Project import generated by Copybara.

PiperOrigin-RevId: 237361882
Change-Id: I109a68f44db867b20f8c6a7732b0ce657133e52a
diff --git a/quic/core/quic_connection_test.cc b/quic/core/quic_connection_test.cc
new file mode 100644
index 0000000..62056a0
--- /dev/null
+++ b/quic/core/quic_connection_test.cc
@@ -0,0 +1,7826 @@
+// Copyright (c) 2012 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "net/third_party/quiche/src/quic/core/quic_connection.h"
+
+#include <errno.h>
+#include <memory>
+#include <ostream>
+#include <utility>
+
+#include "base/macros.h"
+#include "net/third_party/quiche/src/quic/core/congestion_control/loss_detection_interface.h"
+#include "net/third_party/quiche/src/quic/core/congestion_control/send_algorithm_interface.h"
+#include "net/third_party/quiche/src/quic/core/crypto/null_encrypter.h"
+#include "net/third_party/quiche/src/quic/core/crypto/quic_decrypter.h"
+#include "net/third_party/quiche/src/quic/core/crypto/quic_encrypter.h"
+#include "net/third_party/quiche/src/quic/core/quic_connection_id.h"
+#include "net/third_party/quiche/src/quic/core/quic_packets.h"
+#include "net/third_party/quiche/src/quic/core/quic_simple_buffer_allocator.h"
+#include "net/third_party/quiche/src/quic/core/quic_types.h"
+#include "net/third_party/quiche/src/quic/core/quic_utils.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_error_code_wrappers.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_expect_bug.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_flags.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_logging.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_ptr_util.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_reference_counted.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_str_cat.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_string.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_string_piece.h"
+#include "net/third_party/quiche/src/quic/platform/api/quic_test.h"
+#include "net/third_party/quiche/src/quic/test_tools/mock_clock.h"
+#include "net/third_party/quiche/src/quic/test_tools/mock_random.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_config_peer.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_connection_peer.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_framer_peer.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_packet_creator_peer.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_packet_generator_peer.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_sent_packet_manager_peer.h"
+#include "net/third_party/quiche/src/quic/test_tools/quic_test_utils.h"
+#include "net/third_party/quiche/src/quic/test_tools/simple_data_producer.h"
+#include "net/third_party/quiche/src/quic/test_tools/simple_quic_framer.h"
+#include "net/third_party/quiche/src/quic/test_tools/simple_session_notifier.h"
+
+using testing::_;
+using testing::AnyNumber;
+using testing::AtLeast;
+using testing::DoAll;
+using testing::Exactly;
+using testing::Ge;
+using testing::IgnoreResult;
+using testing::InSequence;
+using testing::Invoke;
+using testing::InvokeWithoutArgs;
+using testing::Lt;
+using testing::Ref;
+using testing::Return;
+using testing::SaveArg;
+using testing::SetArgPointee;
+using testing::StrictMock;
+
+namespace quic {
+namespace test {
+namespace {
+
+const char data1[] = "foo";
+const char data2[] = "bar";
+
+const bool kHasStopWaiting = true;
+
+const int kDefaultRetransmissionTimeMs = 500;
+
+const QuicSocketAddress kPeerAddress =
+    QuicSocketAddress(QuicIpAddress::Loopback6(),
+                      /*port=*/12345);
+const QuicSocketAddress kSelfAddress =
+    QuicSocketAddress(QuicIpAddress::Loopback6(),
+                      /*port=*/443);
+
+Perspective InvertPerspective(Perspective perspective) {
+  return perspective == Perspective::IS_CLIENT ? Perspective::IS_SERVER
+                                               : Perspective::IS_CLIENT;
+}
+
+QuicStreamId GetNthClientInitiatedStreamId(int n,
+                                           QuicTransportVersion version) {
+  return QuicUtils::GetHeadersStreamId(version) + n * 2;
+}
+
+// TaggingEncrypter appends kTagSize bytes of |tag| to the end of each message.
+class TaggingEncrypter : public QuicEncrypter {
+ public:
+  explicit TaggingEncrypter(uint8_t tag) : tag_(tag) {}
+  TaggingEncrypter(const TaggingEncrypter&) = delete;
+  TaggingEncrypter& operator=(const TaggingEncrypter&) = delete;
+
+  ~TaggingEncrypter() override {}
+
+  // QuicEncrypter interface.
+  bool SetKey(QuicStringPiece key) override { return true; }
+
+  bool SetNoncePrefix(QuicStringPiece nonce_prefix) override { return true; }
+
+  bool SetIV(QuicStringPiece iv) override { return true; }
+
+  bool EncryptPacket(uint64_t packet_number,
+                     QuicStringPiece associated_data,
+                     QuicStringPiece plaintext,
+                     char* output,
+                     size_t* output_length,
+                     size_t max_output_length) override {
+    const size_t len = plaintext.size() + kTagSize;
+    if (max_output_length < len) {
+      return false;
+    }
+    // Memmove is safe for inplace encryption.
+    memmove(output, plaintext.data(), plaintext.size());
+    output += plaintext.size();
+    memset(output, tag_, kTagSize);
+    *output_length = len;
+    return true;
+  }
+
+  size_t GetKeySize() const override { return 0; }
+  size_t GetNoncePrefixSize() const override { return 0; }
+  size_t GetIVSize() const override { return 0; }
+
+  size_t GetMaxPlaintextSize(size_t ciphertext_size) const override {
+    return ciphertext_size - kTagSize;
+  }
+
+  size_t GetCiphertextSize(size_t plaintext_size) const override {
+    return plaintext_size + kTagSize;
+  }
+
+  QuicStringPiece GetKey() const override { return QuicStringPiece(); }
+
+  QuicStringPiece GetNoncePrefix() const override { return QuicStringPiece(); }
+
+ private:
+  enum {
+    kTagSize = 12,
+  };
+
+  const uint8_t tag_;
+};
+
+// TaggingDecrypter ensures that the final kTagSize bytes of the message all
+// have the same value and then removes them.
+class TaggingDecrypter : public QuicDecrypter {
+ public:
+  ~TaggingDecrypter() override {}
+
+  // QuicDecrypter interface
+  bool SetKey(QuicStringPiece key) override { return true; }
+
+  bool SetNoncePrefix(QuicStringPiece nonce_prefix) override { return true; }
+
+  bool SetIV(QuicStringPiece iv) override { return true; }
+
+  bool SetPreliminaryKey(QuicStringPiece key) override {
+    QUIC_BUG << "should not be called";
+    return false;
+  }
+
+  bool SetDiversificationNonce(const DiversificationNonce& key) override {
+    return true;
+  }
+
+  bool DecryptPacket(uint64_t packet_number,
+                     QuicStringPiece associated_data,
+                     QuicStringPiece ciphertext,
+                     char* output,
+                     size_t* output_length,
+                     size_t max_output_length) override {
+    if (ciphertext.size() < kTagSize) {
+      return false;
+    }
+    if (!CheckTag(ciphertext, GetTag(ciphertext))) {
+      return false;
+    }
+    *output_length = ciphertext.size() - kTagSize;
+    memcpy(output, ciphertext.data(), *output_length);
+    return true;
+  }
+
+  size_t GetKeySize() const override { return 0; }
+  size_t GetIVSize() const override { return 0; }
+  QuicStringPiece GetKey() const override { return QuicStringPiece(); }
+  QuicStringPiece GetNoncePrefix() const override { return QuicStringPiece(); }
+  // Use a distinct value starting with 0xFFFFFF, which is never used by TLS.
+  uint32_t cipher_id() const override { return 0xFFFFFFF0; }
+
+ protected:
+  virtual uint8_t GetTag(QuicStringPiece ciphertext) {
+    return ciphertext.data()[ciphertext.size() - 1];
+  }
+
+ private:
+  enum {
+    kTagSize = 12,
+  };
+
+  bool CheckTag(QuicStringPiece ciphertext, uint8_t tag) {
+    for (size_t i = ciphertext.size() - kTagSize; i < ciphertext.size(); i++) {
+      if (ciphertext.data()[i] != tag) {
+        return false;
+      }
+    }
+
+    return true;
+  }
+};
+
+// StringTaggingDecrypter ensures that the final kTagSize bytes of the message
+// match the expected value.
+class StrictTaggingDecrypter : public TaggingDecrypter {
+ public:
+  explicit StrictTaggingDecrypter(uint8_t tag) : tag_(tag) {}
+  ~StrictTaggingDecrypter() override {}
+
+  // TaggingQuicDecrypter
+  uint8_t GetTag(QuicStringPiece ciphertext) override { return tag_; }
+
+  // Use a distinct value starting with 0xFFFFFF, which is never used by TLS.
+  uint32_t cipher_id() const override { return 0xFFFFFFF1; }
+
+ private:
+  const uint8_t tag_;
+};
+
+class TestConnectionHelper : public QuicConnectionHelperInterface {
+ public:
+  TestConnectionHelper(MockClock* clock, MockRandom* random_generator)
+      : clock_(clock), random_generator_(random_generator) {
+    clock_->AdvanceTime(QuicTime::Delta::FromSeconds(1));
+  }
+  TestConnectionHelper(const TestConnectionHelper&) = delete;
+  TestConnectionHelper& operator=(const TestConnectionHelper&) = delete;
+
+  // QuicConnectionHelperInterface
+  const QuicClock* GetClock() const override { return clock_; }
+
+  QuicRandom* GetRandomGenerator() override { return random_generator_; }
+
+  QuicBufferAllocator* GetStreamSendBufferAllocator() override {
+    return &buffer_allocator_;
+  }
+
+ private:
+  MockClock* clock_;
+  MockRandom* random_generator_;
+  SimpleBufferAllocator buffer_allocator_;
+};
+
+class TestAlarmFactory : public QuicAlarmFactory {
+ public:
+  class TestAlarm : public QuicAlarm {
+   public:
+    explicit TestAlarm(QuicArenaScopedPtr<QuicAlarm::Delegate> delegate)
+        : QuicAlarm(std::move(delegate)) {}
+
+    void SetImpl() override {}
+    void CancelImpl() override {}
+    using QuicAlarm::Fire;
+  };
+
+  TestAlarmFactory() {}
+  TestAlarmFactory(const TestAlarmFactory&) = delete;
+  TestAlarmFactory& operator=(const TestAlarmFactory&) = delete;
+
+  QuicAlarm* CreateAlarm(QuicAlarm::Delegate* delegate) override {
+    return new TestAlarm(QuicArenaScopedPtr<QuicAlarm::Delegate>(delegate));
+  }
+
+  QuicArenaScopedPtr<QuicAlarm> CreateAlarm(
+      QuicArenaScopedPtr<QuicAlarm::Delegate> delegate,
+      QuicConnectionArena* arena) override {
+    return arena->New<TestAlarm>(std::move(delegate));
+  }
+};
+
+class TestPacketWriter : public QuicPacketWriter {
+ public:
+  TestPacketWriter(ParsedQuicVersion version, MockClock* clock)
+      : version_(version),
+        framer_(SupportedVersions(version_), Perspective::IS_SERVER),
+        last_packet_size_(0),
+        write_blocked_(false),
+        write_should_fail_(false),
+        block_on_next_flush_(false),
+        block_on_next_write_(false),
+        next_packet_too_large_(false),
+        always_get_packet_too_large_(false),
+        is_write_blocked_data_buffered_(false),
+        is_batch_mode_(false),
+        final_bytes_of_last_packet_(0),
+        final_bytes_of_previous_packet_(0),
+        use_tagging_decrypter_(false),
+        packets_write_attempts_(0),
+        clock_(clock),
+        write_pause_time_delta_(QuicTime::Delta::Zero()),
+        max_packet_size_(kMaxPacketSize),
+        supports_release_time_(false) {}
+  TestPacketWriter(const TestPacketWriter&) = delete;
+  TestPacketWriter& operator=(const TestPacketWriter&) = delete;
+
+  // QuicPacketWriter interface
+  WriteResult WritePacket(const char* buffer,
+                          size_t buf_len,
+                          const QuicIpAddress& self_address,
+                          const QuicSocketAddress& peer_address,
+                          PerPacketOptions* options) override {
+    QuicEncryptedPacket packet(buffer, buf_len);
+    ++packets_write_attempts_;
+
+    if (packet.length() >= sizeof(final_bytes_of_last_packet_)) {
+      final_bytes_of_previous_packet_ = final_bytes_of_last_packet_;
+      memcpy(&final_bytes_of_last_packet_, packet.data() + packet.length() - 4,
+             sizeof(final_bytes_of_last_packet_));
+    }
+
+    if (use_tagging_decrypter_) {
+      framer_.framer()->SetDecrypter(ENCRYPTION_NONE,
+                                     QuicMakeUnique<TaggingDecrypter>());
+    }
+    EXPECT_TRUE(framer_.ProcessPacket(packet));
+    if (block_on_next_write_) {
+      write_blocked_ = true;
+      block_on_next_write_ = false;
+    }
+    if (next_packet_too_large_) {
+      next_packet_too_large_ = false;
+      return WriteResult(WRITE_STATUS_ERROR, QUIC_EMSGSIZE);
+    }
+    if (always_get_packet_too_large_) {
+      return WriteResult(WRITE_STATUS_ERROR, QUIC_EMSGSIZE);
+    }
+    if (IsWriteBlocked()) {
+      return WriteResult(is_write_blocked_data_buffered_
+                             ? WRITE_STATUS_BLOCKED_DATA_BUFFERED
+                             : WRITE_STATUS_BLOCKED,
+                         0);
+    }
+
+    if (ShouldWriteFail()) {
+      return WriteResult(WRITE_STATUS_ERROR, 0);
+    }
+
+    last_packet_size_ = packet.length();
+    last_packet_header_ = framer_.header();
+
+    if (!write_pause_time_delta_.IsZero()) {
+      clock_->AdvanceTime(write_pause_time_delta_);
+    }
+    return WriteResult(WRITE_STATUS_OK, last_packet_size_);
+  }
+
+  bool ShouldWriteFail() { return write_should_fail_; }
+
+  bool IsWriteBlocked() const override { return write_blocked_; }
+
+  void SetWriteBlocked() { write_blocked_ = true; }
+
+  void SetWritable() override { write_blocked_ = false; }
+
+  void SetShouldWriteFail() { write_should_fail_ = true; }
+
+  QuicByteCount GetMaxPacketSize(
+      const QuicSocketAddress& /*peer_address*/) const override {
+    return max_packet_size_;
+  }
+
+  bool SupportsReleaseTime() const { return supports_release_time_; }
+
+  bool IsBatchMode() const override { return is_batch_mode_; }
+
+  char* GetNextWriteLocation(const QuicIpAddress& self_address,
+                             const QuicSocketAddress& peer_address) override {
+    return nullptr;
+  }
+
+  WriteResult Flush() override {
+    if (block_on_next_flush_) {
+      block_on_next_flush_ = false;
+      SetWriteBlocked();
+      return WriteResult(WRITE_STATUS_BLOCKED, /*errno*/ -1);
+    }
+    return WriteResult(WRITE_STATUS_OK, 0);
+  }
+
+  void BlockOnNextFlush() { block_on_next_flush_ = true; }
+
+  void BlockOnNextWrite() { block_on_next_write_ = true; }
+
+  void SimulateNextPacketTooLarge() { next_packet_too_large_ = true; }
+
+  void AlwaysGetPacketTooLarge() { always_get_packet_too_large_ = true; }
+
+  // Sets the amount of time that the writer should before the actual write.
+  void SetWritePauseTimeDelta(QuicTime::Delta delta) {
+    write_pause_time_delta_ = delta;
+  }
+
+  void SetBatchMode(bool new_value) { is_batch_mode_ = new_value; }
+
+  const QuicPacketHeader& header() { return framer_.header(); }
+
+  size_t frame_count() const { return framer_.num_frames(); }
+
+  const std::vector<QuicAckFrame>& ack_frames() const {
+    return framer_.ack_frames();
+  }
+
+  const std::vector<QuicStopWaitingFrame>& stop_waiting_frames() const {
+    return framer_.stop_waiting_frames();
+  }
+
+  const std::vector<QuicConnectionCloseFrame>& connection_close_frames() const {
+    return framer_.connection_close_frames();
+  }
+
+  const std::vector<QuicRstStreamFrame>& rst_stream_frames() const {
+    return framer_.rst_stream_frames();
+  }
+
+  const std::vector<std::unique_ptr<QuicStreamFrame>>& stream_frames() const {
+    return framer_.stream_frames();
+  }
+
+  const std::vector<std::unique_ptr<QuicCryptoFrame>>& crypto_frames() const {
+    return framer_.crypto_frames();
+  }
+
+  const std::vector<QuicPingFrame>& ping_frames() const {
+    return framer_.ping_frames();
+  }
+
+  const std::vector<QuicMessageFrame>& message_frames() const {
+    return framer_.message_frames();
+  }
+
+  const std::vector<QuicWindowUpdateFrame>& window_update_frames() const {
+    return framer_.window_update_frames();
+  }
+
+  const std::vector<QuicPaddingFrame>& padding_frames() const {
+    return framer_.padding_frames();
+  }
+
+  const std::vector<QuicPathChallengeFrame>& path_challenge_frames() const {
+    return framer_.path_challenge_frames();
+  }
+
+  const std::vector<QuicPathResponseFrame>& path_response_frames() const {
+    return framer_.path_response_frames();
+  }
+
+  size_t last_packet_size() { return last_packet_size_; }
+
+  const QuicPacketHeader& last_packet_header() const {
+    return last_packet_header_;
+  }
+
+  const QuicVersionNegotiationPacket* version_negotiation_packet() {
+    return framer_.version_negotiation_packet();
+  }
+
+  void set_is_write_blocked_data_buffered(bool buffered) {
+    is_write_blocked_data_buffered_ = buffered;
+  }
+
+  void set_perspective(Perspective perspective) {
+    // We invert perspective here, because the framer needs to parse packets
+    // we send.
+    QuicFramerPeer::SetPerspective(framer_.framer(),
+                                   InvertPerspective(perspective));
+  }
+
+  // final_bytes_of_last_packet_ returns the last four bytes of the previous
+  // packet as a little-endian, uint32_t. This is intended to be used with a
+  // TaggingEncrypter so that tests can determine which encrypter was used for
+  // a given packet.
+  uint32_t final_bytes_of_last_packet() { return final_bytes_of_last_packet_; }
+
+  // Returns the final bytes of the second to last packet.
+  uint32_t final_bytes_of_previous_packet() {
+    return final_bytes_of_previous_packet_;
+  }
+
+  void use_tagging_decrypter() { use_tagging_decrypter_ = true; }
+
+  uint32_t packets_write_attempts() { return packets_write_attempts_; }
+
+  void Reset() { framer_.Reset(); }
+
+  void SetSupportedVersions(const ParsedQuicVersionVector& versions) {
+    framer_.SetSupportedVersions(versions);
+  }
+
+  void set_max_packet_size(QuicByteCount max_packet_size) {
+    max_packet_size_ = max_packet_size;
+  }
+
+  void set_supports_release_time(bool supports_release_time) {
+    supports_release_time_ = supports_release_time;
+  }
+
+  SimpleQuicFramer* framer() { return &framer_; }
+
+ private:
+  ParsedQuicVersion version_;
+  SimpleQuicFramer framer_;
+  size_t last_packet_size_;
+  QuicPacketHeader last_packet_header_;
+  bool write_blocked_;
+  bool write_should_fail_;
+  bool block_on_next_flush_;
+  bool block_on_next_write_;
+  bool next_packet_too_large_;
+  bool always_get_packet_too_large_;
+  bool is_write_blocked_data_buffered_;
+  bool is_batch_mode_;
+  uint32_t final_bytes_of_last_packet_;
+  uint32_t final_bytes_of_previous_packet_;
+  bool use_tagging_decrypter_;
+  uint32_t packets_write_attempts_;
+  MockClock* clock_;
+  // If non-zero, the clock will pause during WritePacket for this amount of
+  // time.
+  QuicTime::Delta write_pause_time_delta_;
+  QuicByteCount max_packet_size_;
+  bool supports_release_time_;
+};
+
+class TestConnection : public QuicConnection {
+ public:
+  TestConnection(QuicConnectionId connection_id,
+                 QuicSocketAddress address,
+                 TestConnectionHelper* helper,
+                 TestAlarmFactory* alarm_factory,
+                 TestPacketWriter* writer,
+                 Perspective perspective,
+                 ParsedQuicVersion version)
+      : QuicConnection(connection_id,
+                       address,
+                       helper,
+                       alarm_factory,
+                       writer,
+                       /* owns_writer= */ false,
+                       perspective,
+                       SupportedVersions(version)),
+        notifier_(nullptr) {
+    writer->set_perspective(perspective);
+    SetEncrypter(ENCRYPTION_FORWARD_SECURE,
+                 QuicMakeUnique<NullEncrypter>(perspective));
+    SetDataProducer(&producer_);
+  }
+  TestConnection(const TestConnection&) = delete;
+  TestConnection& operator=(const TestConnection&) = delete;
+
+  void SendAck() { QuicConnectionPeer::SendAck(this); }
+
+  void SetSendAlgorithm(SendAlgorithmInterface* send_algorithm) {
+    QuicConnectionPeer::SetSendAlgorithm(this, send_algorithm);
+  }
+
+  void SetLossAlgorithm(LossDetectionInterface* loss_algorithm) {
+    QuicConnectionPeer::SetLossAlgorithm(this, loss_algorithm);
+  }
+
+  void SendPacket(EncryptionLevel level,
+                  uint64_t packet_number,
+                  std::unique_ptr<QuicPacket> packet,
+                  HasRetransmittableData retransmittable,
+                  bool has_ack,
+                  bool has_pending_frames) {
+    char buffer[kMaxPacketSize];
+    size_t encrypted_length =
+        QuicConnectionPeer::GetFramer(this)->EncryptPayload(
+            ENCRYPTION_NONE, QuicPacketNumber(packet_number), *packet, buffer,
+            kMaxPacketSize);
+    SerializedPacket serialized_packet(
+        QuicPacketNumber(packet_number), PACKET_4BYTE_PACKET_NUMBER, buffer,
+        encrypted_length, has_ack, has_pending_frames);
+    if (retransmittable == HAS_RETRANSMITTABLE_DATA) {
+      serialized_packet.retransmittable_frames.push_back(
+          QuicFrame(QuicStreamFrame()));
+    }
+    OnSerializedPacket(&serialized_packet);
+  }
+
+  QuicConsumedData SaveAndSendStreamData(QuicStreamId id,
+                                         const struct iovec* iov,
+                                         int iov_count,
+                                         size_t total_length,
+                                         QuicStreamOffset offset,
+                                         StreamSendingState state) {
+    ScopedPacketFlusher flusher(this, NO_ACK);
+    producer_.SaveStreamData(id, iov, iov_count, 0u, total_length);
+    if (notifier_ != nullptr) {
+      return notifier_->WriteOrBufferData(id, total_length, state);
+    }
+    return QuicConnection::SendStreamData(id, total_length, offset, state);
+  }
+
+  QuicConsumedData SendStreamDataWithString(QuicStreamId id,
+                                            QuicStringPiece data,
+                                            QuicStreamOffset offset,
+                                            StreamSendingState state) {
+    ScopedPacketFlusher flusher(this, NO_ACK);
+    if (id != QuicUtils::GetCryptoStreamId(transport_version()) &&
+        this->encryption_level() == ENCRYPTION_NONE) {
+      this->SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+    }
+    struct iovec iov;
+    MakeIOVector(data, &iov);
+    return SaveAndSendStreamData(id, &iov, 1, data.length(), offset, state);
+  }
+
+  QuicConsumedData SendStreamData3() {
+    return SendStreamDataWithString(
+        GetNthClientInitiatedStreamId(1, transport_version()), "food", 0,
+        NO_FIN);
+  }
+
+  QuicConsumedData SendStreamData5() {
+    return SendStreamDataWithString(
+        GetNthClientInitiatedStreamId(2, transport_version()), "food2", 0,
+        NO_FIN);
+  }
+
+  // Ensures the connection can write stream data before writing.
+  QuicConsumedData EnsureWritableAndSendStreamData5() {
+    EXPECT_TRUE(CanWriteStreamData());
+    return SendStreamData5();
+  }
+
+  // The crypto stream has special semantics so that it is not blocked by a
+  // congestion window limitation, and also so that it gets put into a separate
+  // packet (so that it is easier to reason about a crypto frame not being
+  // split needlessly across packet boundaries).  As a result, we have separate
+  // tests for some cases for this stream.
+  QuicConsumedData SendCryptoStreamData() {
+    QuicStreamOffset offset = 0;
+    QuicStringPiece data("chlo");
+    if (transport_version() < QUIC_VERSION_47) {
+      return SendStreamDataWithString(
+          QuicUtils::GetCryptoStreamId(transport_version()), data, offset,
+          NO_FIN);
+    }
+    producer_.SaveCryptoData(ENCRYPTION_NONE, offset, data);
+    size_t bytes_written;
+    if (notifier_) {
+      bytes_written =
+          notifier_->WriteCryptoData(ENCRYPTION_NONE, data.length(), offset);
+    } else {
+      bytes_written = QuicConnection::SendCryptoData(ENCRYPTION_NONE,
+                                                     data.length(), offset);
+    }
+    return QuicConsumedData(bytes_written, /*fin_consumed*/ false);
+  }
+
+  void set_version(ParsedQuicVersion version) {
+    QuicConnectionPeer::GetFramer(this)->set_version(version);
+  }
+
+  void SetSupportedVersions(const ParsedQuicVersionVector& versions) {
+    QuicConnectionPeer::GetFramer(this)->SetSupportedVersions(versions);
+    QuicConnectionPeer::SetNoVersionNegotiation(this, versions.size() == 1);
+    writer()->SetSupportedVersions(versions);
+  }
+
+  void set_perspective(Perspective perspective) {
+    writer()->set_perspective(perspective);
+    QuicConnectionPeer::SetPerspective(this, perspective);
+  }
+
+  // Enable path MTU discovery.  Assumes that the test is performed from the
+  // client perspective and the higher value of MTU target is used.
+  void EnablePathMtuDiscovery(MockSendAlgorithm* send_algorithm) {
+    ASSERT_EQ(Perspective::IS_CLIENT, perspective());
+
+    QuicConfig config;
+    QuicTagVector connection_options;
+    connection_options.push_back(kMTUH);
+    config.SetConnectionOptionsToSend(connection_options);
+    EXPECT_CALL(*send_algorithm, SetFromConfig(_, _));
+    SetFromConfig(config);
+
+    // Normally, the pacing would be disabled in the test, but calling
+    // SetFromConfig enables it.  Set nearly-infinite bandwidth to make the
+    // pacing algorithm work.
+    EXPECT_CALL(*send_algorithm, PacingRate(_))
+        .WillRepeatedly(Return(QuicBandwidth::Infinite()));
+  }
+
+  TestAlarmFactory::TestAlarm* GetAckAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetAckAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetPingAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetPingAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetRetransmissionAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetRetransmissionAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetSendAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetSendAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetTimeoutAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetTimeoutAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetMtuDiscoveryAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetMtuDiscoveryAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetPathDegradingAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetPathDegradingAlarm(this));
+  }
+
+  TestAlarmFactory::TestAlarm* GetProcessUndecryptablePacketsAlarm() {
+    return reinterpret_cast<TestAlarmFactory::TestAlarm*>(
+        QuicConnectionPeer::GetProcessUndecryptablePacketsAlarm(this));
+  }
+
+  void SetMaxTailLossProbes(size_t max_tail_loss_probes) {
+    QuicSentPacketManagerPeer::SetMaxTailLossProbes(
+        QuicConnectionPeer::GetSentPacketManager(this), max_tail_loss_probes);
+  }
+
+  QuicByteCount GetBytesInFlight() {
+    return QuicSentPacketManagerPeer::GetBytesInFlight(
+        QuicConnectionPeer::GetSentPacketManager(this));
+  }
+
+  void set_notifier(SimpleSessionNotifier* notifier) { notifier_ = notifier; }
+
+  void ReturnEffectivePeerAddressForNextPacket(const QuicSocketAddress& addr) {
+    next_effective_peer_addr_ = QuicMakeUnique<QuicSocketAddress>(addr);
+  }
+
+  using QuicConnection::active_effective_peer_migration_type;
+  using QuicConnection::IsCurrentPacketConnectivityProbing;
+  using QuicConnection::SelectMutualVersion;
+  using QuicConnection::SendProbingRetransmissions;
+  using QuicConnection::set_defer_send_in_response_to_packets;
+
+ protected:
+  QuicSocketAddress GetEffectivePeerAddressFromCurrentPacket() const override {
+    if (next_effective_peer_addr_) {
+      return *std::move(next_effective_peer_addr_);
+    }
+    return QuicConnection::GetEffectivePeerAddressFromCurrentPacket();
+  }
+
+ private:
+  TestPacketWriter* writer() {
+    return static_cast<TestPacketWriter*>(QuicConnection::writer());
+  }
+
+  SimpleDataProducer producer_;
+
+  SimpleSessionNotifier* notifier_;
+
+  std::unique_ptr<QuicSocketAddress> next_effective_peer_addr_;
+};
+
+enum class AckResponse { kDefer, kImmediate };
+
+// Run tests with combinations of {ParsedQuicVersion, AckResponse}.
+struct TestParams {
+  TestParams(ParsedQuicVersion version,
+             AckResponse ack_response,
+             bool no_stop_waiting)
+      : version(version),
+        ack_response(ack_response),
+        no_stop_waiting(no_stop_waiting) {}
+
+  friend std::ostream& operator<<(std::ostream& os, const TestParams& p) {
+    os << "{ client_version: " << ParsedQuicVersionToString(p.version)
+       << " ack_response: "
+       << (p.ack_response == AckResponse::kDefer ? "defer" : "immediate")
+       << " no_stop_waiting: " << p.no_stop_waiting << " }";
+    return os;
+  }
+
+  ParsedQuicVersion version;
+  AckResponse ack_response;
+  bool no_stop_waiting;
+};
+
+// Constructs various test permutations.
+std::vector<TestParams> GetTestParams() {
+  QuicFlagSaver flags;
+  SetQuicFlag(&FLAGS_quic_supports_tls_handshake, true);
+  std::vector<TestParams> params;
+  ParsedQuicVersionVector all_supported_versions = AllSupportedVersions();
+  for (size_t i = 0; i < all_supported_versions.size(); ++i) {
+    for (AckResponse ack_response :
+         {AckResponse::kDefer, AckResponse::kImmediate}) {
+      for (bool no_stop_waiting : {true, false}) {
+        // After version 43, never use STOP_WAITING.
+        if (all_supported_versions[i].transport_version <= QUIC_VERSION_43 ||
+            no_stop_waiting) {
+          params.push_back(TestParams(all_supported_versions[i], ack_response,
+                                      no_stop_waiting));
+        }
+      }
+    }
+  }
+  return params;
+}
+
+class QuicConnectionTest : public QuicTestWithParam<TestParams> {
+ protected:
+  QuicConnectionTest()
+      : connection_id_(TestConnectionId()),
+        framer_(SupportedVersions(version()),
+                QuicTime::Zero(),
+                Perspective::IS_CLIENT,
+                connection_id_.length()),
+        send_algorithm_(new StrictMock<MockSendAlgorithm>),
+        loss_algorithm_(new MockLossAlgorithm()),
+        helper_(new TestConnectionHelper(&clock_, &random_generator_)),
+        alarm_factory_(new TestAlarmFactory()),
+        peer_framer_(SupportedVersions(version()),
+                     QuicTime::Zero(),
+                     Perspective::IS_SERVER,
+                     connection_id_.length()),
+        peer_creator_(connection_id_,
+                      &peer_framer_,
+                      /*delegate=*/nullptr),
+        writer_(new TestPacketWriter(version(), &clock_)),
+        connection_(connection_id_,
+                    kPeerAddress,
+                    helper_.get(),
+                    alarm_factory_.get(),
+                    writer_.get(),
+                    Perspective::IS_CLIENT,
+                    version()),
+        creator_(QuicConnectionPeer::GetPacketCreator(&connection_)),
+        generator_(QuicConnectionPeer::GetPacketGenerator(&connection_)),
+        manager_(QuicConnectionPeer::GetSentPacketManager(&connection_)),
+        frame1_(QuicUtils::GetCryptoStreamId(version().transport_version),
+                false,
+                0,
+                QuicStringPiece(data1)),
+        frame2_(QuicUtils::GetCryptoStreamId(version().transport_version),
+                false,
+                3,
+                QuicStringPiece(data2)),
+        packet_number_length_(PACKET_4BYTE_PACKET_NUMBER),
+        connection_id_included_(CONNECTION_ID_PRESENT),
+        notifier_(&connection_) {
+    SetQuicFlag(&FLAGS_quic_supports_tls_handshake, true);
+    connection_.set_defer_send_in_response_to_packets(GetParam().ack_response ==
+                                                      AckResponse::kDefer);
+    QuicFramerPeer::SetLastSerializedConnectionId(
+        QuicConnectionPeer::GetFramer(&connection_), connection_id_);
+    if (version().transport_version > QUIC_VERSION_43) {
+      EXPECT_TRUE(QuicConnectionPeer::GetNoStopWaitingFrames(&connection_));
+    } else {
+      QuicConnectionPeer::SetNoStopWaitingFrames(&connection_,
+                                                 GetParam().no_stop_waiting);
+    }
+    connection_.set_visitor(&visitor_);
+    if (connection_.session_decides_what_to_write()) {
+      connection_.SetSessionNotifier(&notifier_);
+      connection_.set_notifier(&notifier_);
+    }
+    connection_.SetSendAlgorithm(send_algorithm_);
+    connection_.SetLossAlgorithm(loss_algorithm_.get());
+    EXPECT_CALL(*send_algorithm_, CanSend(_)).WillRepeatedly(Return(true));
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+        .Times(AnyNumber());
+    EXPECT_CALL(*send_algorithm_, GetCongestionWindow())
+        .WillRepeatedly(Return(kDefaultTCPMSS));
+    EXPECT_CALL(*send_algorithm_, PacingRate(_))
+        .WillRepeatedly(Return(QuicBandwidth::Zero()));
+    EXPECT_CALL(*send_algorithm_, HasReliableBandwidthEstimate())
+        .Times(AnyNumber());
+    EXPECT_CALL(*send_algorithm_, BandwidthEstimate())
+        .Times(AnyNumber())
+        .WillRepeatedly(Return(QuicBandwidth::Zero()));
+    EXPECT_CALL(*send_algorithm_, InSlowStart()).Times(AnyNumber());
+    EXPECT_CALL(*send_algorithm_, InRecovery()).Times(AnyNumber());
+    EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(AnyNumber());
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite()).Times(AnyNumber());
+    EXPECT_CALL(visitor_, HasPendingHandshake()).Times(AnyNumber());
+    if (connection_.session_decides_what_to_write()) {
+      EXPECT_CALL(visitor_, OnCanWrite())
+          .WillRepeatedly(
+              Invoke(&notifier_, &SimpleSessionNotifier::OnCanWrite));
+    } else {
+      EXPECT_CALL(visitor_, OnCanWrite()).Times(AnyNumber());
+    }
+    EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+        .WillRepeatedly(Return(false));
+    EXPECT_CALL(visitor_, OnCongestionWindowChange(_)).Times(AnyNumber());
+    EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(AnyNumber());
+    EXPECT_CALL(visitor_, OnForwardProgressConfirmed()).Times(AnyNumber());
+
+    EXPECT_CALL(*loss_algorithm_, GetLossTimeout())
+        .WillRepeatedly(Return(QuicTime::Zero()));
+    EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+        .Times(AnyNumber());
+  }
+
+  QuicConnectionTest(const QuicConnectionTest&) = delete;
+  QuicConnectionTest& operator=(const QuicConnectionTest&) = delete;
+
+  ParsedQuicVersion version() { return GetParam().version; }
+
+  QuicAckFrame* outgoing_ack() {
+    QuicFrame ack_frame = QuicConnectionPeer::GetUpdatedAckFrame(&connection_);
+    ack_ = *ack_frame.ack_frame;
+    return &ack_;
+  }
+
+  QuicStopWaitingFrame* stop_waiting() {
+    QuicConnectionPeer::PopulateStopWaitingFrame(&connection_, &stop_waiting_);
+    return &stop_waiting_;
+  }
+
+  QuicPacketNumber least_unacked() {
+    if (writer_->stop_waiting_frames().empty()) {
+      return QuicPacketNumber();
+    }
+    return writer_->stop_waiting_frames()[0].least_unacked;
+  }
+
+  void use_tagging_decrypter() { writer_->use_tagging_decrypter(); }
+
+  void ProcessPacket(uint64_t number) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacket(number);
+    if (connection_.GetSendAlarm()->IsSet()) {
+      connection_.GetSendAlarm()->Fire();
+    }
+  }
+
+  void ProcessReceivedPacket(const QuicSocketAddress& self_address,
+                             const QuicSocketAddress& peer_address,
+                             const QuicReceivedPacket& packet) {
+    connection_.ProcessUdpPacket(self_address, peer_address, packet);
+    if (connection_.GetSendAlarm()->IsSet()) {
+      connection_.GetSendAlarm()->Fire();
+    }
+  }
+
+  void ProcessFramePacket(QuicFrame frame) {
+    ProcessFramePacketWithAddresses(frame, kSelfAddress, kPeerAddress);
+  }
+
+  void ProcessFramePacketWithAddresses(QuicFrame frame,
+                                       QuicSocketAddress self_address,
+                                       QuicSocketAddress peer_address) {
+    QuicFrames frames;
+    frames.push_back(QuicFrame(frame));
+    QuicPacketCreatorPeer::SetSendVersionInPacket(
+        &peer_creator_, connection_.perspective() == Perspective::IS_SERVER);
+    if (QuicPacketCreatorPeer::GetEncryptionLevel(&peer_creator_) >
+        ENCRYPTION_NONE) {
+      // Set peer_framer_'s corresponding encrypter.
+      peer_creator_.SetEncrypter(
+          QuicPacketCreatorPeer::GetEncryptionLevel(&peer_creator_),
+          QuicMakeUnique<NullEncrypter>(peer_framer_.perspective()));
+    }
+
+    char buffer[kMaxPacketSize];
+    SerializedPacket serialized_packet =
+        QuicPacketCreatorPeer::SerializeAllFrames(&peer_creator_, frames,
+                                                  buffer, kMaxPacketSize);
+    connection_.ProcessUdpPacket(
+        self_address, peer_address,
+        QuicReceivedPacket(serialized_packet.encrypted_buffer,
+                           serialized_packet.encrypted_length, clock_.Now()));
+    if (connection_.GetSendAlarm()->IsSet()) {
+      connection_.GetSendAlarm()->Fire();
+    }
+  }
+
+  // Bypassing the packet creator is unrealistic, but allows us to process
+  // packets the QuicPacketCreator won't allow us to create.
+  void ForceProcessFramePacket(QuicFrame frame) {
+    QuicFrames frames;
+    frames.push_back(QuicFrame(frame));
+    QuicPacketCreatorPeer::SetSendVersionInPacket(
+        &peer_creator_, connection_.perspective() == Perspective::IS_SERVER);
+    QuicPacketHeader header;
+    QuicPacketCreatorPeer::FillPacketHeader(&peer_creator_, &header);
+    char encrypted_buffer[kMaxPacketSize];
+    size_t length = peer_framer_.BuildDataPacket(
+        header, frames, encrypted_buffer, kMaxPacketSize, ENCRYPTION_NONE);
+    DCHECK_GT(length, 0u);
+
+    const size_t encrypted_length = peer_framer_.EncryptInPlace(
+        ENCRYPTION_NONE, header.packet_number,
+        GetStartOfEncryptedData(peer_framer_.version().transport_version,
+                                header),
+        length, kMaxPacketSize, encrypted_buffer);
+    DCHECK_GT(encrypted_length, 0u);
+
+    connection_.ProcessUdpPacket(
+        kSelfAddress, kPeerAddress,
+        QuicReceivedPacket(encrypted_buffer, encrypted_length, clock_.Now()));
+  }
+
+  size_t ProcessFramePacketAtLevel(uint64_t number,
+                                   QuicFrame frame,
+                                   EncryptionLevel level) {
+    QuicPacketHeader header;
+    header.destination_connection_id = connection_id_;
+    header.packet_number_length = packet_number_length_;
+    header.destination_connection_id_included = connection_id_included_;
+    if (peer_framer_.transport_version() > QUIC_VERSION_43 &&
+        peer_framer_.perspective() == Perspective::IS_SERVER) {
+      header.destination_connection_id_included = CONNECTION_ID_ABSENT;
+    }
+    header.packet_number = QuicPacketNumber(number);
+    QuicFrames frames;
+    frames.push_back(frame);
+    std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+
+    char buffer[kMaxPacketSize];
+    size_t encrypted_length = framer_.EncryptPayload(
+        level, QuicPacketNumber(number), *packet, buffer, kMaxPacketSize);
+    connection_.ProcessUdpPacket(
+        kSelfAddress, kPeerAddress,
+        QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+    return encrypted_length;
+  }
+
+  size_t ProcessDataPacket(uint64_t number) {
+    return ProcessDataPacketAtLevel(number, false, ENCRYPTION_NONE);
+  }
+
+  size_t ProcessDataPacket(QuicPacketNumber packet_number) {
+    return ProcessDataPacketAtLevel(packet_number, false, ENCRYPTION_NONE);
+  }
+
+  size_t ProcessDataPacketAtLevel(QuicPacketNumber packet_number,
+                                  bool has_stop_waiting,
+                                  EncryptionLevel level) {
+    return ProcessDataPacketAtLevel(packet_number.ToUint64(), has_stop_waiting,
+                                    level);
+  }
+
+  size_t ProcessDataPacketAtLevel(uint64_t number,
+                                  bool has_stop_waiting,
+                                  EncryptionLevel level) {
+    std::unique_ptr<QuicPacket> packet(
+        ConstructDataPacket(number, has_stop_waiting));
+    char buffer[kMaxPacketSize];
+    size_t encrypted_length = peer_framer_.EncryptPayload(
+        level, QuicPacketNumber(number), *packet, buffer, kMaxPacketSize);
+    connection_.ProcessUdpPacket(
+        kSelfAddress, kPeerAddress,
+        QuicReceivedPacket(buffer, encrypted_length, clock_.Now(), false));
+    if (connection_.GetSendAlarm()->IsSet()) {
+      connection_.GetSendAlarm()->Fire();
+    }
+    return encrypted_length;
+  }
+
+  void ProcessClosePacket(uint64_t number) {
+    std::unique_ptr<QuicPacket> packet(ConstructClosePacket(number));
+    char buffer[kMaxPacketSize];
+    size_t encrypted_length =
+        peer_framer_.EncryptPayload(ENCRYPTION_NONE, QuicPacketNumber(number),
+                                    *packet, buffer, kMaxPacketSize);
+    connection_.ProcessUdpPacket(
+        kSelfAddress, kPeerAddress,
+        QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+  }
+
+  QuicByteCount SendStreamDataToPeer(QuicStreamId id,
+                                     QuicStringPiece data,
+                                     QuicStreamOffset offset,
+                                     StreamSendingState state,
+                                     QuicPacketNumber* last_packet) {
+    QuicByteCount packet_size;
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+        .WillOnce(SaveArg<3>(&packet_size));
+    connection_.SendStreamDataWithString(id, data, offset, state);
+    if (last_packet != nullptr) {
+      *last_packet = creator_->packet_number();
+    }
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+        .Times(AnyNumber());
+    return packet_size;
+  }
+
+  void SendAckPacketToPeer() {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+    {
+      QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                  QuicConnection::NO_ACK);
+      connection_.SendAck();
+    }
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+        .Times(AnyNumber());
+  }
+
+  void SendRstStream(QuicStreamId id,
+                     QuicRstStreamErrorCode error,
+                     QuicStreamOffset bytes_written) {
+    if (connection_.session_decides_what_to_write()) {
+      notifier_.WriteOrBufferRstStream(id, error, bytes_written);
+      connection_.OnStreamReset(id, error);
+      return;
+    }
+    std::unique_ptr<QuicRstStreamFrame> rst_stream =
+        QuicMakeUnique<QuicRstStreamFrame>(1, id, error, bytes_written);
+    if (connection_.SendControlFrame(QuicFrame(rst_stream.get()))) {
+      rst_stream.release();
+    }
+    connection_.OnStreamReset(id, error);
+  }
+
+  void ProcessAckPacket(uint64_t packet_number, QuicAckFrame* frame) {
+    if (packet_number > 1) {
+      QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, packet_number - 1);
+    } else {
+      QuicPacketCreatorPeer::ClearPacketNumber(&peer_creator_);
+    }
+    ProcessFramePacket(QuicFrame(frame));
+  }
+
+  void ProcessAckPacket(QuicAckFrame* frame) {
+    ProcessFramePacket(QuicFrame(frame));
+  }
+
+  void ProcessStopWaitingPacket(QuicStopWaitingFrame frame) {
+    ProcessFramePacket(QuicFrame(frame));
+  }
+
+  size_t ProcessStopWaitingPacketAtLevel(uint64_t number,
+                                         QuicStopWaitingFrame frame,
+                                         EncryptionLevel level) {
+    return ProcessFramePacketAtLevel(number, QuicFrame(frame),
+                                     ENCRYPTION_ZERO_RTT);
+  }
+
+  void ProcessGoAwayPacket(QuicGoAwayFrame* frame) {
+    ProcessFramePacket(QuicFrame(frame));
+  }
+
+  bool IsMissing(uint64_t number) {
+    return IsAwaitingPacket(*outgoing_ack(), QuicPacketNumber(number),
+                            QuicPacketNumber());
+  }
+
+  std::unique_ptr<QuicPacket> ConstructPacket(const QuicPacketHeader& header,
+                                              const QuicFrames& frames) {
+    auto packet = BuildUnsizedDataPacket(&peer_framer_, header, frames);
+    EXPECT_NE(nullptr, packet.get());
+    return packet;
+  }
+
+  std::unique_ptr<QuicPacket> ConstructDataPacket(uint64_t number,
+                                                  bool has_stop_waiting) {
+    QuicPacketHeader header;
+    // Set connection_id to peer's in memory representation as this data packet
+    // is created by peer_framer.
+    header.destination_connection_id = connection_id_;
+    header.packet_number_length = packet_number_length_;
+    header.destination_connection_id_included = connection_id_included_;
+    if (peer_framer_.transport_version() > QUIC_VERSION_43 &&
+        peer_framer_.perspective() == Perspective::IS_SERVER) {
+      header.destination_connection_id_included = CONNECTION_ID_ABSENT;
+    }
+    header.packet_number = QuicPacketNumber(number);
+
+    QuicFrames frames;
+    frames.push_back(QuicFrame(frame1_));
+    if (has_stop_waiting) {
+      frames.push_back(QuicFrame(stop_waiting_));
+    }
+    return ConstructPacket(header, frames);
+  }
+
+  OwningSerializedPacketPointer ConstructProbingPacket() {
+    if (version().transport_version == QUIC_VERSION_99) {
+      QuicPathFrameBuffer payload = {
+          {0xde, 0xad, 0xbe, 0xef, 0xba, 0xdc, 0x0f, 0xfe}};
+      return QuicPacketCreatorPeer::
+          SerializePathChallengeConnectivityProbingPacket(&peer_creator_,
+                                                          &payload);
+    }
+    return QuicPacketCreatorPeer::SerializeConnectivityProbingPacket(
+        &peer_creator_);
+  }
+
+  std::unique_ptr<QuicPacket> ConstructClosePacket(uint64_t number) {
+    QuicPacketHeader header;
+    // Set connection_id to peer's in memory representation as this connection
+    // close packet is created by peer_framer.
+    header.destination_connection_id = connection_id_;
+    header.packet_number = QuicPacketNumber(number);
+    if (peer_framer_.transport_version() > QUIC_VERSION_43 &&
+        peer_framer_.perspective() == Perspective::IS_SERVER) {
+      header.destination_connection_id_included = CONNECTION_ID_ABSENT;
+    }
+
+    QuicConnectionCloseFrame qccf;
+    qccf.error_code = QUIC_PEER_GOING_AWAY;
+
+    QuicFrames frames;
+    frames.push_back(QuicFrame(&qccf));
+    return ConstructPacket(header, frames);
+  }
+
+  QuicTime::Delta DefaultRetransmissionTime() {
+    return QuicTime::Delta::FromMilliseconds(kDefaultRetransmissionTimeMs);
+  }
+
+  QuicTime::Delta DefaultDelayedAckTime() {
+    return QuicTime::Delta::FromMilliseconds(kDefaultDelayedAckTimeMs);
+  }
+
+  const QuicStopWaitingFrame InitStopWaitingFrame(uint64_t least_unacked) {
+    QuicStopWaitingFrame frame;
+    frame.least_unacked = QuicPacketNumber(least_unacked);
+    return frame;
+  }
+
+  // Construct a ack_frame that acks all packet numbers between 1 and
+  // |largest_acked|, except |missing|.
+  // REQUIRES: 1 <= |missing| < |largest_acked|
+  QuicAckFrame ConstructAckFrame(uint64_t largest_acked, uint64_t missing) {
+    return ConstructAckFrame(QuicPacketNumber(largest_acked),
+                             QuicPacketNumber(missing));
+  }
+
+  QuicAckFrame ConstructAckFrame(QuicPacketNumber largest_acked,
+                                 QuicPacketNumber missing) {
+    if (missing == QuicPacketNumber(1)) {
+      return InitAckFrame({{missing + 1, largest_acked + 1}});
+    }
+    return InitAckFrame(
+        {{QuicPacketNumber(1), missing}, {missing + 1, largest_acked + 1}});
+  }
+
+  // Undo nacking a packet within the frame.
+  void AckPacket(QuicPacketNumber arrived, QuicAckFrame* frame) {
+    EXPECT_FALSE(frame->packets.Contains(arrived));
+    frame->packets.Add(arrived);
+  }
+
+  void TriggerConnectionClose() {
+    // Send an erroneous packet to close the connection.
+    EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_PACKET_HEADER, _,
+                                             ConnectionCloseSource::FROM_SELF));
+    // Call ProcessDataPacket rather than ProcessPacket, as we should not get a
+    // packet call to the visitor.
+    if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+      ProcessDataPacket(MaxRandomInitialPacketNumber() + 6000);
+    } else {
+      ProcessDataPacket(6000);
+    }
+
+    EXPECT_FALSE(QuicConnectionPeer::GetConnectionClosePacket(&connection_) ==
+                 nullptr);
+  }
+
+  void BlockOnNextWrite() {
+    writer_->BlockOnNextWrite();
+    EXPECT_CALL(visitor_, OnWriteBlocked()).Times(AtLeast(1));
+  }
+
+  void SimulateNextPacketTooLarge() { writer_->SimulateNextPacketTooLarge(); }
+
+  void AlwaysGetPacketTooLarge() { writer_->AlwaysGetPacketTooLarge(); }
+
+  void SetWritePauseTimeDelta(QuicTime::Delta delta) {
+    writer_->SetWritePauseTimeDelta(delta);
+  }
+
+  void CongestionBlockWrites() {
+    EXPECT_CALL(*send_algorithm_, CanSend(_))
+        .WillRepeatedly(testing::Return(false));
+  }
+
+  void CongestionUnblockWrites() {
+    EXPECT_CALL(*send_algorithm_, CanSend(_))
+        .WillRepeatedly(testing::Return(true));
+  }
+
+  void set_perspective(Perspective perspective) {
+    connection_.set_perspective(perspective);
+    if (perspective == Perspective::IS_SERVER) {
+      connection_.set_can_truncate_connection_ids(true);
+    }
+    QuicFramerPeer::SetPerspective(&peer_framer_,
+                                   InvertPerspective(perspective));
+  }
+
+  void set_packets_between_probes_base(
+      const QuicPacketCount packets_between_probes_base) {
+    QuicConnectionPeer::SetPacketsBetweenMtuProbes(&connection_,
+                                                   packets_between_probes_base);
+    QuicConnectionPeer::SetNextMtuProbeAt(
+        &connection_, QuicPacketNumber(packets_between_probes_base));
+  }
+
+  bool IsDefaultTestConfiguration() {
+    TestParams p = GetParam();
+    return p.ack_response == AckResponse::kImmediate &&
+           p.version == AllSupportedVersions()[0] && p.no_stop_waiting;
+  }
+
+  QuicConnectionId connection_id_;
+  QuicFramer framer_;
+
+  MockSendAlgorithm* send_algorithm_;
+  std::unique_ptr<MockLossAlgorithm> loss_algorithm_;
+  MockClock clock_;
+  MockRandom random_generator_;
+  SimpleBufferAllocator buffer_allocator_;
+  std::unique_ptr<TestConnectionHelper> helper_;
+  std::unique_ptr<TestAlarmFactory> alarm_factory_;
+  QuicFramer peer_framer_;
+  QuicPacketCreator peer_creator_;
+  std::unique_ptr<TestPacketWriter> writer_;
+  TestConnection connection_;
+  QuicPacketCreator* creator_;
+  QuicPacketGenerator* generator_;
+  QuicSentPacketManager* manager_;
+  StrictMock<MockQuicConnectionVisitor> visitor_;
+
+  QuicStreamFrame frame1_;
+  QuicStreamFrame frame2_;
+  QuicAckFrame ack_;
+  QuicStopWaitingFrame stop_waiting_;
+  QuicPacketNumberLength packet_number_length_;
+  QuicConnectionIdIncluded connection_id_included_;
+
+  SimpleSessionNotifier notifier_;
+};
+
+// Run all end to end tests with all supported versions.
+INSTANTIATE_TEST_SUITE_P(SupportedVersion,
+                         QuicConnectionTest,
+                         ::testing::ValuesIn(GetTestParams()));
+
+TEST_P(QuicConnectionTest, SelfAddressChangeAtClient) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+  EXPECT_TRUE(connection_.connected());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_));
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  // Cause change in self_address.
+  QuicIpAddress host;
+  host.FromString("1.1.1.1");
+  QuicSocketAddress self_address(host, 123);
+  EXPECT_CALL(visitor_, OnStreamFrame(_));
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), self_address,
+                                  kPeerAddress);
+  EXPECT_TRUE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, SelfAddressChangeAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+  EXPECT_TRUE(connection_.connected());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_));
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  // Cause change in self_address.
+  QuicIpAddress host;
+  host.FromString("1.1.1.1");
+  QuicSocketAddress self_address(host, 123);
+  EXPECT_CALL(visitor_, AllowSelfAddressChange()).WillOnce(Return(false));
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_ERROR_MIGRATING_ADDRESS, _, _));
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), self_address,
+                                  kPeerAddress);
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, AllowSelfAddressChangeToMappedIpv4AddressAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+  EXPECT_TRUE(connection_.connected());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(3);
+  QuicIpAddress host;
+  host.FromString("1.1.1.1");
+  QuicSocketAddress self_address1(host, 443);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), self_address1,
+                                  kPeerAddress);
+  // Cause self_address change to mapped Ipv4 address.
+  QuicIpAddress host2;
+  host2.FromString(
+      QuicStrCat("::ffff:", connection_.self_address().host().ToString()));
+  QuicSocketAddress self_address2(host2, connection_.self_address().port());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), self_address2,
+                                  kPeerAddress);
+  EXPECT_TRUE(connection_.connected());
+  // self_address change back to Ipv4 address.
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), self_address1,
+                                  kPeerAddress);
+  EXPECT_TRUE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, ClientAddressChangeAndPacketReordered) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 5);
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(),
+                        /*port=*/23456);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kNewPeerAddress);
+  EXPECT_EQ(kNewPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewPeerAddress, connection_.effective_peer_address());
+
+  // Decrease packet number to simulate out-of-order packets.
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 4);
+  // This is an old packet, do not migrate.
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kNewPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, PeerAddressChangeAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Process another packet with a different peer address on server side will
+  // start connection migration.
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(1);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kNewPeerAddress);
+  EXPECT_EQ(kNewPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, EffectivePeerAddressChangeAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is different from direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  const QuicSocketAddress kEffectivePeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/43210);
+  connection_.ReturnEffectivePeerAddressForNextPacket(kEffectivePeerAddress);
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kEffectivePeerAddress, connection_.effective_peer_address());
+
+  // Process another packet with the same direct peer address and different
+  // effective peer address on server side will start connection migration.
+  const QuicSocketAddress kNewEffectivePeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/54321);
+  connection_.ReturnEffectivePeerAddressForNextPacket(kNewEffectivePeerAddress);
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(1);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewEffectivePeerAddress, connection_.effective_peer_address());
+
+  // Process another packet with a different direct peer address and the same
+  // effective peer address on server side will not start connection migration.
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+  connection_.ReturnEffectivePeerAddressForNextPacket(kNewEffectivePeerAddress);
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  // ack_frame is used to complete the migration started by the last packet, we
+  // need to make sure a new migration does not start after the previous one is
+  // completed.
+  QuicAckFrame ack_frame = InitAckFrame(1);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(_, _, _, _, _));
+  ProcessFramePacketWithAddresses(QuicFrame(&ack_frame), kSelfAddress,
+                                  kNewPeerAddress);
+  EXPECT_EQ(kNewPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewEffectivePeerAddress, connection_.effective_peer_address());
+
+  // Process another packet with different direct peer address and different
+  // effective peer address on server side will start connection migration.
+  const QuicSocketAddress kNewerEffectivePeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/65432);
+  const QuicSocketAddress kFinalPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/34567);
+  connection_.ReturnEffectivePeerAddressForNextPacket(
+      kNewerEffectivePeerAddress);
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(1);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kFinalPeerAddress);
+  EXPECT_EQ(kFinalPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewerEffectivePeerAddress, connection_.effective_peer_address());
+  EXPECT_EQ(PORT_CHANGE, connection_.active_effective_peer_migration_type());
+
+  // While the previous migration is ongoing, process another packet with the
+  // same direct peer address and different effective peer address on server
+  // side will start a new connection migration.
+  const QuicSocketAddress kNewestEffectivePeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback4(), /*port=*/65430);
+  connection_.ReturnEffectivePeerAddressForNextPacket(
+      kNewestEffectivePeerAddress);
+  EXPECT_CALL(visitor_, OnConnectionMigration(IPV6_TO_IPV4_CHANGE)).Times(1);
+  EXPECT_CALL(*send_algorithm_, OnConnectionMigration()).Times(1);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kFinalPeerAddress);
+  EXPECT_EQ(kFinalPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewestEffectivePeerAddress, connection_.effective_peer_address());
+  EXPECT_EQ(IPV6_TO_IPV4_CHANGE,
+            connection_.active_effective_peer_migration_type());
+}
+
+TEST_P(QuicConnectionTest, ReceivePaddedPingAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(0);
+
+  // Process a padded PING or PATH CHALLENGE packet with no peer address change
+  // on server side will be ignored.
+  OwningSerializedPacketPointer probing_packet;
+  if (version().transport_version == QUIC_VERSION_99) {
+    QuicPathFrameBuffer payload = {
+        {0xde, 0xad, 0xbe, 0xef, 0xba, 0xdc, 0x0f, 0xfe}};
+    probing_packet =
+        QuicPacketCreatorPeer::SerializePathChallengeConnectivityProbingPacket(
+            &peer_creator_, &payload);
+  } else {
+    probing_packet = QuicPacketCreatorPeer::SerializeConnectivityProbingPacket(
+        &peer_creator_);
+  }
+  std::unique_ptr<QuicReceivedPacket> received(ConstructReceivedPacket(
+      QuicEncryptedPacket(probing_packet->encrypted_buffer,
+                          probing_packet->encrypted_length),
+      clock_.Now()));
+
+  uint64_t num_probing_received =
+      connection_.GetStats().num_connectivity_probing_received;
+  ProcessReceivedPacket(kSelfAddress, kPeerAddress, *received);
+
+  EXPECT_EQ(num_probing_received,
+            connection_.GetStats().num_connectivity_probing_received);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, WriteOutOfOrderQueuedPackets) {
+  // EXPECT_QUIC_BUG tests are expensive so only run one instance of them.
+  if (!IsDefaultTestConfiguration()) {
+    return;
+  }
+
+  set_perspective(Perspective::IS_CLIENT);
+
+  BlockOnNextWrite();
+
+  QuicStreamId stream_id = 2;
+  connection_.SendStreamDataWithString(stream_id, "foo", 0, NO_FIN);
+
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+
+  writer_->SetWritable();
+  connection_.SendConnectivityProbingPacket(writer_.get(),
+                                            connection_.peer_address());
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INTERNAL_ERROR,
+                                           "Packet written out of order.",
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_QUIC_BUG(connection_.OnCanWrite(),
+                  "Attempt to write packet:1 after:2");
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, DiscardQueuedPacketsAfterConnectionClose) {
+  // Regression test for b/74073386.
+  {
+    InSequence seq;
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+    EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(1);
+  }
+
+  set_perspective(Perspective::IS_CLIENT);
+
+  writer_->SimulateNextPacketTooLarge();
+
+  // This packet write should fail, which should cause the connection to close
+  // after sending a connection close packet, then the failed packet should be
+  // queued.
+  connection_.SendStreamDataWithString(/*id=*/2, "foo", 0, NO_FIN);
+
+  EXPECT_FALSE(connection_.connected());
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+
+  EXPECT_EQ(0u, connection_.GetStats().packets_discarded);
+  connection_.OnCanWrite();
+  EXPECT_EQ(1u, connection_.GetStats().packets_discarded);
+}
+
+TEST_P(QuicConnectionTest, ReceiveConnectivityProbingAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(1);
+
+  // Process a padded PING packet from a new peer address on server side
+  // is effectively receiving a connectivity probing.
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+
+  OwningSerializedPacketPointer probing_packet = ConstructProbingPacket();
+  std::unique_ptr<QuicReceivedPacket> received(ConstructReceivedPacket(
+      QuicEncryptedPacket(probing_packet->encrypted_buffer,
+                          probing_packet->encrypted_length),
+      clock_.Now()));
+
+  uint64_t num_probing_received =
+      connection_.GetStats().num_connectivity_probing_received;
+  ProcessReceivedPacket(kSelfAddress, kNewPeerAddress, *received);
+
+  EXPECT_EQ(num_probing_received + 1,
+            connection_.GetStats().num_connectivity_probing_received);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Process another packet with the old peer address on server side will not
+  // start peer migration.
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, ReceiveReorderedConnectivityProbingAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 5);
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Decrease packet number to simulate out-of-order packets.
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 4);
+
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(1);
+
+  // Process a padded PING packet from a new peer address on server side
+  // is effectively receiving a connectivity probing, even if a newer packet has
+  // been received before this one.
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+
+  OwningSerializedPacketPointer probing_packet = ConstructProbingPacket();
+  std::unique_ptr<QuicReceivedPacket> received(ConstructReceivedPacket(
+      QuicEncryptedPacket(probing_packet->encrypted_buffer,
+                          probing_packet->encrypted_length),
+      clock_.Now()));
+
+  uint64_t num_probing_received =
+      connection_.GetStats().num_connectivity_probing_received;
+  ProcessReceivedPacket(kSelfAddress, kNewPeerAddress, *received);
+
+  EXPECT_EQ(num_probing_received + 1,
+            connection_.GetStats().num_connectivity_probing_received);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, MigrateAfterProbingAtServer) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  EXPECT_EQ(Perspective::IS_SERVER, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(1);
+
+  // Process a padded PING packet from a new peer address on server side
+  // is effectively receiving a connectivity probing.
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+
+  OwningSerializedPacketPointer probing_packet = ConstructProbingPacket();
+  std::unique_ptr<QuicReceivedPacket> received(ConstructReceivedPacket(
+      QuicEncryptedPacket(probing_packet->encrypted_buffer,
+                          probing_packet->encrypted_length),
+      clock_.Now()));
+  ProcessReceivedPacket(kSelfAddress, kNewPeerAddress, *received);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Process another non-probing packet with the new peer address on server
+  // side will start peer migration.
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(1);
+
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kNewPeerAddress);
+  EXPECT_EQ(kNewPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, ReceivePaddedPingAtClient) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_CLIENT);
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Client takes all padded PING packet as speculative connectivity
+  // probing packet, and reports to visitor.
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(1);
+
+  OwningSerializedPacketPointer probing_packet = ConstructProbingPacket();
+  std::unique_ptr<QuicReceivedPacket> received(ConstructReceivedPacket(
+      QuicEncryptedPacket(probing_packet->encrypted_buffer,
+                          probing_packet->encrypted_length),
+      clock_.Now()));
+  uint64_t num_probing_received =
+      connection_.GetStats().num_connectivity_probing_received;
+  ProcessReceivedPacket(kSelfAddress, kPeerAddress, *received);
+
+  EXPECT_EQ(num_probing_received,
+            connection_.GetStats().num_connectivity_probing_received);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, ReceiveConnectivityProbingAtClient) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_CLIENT);
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Process a padded PING packet with a different self address on client side
+  // is effectively receiving a connectivity probing.
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  EXPECT_CALL(visitor_, OnConnectivityProbeReceived(_, _)).Times(1);
+
+  const QuicSocketAddress kNewSelfAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+
+  OwningSerializedPacketPointer probing_packet = ConstructProbingPacket();
+  std::unique_ptr<QuicReceivedPacket> received(ConstructReceivedPacket(
+      QuicEncryptedPacket(probing_packet->encrypted_buffer,
+                          probing_packet->encrypted_length),
+      clock_.Now()));
+  uint64_t num_probing_received =
+      connection_.GetStats().num_connectivity_probing_received;
+  ProcessReceivedPacket(kNewSelfAddress, kPeerAddress, *received);
+
+  EXPECT_EQ(num_probing_received + 1,
+            connection_.GetStats().num_connectivity_probing_received);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, PeerAddressChangeAtClient) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  set_perspective(Perspective::IS_CLIENT);
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+
+  // Clear direct_peer_address.
+  QuicConnectionPeer::SetDirectPeerAddress(&connection_, QuicSocketAddress());
+  // Clear effective_peer_address, it is the same as direct_peer_address for
+  // this test.
+  QuicConnectionPeer::SetEffectivePeerAddress(&connection_,
+                                              QuicSocketAddress());
+  EXPECT_FALSE(connection_.effective_peer_address().IsInitialized());
+
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(AnyNumber());
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+  EXPECT_EQ(kPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kPeerAddress, connection_.effective_peer_address());
+
+  // Process another packet with a different peer address on client side will
+  // only update peer address.
+  const QuicSocketAddress kNewPeerAddress =
+      QuicSocketAddress(QuicIpAddress::Loopback6(), /*port=*/23456);
+  EXPECT_CALL(visitor_, OnConnectionMigration(PORT_CHANGE)).Times(0);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kNewPeerAddress);
+  EXPECT_EQ(kNewPeerAddress, connection_.peer_address());
+  EXPECT_EQ(kNewPeerAddress, connection_.effective_peer_address());
+}
+
+TEST_P(QuicConnectionTest, MaxPacketSize) {
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+  EXPECT_EQ(1350u, connection_.max_packet_length());
+}
+
+TEST_P(QuicConnectionTest, SmallerServerMaxPacketSize) {
+  TestConnection connection(TestConnectionId(), kPeerAddress, helper_.get(),
+                            alarm_factory_.get(), writer_.get(),
+                            Perspective::IS_SERVER, version());
+  EXPECT_EQ(Perspective::IS_SERVER, connection.perspective());
+  EXPECT_EQ(1000u, connection.max_packet_length());
+}
+
+TEST_P(QuicConnectionTest, IncreaseServerMaxPacketSize) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  set_perspective(Perspective::IS_SERVER);
+  connection_.SetMaxPacketLength(1000);
+
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  header.version_flag = true;
+  header.packet_number = QuicPacketNumber(1);
+
+  if (QuicVersionHasLongHeaderLengths(
+          peer_framer_.version().transport_version)) {
+    header.long_packet_type = INITIAL;
+    header.retry_token_length_length = VARIABLE_LENGTH_INTEGER_LENGTH_1;
+    header.length_length = VARIABLE_LENGTH_INTEGER_LENGTH_2;
+  }
+
+  QuicFrames frames;
+  QuicPaddingFrame padding;
+  frames.push_back(QuicFrame(frame1_));
+  frames.push_back(QuicFrame(padding));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = peer_framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(12), *packet, buffer, kMaxPacketSize);
+  EXPECT_EQ(kMaxPacketSize, encrypted_length);
+
+  framer_.set_version(version());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+
+  EXPECT_EQ(kMaxPacketSize, connection_.max_packet_length());
+}
+
+TEST_P(QuicConnectionTest, IncreaseServerMaxPacketSizeWhileWriterLimited) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  const QuicByteCount lower_max_packet_size = 1240;
+  writer_->set_max_packet_size(lower_max_packet_size);
+  set_perspective(Perspective::IS_SERVER);
+  connection_.SetMaxPacketLength(1000);
+  EXPECT_EQ(1000u, connection_.max_packet_length());
+
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  header.version_flag = true;
+  header.packet_number = QuicPacketNumber(1);
+
+  if (QuicVersionHasLongHeaderLengths(
+          peer_framer_.version().transport_version)) {
+    header.long_packet_type = INITIAL;
+    header.retry_token_length_length = VARIABLE_LENGTH_INTEGER_LENGTH_1;
+    header.length_length = VARIABLE_LENGTH_INTEGER_LENGTH_2;
+  }
+
+  QuicFrames frames;
+  QuicPaddingFrame padding;
+  frames.push_back(QuicFrame(frame1_));
+  frames.push_back(QuicFrame(padding));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = peer_framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(12), *packet, buffer, kMaxPacketSize);
+  EXPECT_EQ(kMaxPacketSize, encrypted_length);
+
+  framer_.set_version(version());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+
+  // Here, the limit imposed by the writer is lower than the size of the packet
+  // received, so the writer max packet size is used.
+  EXPECT_EQ(lower_max_packet_size, connection_.max_packet_length());
+}
+
+TEST_P(QuicConnectionTest, LimitMaxPacketSizeByWriter) {
+  const QuicByteCount lower_max_packet_size = 1240;
+  writer_->set_max_packet_size(lower_max_packet_size);
+
+  static_assert(lower_max_packet_size < kDefaultMaxPacketSize,
+                "Default maximum packet size is too low");
+  connection_.SetMaxPacketLength(kDefaultMaxPacketSize);
+
+  EXPECT_EQ(lower_max_packet_size, connection_.max_packet_length());
+}
+
+TEST_P(QuicConnectionTest, LimitMaxPacketSizeByWriterForNewConnection) {
+  const QuicConnectionId connection_id = TestConnectionId(17);
+  const QuicByteCount lower_max_packet_size = 1240;
+  writer_->set_max_packet_size(lower_max_packet_size);
+  TestConnection connection(connection_id, kPeerAddress, helper_.get(),
+                            alarm_factory_.get(), writer_.get(),
+                            Perspective::IS_CLIENT, version());
+  EXPECT_EQ(Perspective::IS_CLIENT, connection.perspective());
+  EXPECT_EQ(lower_max_packet_size, connection.max_packet_length());
+}
+
+TEST_P(QuicConnectionTest, PacketsInOrder) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  ProcessPacket(1);
+  EXPECT_EQ(QuicPacketNumber(1u), LargestAcked(*outgoing_ack()));
+  EXPECT_EQ(1u, outgoing_ack()->packets.NumIntervals());
+
+  ProcessPacket(2);
+  EXPECT_EQ(QuicPacketNumber(2u), LargestAcked(*outgoing_ack()));
+  EXPECT_EQ(1u, outgoing_ack()->packets.NumIntervals());
+
+  ProcessPacket(3);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_EQ(1u, outgoing_ack()->packets.NumIntervals());
+}
+
+TEST_P(QuicConnectionTest, PacketsOutOfOrder) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  ProcessPacket(3);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_TRUE(IsMissing(2));
+  EXPECT_TRUE(IsMissing(1));
+
+  ProcessPacket(2);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_FALSE(IsMissing(2));
+  EXPECT_TRUE(IsMissing(1));
+
+  ProcessPacket(1);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_FALSE(IsMissing(2));
+  EXPECT_FALSE(IsMissing(1));
+}
+
+TEST_P(QuicConnectionTest, DuplicatePacket) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  ProcessPacket(3);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_TRUE(IsMissing(2));
+  EXPECT_TRUE(IsMissing(1));
+
+  // Send packet 3 again, but do not set the expectation that
+  // the visitor OnStreamFrame() will be called.
+  ProcessDataPacket(3);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_TRUE(IsMissing(2));
+  EXPECT_TRUE(IsMissing(1));
+}
+
+TEST_P(QuicConnectionTest, PacketsOutOfOrderWithAdditionsAndLeastAwaiting) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  ProcessPacket(3);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_TRUE(IsMissing(2));
+  EXPECT_TRUE(IsMissing(1));
+
+  ProcessPacket(2);
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(*outgoing_ack()));
+  EXPECT_TRUE(IsMissing(1));
+
+  ProcessPacket(5);
+  EXPECT_EQ(QuicPacketNumber(5u), LargestAcked(*outgoing_ack()));
+  EXPECT_TRUE(IsMissing(1));
+  EXPECT_TRUE(IsMissing(4));
+
+  // Pretend at this point the client has gotten acks for 2 and 3 and 1 is a
+  // packet the peer will not retransmit.  It indicates this by sending 'least
+  // awaiting' is 4.  The connection should then realize 1 will not be
+  // retransmitted, and will remove it from the missing list.
+  QuicAckFrame frame = InitAckFrame(1);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(_, _, _, _, _));
+  ProcessAckPacket(6, &frame);
+
+  // Force an ack to be sent.
+  SendAckPacketToPeer();
+  EXPECT_TRUE(IsMissing(4));
+}
+
+TEST_P(QuicConnectionTest, RejectPacketTooFarOut) {
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_PACKET_HEADER, _,
+                                           ConnectionCloseSource::FROM_SELF));
+
+  // Call ProcessDataPacket rather than ProcessPacket, as we should not get a
+  // packet call to the visitor.
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    ProcessDataPacket(MaxRandomInitialPacketNumber() + 6000);
+  } else {
+    ProcessDataPacket(6000);
+  }
+  EXPECT_FALSE(QuicConnectionPeer::GetConnectionClosePacket(&connection_) ==
+               nullptr);
+}
+
+TEST_P(QuicConnectionTest, RejectUnencryptedStreamData) {
+  // EXPECT_QUIC_BUG tests are expensive so only run one instance of them.
+  if (!IsDefaultTestConfiguration()) {
+    return;
+  }
+
+  // Process an unencrypted packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_UNENCRYPTED_STREAM_DATA, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_QUIC_PEER_BUG(ProcessDataPacket(1), "");
+  EXPECT_FALSE(QuicConnectionPeer::GetConnectionClosePacket(&connection_) ==
+               nullptr);
+  const std::vector<QuicConnectionCloseFrame>& connection_close_frames =
+      writer_->connection_close_frames();
+  EXPECT_EQ(1u, connection_close_frames.size());
+  EXPECT_EQ(QUIC_UNENCRYPTED_STREAM_DATA,
+            connection_close_frames[0].error_code);
+}
+
+TEST_P(QuicConnectionTest, OutOfOrderReceiptCausesAckSend) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  ProcessPacket(3);
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    // Should not cause an ack.
+    EXPECT_EQ(0u, writer_->packets_write_attempts());
+  } else {
+    // Should ack immediately since we have missing packets.
+    EXPECT_EQ(1u, writer_->packets_write_attempts());
+  }
+
+  ProcessPacket(2);
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    // Should ack immediately, since this fills the last hole.
+    EXPECT_EQ(1u, writer_->packets_write_attempts());
+  } else {
+    // Should ack immediately since we have missing packets.
+    EXPECT_EQ(2u, writer_->packets_write_attempts());
+  }
+
+  ProcessPacket(1);
+  // Should ack immediately, since this fills the last hole.
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    EXPECT_EQ(2u, writer_->packets_write_attempts());
+  } else {
+    EXPECT_EQ(3u, writer_->packets_write_attempts());
+  }
+
+  ProcessPacket(4);
+  // Should not cause an ack.
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    EXPECT_EQ(2u, writer_->packets_write_attempts());
+  } else {
+    EXPECT_EQ(3u, writer_->packets_write_attempts());
+  }
+}
+
+TEST_P(QuicConnectionTest, OutOfOrderAckReceiptCausesNoAck) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, nullptr);
+  SendStreamDataToPeer(1, "bar", 3, NO_FIN, nullptr);
+  EXPECT_EQ(2u, writer_->packets_write_attempts());
+
+  QuicAckFrame ack1 = InitAckFrame(1);
+  QuicAckFrame ack2 = InitAckFrame(2);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(2, &ack2);
+  // Should ack immediately since we have missing packets.
+  EXPECT_EQ(2u, writer_->packets_write_attempts());
+
+  ProcessAckPacket(1, &ack1);
+  // Should not ack an ack filling a missing packet.
+  EXPECT_EQ(2u, writer_->packets_write_attempts());
+}
+
+TEST_P(QuicConnectionTest, AckReceiptCausesAckSend) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  QuicPacketNumber original, second;
+
+  QuicByteCount packet_size =
+      SendStreamDataToPeer(3, "foo", 0, NO_FIN, &original);  // 1st packet.
+  SendStreamDataToPeer(3, "bar", 3, NO_FIN, &second);        // 2nd packet.
+
+  QuicAckFrame frame = InitAckFrame({{second, second + 1}});
+  // First nack triggers early retransmit.
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(original, kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicPacketNumber retransmission;
+  // Packet 1 is short header for IETF QUIC because the encryption level
+  // switched to ENCRYPTION_FORWARD_SECURE in SendStreamDataToPeer.
+  EXPECT_CALL(
+      *send_algorithm_,
+      OnPacketSent(_, _, _,
+                   GetParam().version.transport_version > QUIC_VERSION_43
+                       ? packet_size
+                       : packet_size - kQuicVersionSize,
+                   _))
+      .WillOnce(SaveArg<2>(&retransmission));
+
+  ProcessAckPacket(&frame);
+
+  QuicAckFrame frame2 = ConstructAckFrame(retransmission, original);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  ProcessAckPacket(&frame2);
+
+  // Now if the peer sends an ack which still reports the retransmitted packet
+  // as missing, that will bundle an ack with data after two acks in a row
+  // indicate the high water mark needs to be raised.
+  EXPECT_CALL(*send_algorithm_,
+              OnPacketSent(_, _, _, _, HAS_RETRANSMITTABLE_DATA));
+  connection_.SendStreamDataWithString(3, "foo", 6, NO_FIN);
+  // No ack sent.
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+
+  // No more packet loss for the rest of the test.
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .Times(AnyNumber());
+  ProcessAckPacket(&frame2);
+  EXPECT_CALL(*send_algorithm_,
+              OnPacketSent(_, _, _, _, HAS_RETRANSMITTABLE_DATA));
+  connection_.SendStreamDataWithString(3, "foo", 9, NO_FIN);
+  // Ack bundled.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(2u, writer_->frame_count());
+  } else {
+    EXPECT_EQ(3u, writer_->frame_count());
+  }
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_FALSE(writer_->ack_frames().empty());
+
+  // But an ack with no missing packets will not send an ack.
+  AckPacket(original, &frame2);
+  ProcessAckPacket(&frame2);
+  ProcessAckPacket(&frame2);
+}
+
+TEST_P(QuicConnectionTest, AckSentEveryNthPacket) {
+  connection_.set_ack_frequency_before_ack_decimation(3);
+
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(39);
+
+  // Expect 13 acks, every 3rd packet.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(13);
+  // Receives packets 1 - 39.
+  for (size_t i = 1; i <= 39; ++i) {
+    ProcessDataPacket(i);
+  }
+}
+
+TEST_P(QuicConnectionTest, AckDecimationReducesAcks) {
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION_WITH_REORDERING);
+
+  // Start ack decimation from 10th packet.
+  connection_.set_min_received_before_ack_decimation(10);
+
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(30);
+
+  // Expect 6 acks: 5 acks between packets 1-10, and ack at 20.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(6);
+  // Receives packets 1 - 29.
+  for (size_t i = 1; i <= 29; ++i) {
+    ProcessDataPacket(i);
+  }
+
+  // We now receive the 30th packet, and so we send an ack.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  ProcessDataPacket(30);
+}
+
+TEST_P(QuicConnectionTest, AckNeedsRetransmittableFrames) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(99);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(19);
+  // Receives packets 1 - 39.
+  for (size_t i = 1; i <= 39; ++i) {
+    ProcessDataPacket(i);
+  }
+  // Receiving Packet 40 causes 20th ack to send. Session is informed and adds
+  // WINDOW_UPDATE.
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame())
+      .WillOnce(Invoke([this]() {
+        connection_.SendControlFrame(
+            QuicFrame(new QuicWindowUpdateFrame(1, 0, 0)));
+      }));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  EXPECT_EQ(0u, writer_->window_update_frames().size());
+  ProcessDataPacket(40);
+  EXPECT_EQ(1u, writer_->window_update_frames().size());
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(9);
+  // Receives packets 41 - 59.
+  for (size_t i = 41; i <= 59; ++i) {
+    ProcessDataPacket(i);
+  }
+  // Send a packet containing stream frame.
+  SendStreamDataToPeer(1, "bar", 0, NO_FIN, nullptr);
+
+  // Session will not be informed until receiving another 20 packets.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(19);
+  for (size_t i = 60; i <= 98; ++i) {
+    ProcessDataPacket(i);
+    EXPECT_EQ(0u, writer_->window_update_frames().size());
+  }
+  // Session does not add a retransmittable frame.
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame())
+      .WillOnce(Invoke([this]() {
+        connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+      }));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  EXPECT_EQ(0u, writer_->ping_frames().size());
+  ProcessDataPacket(99);
+  EXPECT_EQ(0u, writer_->window_update_frames().size());
+  // A ping frame will be added.
+  EXPECT_EQ(1u, writer_->ping_frames().size());
+}
+
+TEST_P(QuicConnectionTest, LeastUnackedLower) {
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    return;
+  }
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, nullptr);
+  SendStreamDataToPeer(1, "bar", 3, NO_FIN, nullptr);
+  SendStreamDataToPeer(1, "eep", 6, NO_FIN, nullptr);
+
+  // Start out saying the least unacked is 2.
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 5);
+  ProcessStopWaitingPacket(InitStopWaitingFrame(2));
+
+  // Change it to 1, but lower the packet number to fake out-of-order packets.
+  // This should be fine.
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 1);
+  // The scheduler will not process out of order acks, but all packet processing
+  // causes the connection to try to write.
+  if (!GetParam().no_stop_waiting) {
+    EXPECT_CALL(visitor_, OnCanWrite());
+  }
+  ProcessStopWaitingPacket(InitStopWaitingFrame(1));
+
+  // Now claim it's one, but set the ordering so it was sent "after" the first
+  // one.  This should cause a connection error.
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 7);
+  if (!GetParam().no_stop_waiting) {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_STOP_WAITING_DATA, _,
+                                             ConnectionCloseSource::FROM_SELF));
+  }
+  ProcessStopWaitingPacket(InitStopWaitingFrame(1));
+}
+
+TEST_P(QuicConnectionTest, TooManySentPackets) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  QuicPacketCount max_tracked_packets = 50;
+  QuicConnectionPeer::SetMaxTrackedPackets(&connection_, max_tracked_packets);
+
+  const int num_packets = max_tracked_packets + 5;
+
+  for (int i = 0; i < num_packets; ++i) {
+    SendStreamDataToPeer(1, "foo", 3 * i, NO_FIN, nullptr);
+  }
+
+  // Ack packet 1, which leaves more than the limit outstanding.
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(visitor_,
+              OnConnectionClosed(QUIC_TOO_MANY_OUTSTANDING_SENT_PACKETS, _,
+                                 ConnectionCloseSource::FROM_SELF));
+
+  // Nack the first packet and ack the rest, leaving a huge gap.
+  QuicAckFrame frame1 = ConstructAckFrame(num_packets, 1);
+  ProcessAckPacket(&frame1);
+}
+
+TEST_P(QuicConnectionTest, LargestObservedLower) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, nullptr);
+  SendStreamDataToPeer(1, "bar", 3, NO_FIN, nullptr);
+  SendStreamDataToPeer(1, "eep", 6, NO_FIN, nullptr);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+
+  // Start out saying the largest observed is 2.
+  QuicAckFrame frame1 = InitAckFrame(1);
+  QuicAckFrame frame2 = InitAckFrame(2);
+  ProcessAckPacket(&frame2);
+
+  // Now change it to 1, and it should cause a connection error.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_ACK_DATA, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(visitor_, OnCanWrite()).Times(0);
+  ProcessAckPacket(&frame1);
+}
+
+TEST_P(QuicConnectionTest, AckUnsentData) {
+  // Ack a packet which has not been sent.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_ACK_DATA, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  QuicAckFrame frame = InitAckFrame(1);
+  EXPECT_CALL(visitor_, OnCanWrite()).Times(0);
+  ProcessAckPacket(&frame);
+}
+
+TEST_P(QuicConnectionTest, BasicSending) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(1);
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 2);
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);  // Packet 1
+  EXPECT_EQ(QuicPacketNumber(1u), last_packet);
+  SendAckPacketToPeer();  // Packet 2
+
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(1u), least_unacked());
+  }
+
+  SendAckPacketToPeer();  // Packet 3
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(1u), least_unacked());
+  }
+
+  SendStreamDataToPeer(1, "bar", 3, NO_FIN, &last_packet);  // Packet 4
+  EXPECT_EQ(QuicPacketNumber(4u), last_packet);
+  SendAckPacketToPeer();  // Packet 5
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(1u), least_unacked());
+  }
+
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+
+  // Peer acks up to packet 3.
+  QuicAckFrame frame = InitAckFrame(3);
+  ProcessAckPacket(&frame);
+  SendAckPacketToPeer();  // Packet 6
+
+  // As soon as we've acked one, we skip ack packets 2 and 3 and note lack of
+  // ack for 4.
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(4u), least_unacked());
+  }
+
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+
+  // Peer acks up to packet 4, the last packet.
+  QuicAckFrame frame2 = InitAckFrame(6);
+  ProcessAckPacket(&frame2);  // Acks don't instigate acks.
+
+  // Verify that we did not send an ack.
+  EXPECT_EQ(QuicPacketNumber(6u), writer_->header().packet_number);
+
+  // So the last ack has not changed.
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(4u), least_unacked());
+  }
+
+  // If we force an ack, we shouldn't change our retransmit state.
+  SendAckPacketToPeer();  // Packet 7
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(7u), least_unacked());
+  }
+
+  // But if we send more data it should.
+  SendStreamDataToPeer(1, "eep", 6, NO_FIN, &last_packet);  // Packet 8
+  EXPECT_EQ(QuicPacketNumber(8u), last_packet);
+  SendAckPacketToPeer();  // Packet 9
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(7u), least_unacked());
+  }
+}
+
+// QuicConnection should record the packet sent-time prior to sending the
+// packet.
+TEST_P(QuicConnectionTest, RecordSentTimeBeforePacketSent) {
+  // We're using a MockClock for the tests, so we have complete control over the
+  // time.
+  // Our recorded timestamp for the last packet sent time will be passed in to
+  // the send_algorithm.  Make sure that it is set to the correct value.
+  QuicTime actual_recorded_send_time = QuicTime::Zero();
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<0>(&actual_recorded_send_time));
+
+  // First send without any pause and check the result.
+  QuicTime expected_recorded_send_time = clock_.Now();
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_EQ(expected_recorded_send_time, actual_recorded_send_time)
+      << "Expected time = " << expected_recorded_send_time.ToDebuggingValue()
+      << ".  Actual time = " << actual_recorded_send_time.ToDebuggingValue();
+
+  // Now pause during the write, and check the results.
+  actual_recorded_send_time = QuicTime::Zero();
+  const QuicTime::Delta write_pause_time_delta =
+      QuicTime::Delta::FromMilliseconds(5000);
+  SetWritePauseTimeDelta(write_pause_time_delta);
+  expected_recorded_send_time = clock_.Now();
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<0>(&actual_recorded_send_time));
+  connection_.SendStreamDataWithString(2, "baz", 0, NO_FIN);
+  EXPECT_EQ(expected_recorded_send_time, actual_recorded_send_time)
+      << "Expected time = " << expected_recorded_send_time.ToDebuggingValue()
+      << ".  Actual time = " << actual_recorded_send_time.ToDebuggingValue();
+}
+
+TEST_P(QuicConnectionTest, FramePacking) {
+  // Send two stream frames in 1 packet by queueing them.
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+  {
+    QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                QuicConnection::SEND_ACK);
+    connection_.SendStreamData3();
+    connection_.SendStreamData5();
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  }
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure it's an ack and two stream frames from
+  // two different streams.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  }
+
+  EXPECT_TRUE(writer_->ack_frames().empty());
+
+  ASSERT_EQ(2u, writer_->stream_frames().size());
+  EXPECT_EQ(GetNthClientInitiatedStreamId(1, connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+  EXPECT_EQ(GetNthClientInitiatedStreamId(2, connection_.transport_version()),
+            writer_->stream_frames()[1]->stream_id);
+}
+
+TEST_P(QuicConnectionTest, FramePackingNonCryptoThenCrypto) {
+  // Send two stream frames (one non-crypto, then one crypto) in 2 packets by
+  // queueing them.
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+  {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+    QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                QuicConnection::SEND_ACK);
+    connection_.SendStreamData3();
+    connection_.SendCryptoStreamData();
+  }
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure it's the crypto stream frame.
+  EXPECT_EQ(2u, writer_->frame_count());
+  ASSERT_EQ(1u, writer_->padding_frames().size());
+  if (connection_.transport_version() < QUIC_VERSION_47) {
+    ASSERT_EQ(1u, writer_->stream_frames().size());
+    EXPECT_EQ(QuicUtils::GetCryptoStreamId(connection_.transport_version()),
+              writer_->stream_frames()[0]->stream_id);
+  } else {
+    EXPECT_EQ(1u, writer_->crypto_frames().size());
+  }
+}
+
+TEST_P(QuicConnectionTest, FramePackingCryptoThenNonCrypto) {
+  // Send two stream frames (one crypto, then one non-crypto) in 2 packets by
+  // queueing them.
+  {
+    connection_.SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+    QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                QuicConnection::SEND_ACK);
+    connection_.SendCryptoStreamData();
+    connection_.SendStreamData3();
+  }
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure it's the stream frame from stream 3.
+  EXPECT_EQ(1u, writer_->frame_count());
+  ASSERT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_EQ(GetNthClientInitiatedStreamId(1, connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+}
+
+TEST_P(QuicConnectionTest, FramePackingAckResponse) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  // Process a data packet to queue up a pending ack.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(1);
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);
+  // Verify ack is bundled with outging packet.
+  EXPECT_FALSE(writer_->ack_frames().empty());
+
+  EXPECT_CALL(visitor_, OnCanWrite())
+      .WillOnce(DoAll(IgnoreResult(InvokeWithoutArgs(
+                          &connection_, &TestConnection::SendStreamData3)),
+                      IgnoreResult(InvokeWithoutArgs(
+                          &connection_, &TestConnection::SendStreamData5))));
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+
+  // Process a data packet to cause the visitor's OnCanWrite to be invoked.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(2);
+
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure it's an ack and two stream frames from
+  // two different streams.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(3u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(4u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  ASSERT_EQ(2u, writer_->stream_frames().size());
+  EXPECT_EQ(GetNthClientInitiatedStreamId(1, connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+  EXPECT_EQ(GetNthClientInitiatedStreamId(2, connection_.transport_version()),
+            writer_->stream_frames()[1]->stream_id);
+}
+
+TEST_P(QuicConnectionTest, FramePackingSendv) {
+  // Send data in 1 packet by writing multiple blocks in a single iovector
+  // using writev.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+
+  char data[] = "ABCDEF";
+  struct iovec iov[2];
+  iov[0].iov_base = data;
+  iov[0].iov_len = 4;
+  iov[1].iov_base = data + 4;
+  iov[1].iov_len = 2;
+  connection_.SaveAndSendStreamData(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), iov, 2, 6,
+      0, NO_FIN);
+
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure multiple iovector blocks have
+  // been packed into a single stream frame from one stream.
+  EXPECT_EQ(2u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_EQ(1u, writer_->padding_frames().size());
+  QuicStreamFrame* frame = writer_->stream_frames()[0].get();
+  EXPECT_EQ(QuicUtils::GetCryptoStreamId(connection_.transport_version()),
+            frame->stream_id);
+  EXPECT_EQ("ABCDEF", QuicStringPiece(frame->data_buffer, frame->data_length));
+}
+
+TEST_P(QuicConnectionTest, FramePackingSendvQueued) {
+  // Try to send two stream frames in 1 packet by using writev.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+
+  BlockOnNextWrite();
+  char data[] = "ABCDEF";
+  struct iovec iov[2];
+  iov[0].iov_base = data;
+  iov[0].iov_len = 4;
+  iov[1].iov_base = data + 4;
+  iov[1].iov_len = 2;
+  connection_.SaveAndSendStreamData(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), iov, 2, 6,
+      0, NO_FIN);
+
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+  EXPECT_TRUE(connection_.HasQueuedData());
+
+  // Unblock the writes and actually send.
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+
+  // Parse the last packet and ensure it's one stream frame from one stream.
+  EXPECT_EQ(2u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_EQ(1u, writer_->padding_frames().size());
+  EXPECT_EQ(QuicUtils::GetCryptoStreamId(connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+}
+
+TEST_P(QuicConnectionTest, SendingZeroBytes) {
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+  // Send a zero byte write with a fin using writev.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.SaveAndSendStreamData(
+      QuicUtils::GetHeadersStreamId(connection_.transport_version()), nullptr,
+      0, 0, 0, FIN);
+
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure it's one stream frame from one stream.
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_EQ(QuicUtils::GetHeadersStreamId(connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+  EXPECT_TRUE(writer_->stream_frames()[0]->fin);
+}
+
+TEST_P(QuicConnectionTest, LargeSendWithPendingAck) {
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+  // Set the ack alarm by processing a ping frame.
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  // Processs a PING frame.
+  ProcessFramePacket(QuicFrame(QuicPingFrame()));
+  // Ensure that this has caused the ACK alarm to be set.
+  QuicAlarm* ack_alarm = QuicConnectionPeer::GetAckAlarm(&connection_);
+  EXPECT_TRUE(ack_alarm->IsSet());
+
+  // Send data and ensure the ack is bundled.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(8);
+  size_t len = 10000;
+  std::unique_ptr<char[]> data_array(new char[len]);
+  memset(data_array.get(), '?', len);
+  struct iovec iov;
+  iov.iov_base = data_array.get();
+  iov.iov_len = len;
+  QuicConsumedData consumed = connection_.SaveAndSendStreamData(
+      QuicUtils::GetHeadersStreamId(connection_.transport_version()), &iov, 1,
+      len, 0, FIN);
+  EXPECT_EQ(len, consumed.bytes_consumed);
+  EXPECT_TRUE(consumed.fin_consumed);
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.HasQueuedData());
+
+  // Parse the last packet and ensure it's one stream frame with a fin.
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_EQ(QuicUtils::GetHeadersStreamId(connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+  EXPECT_TRUE(writer_->stream_frames()[0]->fin);
+  // Ensure the ack alarm was cancelled when the ack was sent.
+  EXPECT_FALSE(ack_alarm->IsSet());
+}
+
+TEST_P(QuicConnectionTest, OnCanWrite) {
+  // Visitor's OnCanWrite will send data, but will have more pending writes.
+  EXPECT_CALL(visitor_, OnCanWrite())
+      .WillOnce(DoAll(IgnoreResult(InvokeWithoutArgs(
+                          &connection_, &TestConnection::SendStreamData3)),
+                      IgnoreResult(InvokeWithoutArgs(
+                          &connection_, &TestConnection::SendStreamData5))));
+  {
+    InSequence seq;
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite()).WillOnce(Return(true));
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite())
+        .WillRepeatedly(Return(false));
+  }
+
+  EXPECT_CALL(*send_algorithm_, CanSend(_))
+      .WillRepeatedly(testing::Return(true));
+
+  connection_.OnCanWrite();
+
+  // Parse the last packet and ensure it's the two stream frames from
+  // two different streams.
+  EXPECT_EQ(2u, writer_->frame_count());
+  EXPECT_EQ(2u, writer_->stream_frames().size());
+  EXPECT_EQ(GetNthClientInitiatedStreamId(1, connection_.transport_version()),
+            writer_->stream_frames()[0]->stream_id);
+  EXPECT_EQ(GetNthClientInitiatedStreamId(2, connection_.transport_version()),
+            writer_->stream_frames()[1]->stream_id);
+}
+
+TEST_P(QuicConnectionTest, RetransmitOnNack) {
+  QuicPacketNumber last_packet;
+  QuicByteCount second_packet_size;
+  SendStreamDataToPeer(3, "foo", 0, NO_FIN, &last_packet);  // Packet 1
+  second_packet_size =
+      SendStreamDataToPeer(3, "foos", 3, NO_FIN, &last_packet);  // Packet 2
+  SendStreamDataToPeer(3, "fooos", 7, NO_FIN, &last_packet);     // Packet 3
+
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  // Don't lose a packet on an ack, and nothing is retransmitted.
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame ack_one = InitAckFrame(1);
+  ProcessAckPacket(&ack_one);
+
+  // Lose a packet and ensure it triggers retransmission.
+  QuicAckFrame nack_two = ConstructAckFrame(3, 2);
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(QuicPacketNumber(2), kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  EXPECT_FALSE(QuicPacketCreatorPeer::SendVersionInPacket(creator_));
+  ProcessAckPacket(&nack_two);
+}
+
+TEST_P(QuicConnectionTest, DoNotSendQueuedPacketForResetStream) {
+  // Block the connection to queue the packet.
+  BlockOnNextWrite();
+
+  QuicStreamId stream_id = 2;
+  connection_.SendStreamDataWithString(stream_id, "foo", 0, NO_FIN);
+
+  // Now that there is a queued packet, reset the stream.
+  SendRstStream(stream_id, QUIC_ERROR_PROCESSING_STREAM, 3);
+
+  // Unblock the connection and verify that only the RST_STREAM is sent.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  if (!connection_.session_decides_what_to_write()) {
+    // OnCanWrite will cause RST_STREAM be sent again.
+    connection_.SendControlFrame(QuicFrame(new QuicRstStreamFrame(
+        1, stream_id, QUIC_ERROR_PROCESSING_STREAM, 14)));
+  }
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->rst_stream_frames().size());
+}
+
+TEST_P(QuicConnectionTest, SendQueuedPacketForQuicRstStreamNoError) {
+  // Block the connection to queue the packet.
+  BlockOnNextWrite();
+
+  QuicStreamId stream_id = 2;
+  connection_.SendStreamDataWithString(stream_id, "foo", 0, NO_FIN);
+
+  // Now that there is a queued packet, reset the stream.
+  SendRstStream(stream_id, QUIC_STREAM_NO_ERROR, 3);
+
+  // Unblock the connection and verify that the RST_STREAM is sent and the data
+  // packet is sent.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AtLeast(2));
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  if (!connection_.session_decides_what_to_write()) {
+    // OnCanWrite will cause RST_STREAM be sent again.
+    connection_.SendControlFrame(QuicFrame(
+        new QuicRstStreamFrame(1, stream_id, QUIC_STREAM_NO_ERROR, 14)));
+  }
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->rst_stream_frames().size());
+}
+
+TEST_P(QuicConnectionTest, DoNotRetransmitForResetStreamOnNack) {
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "foos", 3, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "fooos", 7, NO_FIN, &last_packet);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  SendRstStream(stream_id, QUIC_ERROR_PROCESSING_STREAM, 12);
+
+  // Lose a packet and ensure it does not trigger retransmission.
+  QuicAckFrame nack_two = ConstructAckFrame(last_packet, last_packet - 1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  ProcessAckPacket(&nack_two);
+}
+
+TEST_P(QuicConnectionTest, RetransmitForQuicRstStreamNoErrorOnNack) {
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "foos", 3, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "fooos", 7, NO_FIN, &last_packet);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  SendRstStream(stream_id, QUIC_STREAM_NO_ERROR, 12);
+
+  // Lose a packet, ensure it triggers retransmission.
+  QuicAckFrame nack_two = ConstructAckFrame(last_packet, last_packet - 1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(last_packet - 1, kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AtLeast(1));
+  ProcessAckPacket(&nack_two);
+}
+
+TEST_P(QuicConnectionTest, DoNotRetransmitForResetStreamOnRTO) {
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  SendRstStream(stream_id, QUIC_ERROR_PROCESSING_STREAM, 3);
+
+  // Fire the RTO and verify that the RST_STREAM is resent, not stream data.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  connection_.GetRetransmissionAlarm()->Fire();
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->rst_stream_frames().size());
+  EXPECT_EQ(stream_id, writer_->rst_stream_frames().front().stream_id);
+}
+
+// Ensure that if the only data in flight is non-retransmittable, the
+// retransmission alarm is not set.
+TEST_P(QuicConnectionTest, CancelRetransmissionAlarmAfterResetStream) {
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_data_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_data_packet);
+
+  // Cancel the stream.
+  const QuicPacketNumber rst_packet = last_data_packet + 1;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, rst_packet, _, _)).Times(1);
+  SendRstStream(stream_id, QUIC_ERROR_PROCESSING_STREAM, 3);
+
+  // Ack the RST_STREAM frame (since it's retransmittable), but not the data
+  // packet, which is no longer retransmittable since the stream was cancelled.
+  QuicAckFrame nack_stream_data =
+      ConstructAckFrame(rst_packet, last_data_packet);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  ProcessAckPacket(&nack_stream_data);
+
+  // Ensure that the data is still in flight, but the retransmission alarm is no
+  // longer set.
+  EXPECT_GT(QuicSentPacketManagerPeer::GetBytesInFlight(manager_), 0u);
+  if (GetQuicReloadableFlag(quic_optimize_inflight_check)) {
+    EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+    // Firing the alarm should remove all bytes_in_flight.
+    connection_.GetRetransmissionAlarm()->Fire();
+    EXPECT_EQ(0u, QuicSentPacketManagerPeer::GetBytesInFlight(manager_));
+  }
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, RetransmitForQuicRstStreamNoErrorOnRTO) {
+  connection_.SetMaxTailLossProbes(0);
+
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  SendRstStream(stream_id, QUIC_STREAM_NO_ERROR, 3);
+
+  // Fire the RTO and verify that the RST_STREAM is resent, the stream data
+  // is sent.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AtLeast(2));
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  connection_.GetRetransmissionAlarm()->Fire();
+  EXPECT_EQ(1u, writer_->frame_count());
+  ASSERT_EQ(1u, writer_->rst_stream_frames().size());
+  EXPECT_EQ(stream_id, writer_->rst_stream_frames().front().stream_id);
+}
+
+TEST_P(QuicConnectionTest, DoNotSendPendingRetransmissionForResetStream) {
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "foos", 3, NO_FIN, &last_packet);
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(stream_id, "fooos", 7, NO_FIN);
+
+  // Lose a packet which will trigger a pending retransmission.
+  QuicAckFrame ack = ConstructAckFrame(last_packet, last_packet - 1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  ProcessAckPacket(&ack);
+
+  SendRstStream(stream_id, QUIC_ERROR_PROCESSING_STREAM, 12);
+
+  // Unblock the connection and verify that the RST_STREAM is sent but not the
+  // second data packet nor a retransmit.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  if (!connection_.session_decides_what_to_write()) {
+    // OnCanWrite will cause this RST_STREAM_FRAME be sent again.
+    connection_.SendControlFrame(QuicFrame(new QuicRstStreamFrame(
+        1, stream_id, QUIC_ERROR_PROCESSING_STREAM, 14)));
+  }
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->rst_stream_frames().size());
+  EXPECT_EQ(stream_id, writer_->rst_stream_frames().front().stream_id);
+}
+
+TEST_P(QuicConnectionTest, SendPendingRetransmissionForQuicRstStreamNoError) {
+  QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "foos", 3, NO_FIN, &last_packet);
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(stream_id, "fooos", 7, NO_FIN);
+
+  // Lose a packet which will trigger a pending retransmission.
+  QuicAckFrame ack = ConstructAckFrame(last_packet, last_packet - 1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(last_packet - 1, kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  ProcessAckPacket(&ack);
+
+  SendRstStream(stream_id, QUIC_STREAM_NO_ERROR, 12);
+
+  // Unblock the connection and verify that the RST_STREAM is sent and the
+  // second data packet or a retransmit is sent.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AtLeast(2));
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  // The RST_STREAM_FRAME is sent after queued packets and pending
+  // retransmission.
+  connection_.SendControlFrame(QuicFrame(
+      new QuicRstStreamFrame(1, stream_id, QUIC_STREAM_NO_ERROR, 14)));
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->rst_stream_frames().size());
+}
+
+TEST_P(QuicConnectionTest, RetransmitAckedPacket) {
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);    // Packet 1
+  SendStreamDataToPeer(1, "foos", 3, NO_FIN, &last_packet);   // Packet 2
+  SendStreamDataToPeer(1, "fooos", 7, NO_FIN, &last_packet);  // Packet 3
+
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  // Instigate a loss with an ack.
+  QuicAckFrame nack_two = ConstructAckFrame(3, 2);
+  // The first nack should trigger a fast retransmission, but we'll be
+  // write blocked, so the packet will be queued.
+  BlockOnNextWrite();
+
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(QuicPacketNumber(2), kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&nack_two);
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+
+  // Now, ack the previous transmission.
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(false, _, _, _, _));
+  QuicAckFrame ack_all = InitAckFrame(3);
+  ProcessAckPacket(&ack_all);
+
+  // Unblock the socket and attempt to send the queued packets. We will always
+  // send the retransmission.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(4), _, _))
+      .Times(1);
+
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  // We do not store retransmittable frames of this retransmission.
+  EXPECT_FALSE(QuicConnectionPeer::HasRetransmittableFrames(&connection_, 4));
+}
+
+TEST_P(QuicConnectionTest, RetransmitNackedLargestObserved) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  QuicPacketNumber original, second;
+
+  QuicByteCount packet_size =
+      SendStreamDataToPeer(3, "foo", 0, NO_FIN, &original);  // 1st packet.
+  SendStreamDataToPeer(3, "bar", 3, NO_FIN, &second);        // 2nd packet.
+
+  QuicAckFrame frame = InitAckFrame({{second, second + 1}});
+  // The first nack should retransmit the largest observed packet.
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(original, kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  // Packet 1 is short header for IETF QUIC because the encryption level
+  // switched to ENCRYPTION_FORWARD_SECURE in SendStreamDataToPeer.
+  EXPECT_CALL(
+      *send_algorithm_,
+      OnPacketSent(_, _, _,
+                   GetParam().version.transport_version > QUIC_VERSION_43
+                       ? packet_size
+                       : packet_size - kQuicVersionSize,
+                   _));
+  ProcessAckPacket(&frame);
+}
+
+TEST_P(QuicConnectionTest, QueueAfterTwoRTOs) {
+  connection_.SetMaxTailLossProbes(0);
+
+  for (int i = 0; i < 10; ++i) {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+    connection_.SendStreamDataWithString(3, "foo", i * 3, NO_FIN);
+  }
+
+  // Block the writer and ensure they're queued.
+  BlockOnNextWrite();
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  // Only one packet should be retransmitted.
+  connection_.GetRetransmissionAlarm()->Fire();
+  EXPECT_TRUE(connection_.HasQueuedData());
+
+  // Unblock the writer.
+  writer_->SetWritable();
+  clock_.AdvanceTime(QuicTime::Delta::FromMicroseconds(
+      2 * DefaultRetransmissionTime().ToMicroseconds()));
+  // Retransmit already retransmitted packets event though the packet number
+  // greater than the largest observed.
+  if (connection_.session_decides_what_to_write()) {
+    // 2 RTOs + 1 TLP.
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(3);
+  } else {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+  }
+  connection_.GetRetransmissionAlarm()->Fire();
+  connection_.OnCanWrite();
+}
+
+TEST_P(QuicConnectionTest, WriteBlockedBufferedThenSent) {
+  BlockOnNextWrite();
+  writer_->set_is_write_blocked_data_buffered(true);
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, WriteBlockedThenSent) {
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+
+  // The second packet should also be queued, in order to ensure packets are
+  // never sent out of order.
+  writer_->SetWritable();
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_EQ(2u, connection_.NumQueuedPackets());
+
+  // Now both are sent in order when we unblock.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+  connection_.OnCanWrite();
+  EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, RetransmitWriteBlockedAckedOriginalThenSent) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+  EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+
+  BlockOnNextWrite();
+  writer_->set_is_write_blocked_data_buffered(true);
+  // Simulate the retransmission alarm firing.
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // Ack the sent packet before the callback returns, which happens in
+  // rare circumstances with write blocked sockets.
+  QuicAckFrame ack = InitAckFrame(1);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&ack);
+
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  // There is now a pending packet, but with no retransmittable frames.
+  if (GetQuicReloadableFlag(quic_optimize_inflight_check)) {
+    EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+    // Firing the alarm should remove all bytes_in_flight.
+    connection_.GetRetransmissionAlarm()->Fire();
+    EXPECT_EQ(0u, QuicSentPacketManagerPeer::GetBytesInFlight(manager_));
+  }
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  EXPECT_FALSE(QuicConnectionPeer::HasRetransmittableFrames(&connection_, 2));
+}
+
+TEST_P(QuicConnectionTest, AlarmsWhenWriteBlocked) {
+  // Block the connection.
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+  EXPECT_EQ(1u, writer_->packets_write_attempts());
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+
+  // Set the send alarm. Fire the alarm and ensure it doesn't attempt to write.
+  connection_.GetSendAlarm()->Set(clock_.ApproximateNow());
+  connection_.GetSendAlarm()->Fire();
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+  EXPECT_EQ(1u, writer_->packets_write_attempts());
+}
+
+TEST_P(QuicConnectionTest, NoSendAlarmAfterProcessPacketWhenWriteBlocked) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  // Block the connection.
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  // Process packet number 1. Can not call ProcessPacket or ProcessDataPacket
+  // here, because they will fire the alarm after QuicConnection::ProcessPacket
+  // is returned.
+  const uint64_t received_packet_num = 1;
+  const bool has_stop_waiting = false;
+  const EncryptionLevel level = ENCRYPTION_NONE;
+  std::unique_ptr<QuicPacket> packet(
+      ConstructDataPacket(received_packet_num, has_stop_waiting));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length =
+      peer_framer_.EncryptPayload(level, QuicPacketNumber(received_packet_num),
+                                  *packet, buffer, kMaxPacketSize);
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, clock_.Now(), false));
+
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, AddToWriteBlockedListIfWriterBlockedWhenProcessing) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, nullptr);
+
+  // Simulate the case where a shared writer gets blocked by another connection.
+  writer_->SetWriteBlocked();
+
+  // Process an ACK, make sure the connection calls visitor_->OnWriteBlocked().
+  QuicAckFrame ack1 = InitAckFrame(1);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(_, _, _, _, _));
+  EXPECT_CALL(visitor_, OnWriteBlocked()).Times(1);
+  ProcessAckPacket(1, &ack1);
+}
+
+TEST_P(QuicConnectionTest, DoNotAddToWriteBlockedListAfterDisconnect) {
+  writer_->SetBatchMode(true);
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PEER_GOING_AWAY, _,
+                                           ConnectionCloseSource::FROM_SELF));
+
+  EXPECT_CALL(visitor_, OnWriteBlocked()).Times(0);
+
+  {
+    QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                QuicConnection::NO_ACK);
+    connection_.CloseConnection(QUIC_PEER_GOING_AWAY, "no reason",
+                                ConnectionCloseBehavior::SILENT_CLOSE);
+
+    EXPECT_FALSE(connection_.connected());
+    writer_->SetWriteBlocked();
+  }
+}
+
+TEST_P(QuicConnectionTest, AddToWriteBlockedListIfBlockedOnFlushPackets) {
+  writer_->SetBatchMode(true);
+  writer_->BlockOnNextFlush();
+
+  EXPECT_CALL(visitor_, OnWriteBlocked()).Times(1);
+  {
+    QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                QuicConnection::NO_ACK);
+    // flusher's destructor will call connection_.FlushPackets, which should add
+    // the connection to the write blocked list.
+  }
+}
+
+TEST_P(QuicConnectionTest, NoLimitPacketsPerNack) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  int offset = 0;
+  // Send packets 1 to 15.
+  for (int i = 0; i < 15; ++i) {
+    SendStreamDataToPeer(1, "foo", offset, NO_FIN, nullptr);
+    offset += 3;
+  }
+
+  // Ack 15, nack 1-14.
+
+  QuicAckFrame nack =
+      InitAckFrame({{QuicPacketNumber(15), QuicPacketNumber(16)}});
+
+  // 14 packets have been NACK'd and lost.
+  LostPacketVector lost_packets;
+  for (int i = 1; i < 15; ++i) {
+    lost_packets.push_back(LostPacket(QuicPacketNumber(i), kMaxPacketSize));
+  }
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  if (connection_.session_decides_what_to_write()) {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  } else {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(14);
+  }
+  ProcessAckPacket(&nack);
+}
+
+// Test sending multiple acks from the connection to the session.
+TEST_P(QuicConnectionTest, MultipleAcks) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(1);
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 2);
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);  // Packet 1
+  EXPECT_EQ(QuicPacketNumber(1u), last_packet);
+  SendStreamDataToPeer(3, "foo", 0, NO_FIN, &last_packet);  // Packet 2
+  EXPECT_EQ(QuicPacketNumber(2u), last_packet);
+  SendAckPacketToPeer();                                    // Packet 3
+  SendStreamDataToPeer(5, "foo", 0, NO_FIN, &last_packet);  // Packet 4
+  EXPECT_EQ(QuicPacketNumber(4u), last_packet);
+  SendStreamDataToPeer(1, "foo", 3, NO_FIN, &last_packet);  // Packet 5
+  EXPECT_EQ(QuicPacketNumber(5u), last_packet);
+  SendStreamDataToPeer(3, "foo", 3, NO_FIN, &last_packet);  // Packet 6
+  EXPECT_EQ(QuicPacketNumber(6u), last_packet);
+
+  // Client will ack packets 1, 2, [!3], 4, 5.
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame1 = ConstructAckFrame(5, 3);
+  ProcessAckPacket(&frame1);
+
+  // Now the client implicitly acks 3, and explicitly acks 6.
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame2 = InitAckFrame(6);
+  ProcessAckPacket(&frame2);
+}
+
+TEST_P(QuicConnectionTest, DontLatchUnackedPacket) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(1);
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 2);
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, nullptr);  // Packet 1;
+  // From now on, we send acks, so the send algorithm won't mark them pending.
+  SendAckPacketToPeer();  // Packet 2
+
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame = InitAckFrame(1);
+  ProcessAckPacket(&frame);
+
+  // Verify that our internal state has least-unacked as 2, because we're still
+  // waiting for a potential ack for 2.
+
+  EXPECT_EQ(QuicPacketNumber(2u), stop_waiting()->least_unacked);
+
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  frame = InitAckFrame(2);
+  ProcessAckPacket(&frame);
+  EXPECT_EQ(QuicPacketNumber(3u), stop_waiting()->least_unacked);
+
+  // When we send an ack, we make sure our least-unacked makes sense.  In this
+  // case since we're not waiting on an ack for 2 and all packets are acked, we
+  // set it to 3.
+  SendAckPacketToPeer();  // Packet 3
+  // Least_unacked remains at 3 until another ack is received.
+  EXPECT_EQ(QuicPacketNumber(3u), stop_waiting()->least_unacked);
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    // Check that the outgoing ack had its packet number as least_unacked.
+    EXPECT_EQ(QuicPacketNumber(3u), least_unacked());
+  }
+
+  // Ack the ack, which updates the rtt and raises the least unacked.
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  frame = InitAckFrame(3);
+  ProcessAckPacket(&frame);
+
+  SendStreamDataToPeer(1, "bar", 3, NO_FIN, nullptr);  // Packet 4
+  EXPECT_EQ(QuicPacketNumber(4u), stop_waiting()->least_unacked);
+  SendAckPacketToPeer();  // Packet 5
+  if (GetParam().no_stop_waiting) {
+    // Expect no stop waiting frame is sent.
+    EXPECT_FALSE(least_unacked().IsInitialized());
+  } else {
+    EXPECT_EQ(QuicPacketNumber(4u), least_unacked());
+  }
+
+  // Send two data packets at the end, and ensure if the last one is acked,
+  // the least unacked is raised above the ack packets.
+  SendStreamDataToPeer(1, "bar", 6, NO_FIN, nullptr);  // Packet 6
+  SendStreamDataToPeer(1, "bar", 9, NO_FIN, nullptr);  // Packet 7
+
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  frame = InitAckFrame({{QuicPacketNumber(1), QuicPacketNumber(5)},
+                        {QuicPacketNumber(7), QuicPacketNumber(8)}});
+  ProcessAckPacket(&frame);
+
+  EXPECT_EQ(QuicPacketNumber(6u), stop_waiting()->least_unacked);
+}
+
+TEST_P(QuicConnectionTest, TLP) {
+  connection_.SetMaxTailLossProbes(1);
+
+  SendStreamDataToPeer(3, "foo", 0, NO_FIN, nullptr);
+  EXPECT_EQ(QuicPacketNumber(1u), stop_waiting()->least_unacked);
+  QuicTime retransmission_time =
+      connection_.GetRetransmissionAlarm()->deadline();
+  EXPECT_NE(QuicTime::Zero(), retransmission_time);
+
+  EXPECT_EQ(QuicPacketNumber(1u), writer_->header().packet_number);
+  // Simulate the retransmission alarm firing and sending a tlp,
+  // so send algorithm's OnRetransmissionTimeout is not called.
+  clock_.AdvanceTime(retransmission_time - clock_.Now());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(2), _, _));
+  connection_.GetRetransmissionAlarm()->Fire();
+  EXPECT_EQ(QuicPacketNumber(2u), writer_->header().packet_number);
+  // We do not raise the high water mark yet.
+  EXPECT_EQ(QuicPacketNumber(1u), stop_waiting()->least_unacked);
+}
+
+TEST_P(QuicConnectionTest, RTO) {
+  connection_.SetMaxTailLossProbes(0);
+
+  QuicTime default_retransmission_time =
+      clock_.ApproximateNow() + DefaultRetransmissionTime();
+  SendStreamDataToPeer(3, "foo", 0, NO_FIN, nullptr);
+  EXPECT_EQ(QuicPacketNumber(1u), stop_waiting()->least_unacked);
+
+  EXPECT_EQ(QuicPacketNumber(1u), writer_->header().packet_number);
+  EXPECT_EQ(default_retransmission_time,
+            connection_.GetRetransmissionAlarm()->deadline());
+  // Simulate the retransmission alarm firing.
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(2), _, _));
+  connection_.GetRetransmissionAlarm()->Fire();
+  EXPECT_EQ(QuicPacketNumber(2u), writer_->header().packet_number);
+  // We do not raise the high water mark yet.
+  EXPECT_EQ(QuicPacketNumber(1u), stop_waiting()->least_unacked);
+}
+
+TEST_P(QuicConnectionTest, RetransmitWithSameEncryptionLevel) {
+  use_tagging_decrypter();
+
+  // A TaggingEncrypter puts kTagSize copies of the given byte (0x01 here) at
+  // the end of the packet. We can test this to check which encrypter was used.
+  connection_.SetEncrypter(ENCRYPTION_NONE,
+                           QuicMakeUnique<TaggingEncrypter>(0x01));
+  SendStreamDataToPeer(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), "foo", 0,
+      NO_FIN, nullptr);
+  EXPECT_EQ(0x01010101u, writer_->final_bytes_of_last_packet());
+
+  connection_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<TaggingEncrypter>(0x02));
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_ZERO_RTT);
+  SendStreamDataToPeer(3, "foo", 0, NO_FIN, nullptr);
+  EXPECT_EQ(0x02020202u, writer_->final_bytes_of_last_packet());
+
+  {
+    InSequence s;
+    EXPECT_CALL(*send_algorithm_,
+                OnPacketSent(_, _, QuicPacketNumber(3), _, _));
+    EXPECT_CALL(*send_algorithm_,
+                OnPacketSent(_, _, QuicPacketNumber(4), _, _));
+  }
+
+  // Manually mark both packets for retransmission.
+  connection_.RetransmitUnackedPackets(ALL_UNACKED_RETRANSMISSION);
+
+  // Packet should have been sent with ENCRYPTION_NONE.
+  EXPECT_EQ(0x01010101u, writer_->final_bytes_of_previous_packet());
+
+  // Packet should have been sent with ENCRYPTION_ZERO_RTT.
+  EXPECT_EQ(0x02020202u, writer_->final_bytes_of_last_packet());
+}
+
+TEST_P(QuicConnectionTest, SendHandshakeMessages) {
+  use_tagging_decrypter();
+  // A TaggingEncrypter puts kTagSize copies of the given byte (0x01 here) at
+  // the end of the packet. We can test this to check which encrypter was used.
+  connection_.SetEncrypter(ENCRYPTION_NONE,
+                           QuicMakeUnique<TaggingEncrypter>(0x01));
+
+  // Attempt to send a handshake message and have the socket block.
+  EXPECT_CALL(*send_algorithm_, CanSend(_)).WillRepeatedly(Return(true));
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), "foo", 0,
+      NO_FIN);
+  // The packet should be serialized, but not queued.
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+
+  // Switch to the new encrypter.
+  connection_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<TaggingEncrypter>(0x02));
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_ZERO_RTT);
+
+  // Now become writeable and flush the packets.
+  writer_->SetWritable();
+  EXPECT_CALL(visitor_, OnCanWrite());
+  connection_.OnCanWrite();
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+
+  // Verify that the handshake packet went out at the null encryption.
+  EXPECT_EQ(0x01010101u, writer_->final_bytes_of_last_packet());
+}
+
+TEST_P(QuicConnectionTest,
+       DropRetransmitsForNullEncryptedPacketAfterForwardSecure) {
+  use_tagging_decrypter();
+  connection_.SetEncrypter(ENCRYPTION_NONE,
+                           QuicMakeUnique<TaggingEncrypter>(0x01));
+  QuicPacketNumber packet_number;
+  connection_.SendCryptoStreamData();
+
+  // Simulate the retransmission alarm firing and the socket blocking.
+  BlockOnNextWrite();
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // Go forward secure.
+  connection_.SetEncrypter(ENCRYPTION_FORWARD_SECURE,
+                           QuicMakeUnique<TaggingEncrypter>(0x02));
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_FORWARD_SECURE);
+  notifier_.NeuterUnencryptedData();
+  connection_.NeuterUnencryptedPackets();
+
+  EXPECT_EQ(QuicTime::Zero(), connection_.GetRetransmissionAlarm()->deadline());
+  // Unblock the socket and ensure that no packets are sent.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+}
+
+TEST_P(QuicConnectionTest, RetransmitPacketsWithInitialEncryption) {
+  use_tagging_decrypter();
+  connection_.SetEncrypter(ENCRYPTION_NONE,
+                           QuicMakeUnique<TaggingEncrypter>(0x01));
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_NONE);
+
+  SendStreamDataToPeer(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), "foo", 0,
+      NO_FIN, nullptr);
+
+  connection_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<TaggingEncrypter>(0x02));
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_ZERO_RTT);
+
+  SendStreamDataToPeer(2, "bar", 0, NO_FIN, nullptr);
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+
+  connection_.RetransmitUnackedPackets(ALL_INITIAL_RETRANSMISSION);
+}
+
+TEST_P(QuicConnectionTest, BufferNonDecryptablePackets) {
+  // SetFromConfig is always called after construction from InitializeSession.
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  use_tagging_decrypter();
+
+  const uint8_t tag = 0x07;
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+
+  // Process an encrypted packet which can not yet be decrypted which should
+  // result in the packet being buffered.
+  ProcessDataPacketAtLevel(1, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Transition to the new encryption state and process another encrypted packet
+  // which should result in the original packet being processed.
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_ZERO_RTT);
+  connection_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<TaggingEncrypter>(tag));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(2);
+  ProcessDataPacketAtLevel(2, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Finally, process a third packet and note that we do not reprocess the
+  // buffered packet.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(3, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+}
+
+TEST_P(QuicConnectionTest, TestRetransmitOrder) {
+  connection_.SetMaxTailLossProbes(0);
+
+  QuicByteCount first_packet_size;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<3>(&first_packet_size));
+
+  connection_.SendStreamDataWithString(3, "first_packet", 0, NO_FIN);
+  QuicByteCount second_packet_size;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<3>(&second_packet_size));
+  connection_.SendStreamDataWithString(3, "second_packet", 12, NO_FIN);
+  EXPECT_NE(first_packet_size, second_packet_size);
+  // Advance the clock by huge time to make sure packets will be retransmitted.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(10));
+  {
+    InSequence s;
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, first_packet_size, _));
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, second_packet_size, _));
+  }
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // Advance again and expect the packets to be sent again in the same order.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(20));
+  {
+    InSequence s;
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, first_packet_size, _));
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, second_packet_size, _));
+  }
+  connection_.GetRetransmissionAlarm()->Fire();
+}
+
+TEST_P(QuicConnectionTest, Buffer100NonDecryptablePacketsThenKeyChange) {
+  // SetFromConfig is always called after construction from InitializeSession.
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  config.set_max_undecryptable_packets(100);
+  connection_.SetFromConfig(config);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  use_tagging_decrypter();
+
+  const uint8_t tag = 0x07;
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+
+  // Process an encrypted packet which can not yet be decrypted which should
+  // result in the packet being buffered.
+  for (uint64_t i = 1; i <= 100; ++i) {
+    ProcessDataPacketAtLevel(i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+
+  // Transition to the new encryption state and process another encrypted packet
+  // which should result in the original packets being processed.
+  EXPECT_FALSE(connection_.GetProcessUndecryptablePacketsAlarm()->IsSet());
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  EXPECT_TRUE(connection_.GetProcessUndecryptablePacketsAlarm()->IsSet());
+  connection_.SetDefaultEncryptionLevel(ENCRYPTION_ZERO_RTT);
+  connection_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<TaggingEncrypter>(tag));
+
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(100);
+  connection_.GetProcessUndecryptablePacketsAlarm()->Fire();
+
+  // Finally, process a third packet and note that we do not reprocess the
+  // buffered packet.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(102, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+}
+
+TEST_P(QuicConnectionTest, SetRTOAfterWritingToSocket) {
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  // Make sure that RTO is not started when the packet is queued.
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+
+  // Test that RTO is started once we write to the socket.
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, DelayRTOWithAckReceipt) {
+  connection_.SetMaxTailLossProbes(0);
+
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+  connection_.SendStreamDataWithString(2, "foo", 0, NO_FIN);
+  connection_.SendStreamDataWithString(3, "bar", 0, NO_FIN);
+  QuicAlarm* retransmission_alarm = connection_.GetRetransmissionAlarm();
+  EXPECT_TRUE(retransmission_alarm->IsSet());
+  EXPECT_EQ(clock_.Now() + DefaultRetransmissionTime(),
+            retransmission_alarm->deadline());
+
+  // Advance the time right before the RTO, then receive an ack for the first
+  // packet to delay the RTO.
+  clock_.AdvanceTime(DefaultRetransmissionTime());
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame ack = InitAckFrame(1);
+  ProcessAckPacket(&ack);
+  // Now we have an RTT sample of DefaultRetransmissionTime(500ms),
+  // so the RTO has increased to 2 * SRTT.
+  EXPECT_TRUE(retransmission_alarm->IsSet());
+  EXPECT_EQ(retransmission_alarm->deadline(),
+            clock_.Now() + 2 * DefaultRetransmissionTime());
+
+  // Move forward past the original RTO and ensure the RTO is still pending.
+  clock_.AdvanceTime(2 * DefaultRetransmissionTime());
+
+  // Ensure the second packet gets retransmitted when it finally fires.
+  EXPECT_TRUE(retransmission_alarm->IsSet());
+  EXPECT_EQ(retransmission_alarm->deadline(), clock_.ApproximateNow());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  // Manually cancel the alarm to simulate a real test.
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // The new retransmitted packet number should set the RTO to a larger value
+  // than previously.
+  EXPECT_TRUE(retransmission_alarm->IsSet());
+  QuicTime next_rto_time = retransmission_alarm->deadline();
+  QuicTime expected_rto_time =
+      connection_.sent_packet_manager().GetRetransmissionTime();
+  EXPECT_EQ(next_rto_time, expected_rto_time);
+}
+
+TEST_P(QuicConnectionTest, TestQueued) {
+  connection_.SetMaxTailLossProbes(0);
+
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+
+  // Unblock the writes and actually send.
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+}
+
+TEST_P(QuicConnectionTest, InitialTimeout) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AnyNumber());
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+
+  // SetFromConfig sets the initial timeouts before negotiation.
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  // Subtract a second from the idle timeout on the client side.
+  QuicTime default_timeout =
+      clock_.ApproximateNow() +
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  // Simulate the timeout alarm firing.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1));
+  connection_.GetTimeoutAlarm()->Fire();
+
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, IdleTimeoutAfterFirstSentPacket) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AnyNumber());
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  QuicTime initial_ddl =
+      clock_.ApproximateNow() +
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  EXPECT_EQ(initial_ddl, connection_.GetTimeoutAlarm()->deadline());
+  EXPECT_TRUE(connection_.connected());
+
+  // Advance the time and send the first packet to the peer.
+  clock_.AdvanceTime(QuicTime::Delta::FromMicroseconds(20));
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);
+  EXPECT_EQ(QuicPacketNumber(1u), last_packet);
+  // This will be the updated deadline for the connection to idle time out.
+  QuicTime new_ddl = clock_.ApproximateNow() +
+                     QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+
+  // Simulate the timeout alarm firing, the connection should not be closed as
+  // a new packet has been sent.
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(0);
+  QuicTime::Delta delay = initial_ddl - clock_.ApproximateNow();
+  clock_.AdvanceTime(delay);
+  connection_.GetTimeoutAlarm()->Fire();
+  // Verify the timeout alarm deadline is updated.
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_EQ(new_ddl, connection_.GetTimeoutAlarm()->deadline());
+
+  // Simulate the timeout alarm firing again, the connection now should be
+  // closed.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  clock_.AdvanceTime(new_ddl - clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, IdleTimeoutAfterSendTwoPackets) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AnyNumber());
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  QuicTime initial_ddl =
+      clock_.ApproximateNow() +
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  EXPECT_EQ(initial_ddl, connection_.GetTimeoutAlarm()->deadline());
+  EXPECT_TRUE(connection_.connected());
+
+  // Immediately send the first packet, this is a rare case but test code will
+  // hit this issue often as MockClock used for tests doesn't move with code
+  // execution until manually adjusted.
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);
+  EXPECT_EQ(QuicPacketNumber(1u), last_packet);
+
+  // Advance the time and send the second packet to the peer.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(20));
+  SendStreamDataToPeer(1, "foo", 0, NO_FIN, &last_packet);
+  EXPECT_EQ(QuicPacketNumber(2u), last_packet);
+
+  if (GetQuicReloadableFlag(
+          quic_fix_time_of_first_packet_sent_after_receiving)) {
+    // Simulate the timeout alarm firing, the connection will be closed.
+    EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                             ConnectionCloseSource::FROM_SELF));
+    clock_.AdvanceTime(initial_ddl - clock_.ApproximateNow());
+    connection_.GetTimeoutAlarm()->Fire();
+  } else {
+    // Simulate the timeout alarm firing, the connection will not be closed.
+    EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(0);
+    clock_.AdvanceTime(initial_ddl - clock_.ApproximateNow());
+    connection_.GetTimeoutAlarm()->Fire();
+    EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+    EXPECT_TRUE(connection_.connected());
+
+    // Advance another 20ms, and fire the alarm again. The connection will be
+    // closed.
+    EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                             ConnectionCloseSource::FROM_SELF));
+    clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(20));
+    connection_.GetTimeoutAlarm()->Fire();
+  }
+
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, HandshakeTimeout) {
+  // Use a shorter handshake timeout than idle timeout for this test.
+  const QuicTime::Delta timeout = QuicTime::Delta::FromSeconds(5);
+  connection_.SetNetworkTimeouts(timeout, timeout);
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AnyNumber());
+
+  QuicTime handshake_timeout =
+      clock_.ApproximateNow() + timeout - QuicTime::Delta::FromSeconds(1);
+  EXPECT_EQ(handshake_timeout, connection_.GetTimeoutAlarm()->deadline());
+  EXPECT_TRUE(connection_.connected());
+
+  // Send and ack new data 3 seconds later to lengthen the idle timeout.
+  SendStreamDataToPeer(
+      QuicUtils::GetHeadersStreamId(connection_.transport_version()), "GET /",
+      0, FIN, nullptr);
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(3));
+  QuicAckFrame frame = InitAckFrame(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&frame);
+
+  // Fire early to verify it wouldn't timeout yet.
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_TRUE(connection_.connected());
+
+  clock_.AdvanceTime(timeout - QuicTime::Delta::FromSeconds(2));
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_HANDSHAKE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  // Simulate the timeout alarm firing.
+  connection_.GetTimeoutAlarm()->Fire();
+
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, PingAfterSend) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(true));
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+
+  // Advance to 5ms, and send a packet to the peer, which will set
+  // the ping alarm.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  SendStreamDataToPeer(
+      QuicUtils::GetHeadersStreamId(connection_.transport_version()), "GET /",
+      0, FIN, nullptr);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow() + QuicTime::Delta::FromSeconds(15),
+            connection_.GetPingAlarm()->deadline());
+
+  // Now recevie an ACK of the previous packet, which will move the
+  // ping alarm forward.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  QuicAckFrame frame = InitAckFrame(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  // The ping timer is set slightly less than 15 seconds in the future, because
+  // of the 1s ping timer alarm granularity.
+  EXPECT_EQ(clock_.ApproximateNow() + QuicTime::Delta::FromSeconds(15) -
+                QuicTime::Delta::FromMilliseconds(5),
+            connection_.GetPingAlarm()->deadline());
+
+  writer_->Reset();
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(15));
+  EXPECT_CALL(visitor_, SendPing()).WillOnce(Invoke([this]() {
+    connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+  }));
+  connection_.GetPingAlarm()->Fire();
+  EXPECT_EQ(1u, writer_->frame_count());
+  ASSERT_EQ(1u, writer_->ping_frames().size());
+  writer_->Reset();
+
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(false));
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  SendAckPacketToPeer();
+
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, ReducedPingTimeout) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(true));
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+
+  // Use a reduced ping timeout for this connection.
+  connection_.set_ping_timeout(QuicTime::Delta::FromSeconds(10));
+
+  // Advance to 5ms, and send a packet to the peer, which will set
+  // the ping alarm.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+  SendStreamDataToPeer(
+      QuicUtils::GetHeadersStreamId(connection_.transport_version()), "GET /",
+      0, FIN, nullptr);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow() + QuicTime::Delta::FromSeconds(10),
+            connection_.GetPingAlarm()->deadline());
+
+  // Now recevie an ACK of the previous packet, which will move the
+  // ping alarm forward.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  QuicAckFrame frame = InitAckFrame(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  // The ping timer is set slightly less than 10 seconds in the future, because
+  // of the 1s ping timer alarm granularity.
+  EXPECT_EQ(clock_.ApproximateNow() + QuicTime::Delta::FromSeconds(10) -
+                QuicTime::Delta::FromMilliseconds(5),
+            connection_.GetPingAlarm()->deadline());
+
+  writer_->Reset();
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(10));
+  EXPECT_CALL(visitor_, SendPing()).WillOnce(Invoke([this]() {
+    connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+  }));
+  connection_.GetPingAlarm()->Fire();
+  EXPECT_EQ(1u, writer_->frame_count());
+  ASSERT_EQ(1u, writer_->ping_frames().size());
+  writer_->Reset();
+
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(false));
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  SendAckPacketToPeer();
+
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+}
+
+// Tests whether sending an MTU discovery packet to peer successfully causes the
+// maximum packet size to increase.
+TEST_P(QuicConnectionTest, SendMtuDiscoveryPacket) {
+  EXPECT_TRUE(connection_.connected());
+
+  // Send an MTU probe.
+  const size_t new_mtu = kDefaultMaxPacketSize + 100;
+  QuicByteCount mtu_probe_size;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<3>(&mtu_probe_size));
+  connection_.SendMtuDiscoveryPacket(new_mtu);
+  EXPECT_EQ(new_mtu, mtu_probe_size);
+  EXPECT_EQ(QuicPacketNumber(1u), creator_->packet_number());
+
+  // Send more than MTU worth of data.  No acknowledgement was received so far,
+  // so the MTU should be at its old value.
+  const QuicString data(kDefaultMaxPacketSize + 1, '.');
+  QuicByteCount size_before_mtu_change;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .Times(2)
+      .WillOnce(SaveArg<3>(&size_before_mtu_change))
+      .WillOnce(Return());
+  connection_.SendStreamDataWithString(3, data, 0, FIN);
+  EXPECT_EQ(QuicPacketNumber(3u), creator_->packet_number());
+  EXPECT_EQ(kDefaultMaxPacketSize, size_before_mtu_change);
+
+  // Acknowledge all packets so far.
+  QuicAckFrame probe_ack = InitAckFrame(3);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&probe_ack);
+  EXPECT_EQ(new_mtu, connection_.max_packet_length());
+
+  // Send the same data again.  Check that it fits into a single packet now.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendStreamDataWithString(3, data, 0, FIN);
+  EXPECT_EQ(QuicPacketNumber(4u), creator_->packet_number());
+}
+
+// Tests whether MTU discovery does not happen when it is not explicitly enabled
+// by the connection options.
+TEST_P(QuicConnectionTest, MtuDiscoveryDisabled) {
+  EXPECT_TRUE(connection_.connected());
+
+  const QuicPacketCount packets_between_probes_base = 10;
+  set_packets_between_probes_base(packets_between_probes_base);
+
+  const QuicPacketCount number_of_packets = packets_between_probes_base * 2;
+  for (QuicPacketCount i = 0; i < number_of_packets; i++) {
+    SendStreamDataToPeer(3, ".", i, NO_FIN, nullptr);
+    EXPECT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+    EXPECT_EQ(0u, connection_.mtu_probe_count());
+  }
+}
+
+// Tests whether MTU discovery works when the probe gets acknowledged on the
+// first try.
+TEST_P(QuicConnectionTest, MtuDiscoveryEnabled) {
+  EXPECT_TRUE(connection_.connected());
+
+  connection_.EnablePathMtuDiscovery(send_algorithm_);
+
+  const QuicPacketCount packets_between_probes_base = 5;
+  set_packets_between_probes_base(packets_between_probes_base);
+
+  // Send enough packets so that the next one triggers path MTU discovery.
+  for (QuicPacketCount i = 0; i < packets_between_probes_base - 1; i++) {
+    SendStreamDataToPeer(3, ".", i, NO_FIN, nullptr);
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  // Trigger the probe.
+  SendStreamDataToPeer(3, "!", packets_between_probes_base - 1, NO_FIN,
+                       nullptr);
+  ASSERT_TRUE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  QuicByteCount probe_size;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<3>(&probe_size));
+  connection_.GetMtuDiscoveryAlarm()->Fire();
+  EXPECT_EQ(kMtuDiscoveryTargetPacketSizeHigh, probe_size);
+
+  const QuicPacketNumber probe_packet_number =
+      FirstSendingPacketNumber() + packets_between_probes_base;
+  ASSERT_EQ(probe_packet_number, creator_->packet_number());
+
+  // Acknowledge all packets sent so far.
+  QuicAckFrame probe_ack = InitAckFrame(probe_packet_number);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&probe_ack);
+  EXPECT_EQ(kMtuDiscoveryTargetPacketSizeHigh, connection_.max_packet_length());
+  EXPECT_EQ(0u, connection_.GetBytesInFlight());
+
+  // Send more packets, and ensure that none of them sets the alarm.
+  for (QuicPacketCount i = 0; i < 4 * packets_between_probes_base; i++) {
+    SendStreamDataToPeer(3, ".", packets_between_probes_base + i, NO_FIN,
+                         nullptr);
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  EXPECT_EQ(1u, connection_.mtu_probe_count());
+}
+
+// Tests whether MTU discovery works correctly when the probes never get
+// acknowledged.
+TEST_P(QuicConnectionTest, MtuDiscoveryFailed) {
+  EXPECT_TRUE(connection_.connected());
+
+  connection_.EnablePathMtuDiscovery(send_algorithm_);
+
+  const QuicTime::Delta rtt = QuicTime::Delta::FromMilliseconds(100);
+
+  EXPECT_EQ(kPacketsBetweenMtuProbesBase,
+            QuicConnectionPeer::GetPacketsBetweenMtuProbes(&connection_));
+  // Lower the number of probes between packets in order to make the test go
+  // much faster.
+  const QuicPacketCount packets_between_probes_base = 5;
+  set_packets_between_probes_base(packets_between_probes_base);
+
+  // This tests sends more packets than strictly necessary to make sure that if
+  // the connection was to send more discovery packets than needed, those would
+  // get caught as well.
+  const QuicPacketCount number_of_packets =
+      packets_between_probes_base * (1 << (kMtuDiscoveryAttempts + 1));
+  std::vector<QuicPacketNumber> mtu_discovery_packets;
+  // Called by the first ack.
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  // Called on many acks.
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _))
+      .Times(AnyNumber());
+  for (QuicPacketCount i = 0; i < number_of_packets; i++) {
+    SendStreamDataToPeer(3, "!", i, NO_FIN, nullptr);
+    clock_.AdvanceTime(rtt);
+
+    // Receive an ACK, which marks all data packets as received, and all MTU
+    // discovery packets as missing.
+
+    QuicAckFrame ack;
+
+    if (!mtu_discovery_packets.empty()) {
+      QuicPacketNumber min_packet = *min_element(mtu_discovery_packets.begin(),
+                                                 mtu_discovery_packets.end());
+      QuicPacketNumber max_packet = *max_element(mtu_discovery_packets.begin(),
+                                                 mtu_discovery_packets.end());
+      ack.packets.AddRange(QuicPacketNumber(1), min_packet);
+      ack.packets.AddRange(QuicPacketNumber(max_packet + 1),
+                           creator_->packet_number() + 1);
+      ack.largest_acked = creator_->packet_number();
+
+    } else {
+      ack.packets.AddRange(QuicPacketNumber(1), creator_->packet_number() + 1);
+      ack.largest_acked = creator_->packet_number();
+    }
+
+    ProcessAckPacket(&ack);
+
+    // Trigger MTU probe if it would be scheduled now.
+    if (!connection_.GetMtuDiscoveryAlarm()->IsSet()) {
+      continue;
+    }
+
+    // Fire the alarm.  The alarm should cause a packet to be sent.
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    connection_.GetMtuDiscoveryAlarm()->Fire();
+    // Record the packet number of the MTU discovery packet in order to
+    // mark it as NACK'd.
+    mtu_discovery_packets.push_back(creator_->packet_number());
+  }
+
+  // Ensure the number of packets between probes grows exponentially by checking
+  // it against the closed-form expression for the packet number.
+  ASSERT_EQ(kMtuDiscoveryAttempts, mtu_discovery_packets.size());
+  for (uint64_t i = 0; i < kMtuDiscoveryAttempts; i++) {
+    // 2^0 + 2^1 + 2^2 + ... + 2^n = 2^(n + 1) - 1
+    const QuicPacketCount packets_between_probes =
+        packets_between_probes_base * ((1 << (i + 1)) - 1);
+    EXPECT_EQ(QuicPacketNumber(packets_between_probes + (i + 1)),
+              mtu_discovery_packets[i]);
+  }
+
+  EXPECT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  EXPECT_EQ(kDefaultMaxPacketSize, connection_.max_packet_length());
+  EXPECT_EQ(kMtuDiscoveryAttempts, connection_.mtu_probe_count());
+}
+
+// Tests whether MTU discovery works when the writer has a limit on how large a
+// packet can be.
+TEST_P(QuicConnectionTest, MtuDiscoveryWriterLimited) {
+  EXPECT_TRUE(connection_.connected());
+
+  const QuicByteCount mtu_limit = kMtuDiscoveryTargetPacketSizeHigh - 1;
+  writer_->set_max_packet_size(mtu_limit);
+  connection_.EnablePathMtuDiscovery(send_algorithm_);
+
+  const QuicPacketCount packets_between_probes_base = 5;
+  set_packets_between_probes_base(packets_between_probes_base);
+
+  // Send enough packets so that the next one triggers path MTU discovery.
+  for (QuicPacketCount i = 0; i < packets_between_probes_base - 1; i++) {
+    SendStreamDataToPeer(3, ".", i, NO_FIN, nullptr);
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  // Trigger the probe.
+  SendStreamDataToPeer(3, "!", packets_between_probes_base - 1, NO_FIN,
+                       nullptr);
+  ASSERT_TRUE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  QuicByteCount probe_size;
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .WillOnce(SaveArg<3>(&probe_size));
+  connection_.GetMtuDiscoveryAlarm()->Fire();
+  EXPECT_EQ(mtu_limit, probe_size);
+
+  const QuicPacketNumber probe_sequence_number =
+      FirstSendingPacketNumber() + packets_between_probes_base;
+  ASSERT_EQ(probe_sequence_number, creator_->packet_number());
+
+  // Acknowledge all packets sent so far.
+  QuicAckFrame probe_ack = InitAckFrame(probe_sequence_number);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&probe_ack);
+  EXPECT_EQ(mtu_limit, connection_.max_packet_length());
+  EXPECT_EQ(0u, connection_.GetBytesInFlight());
+
+  // Send more packets, and ensure that none of them sets the alarm.
+  for (QuicPacketCount i = 0; i < 4 * packets_between_probes_base; i++) {
+    SendStreamDataToPeer(3, ".", packets_between_probes_base + i, NO_FIN,
+                         nullptr);
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  EXPECT_EQ(1u, connection_.mtu_probe_count());
+}
+
+// Tests whether MTU discovery works when the writer returns an error despite
+// advertising higher packet length.
+TEST_P(QuicConnectionTest, MtuDiscoveryWriterFailed) {
+  EXPECT_TRUE(connection_.connected());
+
+  const QuicByteCount mtu_limit = kMtuDiscoveryTargetPacketSizeHigh - 1;
+  const QuicByteCount initial_mtu = connection_.max_packet_length();
+  EXPECT_LT(initial_mtu, mtu_limit);
+  writer_->set_max_packet_size(mtu_limit);
+  connection_.EnablePathMtuDiscovery(send_algorithm_);
+
+  const QuicPacketCount packets_between_probes_base = 5;
+  set_packets_between_probes_base(packets_between_probes_base);
+
+  // Send enough packets so that the next one triggers path MTU discovery.
+  for (QuicPacketCount i = 0; i < packets_between_probes_base - 1; i++) {
+    SendStreamDataToPeer(3, ".", i, NO_FIN, nullptr);
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  // Trigger the probe.
+  SendStreamDataToPeer(3, "!", packets_between_probes_base - 1, NO_FIN,
+                       nullptr);
+  ASSERT_TRUE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  writer_->SimulateNextPacketTooLarge();
+  connection_.GetMtuDiscoveryAlarm()->Fire();
+  ASSERT_TRUE(connection_.connected());
+
+  // Send more data.
+  QuicPacketNumber probe_number = creator_->packet_number();
+  QuicPacketCount extra_packets = packets_between_probes_base * 3;
+  for (QuicPacketCount i = 0; i < extra_packets; i++) {
+    connection_.EnsureWritableAndSendStreamData5();
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  // Acknowledge all packets sent so far, except for the lost probe.
+  QuicAckFrame probe_ack =
+      ConstructAckFrame(creator_->packet_number(), probe_number);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&probe_ack);
+  EXPECT_EQ(initial_mtu, connection_.max_packet_length());
+
+  // Send more packets, and ensure that none of them sets the alarm.
+  for (QuicPacketCount i = 0; i < 4 * packets_between_probes_base; i++) {
+    connection_.EnsureWritableAndSendStreamData5();
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  EXPECT_EQ(initial_mtu, connection_.max_packet_length());
+  EXPECT_EQ(1u, connection_.mtu_probe_count());
+}
+
+TEST_P(QuicConnectionTest, NoMtuDiscoveryAfterConnectionClosed) {
+  EXPECT_TRUE(connection_.connected());
+
+  connection_.EnablePathMtuDiscovery(send_algorithm_);
+
+  const QuicPacketCount packets_between_probes_base = 10;
+  set_packets_between_probes_base(packets_between_probes_base);
+
+  // Send enough packets so that the next one triggers path MTU discovery.
+  for (QuicPacketCount i = 0; i < packets_between_probes_base - 1; i++) {
+    SendStreamDataToPeer(3, ".", i, NO_FIN, nullptr);
+    ASSERT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+  }
+
+  SendStreamDataToPeer(3, "!", packets_between_probes_base - 1, NO_FIN,
+                       nullptr);
+  EXPECT_TRUE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _));
+  connection_.CloseConnection(QUIC_PEER_GOING_AWAY, "no reason",
+                              ConnectionCloseBehavior::SILENT_CLOSE);
+  EXPECT_FALSE(connection_.GetMtuDiscoveryAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfterSend) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_FALSE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime::Delta initial_idle_timeout =
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  QuicTime default_timeout = clock_.ApproximateNow() + initial_idle_timeout;
+
+  // When we send a packet, the timeout will change to 5ms +
+  // kInitialIdleTimeoutSecs.
+  clock_.AdvanceTime(five_ms);
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // Now send more data. This will not move the timeout because
+  // no data has been received since the previous write.
+  clock_.AdvanceTime(five_ms);
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      3, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // The original alarm will fire.  We should not time out because we had a
+  // network event at t=5ms.  The alarm will reregister.
+  clock_.AdvanceTime(initial_idle_timeout - five_ms - five_ms);
+  EXPECT_EQ(default_timeout, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_EQ(default_timeout + five_ms,
+            connection_.GetTimeoutAlarm()->deadline());
+
+  // This time, we should time out.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  clock_.AdvanceTime(five_ms);
+  EXPECT_EQ(default_timeout + five_ms, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfterRetransmission) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_FALSE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime start_time = clock_.Now();
+  const QuicTime::Delta initial_idle_timeout =
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  QuicTime default_timeout = clock_.Now() + initial_idle_timeout;
+
+  connection_.SetMaxTailLossProbes(0);
+  const QuicTime default_retransmission_time =
+      start_time + DefaultRetransmissionTime();
+
+  ASSERT_LT(default_retransmission_time, default_timeout);
+
+  // When we send a packet, the timeout will change to 5 ms +
+  // kInitialIdleTimeoutSecs (but it will not reschedule the alarm).
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  const QuicTime send_time = start_time + five_ms;
+  clock_.AdvanceTime(five_ms);
+  ASSERT_EQ(send_time, clock_.Now());
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // Move forward 5 ms and receive a packet, which will move the timeout
+  // forward 5 ms more (but will not reschedule the alarm).
+  const QuicTime receive_time = send_time + five_ms;
+  clock_.AdvanceTime(receive_time - clock_.Now());
+  ASSERT_EQ(receive_time, clock_.Now());
+  ProcessPacket(1);
+
+  // Now move forward to the retransmission time and retransmit the
+  // packet, which should move the timeout forward again (but will not
+  // reschedule the alarm).
+  EXPECT_EQ(default_retransmission_time + five_ms,
+            connection_.GetRetransmissionAlarm()->deadline());
+  // Simulate the retransmission alarm firing.
+  const QuicTime rto_time = send_time + DefaultRetransmissionTime();
+  const QuicTime final_timeout = rto_time + initial_idle_timeout;
+  clock_.AdvanceTime(rto_time - clock_.Now());
+  ASSERT_EQ(rto_time, clock_.Now());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(2u), _, _));
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // Advance to the original timeout and fire the alarm. The connection should
+  // timeout, and the alarm should be registered based on the time of the
+  // retransmission.
+  clock_.AdvanceTime(default_timeout - clock_.Now());
+  ASSERT_EQ(default_timeout.ToDebuggingValue(),
+            clock_.Now().ToDebuggingValue());
+  EXPECT_EQ(default_timeout, clock_.Now());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_TRUE(connection_.connected());
+  ASSERT_EQ(final_timeout.ToDebuggingValue(),
+            connection_.GetTimeoutAlarm()->deadline().ToDebuggingValue());
+
+  // This time, we should time out.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  clock_.AdvanceTime(final_timeout - clock_.Now());
+  EXPECT_EQ(connection_.GetTimeoutAlarm()->deadline(), clock_.Now());
+  EXPECT_EQ(final_timeout, clock_.Now());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, NewTimeoutAfterSendSilentClose) {
+  // Same test as above, but complete a handshake which enables silent close,
+  // causing no connection close packet to be sent.
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+
+  // Create a handshake message that also enables silent close.
+  CryptoHandshakeMessage msg;
+  QuicString error_details;
+  QuicConfig client_config;
+  client_config.SetInitialStreamFlowControlWindowToSend(
+      kInitialStreamFlowControlWindowForTest);
+  client_config.SetInitialSessionFlowControlWindowToSend(
+      kInitialSessionFlowControlWindowForTest);
+  client_config.SetIdleNetworkTimeout(
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs),
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs));
+  client_config.ToHandshakeMessage(&msg);
+  const QuicErrorCode error =
+      config.ProcessPeerHello(msg, CLIENT, &error_details);
+  EXPECT_EQ(QUIC_NO_ERROR, error);
+
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime::Delta default_idle_timeout =
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs - 1);
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  QuicTime default_timeout = clock_.ApproximateNow() + default_idle_timeout;
+
+  // When we send a packet, the timeout will change to 5ms +
+  // kInitialIdleTimeoutSecs.
+  clock_.AdvanceTime(five_ms);
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // Now send more data. This will not move the timeout because
+  // no data has been received since the previous write.
+  clock_.AdvanceTime(five_ms);
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      3, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // The original alarm will fire.  We should not time out because we had a
+  // network event at t=5ms.  The alarm will reregister.
+  clock_.AdvanceTime(default_idle_timeout - five_ms - five_ms);
+  EXPECT_EQ(default_timeout, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_EQ(default_timeout + five_ms,
+            connection_.GetTimeoutAlarm()->deadline());
+
+  // This time, we should time out.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  clock_.AdvanceTime(five_ms);
+  EXPECT_EQ(default_timeout + five_ms, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfterSendSilentCloseAndTLP) {
+  // Same test as above, but complete a handshake which enables silent close,
+  // but sending TLPs causes the connection close to be sent.
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+
+  // Create a handshake message that also enables silent close.
+  CryptoHandshakeMessage msg;
+  QuicString error_details;
+  QuicConfig client_config;
+  client_config.SetInitialStreamFlowControlWindowToSend(
+      kInitialStreamFlowControlWindowForTest);
+  client_config.SetInitialSessionFlowControlWindowToSend(
+      kInitialSessionFlowControlWindowForTest);
+  client_config.SetIdleNetworkTimeout(
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs),
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs));
+  client_config.ToHandshakeMessage(&msg);
+  const QuicErrorCode error =
+      config.ProcessPeerHello(msg, CLIENT, &error_details);
+  EXPECT_EQ(QUIC_NO_ERROR, error);
+
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime::Delta default_idle_timeout =
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs - 1);
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  QuicTime default_timeout = clock_.ApproximateNow() + default_idle_timeout;
+
+  // When we send a packet, the timeout will change to 5ms +
+  // kInitialIdleTimeoutSecs.
+  clock_.AdvanceTime(five_ms);
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // Retransmit the packet via tail loss probe.
+  clock_.AdvanceTime(connection_.GetRetransmissionAlarm()->deadline() -
+                     clock_.Now());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(2u), _, _));
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // This time, we should time out and send a connection close due to the TLP.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  clock_.AdvanceTime(connection_.GetTimeoutAlarm()->deadline() -
+                     clock_.ApproximateNow() + five_ms);
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfterSendSilentCloseWithOpenStreams) {
+  // Same test as above, but complete a handshake which enables silent close,
+  // but having open streams causes the connection close to be sent.
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+
+  // Create a handshake message that also enables silent close.
+  CryptoHandshakeMessage msg;
+  QuicString error_details;
+  QuicConfig client_config;
+  client_config.SetInitialStreamFlowControlWindowToSend(
+      kInitialStreamFlowControlWindowForTest);
+  client_config.SetInitialSessionFlowControlWindowToSend(
+      kInitialSessionFlowControlWindowForTest);
+  client_config.SetIdleNetworkTimeout(
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs),
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs));
+  client_config.ToHandshakeMessage(&msg);
+  const QuicErrorCode error =
+      config.ProcessPeerHello(msg, CLIENT, &error_details);
+  EXPECT_EQ(QUIC_NO_ERROR, error);
+
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime::Delta default_idle_timeout =
+      QuicTime::Delta::FromSeconds(kDefaultIdleTimeoutSecs - 1);
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  QuicTime default_timeout = clock_.ApproximateNow() + default_idle_timeout;
+
+  // When we send a packet, the timeout will change to 5ms +
+  // kInitialIdleTimeoutSecs.
+  clock_.AdvanceTime(five_ms);
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, FIN, nullptr);
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  // Indicate streams are still open.
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(true));
+
+  // This time, we should time out and send a connection close due to the TLP.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  clock_.AdvanceTime(connection_.GetTimeoutAlarm()->deadline() -
+                     clock_.ApproximateNow() + five_ms);
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfterReceive) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_FALSE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime::Delta initial_idle_timeout =
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  QuicTime default_timeout = clock_.ApproximateNow() + initial_idle_timeout;
+
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, NO_FIN);
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      3, NO_FIN);
+
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+  clock_.AdvanceTime(five_ms);
+
+  // When we receive a packet, the timeout will change to 5ms +
+  // kInitialIdleTimeoutSecs.
+  QuicAckFrame ack = InitAckFrame(2);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&ack);
+
+  // The original alarm will fire.  We should not time out because we had a
+  // network event at t=5ms.  The alarm will reregister.
+  clock_.AdvanceTime(initial_idle_timeout - five_ms);
+  EXPECT_EQ(default_timeout, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_EQ(default_timeout + five_ms,
+            connection_.GetTimeoutAlarm()->deadline());
+
+  // This time, we should time out.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  clock_.AdvanceTime(five_ms);
+  EXPECT_EQ(default_timeout + five_ms, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfterReceiveNotSendWhenUnacked) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+  EXPECT_FALSE(QuicConnectionPeer::IsSilentCloseEnabled(&connection_));
+
+  const QuicTime::Delta initial_idle_timeout =
+      QuicTime::Delta::FromSeconds(kInitialIdleTimeoutSecs - 1);
+  connection_.SetNetworkTimeouts(
+      QuicTime::Delta::Infinite(),
+      initial_idle_timeout + QuicTime::Delta::FromSeconds(1));
+  const QuicTime::Delta five_ms = QuicTime::Delta::FromMilliseconds(5);
+  QuicTime default_timeout = clock_.ApproximateNow() + initial_idle_timeout;
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, NO_FIN);
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      3, NO_FIN);
+
+  EXPECT_EQ(default_timeout, connection_.GetTimeoutAlarm()->deadline());
+
+  clock_.AdvanceTime(five_ms);
+
+  // When we receive a packet, the timeout will change to 5ms +
+  // kInitialIdleTimeoutSecs.
+  QuicAckFrame ack = InitAckFrame(2);
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&ack);
+
+  // The original alarm will fire.  We should not time out because we had a
+  // network event at t=5ms.  The alarm will reregister.
+  clock_.AdvanceTime(initial_idle_timeout - five_ms);
+  EXPECT_EQ(default_timeout, clock_.ApproximateNow());
+  connection_.GetTimeoutAlarm()->Fire();
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_EQ(default_timeout + five_ms,
+            connection_.GetTimeoutAlarm()->deadline());
+
+  // Now, send packets while advancing the time and verify that the connection
+  // eventually times out.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_NETWORK_IDLE_TIMEOUT, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(AnyNumber());
+  for (int i = 0; i < 100 && connection_.connected(); ++i) {
+    QUIC_LOG(INFO) << "sending data packet";
+    connection_.SendStreamDataWithString(
+        GetNthClientInitiatedStreamId(1, connection_.transport_version()),
+        "foo", 0, NO_FIN);
+    connection_.GetTimeoutAlarm()->Fire();
+    clock_.AdvanceTime(QuicTime::Delta::FromSeconds(1));
+  }
+  EXPECT_FALSE(connection_.connected());
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, TimeoutAfter5ClientRTOs) {
+  connection_.SetMaxTailLossProbes(2);
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  QuicTagVector connection_options;
+  connection_options.push_back(k5RTO);
+  config.SetConnectionOptionsToSend(connection_options);
+  connection_.SetFromConfig(config);
+
+  // Send stream data.
+  SendStreamDataToPeer(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, FIN, nullptr);
+
+  // Fire the retransmission alarm 6 times, twice for TLP and 4 times for RTO.
+  for (int i = 0; i < 6; ++i) {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    connection_.GetRetransmissionAlarm()->Fire();
+    EXPECT_TRUE(connection_.GetTimeoutAlarm()->IsSet());
+    EXPECT_TRUE(connection_.connected());
+  }
+
+  EXPECT_EQ(2u, connection_.sent_packet_manager().GetConsecutiveTlpCount());
+  EXPECT_EQ(4u, connection_.sent_packet_manager().GetConsecutiveRtoCount());
+  // This time, we should time out.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_TOO_MANY_RTOS, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.GetRetransmissionAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetTimeoutAlarm()->IsSet());
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, SendScheduler) {
+  // Test that if we send a packet without delay, it is not queued.
+  QuicFramerPeer::SetPerspective(&peer_framer_, Perspective::IS_CLIENT);
+  std::unique_ptr<QuicPacket> packet = ConstructDataPacket(1, !kHasStopWaiting);
+  QuicPacketCreatorPeer::SetPacketNumber(creator_, 1);
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.SendPacket(ENCRYPTION_NONE, 1, std::move(packet),
+                         HAS_RETRANSMITTABLE_DATA, false, false);
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+}
+
+TEST_P(QuicConnectionTest, FailToSendFirstPacket) {
+  // Test that the connection does not crash when it fails to send the first
+  // packet at which point self_address_ might be uninitialized.
+  QuicFramerPeer::SetPerspective(&peer_framer_, Perspective::IS_CLIENT);
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(1);
+  std::unique_ptr<QuicPacket> packet = ConstructDataPacket(1, !kHasStopWaiting);
+  QuicPacketCreatorPeer::SetPacketNumber(creator_, 1);
+  writer_->SetShouldWriteFail();
+  connection_.SendPacket(ENCRYPTION_NONE, 1, std::move(packet),
+                         HAS_RETRANSMITTABLE_DATA, false, false);
+}
+
+TEST_P(QuicConnectionTest, SendSchedulerEAGAIN) {
+  QuicFramerPeer::SetPerspective(&peer_framer_, Perspective::IS_CLIENT);
+  std::unique_ptr<QuicPacket> packet = ConstructDataPacket(1, !kHasStopWaiting);
+  QuicPacketCreatorPeer::SetPacketNumber(creator_, 1);
+  BlockOnNextWrite();
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(2u), _, _))
+      .Times(0);
+  connection_.SendPacket(ENCRYPTION_NONE, 1, std::move(packet),
+                         HAS_RETRANSMITTABLE_DATA, false, false);
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+}
+
+TEST_P(QuicConnectionTest, TestQueueLimitsOnSendStreamData) {
+  // All packets carry version info till version is negotiated.
+  size_t payload_length;
+  size_t length = GetPacketLengthForOneStream(
+      connection_.version().transport_version, kIncludeVersion,
+      !kIncludeDiversificationNonce, PACKET_8BYTE_CONNECTION_ID,
+      PACKET_0BYTE_CONNECTION_ID,
+      QuicPacketCreatorPeer::GetPacketNumberLength(creator_),
+      QuicPacketCreatorPeer::GetRetryTokenLengthLength(creator_),
+      QuicPacketCreatorPeer::GetLengthLength(creator_), &payload_length);
+  connection_.SetMaxPacketLength(length);
+
+  // Queue the first packet.
+  EXPECT_CALL(*send_algorithm_, CanSend(_)).WillOnce(testing::Return(false));
+  const QuicString payload(payload_length, 'a');
+  EXPECT_EQ(0u, connection_.SendStreamDataWithString(3, payload, 0, NO_FIN)
+                    .bytes_consumed);
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+}
+
+TEST_P(QuicConnectionTest, LoopThroughSendingPackets) {
+  // All packets carry version info till version is negotiated.
+  size_t payload_length;
+
+  // Number of packets this test generates. The goal is to have
+  // kPacketCount packets, each the same size (overhead and payload).
+  // The payload will vary depending on the overhead (which in turn
+  // varies per the QUIC packet encoding rules).
+  const int kPacketCount = 7;
+
+  // Get the basic packet size. This assumes, among other things, a
+  // stream offset of 0.
+  size_t length = GetPacketLengthForOneStream(
+      connection_.version().transport_version, kIncludeVersion,
+      !kIncludeDiversificationNonce, PACKET_8BYTE_CONNECTION_ID,
+      PACKET_0BYTE_CONNECTION_ID,
+      QuicPacketCreatorPeer::GetPacketNumberLength(creator_),
+      QuicPacketCreatorPeer::GetRetryTokenLengthLength(creator_),
+      QuicPacketCreatorPeer::GetLengthLength(creator_), &payload_length);
+  // GetPacketLengthForOneStream() assumes a stream offset of 0 in determining
+  // packet length. The size of the offset field in a stream frame is
+  // 0 for offset 0, and 2 for non-zero offsets up through 16K (for
+  // versions other than 99) and 1 for non-zero offsets through 16K
+  // for version 99. Increase the length by 1 or 2, as apporpriate, so
+  // that subsequent packets containing subsequent stream frames with
+  // non-zero offsets will fit within the packet length.
+  if (connection_.version().transport_version == QUIC_VERSION_99) {
+    length = length + 1;
+  } else {
+    length = length + 2;
+  }
+
+  connection_.SetMaxPacketLength(length);
+
+  // Queue the first packet.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _))
+      .Times(kPacketCount);
+
+  size_t total_payload_length = payload_length * kPacketCount;
+  // The first frame of the stream is at offset 0. When the offset is
+  // 0, it is not included in the stream frame. Increase the total
+  // payload so that the "missing" offset byte in the first packet is
+  // occupied by a payload byte. The net result is that each of the N
+  // packets of the test will contain a single stream frame, each of
+  // which will be the same size (overhead + data).
+  if (connection_.version().transport_version == QUIC_VERSION_99) {
+    // Version 99 encodes the offset in 1 byte for the scope of this test.
+    total_payload_length = total_payload_length + 1;
+  } else {
+    // Versions other than 99 encode the offset in 2 bytes for the
+    // scope of this test.
+    total_payload_length = total_payload_length + 2;
+  }
+  const QuicString payload(total_payload_length, 'a');
+
+  EXPECT_EQ(payload.size(),
+            connection_
+                .SendStreamDataWithString(QuicUtils::GetCryptoStreamId(
+                                              connection_.transport_version()),
+                                          payload, 0, NO_FIN)
+                .bytes_consumed);
+}
+
+TEST_P(QuicConnectionTest, LoopThroughSendingPacketsWithTruncation) {
+  set_perspective(Perspective::IS_SERVER);
+  if (GetParam().version.transport_version <= QUIC_VERSION_43) {
+    // For IETF QUIC, encryption level will be switched to FORWARD_SECURE in
+    // SendStreamDataWithString.
+    QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+  }
+  // Set up a larger payload than will fit in one packet.
+  const QuicString payload(connection_.max_packet_length(), 'a');
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _)).Times(AnyNumber());
+
+  // Now send some packets with no truncation.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+  EXPECT_EQ(payload.size(),
+            connection_.SendStreamDataWithString(3, payload, 0, NO_FIN)
+                .bytes_consumed);
+  // Track the size of the second packet here.  The overhead will be the largest
+  // we see in this test, due to the non-truncated connection id.
+  size_t non_truncated_packet_size = writer_->last_packet_size();
+
+  // Change to a 0 byte connection id.
+  QuicConfig config;
+  QuicConfigPeer::SetReceivedBytesForConnectionId(&config, 0);
+  connection_.SetFromConfig(config);
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+  EXPECT_EQ(payload.size(),
+            connection_.SendStreamDataWithString(3, payload, 1350, NO_FIN)
+                .bytes_consumed);
+  if (connection_.transport_version() > QUIC_VERSION_43) {
+    // Short header packets sent from server omit connection ID already, and
+    // stream offset size increases from 0 to 2.
+    EXPECT_EQ(non_truncated_packet_size, writer_->last_packet_size() - 2);
+  } else {
+    // Just like above, we save 8 bytes on payload, and 8 on truncation. -2
+    // because stream offset size is 2 instead of 0.
+    EXPECT_EQ(non_truncated_packet_size,
+              writer_->last_packet_size() + 8 * 2 - 2);
+  }
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAck) {
+  QuicTime ack_time = clock_.ApproximateNow() + DefaultDelayedAckTime();
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(1, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+  // Simulate delayed ack alarm firing.
+  connection_.GetAckAlarm()->Fire();
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAfterQuiescence) {
+  QuicConnectionPeer::SetFastAckAfterQuiescence(&connection_, true);
+
+  // The beginning of the connection counts as quiescence.
+  QuicTime ack_time =
+      clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(1, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+  // Simulate delayed ack alarm firing.
+  connection_.GetAckAlarm()->Fire();
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Process another packet immedately after sending the ack and expect the
+  // ack alarm to be set delayed ack time in the future.
+  ack_time = clock_.ApproximateNow() + DefaultDelayedAckTime();
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(2, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+  // Simulate delayed ack alarm firing.
+  connection_.GetAckAlarm()->Fire();
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Wait 1 second and enesure the ack alarm is set to 1ms in the future.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(1));
+  ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(1);
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(3, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckDecimation) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/4, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 4);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // The 10th received packet causes an ack to be sent.
+  for (int i = 0; i < 9; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckAckDecimationAfterQuiescence) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION);
+  QuicConnectionPeer::SetFastAckAfterQuiescence(&connection_, true);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+
+  // The beginning of the connection counts as quiescence.
+  QuicTime ack_time =
+      clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(1, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+  // Simulate delayed ack alarm firing.
+  connection_.GetAckAlarm()->Fire();
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Process another packet immedately after sending the ack and expect the
+  // ack alarm to be set delayed ack time in the future.
+  ack_time = clock_.ApproximateNow() + DefaultDelayedAckTime();
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(2, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+  // Simulate delayed ack alarm firing.
+  connection_.GetAckAlarm()->Fire();
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Wait 1 second and enesure the ack alarm is set to 1ms in the future.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(1));
+  ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(1);
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(3, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // Process enough packets to get into ack decimation behavior.
+  // The ack time should be based on min_rtt/4, since it's less than the
+  // default delayed ack time.
+  ack_time = clock_.ApproximateNow() +
+             QuicTime::Delta::FromMilliseconds(kMinRttMs / 4);
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 4; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(4 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // The 10th received packet causes an ack to be sent.
+  for (int i = 0; i < 9; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Wait 1 second and enesure the ack alarm is set to 1ms in the future.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(1));
+  ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(1);
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket + 10, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckDecimationUnlimitedAggregation) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  QuicTagVector connection_options;
+  connection_options.push_back(kACKD);
+  // No limit on the number of packets received before sending an ack.
+  connection_options.push_back(kAKDU);
+  config.SetConnectionOptionsToSend(connection_options);
+  connection_.SetFromConfig(config);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/4, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 4);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // 18 packets will not cause an ack to be sent.  19 will because when
+  // stop waiting frames are in use, we ack every 20 packets no matter what.
+  for (int i = 0; i < 18; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // The delayed ack timer should still be set to the expected deadline.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckDecimationEighthRtt) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION);
+  QuicConnectionPeer::SetAckDecimationDelay(&connection_, 0.125);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/8, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 8);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // The 10th received packet causes an ack to be sent.
+  for (int i = 0; i < 9; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckDecimationWithReordering) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION_WITH_REORDERING);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/4, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 4);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Receive one packet out of order and then the rest in order.
+  // The loop leaves a one packet gap between acks sent to simulate some loss.
+  for (int j = 0; j < 3; ++j) {
+    // Process packet 10 first and ensure the alarm is one eighth min_rtt.
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 9 + (j * 11),
+                             !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+    ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(5);
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+    // The 10th received packet causes an ack to be sent.
+    writer_->Reset();
+    for (int i = 0; i < 9; ++i) {
+      EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+      EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+      // The ACK shouldn't be sent until the 10th packet is processed.
+      EXPECT_TRUE(writer_->ack_frames().empty());
+      ProcessDataPacketAtLevel(kFirstDecimatedPacket + i + (j * 11),
+                               !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+    }
+    // Check that ack is sent and that delayed ack alarm is reset.
+    if (GetParam().no_stop_waiting) {
+      EXPECT_EQ(1u, writer_->frame_count());
+      EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+    } else {
+      EXPECT_EQ(2u, writer_->frame_count());
+      EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+    }
+    EXPECT_FALSE(writer_->ack_frames().empty());
+    EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  }
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckDecimationWithLargeReordering) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION_WITH_REORDERING);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/4, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 4);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // Process packet 10 first and ensure the alarm is one eighth min_rtt.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket + 19, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+  ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(5);
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // The 10th received packet causes an ack to be sent.
+  for (int i = 0; i < 8; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // The next packet received in order will cause an immediate ack,
+  // because it fills a hole.
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket + 10, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckDecimationWithReorderingEighthRtt) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION_WITH_REORDERING);
+  QuicConnectionPeer::SetAckDecimationDelay(&connection_, 0.125);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/8, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 8);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // Process packet 10 first and ensure the alarm is one eighth min_rtt.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket + 9, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+  ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(5);
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // The 10th received packet causes an ack to be sent.
+  for (int i = 0; i < 8; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest,
+       SendDelayedAckDecimationWithLargeReorderingEighthRtt) {
+  EXPECT_CALL(visitor_, OnAckNeedsRetransmittableFrame()).Times(AnyNumber());
+  QuicConnectionPeer::SetAckMode(&connection_, ACK_DECIMATION_WITH_REORDERING);
+  QuicConnectionPeer::SetAckDecimationDelay(&connection_, 0.125);
+
+  const size_t kMinRttMs = 40;
+  RttStats* rtt_stats = const_cast<RttStats*>(manager_->GetRttStats());
+  rtt_stats->UpdateRtt(QuicTime::Delta::FromMilliseconds(kMinRttMs),
+                       QuicTime::Delta::Zero(), QuicTime::Zero());
+  // The ack time should be based on min_rtt/8, since it's less than the
+  // default delayed ack time.
+  QuicTime ack_time = clock_.ApproximateNow() +
+                      QuicTime::Delta::FromMilliseconds(kMinRttMs / 8);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  const uint8_t tag = 0x07;
+  connection_.SetDecrypter(ENCRYPTION_ZERO_RTT,
+                           QuicMakeUnique<StrictTaggingDecrypter>(tag));
+  peer_framer_.SetEncrypter(ENCRYPTION_ZERO_RTT,
+                            QuicMakeUnique<TaggingEncrypter>(tag));
+  // Process a packet from the non-crypto stream.
+  frame1_.stream_id = 3;
+
+  // Process all the initial packets in order so there aren't missing packets.
+  uint64_t kFirstDecimatedPacket = 101;
+  for (unsigned int i = 0; i < kFirstDecimatedPacket - 1; ++i) {
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(1 + i, !kHasStopWaiting, ENCRYPTION_ZERO_RTT);
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  // The same as ProcessPacket(1) except that ENCRYPTION_ZERO_RTT is used
+  // instead of ENCRYPTION_NONE.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+
+  // Check if delayed ack timer is running for the expected interval.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // Process packet 10 first and ensure the alarm is one eighth min_rtt.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket + 19, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+  ack_time = clock_.ApproximateNow() + QuicTime::Delta::FromMilliseconds(5);
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // The 10th received packet causes an ack to be sent.
+  for (int i = 0; i < 8; ++i) {
+    EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+    EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+    ProcessDataPacketAtLevel(kFirstDecimatedPacket + 1 + i, !kHasStopWaiting,
+                             ENCRYPTION_ZERO_RTT);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // The next packet received in order will cause an immediate ack,
+  // because it fills a hole.
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacketAtLevel(kFirstDecimatedPacket + 10, !kHasStopWaiting,
+                           ENCRYPTION_ZERO_RTT);
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckOnHandshakeConfirmed) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessPacket(1);
+  // Check that ack is sent and that delayed ack alarm is set.
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  QuicTime ack_time = clock_.ApproximateNow() + DefaultDelayedAckTime();
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // Completing the handshake as the server does nothing.
+  QuicConnectionPeer::SetPerspective(&connection_, Perspective::IS_SERVER);
+  connection_.OnHandshakeComplete();
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(ack_time, connection_.GetAckAlarm()->deadline());
+
+  // Complete the handshake as the client decreases the delayed ack time to 0ms.
+  QuicConnectionPeer::SetPerspective(&connection_, Perspective::IS_CLIENT);
+  connection_.OnHandshakeComplete();
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow(), connection_.GetAckAlarm()->deadline());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckOnSecondPacket) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessPacket(1);
+  ProcessPacket(2);
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(1u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, NoAckOnOldNacks) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  // Drop one packet, triggering a sequence of acks.
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  } else {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  }
+  ProcessPacket(2);
+  size_t frames_per_ack = GetParam().no_stop_waiting ? 1 : 2;
+  if (!GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    EXPECT_EQ(frames_per_ack, writer_->frame_count());
+    EXPECT_FALSE(writer_->ack_frames().empty());
+    writer_->Reset();
+  }
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  ProcessPacket(3);
+  EXPECT_EQ(frames_per_ack, writer_->frame_count());
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  writer_->Reset();
+
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  } else {
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  }
+  ProcessPacket(4);
+  if (GetQuicRestartFlag(quic_enable_accept_random_ipn)) {
+    EXPECT_EQ(0u, writer_->frame_count());
+  } else {
+    EXPECT_EQ(frames_per_ack, writer_->frame_count());
+    EXPECT_FALSE(writer_->ack_frames().empty());
+    writer_->Reset();
+  }
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  ProcessPacket(5);
+  EXPECT_EQ(frames_per_ack, writer_->frame_count());
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  writer_->Reset();
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  // Now only set the timer on the 6th packet, instead of sending another ack.
+  ProcessPacket(6);
+  EXPECT_EQ(0u, writer_->frame_count());
+  EXPECT_TRUE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckOnOutgoingPacket) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessPacket(1);
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, NO_FIN);
+  // Check that ack is bundled with outgoing data and that delayed ack
+  // alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(3u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, SendDelayedAckOnOutgoingCryptoPacket) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessPacket(1);
+  connection_.SendStreamDataWithString(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), "foo", 0,
+      NO_FIN);
+  // Check that ack is bundled with outgoing crypto data.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(3u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(4u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, BlockAndBufferOnFirstCHLOPacketOfTwo) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessPacket(1);
+  BlockOnNextWrite();
+  writer_->set_is_write_blocked_data_buffered(true);
+  connection_.SendStreamDataWithString(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), "foo", 0,
+      NO_FIN);
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+  EXPECT_FALSE(connection_.HasQueuedData());
+  connection_.SendStreamDataWithString(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), "bar", 3,
+      NO_FIN);
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+  EXPECT_TRUE(connection_.HasQueuedData());
+}
+
+TEST_P(QuicConnectionTest, BundleAckForSecondCHLO) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  EXPECT_CALL(visitor_, OnCanWrite())
+      .WillOnce(IgnoreResult(InvokeWithoutArgs(
+          &connection_, &TestConnection::SendCryptoStreamData)));
+  // Process a packet from the crypto stream, which is frame1_'s default.
+  // Receiving the CHLO as packet 2 first will cause the connection to
+  // immediately send an ack, due to the packet gap.
+  ProcessPacket(2);
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(3u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(4u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  if (connection_.transport_version() < QUIC_VERSION_47) {
+    EXPECT_EQ(1u, writer_->stream_frames().size());
+  } else {
+    EXPECT_EQ(1u, writer_->crypto_frames().size());
+  }
+  EXPECT_EQ(1u, writer_->padding_frames().size());
+  ASSERT_FALSE(writer_->ack_frames().empty());
+  EXPECT_EQ(QuicPacketNumber(2u), LargestAcked(writer_->ack_frames().front()));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, BundleAckForSecondCHLOTwoPacketReject) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+
+  // Process two packets from the crypto stream, which is frame1_'s default,
+  // simulating a 2 packet reject.
+  {
+    ProcessPacket(1);
+    // Send the new CHLO when the REJ is processed.
+    EXPECT_CALL(visitor_, OnStreamFrame(_))
+        .WillOnce(IgnoreResult(InvokeWithoutArgs(
+            &connection_, &TestConnection::SendCryptoStreamData)));
+    ProcessDataPacket(2);
+  }
+  // Check that ack is sent and that delayed ack alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(3u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(4u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  if (connection_.transport_version() < QUIC_VERSION_47) {
+    EXPECT_EQ(1u, writer_->stream_frames().size());
+  } else {
+    EXPECT_EQ(1u, writer_->crypto_frames().size());
+  }
+  EXPECT_EQ(1u, writer_->padding_frames().size());
+  ASSERT_FALSE(writer_->ack_frames().empty());
+  EXPECT_EQ(QuicPacketNumber(2u), LargestAcked(writer_->ack_frames().front()));
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, BundleAckWithDataOnIncomingAck) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, NO_FIN);
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      3, NO_FIN);
+  // Ack the second packet, which will retransmit the first packet.
+  QuicAckFrame ack = ConstructAckFrame(2, 1);
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(QuicPacketNumber(1), kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&ack);
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  writer_->Reset();
+
+  // Now ack the retransmission, which will both raise the high water mark
+  // and see if there is more data to send.
+  ack = ConstructAckFrame(3, 1);
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  ProcessAckPacket(&ack);
+
+  // Check that no packet is sent and the ack alarm isn't set.
+  EXPECT_EQ(0u, writer_->frame_count());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+  writer_->Reset();
+
+  // Send the same ack, but send both data and an ack together.
+  ack = ConstructAckFrame(3, 1);
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(visitor_, OnCanWrite())
+      .WillOnce(IgnoreResult(InvokeWithoutArgs(
+          &connection_, &TestConnection::EnsureWritableAndSendStreamData5)));
+  ProcessAckPacket(&ack);
+
+  // Check that ack is bundled with outgoing data and the delayed ack
+  // alarm is reset.
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(2u, writer_->frame_count());
+    EXPECT_TRUE(writer_->stop_waiting_frames().empty());
+  } else {
+    EXPECT_EQ(3u, writer_->frame_count());
+    EXPECT_FALSE(writer_->stop_waiting_frames().empty());
+  }
+  EXPECT_FALSE(writer_->ack_frames().empty());
+  EXPECT_EQ(QuicPacketNumber(3u), LargestAcked(writer_->ack_frames().front()));
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_FALSE(connection_.GetAckAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, NoAckSentForClose) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessPacket(1);
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PEER_GOING_AWAY, _,
+                                           ConnectionCloseSource::FROM_PEER));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  ProcessClosePacket(2);
+}
+
+TEST_P(QuicConnectionTest, SendWhenDisconnected) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PEER_GOING_AWAY, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  connection_.CloseConnection(QUIC_PEER_GOING_AWAY, "no reason",
+                              ConnectionCloseBehavior::SILENT_CLOSE);
+  EXPECT_FALSE(connection_.connected());
+  EXPECT_FALSE(connection_.CanWriteStreamData());
+  std::unique_ptr<QuicPacket> packet = ConstructDataPacket(1, !kHasStopWaiting);
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(1), _, _))
+      .Times(0);
+  connection_.SendPacket(ENCRYPTION_NONE, 1, std::move(packet),
+                         HAS_RETRANSMITTABLE_DATA, false, false);
+}
+
+TEST_P(QuicConnectionTest, SendConnectivityProbingWhenDisconnected) {
+  // EXPECT_QUIC_BUG tests are expensive so only run one instance of them.
+  if (!IsDefaultTestConfiguration()) {
+    return;
+  }
+
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PEER_GOING_AWAY, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  connection_.CloseConnection(QUIC_PEER_GOING_AWAY, "no reason",
+                              ConnectionCloseBehavior::SILENT_CLOSE);
+  EXPECT_FALSE(connection_.connected());
+  EXPECT_FALSE(connection_.CanWriteStreamData());
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(1), _, _))
+      .Times(0);
+
+  EXPECT_QUIC_BUG(connection_.SendConnectivityProbingPacket(
+                      writer_.get(), connection_.peer_address()),
+                  "Not sending connectivity probing packet as connection is "
+                  "disconnected.");
+}
+
+TEST_P(QuicConnectionTest, WriteBlockedAfterClientSendsConnectivityProbe) {
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+  TestPacketWriter probing_writer(version(), &clock_);
+  // Block next write so that sending connectivity probe will encounter a
+  // blocked write when send a connectivity probe to the peer.
+  probing_writer.BlockOnNextWrite();
+  // Connection will not be marked as write blocked as connectivity probe only
+  // affects the probing_writer which is not the default.
+  EXPECT_CALL(visitor_, OnWriteBlocked()).Times(0);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(1), _, _))
+      .Times(1);
+  connection_.SendConnectivityProbingPacket(&probing_writer,
+                                            connection_.peer_address());
+}
+
+TEST_P(QuicConnectionTest, WriterBlockedAfterServerSendsConnectivityProbe) {
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  // Block next write so that sending connectivity probe will encounter a
+  // blocked write when send a connectivity probe to the peer.
+  writer_->BlockOnNextWrite();
+  // Connection will be marked as write blocked as server uses the default
+  // writer to send connectivity probes.
+  EXPECT_CALL(visitor_, OnWriteBlocked()).Times(1);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(1), _, _))
+      .Times(1);
+  connection_.SendConnectivityProbingPacket(writer_.get(),
+                                            connection_.peer_address());
+}
+
+TEST_P(QuicConnectionTest, WriterErrorWhenClientSendsConnectivityProbe) {
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+  TestPacketWriter probing_writer(version(), &clock_);
+  probing_writer.SetShouldWriteFail();
+
+  // Connection should not be closed if a connectivity probe is failed to be
+  // sent.
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(0);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(1), _, _))
+      .Times(0);
+  connection_.SendConnectivityProbingPacket(&probing_writer,
+                                            connection_.peer_address());
+}
+
+TEST_P(QuicConnectionTest, WriterErrorWhenServerSendsConnectivityProbe) {
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  writer_->SetShouldWriteFail();
+  // Connection should not be closed if a connectivity probe is failed to be
+  // sent.
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(0);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, QuicPacketNumber(1), _, _))
+      .Times(0);
+  connection_.SendConnectivityProbingPacket(writer_.get(),
+                                            connection_.peer_address());
+}
+
+TEST_P(QuicConnectionTest, PublicReset) {
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    return;
+  }
+  QuicPublicResetPacket header;
+  // Public reset packet in only built by server.
+  header.connection_id = connection_id_;
+  std::unique_ptr<QuicEncryptedPacket> packet(
+      framer_.BuildPublicResetPacket(header));
+  std::unique_ptr<QuicReceivedPacket> received(
+      ConstructReceivedPacket(*packet, QuicTime::Zero()));
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PUBLIC_RESET, _,
+                                           ConnectionCloseSource::FROM_PEER));
+  connection_.ProcessUdpPacket(kSelfAddress, kPeerAddress, *received);
+}
+
+TEST_P(QuicConnectionTest, IetfStatelessReset) {
+  if (GetParam().version.transport_version <= QUIC_VERSION_43) {
+    return;
+  }
+  const QuicUint128 kTestStatelessResetToken = 1010101;
+  QuicConfig config;
+  QuicConfigPeer::SetReceivedStatelessResetToken(&config,
+                                                 kTestStatelessResetToken);
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  connection_.SetFromConfig(config);
+  std::unique_ptr<QuicEncryptedPacket> packet(
+      QuicFramer::BuildIetfStatelessResetPacket(connection_id_,
+                                                kTestStatelessResetToken));
+  std::unique_ptr<QuicReceivedPacket> received(
+      ConstructReceivedPacket(*packet, QuicTime::Zero()));
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PUBLIC_RESET, _,
+                                           ConnectionCloseSource::FROM_PEER));
+  connection_.ProcessUdpPacket(kSelfAddress, kPeerAddress, *received);
+}
+
+TEST_P(QuicConnectionTest, GoAway) {
+  if (GetParam().version.transport_version == QUIC_VERSION_99) {
+    // GoAway is not available in version 99.
+    return;
+  }
+
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  QuicGoAwayFrame goaway;
+  goaway.last_good_stream_id = 1;
+  goaway.error_code = QUIC_PEER_GOING_AWAY;
+  goaway.reason_phrase = "Going away.";
+  EXPECT_CALL(visitor_, OnGoAway(_));
+  ProcessGoAwayPacket(&goaway);
+}
+
+TEST_P(QuicConnectionTest, WindowUpdate) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  QuicWindowUpdateFrame window_update;
+  window_update.stream_id = 3;
+  window_update.byte_offset = 1234;
+  EXPECT_CALL(visitor_, OnWindowUpdateFrame(_));
+  ProcessFramePacket(QuicFrame(&window_update));
+}
+
+TEST_P(QuicConnectionTest, Blocked) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  QuicBlockedFrame blocked;
+  blocked.stream_id = 3;
+  EXPECT_CALL(visitor_, OnBlockedFrame(_));
+  ProcessFramePacket(QuicFrame(&blocked));
+  EXPECT_EQ(1u, connection_.GetStats().blocked_frames_received);
+  EXPECT_EQ(0u, connection_.GetStats().blocked_frames_sent);
+}
+
+TEST_P(QuicConnectionTest, ZeroBytePacket) {
+  // Don't close the connection for zero byte packets.
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(0);
+  QuicReceivedPacket encrypted(nullptr, 0, QuicTime::Zero());
+  connection_.ProcessUdpPacket(kSelfAddress, kPeerAddress, encrypted);
+}
+
+TEST_P(QuicConnectionTest, MissingPacketsBeforeLeastUnacked) {
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    return;
+  }
+  // Set the packet number of the ack packet to be least unacked (4).
+  QuicPacketCreatorPeer::SetPacketNumber(&peer_creator_, 3);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessStopWaitingPacket(InitStopWaitingFrame(4));
+  EXPECT_FALSE(outgoing_ack()->packets.Empty());
+}
+
+TEST_P(QuicConnectionTest, ServerSendsVersionNegotiationPacket) {
+  // Turn off QUIC_VERSION_99.
+  SetQuicReloadableFlag(quic_enable_version_99, false);
+  connection_.SetSupportedVersions(CurrentSupportedVersions());
+  set_perspective(Perspective::IS_SERVER);
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    peer_framer_.set_version_for_tests(
+        ParsedQuicVersion(PROTOCOL_QUIC_CRYPTO, QUIC_VERSION_99));
+  } else {
+    peer_framer_.set_version_for_tests(UnsupportedQuicVersion());
+  }
+
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  header.version_flag = true;
+  header.packet_number = QuicPacketNumber(12);
+
+  if (QuicVersionHasLongHeaderLengths(
+          peer_framer_.version().transport_version)) {
+    header.long_packet_type = INITIAL;
+    header.retry_token_length_length = VARIABLE_LENGTH_INTEGER_LENGTH_1;
+    header.length_length = VARIABLE_LENGTH_INTEGER_LENGTH_2;
+  }
+
+  QuicFrames frames;
+  frames.push_back(QuicFrame(frame1_));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(12), *packet, buffer, kMaxPacketSize);
+
+  framer_.set_version(version());
+  // Writer's framer's perspective is client, so that it needs to have the right
+  // version to process either IETF or GQUIC version negotiation packet.
+  writer_->SetSupportedVersions({version()});
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+  EXPECT_TRUE(writer_->version_negotiation_packet() != nullptr);
+
+  ParsedQuicVersionVector supported_versions = CurrentSupportedVersions();
+  ASSERT_EQ(supported_versions.size(),
+            writer_->version_negotiation_packet()->versions.size());
+
+  // We expect all versions in supported_versions to be
+  // included in the packet.
+  for (size_t i = 0; i < supported_versions.size(); ++i) {
+    EXPECT_EQ(supported_versions[i],
+              writer_->version_negotiation_packet()->versions[i]);
+  }
+}
+
+TEST_P(QuicConnectionTest, ServerSendsVersionNegotiationPacketSocketBlocked) {
+  // Turn off QUIC_VERSION_99.
+  SetQuicReloadableFlag(quic_enable_version_99, false);
+  connection_.SetSupportedVersions(CurrentSupportedVersions());
+  set_perspective(Perspective::IS_SERVER);
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    peer_framer_.set_version_for_tests(
+        ParsedQuicVersion(PROTOCOL_QUIC_CRYPTO, QUIC_VERSION_99));
+  } else {
+    peer_framer_.set_version_for_tests(UnsupportedQuicVersion());
+  }
+
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  header.version_flag = true;
+  header.packet_number = QuicPacketNumber(12);
+
+  if (QuicVersionHasLongHeaderLengths(
+          peer_framer_.version().transport_version)) {
+    header.long_packet_type = INITIAL;
+    header.retry_token_length_length = VARIABLE_LENGTH_INTEGER_LENGTH_1;
+    header.length_length = VARIABLE_LENGTH_INTEGER_LENGTH_2;
+  }
+
+  QuicFrames frames;
+  frames.push_back(QuicFrame(frame1_));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(12), *packet, buffer, kMaxPacketSize);
+
+  framer_.set_version(version());
+  BlockOnNextWrite();
+  // Writer's framer's perspective is client, so that it needs to have the right
+  // version to process either IETF or GQUIC version negotiation packet.
+  writer_->SetSupportedVersions({version()});
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+  EXPECT_EQ(0u, writer_->last_packet_size());
+  EXPECT_TRUE(connection_.HasQueuedData());
+
+  writer_->SetWritable();
+  connection_.OnCanWrite();
+  EXPECT_TRUE(writer_->version_negotiation_packet() != nullptr);
+
+  ParsedQuicVersionVector supported_versions = CurrentSupportedVersions();
+  ASSERT_EQ(supported_versions.size(),
+            writer_->version_negotiation_packet()->versions.size());
+
+  // We expect all versions in supported_versions to be
+  // included in the packet.
+  for (size_t i = 0; i < supported_versions.size(); ++i) {
+    EXPECT_EQ(supported_versions[i],
+              writer_->version_negotiation_packet()->versions[i]);
+  }
+}
+
+TEST_P(QuicConnectionTest,
+       ServerSendsVersionNegotiationPacketSocketBlockedDataBuffered) {
+  // Turn off QUIC_VERSION_99.
+  SetQuicReloadableFlag(quic_enable_version_99, false);
+  connection_.SetSupportedVersions(CurrentSupportedVersions());
+  set_perspective(Perspective::IS_SERVER);
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    peer_framer_.set_version_for_tests(
+        ParsedQuicVersion(PROTOCOL_QUIC_CRYPTO, QUIC_VERSION_99));
+  } else {
+    peer_framer_.set_version_for_tests(UnsupportedQuicVersion());
+  }
+
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  header.version_flag = true;
+  header.packet_number = QuicPacketNumber(12);
+
+  if (QuicVersionHasLongHeaderLengths(
+          peer_framer_.version().transport_version)) {
+    header.long_packet_type = INITIAL;
+    header.retry_token_length_length = VARIABLE_LENGTH_INTEGER_LENGTH_1;
+    header.length_length = VARIABLE_LENGTH_INTEGER_LENGTH_2;
+  }
+
+  QuicFrames frames;
+  frames.push_back(QuicFrame(frame1_));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  char buffer[kMaxPacketSize];
+  size_t encryped_length = framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(12), *packet, buffer, kMaxPacketSize);
+
+  framer_.set_version(version());
+  set_perspective(Perspective::IS_SERVER);
+  BlockOnNextWrite();
+  writer_->set_is_write_blocked_data_buffered(true);
+  // Writer's framer's perspective is client, so that it needs to have the right
+  // version to process either IETF or GQUIC version negotiation packet.
+  writer_->SetSupportedVersions({version()});
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encryped_length, QuicTime::Zero(), false));
+  EXPECT_EQ(0u, writer_->last_packet_size());
+  EXPECT_FALSE(connection_.HasQueuedData());
+}
+
+TEST_P(QuicConnectionTest, ClientHandlesVersionNegotiation) {
+  // Start out with some unsupported version.
+  QuicConnectionPeer::GetFramer(&connection_)
+      ->set_version_for_tests(ParsedQuicVersion(
+          PROTOCOL_UNSUPPORTED,
+          GetParam().version.transport_version == QUIC_VERSION_99
+              ? QUIC_VERSION_99
+              : QUIC_VERSION_UNSUPPORTED));
+
+  // Send a version negotiation packet.
+  std::unique_ptr<QuicEncryptedPacket> encrypted(
+      peer_framer_.BuildVersionNegotiationPacket(
+          connection_id_, connection_.transport_version() > QUIC_VERSION_43,
+          AllSupportedVersions()));
+  std::unique_ptr<QuicReceivedPacket> received(
+      ConstructReceivedPacket(*encrypted, QuicTime::Zero()));
+  if (GetQuicReloadableFlag(quic_no_client_conn_ver_negotiation)) {
+    EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_VERSION, _,
+                                             ConnectionCloseSource::FROM_SELF));
+  }
+  connection_.ProcessUdpPacket(kSelfAddress, kPeerAddress, *received);
+  if (GetQuicReloadableFlag(quic_no_client_conn_ver_negotiation)) {
+    EXPECT_FALSE(connection_.connected());
+    return;
+  }
+
+  // Now force another packet.  The connection should transition into
+  // NEGOTIATED_VERSION state and tell the packet creator to StopSendingVersion.
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  header.destination_connection_id_included = CONNECTION_ID_ABSENT;
+  header.packet_number = QuicPacketNumber(12);
+  header.version_flag = false;
+  QuicFrames frames;
+  frames.push_back(QuicFrame(frame1_));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = peer_framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(12), *packet, buffer, kMaxPacketSize);
+  ASSERT_NE(0u, encrypted_length);
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    // IETF QUIC stops sending version when switch to FORWARD_SECURE.
+    EXPECT_NE(ENCRYPTION_FORWARD_SECURE, connection_.encryption_level());
+    ASSERT_TRUE(QuicPacketCreatorPeer::SendVersionInPacket(creator_));
+  } else {
+    ASSERT_FALSE(QuicPacketCreatorPeer::SendVersionInPacket(creator_));
+  }
+}
+
+TEST_P(QuicConnectionTest, BadVersionNegotiation) {
+  // Send a version negotiation packet with the version the client started with.
+  // It should be rejected.
+  EXPECT_CALL(visitor_,
+              OnConnectionClosed(QUIC_INVALID_VERSION_NEGOTIATION_PACKET, _,
+                                 ConnectionCloseSource::FROM_SELF));
+  std::unique_ptr<QuicEncryptedPacket> encrypted(
+      framer_.BuildVersionNegotiationPacket(
+          connection_id_, connection_.transport_version() > QUIC_VERSION_43,
+          AllSupportedVersions()));
+  std::unique_ptr<QuicReceivedPacket> received(
+      ConstructReceivedPacket(*encrypted, QuicTime::Zero()));
+  connection_.ProcessUdpPacket(kSelfAddress, kPeerAddress, *received);
+}
+
+TEST_P(QuicConnectionTest, CheckSendStats) {
+  connection_.SetMaxTailLossProbes(0);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.SendStreamDataWithString(3, "first", 0, NO_FIN);
+  size_t first_packet_size = writer_->last_packet_size();
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+  connection_.SendStreamDataWithString(5, "second", 0, NO_FIN);
+  size_t second_packet_size = writer_->last_packet_size();
+
+  // 2 retransmissions due to rto, 1 due to explicit nack.
+  EXPECT_CALL(*send_algorithm_, OnRetransmissionTimeout(true));
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(3);
+
+  // Retransmit due to RTO.
+  clock_.AdvanceTime(QuicTime::Delta::FromSeconds(10));
+  connection_.GetRetransmissionAlarm()->Fire();
+
+  // Retransmit due to explicit nacks.
+  QuicAckFrame nack_three =
+      InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)},
+                    {QuicPacketNumber(4), QuicPacketNumber(5)}});
+
+  LostPacketVector lost_packets;
+  lost_packets.push_back(LostPacket(QuicPacketNumber(1), kMaxPacketSize));
+  lost_packets.push_back(LostPacket(QuicPacketNumber(3), kMaxPacketSize));
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _))
+      .WillOnce(SetArgPointee<5>(lost_packets));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  if (!connection_.session_decides_what_to_write()) {
+    EXPECT_CALL(visitor_, OnCanWrite());
+  }
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  ProcessAckPacket(&nack_three);
+
+  EXPECT_CALL(*send_algorithm_, BandwidthEstimate())
+      .WillOnce(Return(QuicBandwidth::Zero()));
+
+  const QuicConnectionStats& stats = connection_.GetStats();
+  // For IETF QUIC, version is not included as the encryption level switches to
+  // FORWARD_SECURE in SendStreamDataWithString.
+  size_t save_on_version =
+      GetParam().version.transport_version > QUIC_VERSION_43 ? 0
+                                                             : kQuicVersionSize;
+  EXPECT_EQ(3 * first_packet_size + 2 * second_packet_size - save_on_version,
+            stats.bytes_sent);
+  EXPECT_EQ(5u, stats.packets_sent);
+  EXPECT_EQ(2 * first_packet_size + second_packet_size - save_on_version,
+            stats.bytes_retransmitted);
+  EXPECT_EQ(3u, stats.packets_retransmitted);
+  EXPECT_EQ(1u, stats.rto_count);
+  EXPECT_EQ(kDefaultMaxPacketSize, stats.max_packet_size);
+}
+
+TEST_P(QuicConnectionTest, ProcessFramesIfPacketClosedConnection) {
+  // Construct a packet with stream frame and connection close frame.
+  QuicPacketHeader header;
+  header.destination_connection_id = connection_id_;
+  if (peer_framer_.transport_version() > QUIC_VERSION_43) {
+    header.destination_connection_id_included = CONNECTION_ID_ABSENT;
+  }
+  header.packet_number = QuicPacketNumber(1);
+  header.version_flag = false;
+
+  QuicConnectionCloseFrame qccf;
+  qccf.error_code = QUIC_PEER_GOING_AWAY;
+
+  QuicFrames frames;
+  frames.push_back(QuicFrame(frame1_));
+  frames.push_back(QuicFrame(&qccf));
+  std::unique_ptr<QuicPacket> packet(ConstructPacket(header, frames));
+  EXPECT_TRUE(nullptr != packet);
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = peer_framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(1), *packet, buffer, kMaxPacketSize);
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PEER_GOING_AWAY, _,
+                                           ConnectionCloseSource::FROM_PEER));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, QuicTime::Zero(), false));
+}
+
+TEST_P(QuicConnectionTest, SelectMutualVersion) {
+  connection_.SetSupportedVersions(AllSupportedVersions());
+  // Set the connection to speak the lowest quic version.
+  connection_.set_version(QuicVersionMin());
+  EXPECT_EQ(QuicVersionMin(), connection_.version());
+
+  // Pass in available versions which includes a higher mutually supported
+  // version.  The higher mutually supported version should be selected.
+  ParsedQuicVersionVector supported_versions = AllSupportedVersions();
+  EXPECT_TRUE(connection_.SelectMutualVersion(supported_versions));
+  EXPECT_EQ(QuicVersionMax(), connection_.version());
+
+  // Expect that the lowest version is selected.
+  // Ensure the lowest supported version is less than the max, unless they're
+  // the same.
+  ParsedQuicVersionVector lowest_version_vector;
+  lowest_version_vector.push_back(QuicVersionMin());
+  EXPECT_TRUE(connection_.SelectMutualVersion(lowest_version_vector));
+  EXPECT_EQ(QuicVersionMin(), connection_.version());
+
+  // Shouldn't be able to find a mutually supported version.
+  ParsedQuicVersionVector unsupported_version;
+  unsupported_version.push_back(UnsupportedQuicVersion());
+  EXPECT_FALSE(connection_.SelectMutualVersion(unsupported_version));
+}
+
+TEST_P(QuicConnectionTest, ConnectionCloseWhenWritable) {
+  EXPECT_FALSE(writer_->IsWriteBlocked());
+
+  // Send a packet.
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_EQ(0u, connection_.NumQueuedPackets());
+  EXPECT_EQ(1u, writer_->packets_write_attempts());
+
+  TriggerConnectionClose();
+  EXPECT_EQ(2u, writer_->packets_write_attempts());
+}
+
+TEST_P(QuicConnectionTest, ConnectionCloseGettingWriteBlocked) {
+  BlockOnNextWrite();
+  TriggerConnectionClose();
+  EXPECT_EQ(1u, writer_->packets_write_attempts());
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+}
+
+TEST_P(QuicConnectionTest, ConnectionCloseWhenWriteBlocked) {
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+  EXPECT_EQ(1u, connection_.NumQueuedPackets());
+  EXPECT_EQ(1u, writer_->packets_write_attempts());
+  EXPECT_TRUE(writer_->IsWriteBlocked());
+  TriggerConnectionClose();
+  EXPECT_EQ(1u, writer_->packets_write_attempts());
+}
+
+TEST_P(QuicConnectionTest, OnPacketSentDebugVisitor) {
+  MockQuicConnectionDebugVisitor debug_visitor;
+  connection_.set_debug_visitor(&debug_visitor);
+
+  EXPECT_CALL(debug_visitor, OnPacketSent(_, _, _, _)).Times(1);
+  connection_.SendStreamDataWithString(1, "foo", 0, NO_FIN);
+
+  EXPECT_CALL(debug_visitor, OnPacketSent(_, _, _, _)).Times(1);
+  connection_.SendConnectivityProbingPacket(writer_.get(),
+                                            connection_.peer_address());
+}
+
+TEST_P(QuicConnectionTest, OnPacketHeaderDebugVisitor) {
+  QuicPacketHeader header;
+  header.packet_number = QuicPacketNumber(1);
+  if (GetParam().version.transport_version > QUIC_VERSION_43) {
+    header.form = IETF_QUIC_LONG_HEADER_PACKET;
+  }
+
+  MockQuicConnectionDebugVisitor debug_visitor;
+  connection_.set_debug_visitor(&debug_visitor);
+  EXPECT_CALL(debug_visitor, OnPacketHeader(Ref(header))).Times(1);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_)).Times(1);
+  EXPECT_CALL(debug_visitor, OnSuccessfulVersionNegotiation(_)).Times(1);
+  connection_.OnPacketHeader(header);
+}
+
+TEST_P(QuicConnectionTest, Pacing) {
+  TestConnection server(connection_id_, kSelfAddress, helper_.get(),
+                        alarm_factory_.get(), writer_.get(),
+                        Perspective::IS_SERVER, version());
+  TestConnection client(connection_id_, kPeerAddress, helper_.get(),
+                        alarm_factory_.get(), writer_.get(),
+                        Perspective::IS_CLIENT, version());
+  EXPECT_FALSE(QuicSentPacketManagerPeer::UsingPacing(
+      static_cast<const QuicSentPacketManager*>(
+          &client.sent_packet_manager())));
+  EXPECT_FALSE(QuicSentPacketManagerPeer::UsingPacing(
+      static_cast<const QuicSentPacketManager*>(
+          &server.sent_packet_manager())));
+}
+
+TEST_P(QuicConnectionTest, WindowUpdateInstigateAcks) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  // Send a WINDOW_UPDATE frame.
+  QuicWindowUpdateFrame window_update;
+  window_update.stream_id = 3;
+  window_update.byte_offset = 1234;
+  EXPECT_CALL(visitor_, OnWindowUpdateFrame(_));
+  ProcessFramePacket(QuicFrame(&window_update));
+
+  // Ensure that this has caused the ACK alarm to be set.
+  QuicAlarm* ack_alarm = QuicConnectionPeer::GetAckAlarm(&connection_);
+  EXPECT_TRUE(ack_alarm->IsSet());
+}
+
+TEST_P(QuicConnectionTest, BlockedFrameInstigateAcks) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  // Send a BLOCKED frame.
+  QuicBlockedFrame blocked;
+  blocked.stream_id = 3;
+  EXPECT_CALL(visitor_, OnBlockedFrame(_));
+  ProcessFramePacket(QuicFrame(&blocked));
+
+  // Ensure that this has caused the ACK alarm to be set.
+  QuicAlarm* ack_alarm = QuicConnectionPeer::GetAckAlarm(&connection_);
+  EXPECT_TRUE(ack_alarm->IsSet());
+}
+
+TEST_P(QuicConnectionTest, ReevaluateTimeUntilSendOnAck) {
+  // Enable pacing.
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  connection_.SetFromConfig(config);
+
+  // Send two packets.  One packet is not sufficient because if it gets acked,
+  // there will be no packets in flight after that and the pacer will always
+  // allow the next packet in that situation.
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, CanSend(_)).WillRepeatedly(Return(true));
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "foo",
+      0, NO_FIN);
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "bar",
+      3, NO_FIN);
+  connection_.OnCanWrite();
+
+  // Schedule the next packet for a few milliseconds in future.
+  QuicSentPacketManagerPeer::DisablePacerBursts(manager_);
+  QuicTime scheduled_pacing_time =
+      clock_.Now() + QuicTime::Delta::FromMilliseconds(5);
+  QuicSentPacketManagerPeer::SetNextPacedPacketTime(manager_,
+                                                    scheduled_pacing_time);
+
+  // Send a packet and have it be blocked by congestion control.
+  EXPECT_CALL(*send_algorithm_, CanSend(_)).WillRepeatedly(Return(false));
+  connection_.SendStreamDataWithString(
+      GetNthClientInitiatedStreamId(1, connection_.transport_version()), "baz",
+      6, NO_FIN);
+  EXPECT_FALSE(connection_.GetSendAlarm()->IsSet());
+
+  // Process an ack and the send alarm will be set to the new 5ms delay.
+  QuicAckFrame ack = InitAckFrame(1);
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, CanSend(_)).WillRepeatedly(Return(true));
+  ProcessAckPacket(&ack);
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_EQ(1u, writer_->stream_frames().size());
+  EXPECT_TRUE(connection_.GetSendAlarm()->IsSet());
+  EXPECT_EQ(scheduled_pacing_time, connection_.GetSendAlarm()->deadline());
+  writer_->Reset();
+}
+
+TEST_P(QuicConnectionTest, SendAcksImmediately) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(1);
+  CongestionBlockWrites();
+  SendAckPacketToPeer();
+}
+
+TEST_P(QuicConnectionTest, SendPingImmediately) {
+  MockQuicConnectionDebugVisitor debug_visitor;
+  connection_.set_debug_visitor(&debug_visitor);
+
+  CongestionBlockWrites();
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  EXPECT_CALL(debug_visitor, OnPacketSent(_, _, _, _)).Times(1);
+  EXPECT_CALL(debug_visitor, OnPingSent()).Times(1);
+  connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+  EXPECT_FALSE(connection_.HasQueuedData());
+}
+
+TEST_P(QuicConnectionTest, SendBlockedImmediately) {
+  MockQuicConnectionDebugVisitor debug_visitor;
+  connection_.set_debug_visitor(&debug_visitor);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  EXPECT_CALL(debug_visitor, OnPacketSent(_, _, _, _)).Times(1);
+  EXPECT_EQ(0u, connection_.GetStats().blocked_frames_sent);
+  connection_.SendControlFrame(QuicFrame(new QuicBlockedFrame(1, 3)));
+  EXPECT_EQ(1u, connection_.GetStats().blocked_frames_sent);
+  EXPECT_FALSE(connection_.HasQueuedData());
+}
+
+TEST_P(QuicConnectionTest, SendingUnencryptedStreamDataFails) {
+  // EXPECT_QUIC_BUG tests are expensive so only run one instance of them.
+  if (!IsDefaultTestConfiguration()) {
+    return;
+  }
+
+  EXPECT_CALL(visitor_,
+              OnConnectionClosed(QUIC_ATTEMPT_TO_SEND_UNENCRYPTED_STREAM_DATA,
+                                 _, ConnectionCloseSource::FROM_SELF));
+  struct iovec iov;
+  MakeIOVector("", &iov);
+  EXPECT_QUIC_BUG(connection_.SaveAndSendStreamData(3, &iov, 1, 0, 0, FIN),
+                  "Cannot send stream data without encryption.");
+  EXPECT_FALSE(connection_.connected());
+}
+
+TEST_P(QuicConnectionTest, SetRetransmissionAlarmForCryptoPacket) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_FALSE(connection_.GetRetransmissionAlarm()->IsSet());
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendCryptoStreamData();
+
+  // Verify retransmission timer is correctly set after crypto packet has been
+  // sent.
+  EXPECT_TRUE(connection_.GetRetransmissionAlarm()->IsSet());
+  QuicTime retransmission_time =
+      QuicConnectionPeer::GetSentPacketManager(&connection_)
+          ->GetRetransmissionTime();
+  EXPECT_NE(retransmission_time, clock_.ApproximateNow());
+  EXPECT_EQ(retransmission_time,
+            connection_.GetRetransmissionAlarm()->deadline());
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.GetRetransmissionAlarm()->Fire();
+}
+
+TEST_P(QuicConnectionTest, PathDegradingAlarmForCryptoPacket) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendCryptoStreamData();
+
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  QuicTime::Delta delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+                              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+
+  // Fire the path degrading alarm, path degrading signal should be sent to
+  // the visitor.
+  EXPECT_CALL(visitor_, OnPathDegrading());
+  clock_.AdvanceTime(delay);
+  connection_.GetPathDegradingAlarm()->Fire();
+  EXPECT_TRUE(connection_.IsPathDegrading());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+}
+
+// Includes regression test for https://b.corp.google.com/issues/69979024.
+TEST_P(QuicConnectionTest, PathDegradingAlarmForNonCryptoPackets) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  for (int i = 0; i < 2; ++i) {
+    // Send a packet. Now there's a retransmittable packet on the wire, so the
+    // path degrading alarm should be set.
+    connection_.SendStreamDataWithString(
+        GetNthClientInitiatedStreamId(1, connection_.transport_version()), data,
+        offset, NO_FIN);
+    offset += data_size;
+    EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+    // Check the deadline of the path degrading alarm.
+    QuicTime::Delta delay =
+        QuicConnectionPeer::GetSentPacketManager(&connection_)
+            ->GetPathDegradingDelay();
+    EXPECT_EQ(clock_.ApproximateNow() + delay,
+              connection_.GetPathDegradingAlarm()->deadline());
+
+    // Send a second packet. The path degrading alarm's deadline should remain
+    // the same.
+    // Regression test for https://b.corp.google.com/issues/69979024.
+    clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+    QuicTime prev_deadline = connection_.GetPathDegradingAlarm()->deadline();
+    connection_.SendStreamDataWithString(
+        GetNthClientInitiatedStreamId(1, connection_.transport_version()), data,
+        offset, NO_FIN);
+    offset += data_size;
+    EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+    EXPECT_EQ(prev_deadline, connection_.GetPathDegradingAlarm()->deadline());
+
+    // Now receive an ACK of the first packet. This should advance the path
+    // degrading alarm's deadline since forward progress has been made.
+    clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+    if (i == 0) {
+      EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+    }
+    EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+    QuicAckFrame frame = InitAckFrame(
+        {{QuicPacketNumber(1u + 2u * i), QuicPacketNumber(2u + 2u * i)}});
+    ProcessAckPacket(&frame);
+    EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+    // Check the deadline of the path degrading alarm.
+    delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+                ->GetPathDegradingDelay();
+    EXPECT_EQ(clock_.ApproximateNow() + delay,
+              connection_.GetPathDegradingAlarm()->deadline());
+
+    if (i == 0) {
+      // Now receive an ACK of the second packet. Since there are no more
+      // retransmittable packets on the wire, this should cancel the path
+      // degrading alarm.
+      clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+      EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+      frame = InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)}});
+      ProcessAckPacket(&frame);
+      EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+    } else {
+      // Advance time to the path degrading alarm's deadline and simulate
+      // firing the alarm.
+      clock_.AdvanceTime(delay);
+      EXPECT_CALL(visitor_, OnPathDegrading());
+      connection_.GetPathDegradingAlarm()->Fire();
+      EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+    }
+  }
+  EXPECT_TRUE(connection_.IsPathDegrading());
+}
+
+TEST_P(QuicConnectionTest, RetransmittableOnWireSetsPingAlarm) {
+  const QuicTime::Delta retransmittable_on_wire_timeout =
+      QuicTime::Delta::FromMilliseconds(50);
+  connection_.set_retransmittable_on_wire_timeout(
+      retransmittable_on_wire_timeout);
+
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(true));
+
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  // Send a packet.
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  // Now there's a retransmittable packet on the wire, so the path degrading
+  // alarm should be set.
+  // The retransmittable-on-wire alarm should not be set.
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  QuicTime::Delta delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+                              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+  ASSERT_TRUE(connection_.sent_packet_manager().HasInFlightPackets());
+  // The ping alarm is set for the ping timeout, not the shorter
+  // retransmittable_on_wire_timeout.
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  QuicTime::Delta ping_delay = QuicTime::Delta::FromSeconds(kPingTimeoutSecs);
+  EXPECT_EQ((clock_.ApproximateNow() + ping_delay),
+            connection_.GetPingAlarm()->deadline());
+
+  // Now receive an ACK of the packet.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1), QuicPacketNumber(2)}});
+  ProcessAckPacket(&frame);
+  // No more retransmittable packets on the wire, so the path degrading alarm
+  // should be cancelled, and the ping alarm should be set to the
+  // retransmittable_on_wire_timeout.
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow() + retransmittable_on_wire_timeout,
+            connection_.GetPingAlarm()->deadline());
+
+  // Simulate firing the ping alarm and sending a PING.
+  clock_.AdvanceTime(retransmittable_on_wire_timeout);
+  EXPECT_CALL(visitor_, SendPing()).WillOnce(Invoke([this]() {
+    connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+  }));
+  connection_.GetPingAlarm()->Fire();
+
+  // Now there's a retransmittable packet (PING) on the wire, so the path
+  // degrading alarm should be set.
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+}
+
+// This test verifies that the connection marks path as degrading and does not
+// spin timer to detect path degrading when a new packet is sent on the
+// degraded path.
+TEST_P(QuicConnectionTest, NoPathDegradingAlarmIfPathIsDegrading) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  // Send the first packet. Now there's a retransmittable packet on the wire, so
+  // the path degrading alarm should be set.
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  // Check the deadline of the path degrading alarm.
+  QuicTime::Delta delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+                              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+
+  // Send a second packet. The path degrading alarm's deadline should remain
+  // the same.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  QuicTime prev_deadline = connection_.GetPathDegradingAlarm()->deadline();
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_EQ(prev_deadline, connection_.GetPathDegradingAlarm()->deadline());
+
+  // Now receive an ACK of the first packet. This should advance the path
+  // degrading alarm's deadline since forward progress has been made.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1u), QuicPacketNumber(2u)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  // Check the deadline of the path degrading alarm.
+  delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+
+  // Advance time to the path degrading alarm's deadline and simulate
+  // firing the path degrading alarm. This path will be considered as
+  // degrading.
+  clock_.AdvanceTime(delay);
+  EXPECT_CALL(visitor_, OnPathDegrading()).Times(1);
+  connection_.GetPathDegradingAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_TRUE(connection_.IsPathDegrading());
+
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  // Send a third packet. The path degrading alarm is no longer set but path
+  // should still be marked as degrading.
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_TRUE(connection_.IsPathDegrading());
+}
+
+// This test verifies that the connection unmarks path as degrarding and spins
+// the timer to detect future path degrading when forward progress is made
+// after path has been marked degrading.
+TEST_P(QuicConnectionTest, UnmarkPathDegradingOnForwardProgress) {
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  // Send the first packet. Now there's a retransmittable packet on the wire, so
+  // the path degrading alarm should be set.
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  // Check the deadline of the path degrading alarm.
+  QuicTime::Delta delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+                              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+
+  // Send a second packet. The path degrading alarm's deadline should remain
+  // the same.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  QuicTime prev_deadline = connection_.GetPathDegradingAlarm()->deadline();
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_EQ(prev_deadline, connection_.GetPathDegradingAlarm()->deadline());
+
+  // Now receive an ACK of the first packet. This should advance the path
+  // degrading alarm's deadline since forward progress has been made.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1u), QuicPacketNumber(2u)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+  // Check the deadline of the path degrading alarm.
+  delay = QuicConnectionPeer::GetSentPacketManager(&connection_)
+              ->GetPathDegradingDelay();
+  EXPECT_EQ(clock_.ApproximateNow() + delay,
+            connection_.GetPathDegradingAlarm()->deadline());
+
+  // Advance time to the path degrading alarm's deadline and simulate
+  // firing the alarm.
+  clock_.AdvanceTime(delay);
+  EXPECT_CALL(visitor_, OnPathDegrading()).Times(1);
+  connection_.GetPathDegradingAlarm()->Fire();
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_TRUE(connection_.IsPathDegrading());
+
+  // Send a third packet. The path degrading alarm is no longer set but path
+  // should still be marked as degrading.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+  EXPECT_TRUE(connection_.IsPathDegrading());
+
+  // Now receive an ACK of the second packet. This should unmark the path as
+  // degrading. And will set a timer to detect new path degrading.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  frame = InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)}});
+  ProcessAckPacket(&frame);
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  EXPECT_TRUE(connection_.GetPathDegradingAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, NoPathDegradingOnServer) {
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+
+  // Send data.
+  const char data[] = "data";
+  connection_.SendStreamDataWithString(1, data, 0, NO_FIN);
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+
+  // Ack data.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1u), QuicPacketNumber(2u)}});
+  ProcessAckPacket(&frame);
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, NoPathDegradingAfterSendingAck) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessDataPacket(1);
+  SendAckPacketToPeer();
+  EXPECT_FALSE(connection_.sent_packet_manager().unacked_packets().empty());
+  EXPECT_FALSE(connection_.sent_packet_manager().HasInFlightPackets());
+  EXPECT_FALSE(connection_.IsPathDegrading());
+  EXPECT_FALSE(connection_.GetPathDegradingAlarm()->IsSet());
+}
+
+TEST_P(QuicConnectionTest, MultipleCallsToCloseConnection) {
+  // Verifies that multiple calls to CloseConnection do not
+  // result in multiple attempts to close the connection - it will be marked as
+  // disconnected after the first call.
+  EXPECT_CALL(visitor_, OnConnectionClosed(_, _, _)).Times(1);
+  connection_.CloseConnection(QUIC_NO_ERROR, "no reason",
+                              ConnectionCloseBehavior::SILENT_CLOSE);
+  connection_.CloseConnection(QUIC_NO_ERROR, "no reason",
+                              ConnectionCloseBehavior::SILENT_CLOSE);
+}
+
+TEST_P(QuicConnectionTest, ServerReceivesChloOnNonCryptoStream) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  CryptoHandshakeMessage message;
+  CryptoFramer framer;
+  message.set_tag(kCHLO);
+  std::unique_ptr<QuicData> data(framer.ConstructHandshakeMessage(message));
+  frame1_.stream_id = 10;
+  frame1_.data_buffer = data->data();
+  frame1_.data_length = data->length();
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_MAYBE_CORRUPTED_MEMORY, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  ForceProcessFramePacket(QuicFrame(frame1_));
+}
+
+TEST_P(QuicConnectionTest, ClientReceivesRejOnNonCryptoStream) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  CryptoHandshakeMessage message;
+  CryptoFramer framer;
+  message.set_tag(kREJ);
+  std::unique_ptr<QuicData> data(framer.ConstructHandshakeMessage(message));
+  frame1_.stream_id = 10;
+  frame1_.data_buffer = data->data();
+  frame1_.data_length = data->length();
+
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_MAYBE_CORRUPTED_MEMORY, _,
+                                           ConnectionCloseSource::FROM_SELF));
+  ForceProcessFramePacket(QuicFrame(frame1_));
+}
+
+TEST_P(QuicConnectionTest, CloseConnectionOnPacketTooLarge) {
+  SimulateNextPacketTooLarge();
+  // A connection close packet is sent
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PACKET_WRITE_ERROR, _,
+                                           ConnectionCloseSource::FROM_SELF))
+      .Times(1);
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+}
+
+TEST_P(QuicConnectionTest, AlwaysGetPacketTooLarge) {
+  // Test even we always get packet too large, we do not infinitely try to send
+  // close packet.
+  AlwaysGetPacketTooLarge();
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PACKET_WRITE_ERROR, _,
+                                           ConnectionCloseSource::FROM_SELF))
+      .Times(1);
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+}
+
+// Verify that if connection has no outstanding data, it notifies the send
+// algorithm after the write.
+TEST_P(QuicConnectionTest, SendDataAndBecomeApplicationLimited) {
+  EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(1);
+  {
+    InSequence seq;
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite()).WillRepeatedly(Return(true));
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite())
+        .WillRepeatedly(Return(false));
+  }
+
+  connection_.SendStreamData3();
+}
+
+// Verify that the connection does not become app-limited if there is
+// outstanding data to send after the write.
+TEST_P(QuicConnectionTest, NotBecomeApplicationLimitedIfMoreDataAvailable) {
+  EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(0);
+  {
+    InSequence seq;
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite()).WillRepeatedly(Return(true));
+  }
+
+  connection_.SendStreamData3();
+}
+
+// Verify that the connection does not become app-limited after blocked write
+// even if there is outstanding data to send after the write.
+TEST_P(QuicConnectionTest, NotBecomeApplicationLimitedDueToWriteBlock) {
+  EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(0);
+  EXPECT_CALL(visitor_, WillingAndAbleToWrite()).WillRepeatedly(Return(true));
+  BlockOnNextWrite();
+
+  connection_.SendStreamData3();
+
+  // Now unblock the writer, become congestion control blocked,
+  // and ensure we become app-limited after writing.
+  writer_->SetWritable();
+  CongestionBlockWrites();
+  EXPECT_CALL(visitor_, WillingAndAbleToWrite()).WillRepeatedly(Return(false));
+  {
+    InSequence seq;
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(1);
+  }
+  connection_.OnCanWrite();
+}
+
+// Test the mode in which the link is filled up with probing retransmissions if
+// the connection becomes application-limited.
+TEST_P(QuicConnectionTest, SendDataWhenApplicationLimited) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, ShouldSendProbingPacket())
+      .WillRepeatedly(Return(true));
+  {
+    InSequence seq;
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite()).WillRepeatedly(Return(true));
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _));
+    EXPECT_CALL(visitor_, WillingAndAbleToWrite())
+        .WillRepeatedly(Return(false));
+  }
+  // Fix congestion window to be 20,000 bytes.
+  EXPECT_CALL(*send_algorithm_, CanSend(Ge(20000u)))
+      .WillRepeatedly(Return(false));
+  EXPECT_CALL(*send_algorithm_, CanSend(Lt(20000u)))
+      .WillRepeatedly(Return(true));
+
+  EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(0);
+  ASSERT_EQ(0u, connection_.GetStats().packets_sent);
+  connection_.set_fill_up_link_during_probing(true);
+  connection_.OnHandshakeComplete();
+  connection_.SendStreamData3();
+
+  // We expect a lot of packets from a 20 kbyte window.
+  EXPECT_GT(connection_.GetStats().packets_sent, 10u);
+  // Ensure that the packets are padded.
+  QuicByteCount average_packet_size =
+      connection_.GetStats().bytes_sent / connection_.GetStats().packets_sent;
+  EXPECT_GT(average_packet_size, 1000u);
+
+  // Acknowledge all packets sent, except for the last one.
+  QuicAckFrame ack = InitAckFrame(
+      connection_.sent_packet_manager().GetLargestSentPacket() - 1);
+  EXPECT_CALL(*loss_algorithm_, DetectLosses(_, _, _, _, _, _));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+
+  // Ensure that since we no longer have retransmittable bytes in flight, this
+  // will not cause any responses to be sent.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  EXPECT_CALL(*send_algorithm_, OnApplicationLimited(_)).Times(1);
+  ProcessAckPacket(&ack);
+}
+
+TEST_P(QuicConnectionTest, DonotForceSendingAckOnPacketTooLarge) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  // Send an ack by simulating delayed ack alarm firing.
+  ProcessPacket(1);
+  QuicAlarm* ack_alarm = QuicConnectionPeer::GetAckAlarm(&connection_);
+  EXPECT_TRUE(ack_alarm->IsSet());
+  connection_.GetAckAlarm()->Fire();
+  // Simulate data packet causes write error.
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PACKET_WRITE_ERROR, _, _));
+  SimulateNextPacketTooLarge();
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+  EXPECT_EQ(1u, writer_->frame_count());
+  EXPECT_FALSE(writer_->connection_close_frames().empty());
+  // Ack frame is not bundled in connection close packet.
+  EXPECT_TRUE(writer_->ack_frames().empty());
+}
+
+TEST_P(QuicConnectionTest, CloseConnectionForStatelessReject) {
+  QuicString error_details("stateless reject");
+  EXPECT_CALL(visitor_, OnConnectionClosed(
+                            QUIC_CRYPTO_HANDSHAKE_STATELESS_REJECT,
+                            error_details, ConnectionCloseSource::FROM_PEER));
+  connection_.set_perspective(Perspective::IS_CLIENT);
+  connection_.CloseConnection(QUIC_CRYPTO_HANDSHAKE_STATELESS_REJECT,
+                              error_details,
+                              ConnectionCloseBehavior::SILENT_CLOSE);
+}
+
+// Regression test for b/63620844.
+TEST_P(QuicConnectionTest, FailedToWriteHandshakePacket) {
+  SimulateNextPacketTooLarge();
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_PACKET_WRITE_ERROR, _,
+                                           ConnectionCloseSource::FROM_SELF))
+      .Times(1);
+  connection_.SendCryptoStreamData();
+}
+
+TEST_P(QuicConnectionTest, MaxPacingRate) {
+  EXPECT_EQ(0, connection_.MaxPacingRate().ToBytesPerSecond());
+  connection_.SetMaxPacingRate(QuicBandwidth::FromBytesPerSecond(100));
+  EXPECT_EQ(100, connection_.MaxPacingRate().ToBytesPerSecond());
+}
+
+TEST_P(QuicConnectionTest, ClientAlwaysSendConnectionId) {
+  EXPECT_EQ(Perspective::IS_CLIENT, connection_.perspective());
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendStreamDataWithString(3, "foo", 0, NO_FIN);
+  EXPECT_EQ(CONNECTION_ID_PRESENT,
+            writer_->last_packet_header().destination_connection_id_included);
+
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  QuicConfig config;
+  QuicConfigPeer::SetReceivedBytesForConnectionId(&config, 0);
+  connection_.SetFromConfig(config);
+
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendStreamDataWithString(3, "bar", 3, NO_FIN);
+  // Verify connection id is still sent in the packet.
+  EXPECT_EQ(CONNECTION_ID_PRESENT,
+            writer_->last_packet_header().destination_connection_id_included);
+}
+
+TEST_P(QuicConnectionTest, SendProbingRetransmissions) {
+  MockQuicConnectionDebugVisitor debug_visitor;
+  connection_.set_debug_visitor(&debug_visitor);
+
+  const QuicStreamId stream_id = 2;
+  QuicPacketNumber last_packet;
+  SendStreamDataToPeer(stream_id, "foo", 0, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "bar", 3, NO_FIN, &last_packet);
+  SendStreamDataToPeer(stream_id, "test", 6, NO_FIN, &last_packet);
+
+  const QuicByteCount old_bytes_in_flight =
+      connection_.sent_packet_manager().GetBytesInFlight();
+
+  // Allow 9 probing retransmissions to be sent.
+  {
+    InSequence seq;
+    EXPECT_CALL(*send_algorithm_, CanSend(_))
+        .Times(9 * 2)
+        .WillRepeatedly(Return(true));
+    EXPECT_CALL(*send_algorithm_, CanSend(_)).WillOnce(Return(false));
+  }
+  // Expect them retransmitted in cyclic order (foo, bar, test, foo, bar...).
+  QuicPacketCount sent_count = 0;
+  EXPECT_CALL(debug_visitor, OnPacketSent(_, _, _, _))
+      .WillRepeatedly(Invoke([this, &sent_count](const SerializedPacket&,
+                                                 QuicPacketNumber,
+                                                 TransmissionType, QuicTime) {
+        ASSERT_EQ(1u, writer_->stream_frames().size());
+        // Identify the frames by stream offset (0, 3, 6, 0, 3...).
+        EXPECT_EQ(3 * (sent_count % 3), writer_->stream_frames()[0]->offset);
+        sent_count++;
+      }));
+  EXPECT_CALL(*send_algorithm_, ShouldSendProbingPacket())
+      .WillRepeatedly(Return(true));
+
+  connection_.SendProbingRetransmissions();
+
+  // Ensure that the in-flight has increased.
+  const QuicByteCount new_bytes_in_flight =
+      connection_.sent_packet_manager().GetBytesInFlight();
+  EXPECT_GT(new_bytes_in_flight, old_bytes_in_flight);
+}
+
+// Ensure that SendProbingRetransmissions() does not retransmit anything when
+// there are no outstanding packets.
+TEST_P(QuicConnectionTest,
+       SendProbingRetransmissionsFailsWhenNothingToRetransmit) {
+  ASSERT_TRUE(connection_.sent_packet_manager().unacked_packets().empty());
+
+  MockQuicConnectionDebugVisitor debug_visitor;
+  connection_.set_debug_visitor(&debug_visitor);
+  EXPECT_CALL(debug_visitor, OnPacketSent(_, _, _, _)).Times(0);
+  EXPECT_CALL(*send_algorithm_, ShouldSendProbingPacket())
+      .WillRepeatedly(Return(true));
+
+  connection_.SendProbingRetransmissions();
+}
+
+TEST_P(QuicConnectionTest, PingAfterLastRetransmittablePacketAcked) {
+  const QuicTime::Delta retransmittable_on_wire_timeout =
+      QuicTime::Delta::FromMilliseconds(50);
+  connection_.set_retransmittable_on_wire_timeout(
+      retransmittable_on_wire_timeout);
+
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(true));
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  // Advance 5ms, send a retransmittable packet to the peer.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.sent_packet_manager().HasInFlightPackets());
+  // The ping alarm is set for the ping timeout, not the shorter
+  // retransmittable_on_wire_timeout.
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  QuicTime::Delta ping_delay = QuicTime::Delta::FromSeconds(kPingTimeoutSecs);
+  EXPECT_EQ((clock_.ApproximateNow() + ping_delay),
+            connection_.GetPingAlarm()->deadline());
+
+  // Advance 5ms, send a second retransmittable packet to the peer.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+
+  // Now receive an ACK of the first packet. This should not set the
+  // retransmittable-on-wire alarm since packet 2 is still on the wire.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1), QuicPacketNumber(2)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.sent_packet_manager().HasInFlightPackets());
+  // The ping alarm is set for the ping timeout, not the shorter
+  // retransmittable_on_wire_timeout.
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  // The ping alarm has a 1 second granularity, and the clock has been advanced
+  // 10ms since it was originally set.
+  EXPECT_EQ((clock_.ApproximateNow() + ping_delay -
+             QuicTime::Delta::FromMilliseconds(10)),
+            connection_.GetPingAlarm()->deadline());
+
+  // Now receive an ACK of the second packet. This should set the
+  // retransmittable-on-wire alarm now that no retransmittable packets are on
+  // the wire.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  frame = InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow() + retransmittable_on_wire_timeout,
+            connection_.GetPingAlarm()->deadline());
+
+  // Now receive a duplicate ACK of the second packet. This should not update
+  // the ping alarm.
+  QuicTime prev_deadline = connection_.GetPingAlarm()->deadline();
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  frame = InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(prev_deadline, connection_.GetPingAlarm()->deadline());
+
+  // Now receive a non-ACK packet.  This should not update the ping alarm.
+  prev_deadline = connection_.GetPingAlarm()->deadline();
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  ProcessPacket(4);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(prev_deadline, connection_.GetPingAlarm()->deadline());
+
+  // Simulate the alarm firing and check that a PING is sent.
+  EXPECT_CALL(visitor_, SendPing()).WillOnce(Invoke([this]() {
+    connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+  }));
+  connection_.GetPingAlarm()->Fire();
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(2u, writer_->frame_count());
+  } else {
+    EXPECT_EQ(3u, writer_->frame_count());
+  }
+  ASSERT_EQ(1u, writer_->ping_frames().size());
+}
+
+TEST_P(QuicConnectionTest, NoPingIfRetransmittablePacketSent) {
+  const QuicTime::Delta retransmittable_on_wire_timeout =
+      QuicTime::Delta::FromMilliseconds(50);
+  connection_.set_retransmittable_on_wire_timeout(
+      retransmittable_on_wire_timeout);
+
+  EXPECT_TRUE(connection_.connected());
+  EXPECT_CALL(visitor_, ShouldKeepConnectionAlive())
+      .WillRepeatedly(Return(true));
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  // Advance 5ms, send a retransmittable packet to the peer.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_FALSE(connection_.GetPingAlarm()->IsSet());
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.sent_packet_manager().HasInFlightPackets());
+  // The ping alarm is set for the ping timeout, not the shorter
+  // retransmittable_on_wire_timeout.
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  QuicTime::Delta ping_delay = QuicTime::Delta::FromSeconds(kPingTimeoutSecs);
+  EXPECT_EQ((clock_.ApproximateNow() + ping_delay),
+            connection_.GetPingAlarm()->deadline());
+
+  // Now receive an ACK of the first packet. This should set the
+  // retransmittable-on-wire alarm now that no retransmittable packets are on
+  // the wire.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1), QuicPacketNumber(2)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow() + retransmittable_on_wire_timeout,
+            connection_.GetPingAlarm()->deadline());
+
+  // Before the alarm fires, send another retransmittable packet. This should
+  // cancel the retransmittable-on-wire alarm since now there's a
+  // retransmittable packet on the wire.
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+
+  // Now receive an ACK of the second packet. This should set the
+  // retransmittable-on-wire alarm now that no retransmittable packets are on
+  // the wire.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  frame = InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)}});
+  ProcessAckPacket(&frame);
+  EXPECT_TRUE(connection_.GetPingAlarm()->IsSet());
+  EXPECT_EQ(clock_.ApproximateNow() + retransmittable_on_wire_timeout,
+            connection_.GetPingAlarm()->deadline());
+
+  // Simulate the alarm firing and check that a PING is sent.
+  writer_->Reset();
+  EXPECT_CALL(visitor_, SendPing()).WillOnce(Invoke([this]() {
+    connection_.SendControlFrame(QuicFrame(QuicPingFrame(1)));
+  }));
+  connection_.GetPingAlarm()->Fire();
+  if (GetParam().no_stop_waiting) {
+    EXPECT_EQ(2u, writer_->frame_count());
+  } else {
+    EXPECT_EQ(3u, writer_->frame_count());
+  }
+  ASSERT_EQ(1u, writer_->ping_frames().size());
+}
+
+TEST_P(QuicConnectionTest, OnForwardProgressConfirmed) {
+  EXPECT_CALL(visitor_, OnForwardProgressConfirmed()).Times(Exactly(0));
+  EXPECT_TRUE(connection_.connected());
+
+  const char data[] = "data";
+  size_t data_size = strlen(data);
+  QuicStreamOffset offset = 0;
+
+  // Send two packets.
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+  connection_.SendStreamDataWithString(1, data, offset, NO_FIN);
+  offset += data_size;
+
+  // Ack packet 1. This increases the largest_acked to 1, so
+  // OnForwardProgressConfirmed() should be called
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(visitor_, OnForwardProgressConfirmed());
+  QuicAckFrame frame =
+      InitAckFrame({{QuicPacketNumber(1), QuicPacketNumber(2)}});
+  ProcessAckPacket(&frame);
+
+  // Ack packet 1 again. largest_acked remains at 1, so
+  // OnForwardProgressConfirmed() should not be called.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  frame = InitAckFrame({{QuicPacketNumber(1), QuicPacketNumber(2)}});
+  ProcessAckPacket(&frame);
+
+  // Ack packet 2. This increases the largest_acked to 2, so
+  // OnForwardProgressConfirmed() should be called.
+  clock_.AdvanceTime(QuicTime::Delta::FromMilliseconds(5));
+  EXPECT_CALL(*send_algorithm_, OnCongestionEvent(true, _, _, _, _));
+  EXPECT_CALL(visitor_, OnForwardProgressConfirmed());
+  frame = InitAckFrame({{QuicPacketNumber(2), QuicPacketNumber(3)}});
+  ProcessAckPacket(&frame);
+}
+
+TEST_P(QuicConnectionTest, ValidStatelessResetToken) {
+  const QuicUint128 kTestToken = 1010101;
+  const QuicUint128 kWrongTestToken = 1010100;
+  QuicConfig config;
+  // No token has been received.
+  EXPECT_FALSE(connection_.IsValidStatelessResetToken(kTestToken));
+
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _)).Times(2);
+  // Token is different from received token.
+  QuicConfigPeer::SetReceivedStatelessResetToken(&config, kTestToken);
+  connection_.SetFromConfig(config);
+  EXPECT_FALSE(connection_.IsValidStatelessResetToken(kWrongTestToken));
+
+  QuicConfigPeer::SetReceivedStatelessResetToken(&config, kTestToken);
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(connection_.IsValidStatelessResetToken(kTestToken));
+}
+
+TEST_P(QuicConnectionTest, WriteBlockedWithInvalidAck) {
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnConnectionClosed(QUIC_INVALID_ACK_DATA, _, _));
+
+  BlockOnNextWrite();
+  connection_.SendStreamDataWithString(5, "foo", 0, FIN);
+  // This causes connection to be closed because packet 1 has not been sent yet.
+  QuicAckFrame frame = InitAckFrame(1);
+  ProcessAckPacket(1, &frame);
+}
+
+TEST_P(QuicConnectionTest, SendMessage) {
+  if (connection_.transport_version() <= QUIC_VERSION_44) {
+    return;
+  }
+  QuicString message(connection_.GetLargestMessagePayload() * 2, 'a');
+  QuicStringPiece message_data(message);
+  QuicMemSliceStorage storage(nullptr, 0, nullptr, 0);
+  {
+    QuicConnection::ScopedPacketFlusher flusher(&connection_,
+                                                QuicConnection::SEND_ACK);
+    connection_.SendStreamData3();
+    // Send a message which cannot fit into current open packet, and 2 packets
+    // get sent, one contains stream frame, and the other only contains the
+    // message frame.
+    EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(2);
+    EXPECT_EQ(
+        MESSAGE_STATUS_SUCCESS,
+        connection_.SendMessage(
+            1, MakeSpan(connection_.helper()->GetStreamSendBufferAllocator(),
+                        QuicStringPiece(message_data.data(),
+                                        connection_.GetLargestMessagePayload()),
+                        &storage)));
+  }
+  // Fail to send a message if connection is congestion control blocked.
+  EXPECT_CALL(*send_algorithm_, CanSend(_)).WillOnce(Return(false));
+  EXPECT_EQ(
+      MESSAGE_STATUS_BLOCKED,
+      connection_.SendMessage(
+          2, MakeSpan(connection_.helper()->GetStreamSendBufferAllocator(),
+                      "message", &storage)));
+
+  // Always fail to send a message which cannot fit into one packet.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  EXPECT_EQ(
+      MESSAGE_STATUS_TOO_LARGE,
+      connection_.SendMessage(
+          3,
+          MakeSpan(connection_.helper()->GetStreamSendBufferAllocator(),
+                   QuicStringPiece(message_data.data(),
+                                   connection_.GetLargestMessagePayload() + 1),
+                   &storage)));
+}
+
+// Test to check that the path challenge/path response logic works
+// correctly. This test is only for version-99
+TEST_P(QuicConnectionTest, PathChallengeResponse) {
+  if (connection_.version().transport_version != QUIC_VERSION_99) {
+    return;
+  }
+  // First check if we can probe from server to client and back
+  set_perspective(Perspective::IS_SERVER);
+  QuicPacketCreatorPeer::SetSendVersionInPacket(creator_, false);
+
+  // Create and send the probe request (PATH_CHALLENGE frame).
+  // SendConnectivityProbingPacket ends up calling
+  // TestPacketWriter::WritePacket() which in turns receives and parses the
+  // packet by calling framer_.ProcessPacket() -- which in turn calls
+  // SimpleQuicFramer::OnPathChallengeFrame(). SimpleQuicFramer saves
+  // the packet in writer_->path_challenge_frames()
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendConnectivityProbingPacket(writer_.get(),
+                                            connection_.peer_address());
+  // Save the random contents of the challenge for later comparison to the
+  // response.
+  QuicPathFrameBuffer challenge_data =
+      writer_->path_challenge_frames().front().data_buffer;
+
+  // Normally, QuicConnection::OnPathChallengeFrame and OnPaddingFrame would be
+  // called and it will perform actions to ensure that the rest of the protocol
+  // is performed (specifically, call UpdatePacketContent to say that this is a
+  // path challenge so that when QuicConnection::OnPacketComplete is called
+  // (again, out of the framer), the response is generated).  Simulate those
+  // calls so that the right internal state is set up for generating
+  // the response.
+  EXPECT_TRUE(connection_.OnPathChallengeFrame(
+      writer_->path_challenge_frames().front()));
+  EXPECT_TRUE(connection_.OnPaddingFrame(writer_->padding_frames().front()));
+  // Cause the response to be created and sent. Result is that the response
+  // should be stashed in writer's path_response_frames.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  connection_.SendConnectivityProbingResponsePacket(connection_.peer_address());
+
+  // The final check is to ensure that the random data in the response matches
+  // the random data from the challenge.
+  EXPECT_EQ(0, memcmp(&challenge_data,
+                      &(writer_->path_response_frames().front().data_buffer),
+                      sizeof(challenge_data)));
+}
+
+// Regression test for b/110259444
+TEST_P(QuicConnectionTest, DoNotScheduleSpuriousAckAlarm) {
+  SetQuicReloadableFlag(quic_fix_spurious_ack_alarm, true);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnWriteBlocked()).Times(AtLeast(1));
+  writer_->SetWriteBlocked();
+
+  ProcessPacket(1);
+  QuicAlarm* ack_alarm = QuicConnectionPeer::GetAckAlarm(&connection_);
+  // Verify ack alarm is set.
+  EXPECT_TRUE(ack_alarm->IsSet());
+  // Fire the ack alarm, verify no packet is sent because the writer is blocked.
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(0);
+  connection_.GetAckAlarm()->Fire();
+
+  writer_->SetWritable();
+  EXPECT_CALL(*send_algorithm_, OnPacketSent(_, _, _, _, _)).Times(1);
+  ProcessPacket(2);
+  // Verify ack alarm is not set.
+  EXPECT_FALSE(ack_alarm->IsSet());
+}
+
+TEST_P(QuicConnectionTest, DisablePacingOffloadConnectionOptions) {
+  EXPECT_FALSE(QuicConnectionPeer::SupportsReleaseTime(&connection_));
+  writer_->set_supports_release_time(true);
+  QuicConfig config;
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  connection_.SetFromConfig(config);
+  EXPECT_TRUE(QuicConnectionPeer::SupportsReleaseTime(&connection_));
+
+  QuicTagVector connection_options;
+  connection_options.push_back(kNPCO);
+  config.SetConnectionOptionsToSend(connection_options);
+  EXPECT_CALL(*send_algorithm_, SetFromConfig(_, _));
+  connection_.SetFromConfig(config);
+  // Verify pacing offload is disabled.
+  EXPECT_FALSE(QuicConnectionPeer::SupportsReleaseTime(&connection_));
+}
+
+// Regression test for b/110259444
+// Get a path response without having issued a path challenge...
+TEST_P(QuicConnectionTest, OrphanPathResponse) {
+  QuicPathFrameBuffer data = {{0, 1, 2, 3, 4, 5, 6, 7}};
+
+  QuicPathResponseFrame frame(99, data);
+  EXPECT_TRUE(connection_.OnPathResponseFrame(frame));
+  // If PATH_RESPONSE was accepted (payload matches the payload saved
+  // in QuicConnection::transmitted_connectivity_probe_payload_) then
+  // current_packet_content_ would be set to FIRST_FRAME_IS_PING.
+  // Since this PATH_RESPONSE does not match, current_packet_content_
+  // must not be FIRST_FRAME_IS_PING.
+  EXPECT_NE(QuicConnection::FIRST_FRAME_IS_PING,
+            QuicConnectionPeer::GetCurrentPacketContent(&connection_));
+}
+
+// Regression test for b/120791670
+TEST_P(QuicConnectionTest, StopProcessingGQuicPacketInIetfQuicConnection) {
+  // This test mimics a problematic scenario where an IETF QUIC connection
+  // receives a Google QUIC packet and continue processing it using Google QUIC
+  // wire format.
+  if (version().transport_version <= QUIC_VERSION_43) {
+    return;
+  }
+  set_perspective(Perspective::IS_SERVER);
+  QuicStreamFrame stream_frame(
+      QuicUtils::GetCryptoStreamId(connection_.transport_version()), false, 0u,
+      QuicStringPiece());
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(1);
+  ProcessFramePacketWithAddresses(QuicFrame(stream_frame), kSelfAddress,
+                                  kPeerAddress);
+
+  // Let connection process a Google QUIC packet.
+  peer_framer_.set_version_for_tests(
+      ParsedQuicVersion(PROTOCOL_QUIC_CRYPTO, QUIC_VERSION_43));
+  std::unique_ptr<QuicPacket> packet(ConstructDataPacket(2, !kHasStopWaiting));
+  char buffer[kMaxPacketSize];
+  size_t encrypted_length = peer_framer_.EncryptPayload(
+      ENCRYPTION_NONE, QuicPacketNumber(2), *packet, buffer, kMaxPacketSize);
+  // Make sure no stream frame is processed.
+  EXPECT_CALL(visitor_, OnStreamFrame(_)).Times(0);
+  connection_.ProcessUdpPacket(
+      kSelfAddress, kPeerAddress,
+      QuicReceivedPacket(buffer, encrypted_length, clock_.Now(), false));
+
+  EXPECT_EQ(2u, connection_.GetStats().packets_received);
+  EXPECT_EQ(1u, connection_.GetStats().packets_processed);
+}
+
+TEST_P(QuicConnectionTest, AcceptPacketNumberZero) {
+  if (!GetQuicRestartFlag(quic_uint64max_uninitialized_pn) ||
+      version().transport_version != QUIC_VERSION_99) {
+    return;
+  }
+  // Set first_sending_packet_number to be 0 to allow successfully processing
+  // acks which ack packet number 0.
+  QuicFramerPeer::SetFirstSendingPacketNumber(writer_->framer()->framer(), 0);
+  EXPECT_CALL(visitor_, OnSuccessfulVersionNegotiation(_));
+
+  ProcessPacket(0);
+  EXPECT_EQ(QuicPacketNumber(0), LargestAcked(*outgoing_ack()));
+  EXPECT_EQ(1u, outgoing_ack()->packets.NumIntervals());
+
+  ProcessPacket(1);
+  EXPECT_EQ(QuicPacketNumber(1), LargestAcked(*outgoing_ack()));
+  EXPECT_EQ(1u, outgoing_ack()->packets.NumIntervals());
+
+  ProcessPacket(2);
+  EXPECT_EQ(QuicPacketNumber(2), LargestAcked(*outgoing_ack()));
+  EXPECT_EQ(1u, outgoing_ack()->packets.NumIntervals());
+}
+
+}  // namespace
+}  // namespace test
+}  // namespace quic