12 files changed, 617 insertions, 0 deletions
diff --git a/test/packetdrill/BUILD b/test/packetdrill/BUILD
new file mode 100644
index 000000000..49642f282
--- /dev/null
+++ b/test/packetdrill/BUILD
@@ -0,0 +1,45 @@
+load("//tools:defs.bzl", "bzl_library")
+load("//test/packetdrill:defs.bzl", "packetdrill_test")
+
+package(licenses = ["notice"])
+
+packetdrill_test(
+    name = "packetdrill_sanity_test",
+    scripts = ["sanity_test.pkt"],
+)
+
+packetdrill_test(
+    name = "accept_ack_drop_test",
+    scripts = ["accept_ack_drop.pkt"],
+)
+
+packetdrill_test(
+    name = "fin_wait2_timeout_test",
+    scripts = ["fin_wait2_timeout.pkt"],
+)
+
+packetdrill_test(
+    name = "listen_close_before_handshake_complete_test",
+    scripts = ["listen_close_before_handshake_complete.pkt"],
+)
+
+packetdrill_test(
+    name = "no_rst_to_rst_test",
+    scripts = ["no_rst_to_rst.pkt"],
+)
+
+packetdrill_test(
+    name = "tcp_defer_accept_test",
+    scripts = ["tcp_defer_accept.pkt"],
+)
+
+packetdrill_test(
+    name = "tcp_defer_accept_timeout_test",
+    scripts = ["tcp_defer_accept_timeout.pkt"],
+)
+
+bzl_library(
+    name = "defs_bzl",
+    srcs = ["defs.bzl"],
+    visibility = ["//visibility:private"],
+)
diff --git a/test/packetdrill/accept_ack_drop.pkt b/test/packetdrill/accept_ack_drop.pkt
new file mode 100644
index 000000000..76e638fd4
--- /dev/null
+++ b/test/packetdrill/accept_ack_drop.pkt
@@ -0,0 +1,27 @@
+// Test that the accept works if the final ACK is dropped and an ack with data
+// follows the dropped ack.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0.0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0.0 > S. 0:0(0) ack 1    <...>
+
++0.0 < . 1:5(4) ack 1 win 257
++0.0 > . 1:1(0) ack 5 <...>
+
+// This should cause connection to transition to connected state.
++0.000 accept(3, ..., ...) = 4
++0.000 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
+
+// Now read the data and we should get 4 bytes.
++0.000 read(4,..., 4) = 4
++0.000 close(4) = 0
+
++0.0 > F. 1:1(0) ack 5 <...>
++0.0 < F. 5:5(0) ack 2 win 257
++0.01 > . 2:2(0) ack 6 <...>
+\ No newline at end of file
diff --git a/test/packetdrill/defs.bzl b/test/packetdrill/defs.bzl
new file mode 100644
index 000000000..fc28ce9ba
--- /dev/null
+++ b/test/packetdrill/defs.bzl
@@ -0,0 +1,91 @@
+"""Defines a rule for packetdrill test targets."""
+
+def _packetdrill_test_impl(ctx):
+    test_runner = ctx.executable._test_runner
+    runner = ctx.actions.declare_file("%s-runner" % ctx.label.name)
+
+    script_paths = []
+    for script in ctx.files.scripts:
+        script_paths.append(script.short_path)
+    runner_content = "\n".join([
+        "#!/bin/bash",
+        # This test will run part in a distinct user namespace. This can cause
+        # permission problems, because all runfiles may not be owned by the
+        # current user, and no other users will be mapped in that namespace.
+        # Make sure that everything is readable here.
+        "find . -type f -exec chmod a+rx {} \\;",
+        "find . -type d -exec chmod a+rx {} \\;",
+        "%s %s --init_script %s $@ -- %s\n" % (
+            test_runner.short_path,
+            " ".join(ctx.attr.flags),
+            ctx.files._init_script[0].short_path,
+            " ".join(script_paths),
+        ),
+    ])
+    ctx.actions.write(runner, runner_content, is_executable = True)
+
+    transitive_files = depset()
+    if hasattr(ctx.attr._test_runner, "data_runfiles"):
+        transitive_files = ctx.attr._test_runner.data_runfiles.files
+    runfiles = ctx.runfiles(
+        files = [test_runner] + ctx.files._init_script + ctx.files.scripts,
+        transitive_files = transitive_files,
+        collect_default = True,
+        collect_data = True,
+    )
+    return [DefaultInfo(executable = runner, runfiles = runfiles)]
+
+_packetdrill_test = rule(
+    attrs = {
+        "_test_runner": attr.label(
+            executable = True,
+            cfg = "host",
+            allow_files = True,
+            default = "packetdrill_test.sh",
+        ),
+        "_init_script": attr.label(
+            allow_single_file = True,
+            default = "packetdrill_setup.sh",
+        ),
+        "flags": attr.string_list(
+            mandatory = False,
+            default = [],
+        ),
+        "scripts": attr.label_list(
+            mandatory = True,
+            allow_files = True,
+        ),
+    },
+    test = True,
+    implementation = _packetdrill_test_impl,
+)
+
+PACKETDRILL_TAGS = [
+    "local",
+    "manual",
+    "packetdrill",
+]
+
+def packetdrill_linux_test(name, **kwargs):
+    if "tags" not in kwargs:
+        kwargs["tags"] = PACKETDRILL_TAGS
+    _packetdrill_test(
+        name = name,
+        flags = ["--dut_platform", "linux"],
+        **kwargs
+    )
+
+def packetdrill_netstack_test(name, **kwargs):
+    if "tags" not in kwargs:
+        kwargs["tags"] = PACKETDRILL_TAGS
+    _packetdrill_test(
+        name = name,
+        # This is the default runtime unless
+        # "--test_arg=--runtime=OTHER_RUNTIME" is used to override the value.
+        flags = ["--dut_platform", "netstack", "--runtime", "runsc-d"],
+        **kwargs
+    )
+
+def packetdrill_test(name, **kwargs):
+    packetdrill_linux_test(name + "_linux_test", **kwargs)
+    packetdrill_netstack_test(name + "_netstack_test", **kwargs)
diff --git a/test/packetdrill/fin_wait2_timeout.pkt b/test/packetdrill/fin_wait2_timeout.pkt
new file mode 100644
index 000000000..93ab08575
--- /dev/null
+++ b/test/packetdrill/fin_wait2_timeout.pkt
@@ -0,0 +1,23 @@
+// Test that a socket in FIN_WAIT_2 eventually times out and a subsequent
+// packet generates a RST.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
++0 < P. 1:1(0) ack 1 win 257
+
++0.100 accept(3, ..., ...) = 4
+// set FIN_WAIT2 timeout to 1 seconds.
++0.100 setsockopt(4, SOL_TCP, TCP_LINGER2, [1], 4) = 0
++0 close(4) = 0
+
++0 > F. 1:1(0) ack 1 <...>
++0 < . 1:1(0) ack 2 win 257
+
++2 < . 1:1(0) ack 2 win 257
++0 > R  2:2(0) win 0
diff --git a/test/packetdrill/listen_close_before_handshake_complete.pkt b/test/packetdrill/listen_close_before_handshake_complete.pkt
new file mode 100644
index 000000000..51c3f1a32
--- /dev/null
+++ b/test/packetdrill/listen_close_before_handshake_complete.pkt
@@ -0,0 +1,31 @@
+// Test that closing a listening socket closes any connections in SYN-RCVD
+// state and any packets bound for these connections generate a RESET.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
+
++0.100 close(3) = 0
++0.1 < P. 1:1(0) ack 1 win 257
+
+// Linux generates a reset with no ack number/bit set. This is contradictory to
+// what is specified in Rule 1 under Reset Generation in
+// https://tools.ietf.org/html/rfc793#section-3.4.
+//   "1. If the connection does not exist (CLOSED) then a reset is sent
+//    in response to any incoming segment except another reset.  In
+//    particular, SYNs addressed to a non-existent connection are rejected
+//    by this means.
+//
+//    If the incoming segment has an ACK field, the reset takes its
+//    sequence number from the ACK field of the segment, otherwise the
+//    reset has sequence number zero and the ACK field is set to the sum
+//    of the sequence number and segment length of the incoming segment.
+//    The connection remains in the CLOSED state."
+
++0.0  > R 1:1(0) win 0
+\ No newline at end of file
diff --git a/test/packetdrill/no_rst_to_rst.pkt b/test/packetdrill/no_rst_to_rst.pkt
new file mode 100644
index 000000000..612747827
--- /dev/null
+++ b/test/packetdrill/no_rst_to_rst.pkt
@@ -0,0 +1,36 @@
+// Test a RST is not generated in response to a RST and a RST is correctly
+// generated when an accepted endpoint is RST due to an incoming RST.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0 > S. 0:0(0) ack 1 <...>
++0 < P. 1:1(0) ack 1 win 257
+
++0.100 accept(3, ..., ...) = 4
+
++0.200 < R 1:1(0) win 0
+
++0.300 read(4,..., 4) = -1 ECONNRESET (Connection Reset by Peer)
+
++0.00 < . 1:1(0) ack 1 win 257
+
+// Linux generates a reset with no ack number/bit set. This is contradictory to
+// what is specified in Rule 1 under Reset Generation in
+// https://tools.ietf.org/html/rfc793#section-3.4.
+//   "1. If the connection does not exist (CLOSED) then a reset is sent
+//    in response to any incoming segment except another reset.  In
+//    particular, SYNs addressed to a non-existent connection are rejected
+//    by this means.
+//
+//    If the incoming segment has an ACK field, the reset takes its
+//    sequence number from the ACK field of the segment, otherwise the
+//    reset has sequence number zero and the ACK field is set to the sum
+//    of the sequence number and segment length of the incoming segment.
+//    The connection remains in the CLOSED state."
+
++0.00 > R 1:1(0) win 0
+\ No newline at end of file
diff --git a/test/packetdrill/packetdrill_setup.sh b/test/packetdrill/packetdrill_setup.sh
new file mode 100755
index 000000000..b858072f0
--- /dev/null
+++ b/test/packetdrill/packetdrill_setup.sh
@@ -0,0 +1,26 @@
+#!/bin/bash
+
+# Copyright 2018 The gVisor 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.
+
+# This script runs both within the sentry context and natively. It should tweak
+# TCP parameters to match expectations found in the script files.
+sysctl -q net.ipv4.tcp_sack=1
+sysctl -q net.ipv4.tcp_rmem="4096 2097152 $((8*1024*1024))"
+sysctl -q net.ipv4.tcp_wmem="4096 2097152 $((8*1024*1024))"
+
+# There may be errors from the above, but they will show up in the test logs and
+# we always want to proceed from this point. It's possible that values were
+# already set correctly and the nodes were not available in the namespace.
+exit 0
diff --git a/test/packetdrill/packetdrill_test.sh b/test/packetdrill/packetdrill_test.sh
new file mode 100755
index 000000000..922547d65
--- /dev/null
+++ b/test/packetdrill/packetdrill_test.sh
@@ -0,0 +1,226 @@
+#!/bin/bash
+
+# Copyright 2020 The gVisor 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.
+
+# Run a packetdrill test.  Two docker containers are made, one for the
+# Device-Under-Test (DUT) and one for the test runner.  Each is attached with
+# two networks, one for control packets that aid the test and one for test
+# packets which are sent as part of the test and observed for correctness.
+
+set -euxo pipefail
+
+function failure() {
+  local lineno=$1
+  local msg=$2
+  local filename="$0"
+  echo "FAIL: $filename:$lineno: $msg"
+}
+trap 'failure ${LINENO} "$BASH_COMMAND"' ERR
+
+declare -r LONGOPTS="dut_platform:,init_script:,runtime:"
+
+# Don't use declare below so that the error from getopt will end the script.
+PARSED=$(getopt --options "" --longoptions=$LONGOPTS --name "$0" -- "$@")
+
+eval set -- "$PARSED"
+
+while true; do
+  case "$1" in
+    --dut_platform)
+      # Either "linux" or "netstack".
+      declare -r DUT_PLATFORM="$2"
+      shift 2
+      ;;
+    --init_script)
+      declare -r INIT_SCRIPT="$2"
+      shift 2
+      ;;
+    --runtime)
+      # Not readonly because there might be multiple --runtime arguments and we
+      # want to use just the last one.  Only used if --dut_platform is
+      # "netstack".
+      declare RUNTIME="$2"
+      shift 2
+      ;;
+    --)
+      shift
+      break
+      ;;
+    *)
+      echo "Programming error"
+      exit 3
+  esac
+done
+
+# All the other arguments are scripts.
+declare -r scripts="$@"
+
+# Check that the required flags are defined in a way that is safe for "set -u".
+if [[ "${DUT_PLATFORM-}" == "netstack" ]]; then
+  if [[ -z "${RUNTIME-}" ]]; then
+    echo "FAIL: Missing --runtime argument: ${RUNTIME-}"
+    exit 2
+  fi
+  declare -r RUNTIME_ARG="--runtime ${RUNTIME}"
+elif [[ "${DUT_PLATFORM-}" == "linux" ]]; then
+  declare -r RUNTIME_ARG=""
+else
+  echo "FAIL: Bad or missing --dut_platform argument: ${DUT_PLATFORM-}"
+  exit 2
+fi
+if [[ ! -x "${INIT_SCRIPT-}" ]]; then
+  echo "FAIL: Bad or missing --init_script: ${INIT_SCRIPT-}"
+  exit 2
+fi
+
+function new_net_prefix() {
+  # Class C, 192.0.0.0 to 223.255.255.255, transitionally has mask 24.
+  echo "$(shuf -i 192-223 -n 1).$(shuf -i 0-255 -n 1).$(shuf -i 0-255 -n 1)"
+}
+
+# Variables specific to the control network and interface start with CTRL_.
+# Variables specific to the test network and interface start with TEST_.
+# Variables specific to the DUT start with DUT_.
+# Variables specific to the test runner start with TEST_RUNNER_.
+declare -r PACKETDRILL="/packetdrill/gtests/net/packetdrill/packetdrill"
+# Use random numbers so that test networks don't collide.
+declare CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)"
+declare CTRL_NET_PREFIX=$(new_net_prefix)
+declare TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)"
+declare TEST_NET_PREFIX=$(new_net_prefix)
+declare -r tolerance_usecs=100000
+# On both DUT and test runner, testing packets are on the eth2 interface.
+declare -r TEST_DEVICE="eth2"
+# Number of bits in the *_NET_PREFIX variables.
+declare -r NET_MASK="24"
+# Last bits of the DUT's IP address.
+declare -r DUT_NET_SUFFIX=".10"
+# Control port.
+declare -r CTRL_PORT="40000"
+# Last bits of the test runner's IP address.
+declare -r TEST_RUNNER_NET_SUFFIX=".20"
+declare -r TIMEOUT="60"
+declare -r IMAGE_TAG="gcr.io/gvisor-presubmit/packetdrill"
+
+# Make sure that docker is installed.
+docker --version
+
+function finish {
+  local cleanup_success=1
+  for net in "${CTRL_NET}" "${TEST_NET}"; do
+    # Kill all processes attached to ${net}.
+    for docker_command in "kill" "rm"; do
+      (docker network inspect "${net}" \
+        --format '{{range $key, $value := .Containers}}{{$key}} {{end}}' \
+        | xargs -r docker "${docker_command}") || \
+        cleanup_success=0
+    done
+    # Remove the network.
+    docker network rm "${net}" || \
+      cleanup_success=0
+  done
+
+  if ((!$cleanup_success)); then
+    echo "FAIL: Cleanup command failed"
+    exit 4
+  fi
+}
+trap finish EXIT
+
+# Subnet for control packets between test runner and DUT.
+while ! docker network create \
+  "--subnet=${CTRL_NET_PREFIX}.0/${NET_MASK}" "${CTRL_NET}"; do
+  sleep 0.1
+  CTRL_NET_PREFIX=$(new_net_prefix)
+  CTRL_NET="ctrl_net-$(shuf -i 0-99999999 -n 1)"
+done
+
+# Subnet for the packets that are part of the test.
+while ! docker network create \
+  "--subnet=${TEST_NET_PREFIX}.0/${NET_MASK}" "${TEST_NET}"; do
+  sleep 0.1
+  TEST_NET_PREFIX=$(new_net_prefix)
+  TEST_NET="test_net-$(shuf -i 0-99999999 -n 1)"
+done
+
+# Create the DUT container and connect to network.
+DUT=$(docker create ${RUNTIME_ARG} --privileged --rm \
+  --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG})
+docker network connect "${CTRL_NET}" \
+  --ip "${CTRL_NET_PREFIX}${DUT_NET_SUFFIX}" "${DUT}" \
+  || (docker kill ${DUT}; docker rm ${DUT}; false)
+docker network connect "${TEST_NET}" \
+  --ip "${TEST_NET_PREFIX}${DUT_NET_SUFFIX}" "${DUT}" \
+  || (docker kill ${DUT}; docker rm ${DUT}; false)
+docker start "${DUT}"
+
+# Create the test runner container and connect to network.
+TEST_RUNNER=$(docker create --privileged --rm \
+  --stop-timeout ${TIMEOUT} -it ${IMAGE_TAG})
+docker network connect "${CTRL_NET}" \
+  --ip "${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" "${TEST_RUNNER}" \
+  || (docker kill ${TEST_RUNNER}; docker rm ${REST_RUNNER}; false)
+docker network connect "${TEST_NET}" \
+  --ip "${TEST_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" "${TEST_RUNNER}" \
+  || (docker kill ${TEST_RUNNER}; docker rm ${REST_RUNNER}; false)
+docker start "${TEST_RUNNER}"
+
+# Run tcpdump in the test runner unbuffered, without dns resolution, just on the
+# interface with the test packets.
+docker exec -t ${TEST_RUNNER} tcpdump -U -n -i "${TEST_DEVICE}" &
+
+# Start a packetdrill server on the test_runner.  The packetdrill server sends
+# packets and asserts that they are received.
+docker exec -d "${TEST_RUNNER}" \
+  ${PACKETDRILL} --wire_server --wire_server_dev="${TEST_DEVICE}" \
+  --wire_server_ip="${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \
+  --wire_server_port="${CTRL_PORT}" \
+  --local_ip="${TEST_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \
+  --remote_ip="${TEST_NET_PREFIX}${DUT_NET_SUFFIX}"
+
+# Because the Linux kernel receives the SYN-ACK but didn't send the SYN it will
+# issue a RST. To prevent this IPtables can be used to filter those out.
+docker exec "${TEST_RUNNER}" \
+  iptables -A OUTPUT -p tcp --tcp-flags RST RST -j DROP
+
+# Wait for the packetdrill server on the test runner to come.  Attempt to
+# connect to it from the DUT every 100 milliseconds until success.
+while ! docker exec "${DUT}" \
+  nc -zv "${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" "${CTRL_PORT}"; do
+  sleep 0.1
+done
+
+# Copy the packetdrill setup script to the DUT.
+docker cp -L "${INIT_SCRIPT}" "${DUT}:packetdrill_setup.sh"
+
+# Copy the packetdrill scripts to the DUT.
+declare -a dut_scripts
+for script in $scripts; do
+  docker cp -L "${script}" "${DUT}:$(basename ${script})"
+  dut_scripts+=("/$(basename ${script})")
+done
+
+# Start a packetdrill client on the DUT.  The packetdrill client runs POSIX
+# socket commands and also sends instructions to the server.
+docker exec -t "${DUT}" \
+  ${PACKETDRILL} --wire_client --wire_client_dev="${TEST_DEVICE}" \
+  --wire_server_ip="${CTRL_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \
+  --wire_server_port="${CTRL_PORT}" \
+  --local_ip="${TEST_NET_PREFIX}${DUT_NET_SUFFIX}" \
+  --remote_ip="${TEST_NET_PREFIX}${TEST_RUNNER_NET_SUFFIX}" \
+  --init_scripts=/packetdrill_setup.sh \
+  --tolerance_usecs="${tolerance_usecs}" "${dut_scripts[@]}"
+
+echo PASS: No errors.
diff --git a/test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt b/test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt
new file mode 100644
index 000000000..a86b90ce6
--- /dev/null
+++ b/test/packetdrill/reset_for_ack_when_no_syn_cookies_in_use.pkt
@@ -0,0 +1,9 @@
+// Test that a listening socket generates a RST when it receives an
+// ACK and syn cookies are not in use.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 bind(3, ..., ...) = 0
+
++0 listen(3, 1) = 0
++0.1 < . 1:1(0) ack 1 win 32792
++0 > R 1:1(0) ack 0 win 0
+\ No newline at end of file
diff --git a/test/packetdrill/sanity_test.pkt b/test/packetdrill/sanity_test.pkt
new file mode 100644
index 000000000..b3b58c366
--- /dev/null
+++ b/test/packetdrill/sanity_test.pkt
@@ -0,0 +1,7 @@
+// Basic sanity test. One system call.
+//
+// All of the plumbing has to be working however, and the packetdrill wire
+// client needs to be able to connect to the wire server and send the script,
+// probe local interfaces, run through the test w/ timings, etc.
+
+0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
diff --git a/test/packetdrill/tcp_defer_accept.pkt b/test/packetdrill/tcp_defer_accept.pkt
new file mode 100644
index 000000000..a17f946db
--- /dev/null
+++ b/test/packetdrill/tcp_defer_accept.pkt
@@ -0,0 +1,48 @@
+// Test that a bare ACK does not complete a connection when TCP_DEFER_ACCEPT
+// timeout is not hit but an ACK w/ data does complete and deliver the
+// connection to the accept queue.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 setsockopt(3, SOL_TCP, TCP_DEFER_ACCEPT, [5], 4) = 0
++0.000 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0.0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0.0 > S. 0:0(0) ack 1    <...>
+
+// Send a bare ACK this should not complete the connection as we
+// set the TCP_DEFER_ACCEPT above.
++0.0 < . 1:1(0) ack 1 win 257
+
+// The bare ACK should be dropped and no connection should be delivered
+// to the accept queue.
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// Send another bare ACK and it should still fail we set TCP_DEFER_ACCEPT
+// to 5 seconds above.
++2.5 < . 1:1(0) ack 1 win 257
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// set accept socket back to blocking.
++0.000 fcntl(3, F_SETFL, O_RDWR) = 0
+
+// Now send an ACK w/ data. This should complete the connection
+// and deliver the socket to the accept queue.
++0.1 < . 1:5(4) ack 1 win 257
++0.0 > . 1:1(0) ack 5 <...>
+
+// This should cause connection to transition to connected state.
++0.000 accept(3, ..., ...) = 4
++0.000 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
+
+// Now read the data and we should get 4 bytes.
++0.000 read(4,..., 4) = 4
++0.000 close(4) = 0
+
++0.0 > F. 1:1(0) ack 5 <...>
++0.0 < F. 5:5(0) ack 2 win 257
++0.01 > . 2:2(0) ack 6 <...>
+\ No newline at end of file
diff --git a/test/packetdrill/tcp_defer_accept_timeout.pkt b/test/packetdrill/tcp_defer_accept_timeout.pkt
new file mode 100644
index 000000000..201fdeb14
--- /dev/null
+++ b/test/packetdrill/tcp_defer_accept_timeout.pkt
@@ -0,0 +1,48 @@
+// Test that a bare ACK is accepted after TCP_DEFER_ACCEPT timeout
+// is hit and a connection is delivered.
+
+0  socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
++0 setsockopt(3, SOL_TCP, TCP_DEFER_ACCEPT, [3], 4) = 0
++0.000 fcntl(3, F_SETFL, O_RDWR|O_NONBLOCK) = 0
++0 bind(3, ..., ...) = 0
+
+// Set backlog to 1 so that we can easily test.
++0 listen(3, 1) = 0
+
+// Establish a connection without timestamps.
++0.0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
++0.0 > S. 0:0(0) ack 1    <...>
+
+// Send a bare ACK this should not complete the connection as we
+// set the TCP_DEFER_ACCEPT above.
++0.0 < . 1:1(0) ack 1 win 257
+
+// The bare ACK should be dropped and no connection should be delivered
+// to the accept queue.
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// Send another bare ACK and it should still fail we set TCP_DEFER_ACCEPT
+// to 5 seconds above.
++2.5 < . 1:1(0) ack 1 win 257
++0.100 accept(3, ..., ...) = -1 EWOULDBLOCK (operation would block)
+
+// set accept socket back to blocking.
++0.000 fcntl(3, F_SETFL, O_RDWR) = 0
+
+// We should see one more retransmit of the SYN-ACK as a last ditch
+// attempt when TCP_DEFER_ACCEPT timeout is hit to trigger another
+// ACK or a packet with data.
++.35~+2.35 > S. 0:0(0) ack 1 <...>
+
+// Now send another bare ACK after TCP_DEFER_ACCEPT time has been passed.
++0.0 < . 1:1(0) ack 1 win 257
+
+// The ACK above should cause connection to transition to connected state.
++0.000 accept(3, ..., ...) = 4
++0.000 fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0
+
++0.000 close(4) = 0
+
++0.0 > F. 1:1(0) ack 1 <...>
++0.0 < F. 1:1(0) ack 2 win 257
++0.01 > . 2:2(0) ack 2 <...>