 
   if (!connected) {
     connecting_ = true;
   }
 
+  // Libevent only supports Level trigger on Windows.
+#ifdef WIN32
+  Event::FileTriggerType trigger = Event::FileTriggerType::Level;
+#else
+  Event::FileTriggerType trigger = Event::FileTriggerType::Edge;
+#endif
   // We never ask for both early close and read at the same time. If we are reading, we want to
   // consume all available data.
   file_event_ = dispatcher_.createFileEvent(
-      ioHandle().fd(), [this](uint32_t events) -> void { onFileEvent(events); },
-      Event::FileTriggerType::Edge, Event::FileReadyType::Read | Event::FileReadyType::Write);
+      ioHandle().fd(), [this](uint32_t events) -> void { onFileEvent(events); }, trigger,
+      Event::FileReadyType::Read | Event::FileReadyType::Write);
 
   transport_socket_->setTransportSocketCallbacks(*this);
 }
@@ -205,6 +212,10 @@ void ConnectionImpl::closeSocket(ConnectionEvent close_type) {
   connection_stats_.reset();
 
   file_event_.reset();
+
+  // Shutdown the socket so the other side can read any data in flight.
+  Api::OsSysCallsSingleton::get().shutdown(ioHandle().fd(), ENVOY_SHUT_WR);
+
   socket_->close();
 
   raiseEvent(close_type);
@@ -223,27 +234,39 @@ void ConnectionImpl::noDelay(bool enable) {
   }
 
   // Don't set NODELAY for unix domain sockets
-  sockaddr addr;
+  sockaddr_storage addr;
   socklen_t len = sizeof(addr);
-  int rc = getsockname(ioHandle().fd(), &addr, &len);
-  RELEASE_ASSERT(rc == 0, "");
 
-  if (addr.sa_family == AF_UNIX) {
+  auto os_sys_calls = Api::OsSysCallsSingleton::get();
+  Api::SysCallIntResult result =
+      os_sys_calls.getsockname(ioHandle().fd(), reinterpret_cast<struct sockaddr*>(&addr), &len);
+
+  RELEASE_ASSERT(result.rc_ == 0, "");
+
+  if (addr.ss_family == AF_UNIX) {
     return;
   }
 
   // Set NODELAY
   int new_value = enable;
-  rc = setsockopt(ioHandle().fd(), IPPROTO_TCP, TCP_NODELAY, &new_value, sizeof(new_value));
-#ifdef __APPLE__
-  if (-1 == rc && errno == EINVAL) {
+  result = os_sys_calls.setsockopt(ioHandle().fd(), IPPROTO_TCP, TCP_NODELAY, &new_value,
+                                   sizeof(new_value));
+#if defined(__APPLE__)
+  if (SOCKET_FAILURE(result.rc_) && result.errno_ == EINVAL) {
+    // Sometimes occurs when the connection is not yet fully formed. Empirically, TCP_NODELAY is
+    // enabled despite this result.
+    return;
+  }
+#elif defined(WIN32)
+  if (SOCKET_FAILURE(result.rc_) &&
+      (result.errno_ == WSAEWOULDBLOCK || result.errno_ == WSAEINVAL)) {
     // Sometimes occurs when the connection is not yet fully formed. Empirically, TCP_NODELAY is
     // enabled despite this result.
     return;
   }
 #endif
 
-  RELEASE_ASSERT(0 == rc, "");
+  RELEASE_ASSERT(result.rc_ == 0, "");
 }
 
 void ConnectionImpl::onRead(uint64_t read_buffer_size) {
@@ -562,8 +585,10 @@ void ConnectionImpl::onWriteReady() {
   if (connecting_) {
     int error;
     socklen_t error_size = sizeof(error);
-    int rc = getsockopt(ioHandle().fd(), SOL_SOCKET, SO_ERROR, &error, &error_size);
-    ASSERT(0 == rc);
+    RELEASE_ASSERT(Api::OsSysCallsSingleton::get()
+                           .getsockopt(ioHandle().fd(), SOL_SOCKET, SO_ERROR, &error, &error_size)
+                           .rc_ == 0,
+                   "");
 
     if (error == 0) {
       ENVOY_CONN_LOG(debug, "connected", *this);