Kernel/Net: Implement TCP_NODELAY

Author: https://github.com/romch007
Commit: https://github.com/SerenityOS/serenity/commit/61ac554a34
Pull-request: https://github.com/SerenityOS/serenity/pull/21771

Kernel/API/POSIX/netinet/tcp.h

@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2023, Romain Chardiny <romain.chardiny@gmail.com>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+#pragma once
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define TCP_NODELAY 10
+#define TCP_MAXSEG 11
+
+#ifdef __cplusplus
+}
+#endif

Kernel/Net/TCPSocket.cpp

@@ -211,16 +211,17 @@ ErrorOr<size_t> TCPSocket::protocol_send(UserOrKernelBuffer const& data, size_t
         return set_so_error(EHOSTUNREACH);
     size_t mss = routing_decision.adapter->mtu() - sizeof(IPv4Packet) - sizeof(TCPPacket);
 
-    // RFC 896 (Nagle’s algorithm): https://www.ietf.org/rfc/rfc0896
-    // "The solution is to inhibit the sending of new TCP  segments when
-    //  new  outgoing  data  arrives  from  the  user  if  any previously
-    //  transmitted data on the connection remains unacknowledged.   This
-    //  inhibition  is  to be unconditional; no timers, tests for size of
-    //  data received, or other conditions are required."
-    // FIXME: Make this configurable via TCP_NODELAY.
-    auto has_unacked_data = m_unacked_packets.with_shared([&](auto const& packets) { return packets.size > 0; });
-    if (has_unacked_data && data_length < mss)
-        return set_so_error(EAGAIN);
+    if (!m_no_delay) {
+        // RFC 896 (Nagle’s algorithm): https://www.ietf.org/rfc/rfc0896
+        // "The solution is to inhibit the sending of new TCP  segments when
+        //  new  outgoing  data  arrives  from  the  user  if  any previously
+        //  transmitted data on the connection remains unacknowledged.   This
+        //  inhibition  is  to be unconditional; no timers, tests for size of
+        //  data received, or other conditions are required."
+        auto has_unacked_data = m_unacked_packets.with_shared([&](auto const& packets) { return packets.size > 0; });
+        if (has_unacked_data && data_length < mss)
+            return set_so_error(EAGAIN);
+    }
 
     data_length = min(data_length, mss);
     TRY(send_tcp_packet(TCPFlags::PSH | TCPFlags::ACK, &data, data_length, &routing_decision));
@@ -427,6 +428,54 @@ NetworkOrdered<u16> TCPSocket::compute_tcp_checksum(IPv4Address const& source, I
     return ~(checksum & 0xffff);
 }
 
+ErrorOr<void> TCPSocket::setsockopt(int level, int option, Userspace<void const*> user_value, socklen_t user_value_size)
+{
+    if (level != IPPROTO_TCP)
+        return IPv4Socket::setsockopt(level, option, user_value, user_value_size);
+
+    MutexLocker locker(mutex());
+
+    switch (option) {
+    case TCP_NODELAY:
+        if (user_value_size < sizeof(int))
+            return EINVAL;
+        int value;
+        TRY(copy_from_user(&value, static_ptr_cast<int const*>(user_value)));
+        if (value != 0 && value != 1)
+            return EINVAL;
+        m_no_delay = value;
+        return {};
+    default:
+        dbgln("setsockopt({}) at IPPROTO_TCP not implemented.", option);
+        return ENOPROTOOPT;
+    }
+}
+
+ErrorOr<void> TCPSocket::getsockopt(OpenFileDescription& description, int level, int option, Userspace<void*> value, Userspace<socklen_t*> value_size)
+{
+    if (level != IPPROTO_TCP)
+        return IPv4Socket::getsockopt(description, level, option, value, value_size);
+
+    MutexLocker locker(mutex());
+
+    socklen_t size;
+    TRY(copy_from_user(&size, value_size.unsafe_userspace_ptr()));
+
+    switch (option) {
+    case TCP_NODELAY: {
+        int nodelay = m_no_delay ? 1 : 0;
+        if (size < sizeof(nodelay))
+            return EINVAL;
+        TRY(copy_to_user(static_ptr_cast<int*>(value), &nodelay));
+        size = sizeof(nodelay);
+        return copy_to_user(value_size, &size);
+    }
+    default:
+        dbgln("getsockopt({}) at IPPROTO_TCP not implemented.", option);
+        return ENOPROTOOPT;
+    }
+}
+
 ErrorOr<void> TCPSocket::protocol_bind()
 {
     dbgln_if(TCP_SOCKET_DEBUG, "TCPSocket::protocol_bind(), local_port() is {}", local_port());

Kernel/Net/TCPSocket.h

@@ -165,6 +165,9 @@ public:
 
     static NetworkOrdered<u16> compute_tcp_checksum(IPv4Address const& source, IPv4Address const& destination, TCPPacket const&, u16 payload_size);
 
+    virtual ErrorOr<void> setsockopt(int level, int option, Userspace<void const*>, socklen_t) override;
+    virtual ErrorOr<void> getsockopt(OpenFileDescription&, int level, int option, Userspace<void*>, Userspace<socklen_t*>) override;
+
 protected:
     void set_direction(Direction direction) { m_direction = direction; }
 
@@ -227,6 +230,8 @@ private:
     // peer's advertised window size.
     u32 m_send_window_size { 64 * KiB };
 
+    bool m_no_delay { false };
+
     IntrusiveListNode<TCPSocket> m_retransmit_list_node;
 
 public:

Kernel/UnixTypes.h

@@ -13,6 +13,7 @@
 #include <Kernel/API/POSIX/net/if_arp.h>
 #include <Kernel/API/POSIX/net/route.h>
 #include <Kernel/API/POSIX/netinet/in.h>
+#include <Kernel/API/POSIX/netinet/tcp.h>
 #include <Kernel/API/POSIX/poll.h>
 #include <Kernel/API/POSIX/sched.h>
 #include <Kernel/API/POSIX/serenity.h>

Userland/Libraries/LibC/netinet/tcp.h

@@ -6,5 +6,4 @@
 
 #pragma once
 
-#define TCP_NODELAY 10
-#define TCP_MAXSEG 11
+#include <Kernel/API/POSIX/netinet/tcp.h>