Add support for TCP_DEFER_ACCEPT.

PiperOrigin-RevId: 292233574

2 files changed, 211 insertions, 0 deletions

diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc
index 2f9821555..3bf7081b9 100644
--- a/test/syscalls/linux/socket_inet_loopback.cc
+++ b/test/syscalls/linux/socket_inet_loopback.cc
@@ -828,6 +828,164 @@ TEST_P(SocketInetLoopbackTest, AcceptedInheritsTCPUserTimeout) {
   EXPECT_EQ(get, kUserTimeout);
 }
 
+// TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not
+// saved. Enable S/R once issue is fixed.
+TEST_P(SocketInetLoopbackTest, TCPDeferAccept_NoRandomSave) {
+  // TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not
+  // saved. Enable S/R issue is fixed.
+  DisableSave ds;
+
+  auto const& param = GetParam();
+  TestAddress const& listener = param.listener;
+  TestAddress const& connector = param.connector;
+
+  // Create the listening socket.
+  const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP));
+  sockaddr_storage listen_addr = listener.addr;
+  ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast<sockaddr*>(&listen_addr),
+                   listener.addr_len),
+              SyscallSucceeds());
+  ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds());
+
+  // Get the port bound by the listening socket.
+  socklen_t addrlen = listener.addr_len;
+  ASSERT_THAT(getsockname(listen_fd.get(),
+                          reinterpret_cast<sockaddr*>(&listen_addr), &addrlen),
+              SyscallSucceeds());
+
+  const uint16_t port =
+      ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr));
+
+  // Set the TCP_DEFER_ACCEPT on the listening socket.
+  constexpr int kTCPDeferAccept = 3;
+  ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT,
+                         &kTCPDeferAccept, sizeof(kTCPDeferAccept)),
+              SyscallSucceeds());
+
+  // Connect to the listening socket.
+  FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP));
+
+  sockaddr_storage conn_addr = connector.addr;
+  ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port));
+  ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(),
+                                  reinterpret_cast<sockaddr*>(&conn_addr),
+                                  connector.addr_len),
+              SyscallSucceeds());
+
+  // Set the listening socket to nonblock so that we can verify that there is no
+  // connection in queue despite the connect above succeeding since the peer has
+  // sent no data and TCP_DEFER_ACCEPT is set on the listening socket. Set the
+  // FD to O_NONBLOCK.
+  int opts;
+  ASSERT_THAT(opts = fcntl(listen_fd.get(), F_GETFL), SyscallSucceeds());
+  opts |= O_NONBLOCK;
+  ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds());
+
+  ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr),
+              SyscallFailsWithErrno(EWOULDBLOCK));
+
+  // Set FD back to blocking.
+  opts &= ~O_NONBLOCK;
+  ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds());
+
+  // Now write some data to the socket.
+  int data = 0;
+  ASSERT_THAT(RetryEINTR(write)(conn_fd.get(), &data, sizeof(data)),
+              SyscallSucceedsWithValue(sizeof(data)));
+
+  // This should now cause the connection to complete and be delivered to the
+  // accept socket.
+
+  // Accept the connection.
+  auto accepted =
+      ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr));
+
+  // Verify that the accepted socket returns the data written.
+  int get = -1;
+  ASSERT_THAT(RetryEINTR(recv)(accepted.get(), &get, sizeof(get), 0),
+              SyscallSucceedsWithValue(sizeof(get)));
+
+  EXPECT_EQ(get, data);
+}
+
+// TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not
+// saved. Enable S/R once issue is fixed.
+TEST_P(SocketInetLoopbackTest, TCPDeferAcceptTimeout_NoRandomSave) {
+  // TODO(gvisor.dev/issue/1688): Partially completed passive endpoints are not
+  // saved. Enable S/R once issue is fixed.
+  DisableSave ds;
+
+  auto const& param = GetParam();
+  TestAddress const& listener = param.listener;
+  TestAddress const& connector = param.connector;
+
+  // Create the listening socket.
+  const FileDescriptor listen_fd = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP));
+  sockaddr_storage listen_addr = listener.addr;
+  ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast<sockaddr*>(&listen_addr),
+                   listener.addr_len),
+              SyscallSucceeds());
+  ASSERT_THAT(listen(listen_fd.get(), SOMAXCONN), SyscallSucceeds());
+
+  // Get the port bound by the listening socket.
+  socklen_t addrlen = listener.addr_len;
+  ASSERT_THAT(getsockname(listen_fd.get(),
+                          reinterpret_cast<sockaddr*>(&listen_addr), &addrlen),
+              SyscallSucceeds());
+
+  const uint16_t port =
+      ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr));
+
+  // Set the TCP_DEFER_ACCEPT on the listening socket.
+  constexpr int kTCPDeferAccept = 3;
+  ASSERT_THAT(setsockopt(listen_fd.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT,
+                         &kTCPDeferAccept, sizeof(kTCPDeferAccept)),
+              SyscallSucceeds());
+
+  // Connect to the listening socket.
+  FileDescriptor conn_fd = ASSERT_NO_ERRNO_AND_VALUE(
+      Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP));
+
+  sockaddr_storage conn_addr = connector.addr;
+  ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port));
+  ASSERT_THAT(RetryEINTR(connect)(conn_fd.get(),
+                                  reinterpret_cast<sockaddr*>(&conn_addr),
+                                  connector.addr_len),
+              SyscallSucceeds());
+
+  // Set the listening socket to nonblock so that we can verify that there is no
+  // connection in queue despite the connect above succeeding since the peer has
+  // sent no data and TCP_DEFER_ACCEPT is set on the listening socket. Set the
+  // FD to O_NONBLOCK.
+  int opts;
+  ASSERT_THAT(opts = fcntl(listen_fd.get(), F_GETFL), SyscallSucceeds());
+  opts |= O_NONBLOCK;
+  ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds());
+
+  // Verify that there is no acceptable connection before TCP_DEFER_ACCEPT
+  // timeout is hit.
+  absl::SleepFor(absl::Seconds(kTCPDeferAccept - 1));
+  ASSERT_THAT(accept(listen_fd.get(), nullptr, nullptr),
+              SyscallFailsWithErrno(EWOULDBLOCK));
+
+  // Set FD back to blocking.
+  opts &= ~O_NONBLOCK;
+  ASSERT_THAT(fcntl(listen_fd.get(), F_SETFL, opts), SyscallSucceeds());
+
+  // Now sleep for a little over the TCP_DEFER_ACCEPT duration. When the timeout
+  // is hit a SYN-ACK should be retransmitted by the listener as a last ditch
+  // attempt to complete the connection with or without data.
+  absl::SleepFor(absl::Seconds(2));
+
+  // Verify that we have a connection that can be accepted even though no
+  // data was written.
+  auto accepted =
+      ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr));
+}
+
 INSTANTIATE_TEST_SUITE_P(
     All, SocketInetLoopbackTest,
     ::testing::Values(
diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc
index 33a5ac66c..525ccbd88 100644
--- a/test/syscalls/linux/tcp_socket.cc
+++ b/test/syscalls/linux/tcp_socket.cc
@@ -1286,6 +1286,59 @@ TEST_P(SimpleTcpSocketTest, SetTCPUserTimeout) {
   EXPECT_EQ(get, kTCPUserTimeout);
 }
 
+TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptNeg) {
+  FileDescriptor s =
+      ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP));
+
+  // -ve TCP_DEFER_ACCEPT is same as setting it to zero.
+  constexpr int kNeg = -1;
+  EXPECT_THAT(
+      setsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &kNeg, sizeof(kNeg)),
+      SyscallSucceeds());
+  int get = -1;
+  socklen_t get_len = sizeof(get);
+  ASSERT_THAT(
+      getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len),
+      SyscallSucceedsWithValue(0));
+  EXPECT_EQ(get_len, sizeof(get));
+  EXPECT_EQ(get, 0);
+}
+
+TEST_P(SimpleTcpSocketTest, GetTCPDeferAcceptDefault) {
+  FileDescriptor s =
+      ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP));
+
+  int get = -1;
+  socklen_t get_len = sizeof(get);
+  ASSERT_THAT(
+      getsockopt(s.get(), IPPROTO_TCP, TCP_USER_TIMEOUT, &get, &get_len),
+      SyscallSucceedsWithValue(0));
+  EXPECT_EQ(get_len, sizeof(get));
+  EXPECT_EQ(get, 0);
+}
+
+TEST_P(SimpleTcpSocketTest, SetTCPDeferAcceptGreaterThanZero) {
+  FileDescriptor s =
+      ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP));
+  // kTCPDeferAccept is in seconds.
+  // NOTE: linux translates seconds to # of retries and back from
+  //   #of retries to seconds. Which means only certain values
+  //   translate back exactly. That's why we use 3 here, a value of
+  //   5 will result in us getting back 7 instead of 5 in the
+  //   getsockopt.
+  constexpr int kTCPDeferAccept = 3;
+  ASSERT_THAT(setsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT,
+                         &kTCPDeferAccept, sizeof(kTCPDeferAccept)),
+              SyscallSucceeds());
+  int get = -1;
+  socklen_t get_len = sizeof(get);
+  ASSERT_THAT(
+      getsockopt(s.get(), IPPROTO_TCP, TCP_DEFER_ACCEPT, &get, &get_len),
+      SyscallSucceeds());
+  EXPECT_EQ(get_len, sizeof(get));
+  EXPECT_EQ(get, kTCPDeferAccept);
+}
+
 INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest,
                          ::testing::Values(AF_INET, AF_INET6));