diff --git a/quic/api/QuicBatchWriter.cpp b/quic/api/QuicBatchWriter.cpp
index 189434ac4..abb2c4732 100644
--- a/quic/api/QuicBatchWriter.cpp
+++ b/quic/api/QuicBatchWriter.cpp
@@ -227,10 +227,15 @@ void GSOInplacePacketBatchWriter::reset() {
   lastPacketEnd_ = nullptr;
   prevSize_ = 0;
   numPackets_ = 0;
+  nextPacketSize_ = 0;
 }
 
 bool GSOInplacePacketBatchWriter::needsFlush(size_t size) {
-  return prevSize_ && size > prevSize_;
+  auto shouldFlush = prevSize_ && size > prevSize_;
+  if (shouldFlush) {
+    nextPacketSize_ = size;
+  }
+  return shouldFlush;
 }
 
 bool GSOInplacePacketBatchWriter::append(
@@ -274,11 +279,13 @@ ssize_t GSOInplacePacketBatchWriter::write(
   CHECK(lastPacketEnd_ >= buf->data() && lastPacketEnd_ <= buf->tail())
       << "lastPacketEnd_=" << (long)lastPacketEnd_
       << " data=" << (long)buf->data() << " tail=" << (long)buf->tail();
-  auto diffToEnd = buf->tail() - lastPacketEnd_;
+  uint64_t diffToEnd = buf->tail() - lastPacketEnd_;
   CHECK(
-      diffToEnd >= 0 &&
-      static_cast<uint64_t>(diffToEnd) <= conn_.udpSendPacketLen);
-  auto diffToStart = lastPacketEnd_ - buf->data();
+      diffToEnd <= conn_.udpSendPacketLen ||
+      (nextPacketSize_ && diffToEnd == nextPacketSize_))
+      << "diffToEnd=" << diffToEnd << ", pktLimit=" << conn_.udpSendPacketLen
+      << ", nextPacketSize_=" << nextPacketSize_;
+  uint64_t diffToStart = lastPacketEnd_ - buf->data();
   buf->trimEnd(diffToEnd);
   auto bytesWritten = (numPackets_ > 1)
       ? sock.writeGSO(address, buf, static_cast<int>(prevSize_))
@@ -296,8 +303,15 @@ ssize_t GSOInplacePacketBatchWriter::write(
     buf->trimStart(diffToStart);
     buf->append(diffToEnd);
     buf->retreat(diffToStart);
-    CHECK(buf->length() <= conn_.udpSendPacketLen);
-    CHECK(0 == buf->headroom());
+    auto bufLength = buf->length();
+    CHECK_EQ(diffToEnd, bufLength)
+        << "diffToEnd=" << diffToEnd << ", bufLength=" << bufLength;
+    CHECK(
+        bufLength <= conn_.udpSendPacketLen ||
+        (nextPacketSize_ && bufLength == nextPacketSize_))
+        << "bufLength=" << bufLength << ", pktLimit=" << conn_.udpSendPacketLen
+        << ", nextPacketSize_=" << nextPacketSize_;
+    CHECK(0 == buf->headroom()) << "headroom=" << buf->headroom();
   } else {
     buf->clear();
   }
diff --git a/quic/api/QuicBatchWriter.h b/quic/api/QuicBatchWriter.h
index 47f54ec7e..bdc53addc 100644
--- a/quic/api/QuicBatchWriter.h
+++ b/quic/api/QuicBatchWriter.h
@@ -175,6 +175,15 @@ class GSOInplacePacketBatchWriter : public BatchWriter {
   const uint8_t* lastPacketEnd_{nullptr};
   size_t prevSize_{0};
   size_t numPackets_{0};
+
+  /**
+   * If we flush the batch due to the next packet being larger than current GSO
+   * size, we use the following value to keep track of that next packet, and
+   * checks against buffer residue after writes. The reason we cannot just check
+   * the buffer residue against the Quic packet limit is that there may be some
+   * retranmission packets slightly larger than the limit.
+   */
+  size_t nextPacketSize_{0};
 };
 
 class SendmmsgPacketBatchWriter : public BatchWriter {
diff --git a/quic/api/test/QuicBatchWriterTest.cpp b/quic/api/test/QuicBatchWriterTest.cpp
index ad2fe11c9..7acee2660 100644
--- a/quic/api/test/QuicBatchWriterTest.cpp
+++ b/quic/api/test/QuicBatchWriterTest.cpp
@@ -583,6 +583,50 @@ TEST_P(QuicBatchWriterTest, InplaceWriterLastOneTooBig) {
   EXPECT_EQ(0, buf->headroom());
 }
 
+TEST_P(QuicBatchWriterTest, InplaceWriterBufResidueCheck) {
+  bool useThreadLocal = GetParam();
+  folly::EventBase evb;
+  folly::test::MockAsyncUDPSocket sock(&evb);
+  EXPECT_CALL(sock, getGSO()).WillRepeatedly(Return(1));
+
+  uint32_t batchSize = 20;
+  auto bufAccessor =
+      std::make_unique<SimpleBufAccessor>(conn_.udpSendPacketLen * batchSize);
+  conn_.bufAccessor = bufAccessor.get();
+  conn_.udpSendPacketLen = 1000;
+  auto batchWriter = quic::BatchWriterFactory::makeBatchWriter(
+      sock,
+      quic::QuicBatchingMode::BATCHING_MODE_GSO,
+      batchSize,
+      useThreadLocal,
+      quic::kDefaultThreadLocalDelay,
+      DataPathType::ContinuousMemory,
+      conn_);
+  auto buf = bufAccessor->obtain();
+  folly::IOBuf* rawBuf = buf.get();
+  bufAccessor->release(std::move(buf));
+  rawBuf->append(700);
+  ASSERT_FALSE(
+      batchWriter->append(nullptr, 700, folly::SocketAddress(), nullptr));
+
+  size_t packetSizeTooBig = 1200;
+  rawBuf->append(packetSizeTooBig);
+  EXPECT_TRUE(batchWriter->needsFlush(packetSizeTooBig));
+
+  EXPECT_CALL(sock, write(_, _))
+      .Times(1)
+      .WillOnce(Invoke([&](const auto& /* addr */,
+                           const std::unique_ptr<folly::IOBuf>& buf) {
+        EXPECT_EQ(700, buf->length());
+        return 700;
+      }));
+  // No crash:
+  EXPECT_EQ(700, batchWriter->write(sock, folly::SocketAddress()));
+
+  EXPECT_EQ(1200, rawBuf->length());
+  EXPECT_EQ(0, rawBuf->headroom());
+}
+
 INSTANTIATE_TEST_CASE_P(
     QuicBatchWriterTest,
     QuicBatchWriterTest,