net: P5-C — atomic cutover from TcpConnection to Pcb enum-of-structs

Delete TcpConnection, TcpConnectionTable, TCP_TABLE, TcpInputResult,
and the five process_* handler functions.  tcp_input now locks PCB_TABLE
and dispatches through Pcb::on_segment, returning Actions that callers
drain for segments, timer ops, and socket wake-ups.

Architecture: buffers live in a parallel static array
(PcbTable.buffers[64]) separate from the small Pcb structs
(PcbTable.slots[Option<Pcb>; 64]), matching Linux's sk_buff separation
and preventing stack overflow on construct/drop of the ~75 KB
TcpBufferPair.

Key changes:
- All lifecycle/query/data-path functions take ConnId, drop tcp_ prefix
- TcpState kept as derived read-only view via PcbState::tcp_state()
- SegmentBuilder: delete &TcpConnection methods, rename _raw to canon
- socket.rs: ConnId threading, socket_notify_tcp_activity takes &Actions
- sync_socket_state uses observed_socket_state() directly
- tcp_socket shim deleted from lib.rs
- Keepalive scheduling wired into tcp_input glue
- RttEstimator gains consecutive_timeouts for retransmit-max detection
- with_data_state! macro + with_pcb closure API for test ergonomics
- All 24 files updated, 2351/2351 tests green (1 deleted: empty defaults)

Grep audits: TcpInputResult=0, TcpConnection=0, tcp_get_connection=0,
tcp::tcp_=0, pub mod tcp_socket=0 in code.

Author: nil0ft

drivers/src/virtio_net.rs

@@ -653,11 +653,11 @@ fn dispatch_rx_frame(state: &mut VirtioNetState, frame: &[u8]) {
             let options = &ip_payload[tcp::TCP_HEADER_LEN..hdr_len];
             let payload = &ip_payload[hdr_len..];
             let now_ms = slopos_utils::clock::uptime_ms();
-            let result = tcp::tcp_input(src_ip, dst_ip, &hdr, options, payload, now_ms);
-            if let Some(seg) = result.response {
-                let _ = socket::socket_send_tcp_segment(&seg, &[]);
+            let actions = tcp::input(src_ip, dst_ip, &hdr, options, payload, now_ms);
+            for seg in actions.segments() {
+                let _ = socket::socket_send_tcp_segment(seg, &[]);
             }
-            socket::socket_notify_tcp_activity(&result);
+            socket::socket_notify_tcp_activity(&actions);
         }
         p if p == net::IpProtocol::Udp.as_u8() => {
             if let Some((src_port, dst_port, udp_payload)) = parse_udp_header(ip_payload) {

net/src/ipv4.rs

@@ -213,7 +213,7 @@ fn dispatch_tcp(src_ip: [u8; 4], dst_ip: [u8; 4], pkt: &PacketBuf, checksum_rx:
     // 2-tuple (listener). The actual state machine processing still goes
     // through tcp_input() which uses TcpConnectionTable internally.
     {
-        use super::tcp_socket::TCP_DEMUX;
+        use super::tcp::listener::TCP_DEMUX;
         use super::types::{Ipv4Addr, Port};
 
         let demux = TCP_DEMUX.lock();
@@ -242,12 +242,12 @@ fn dispatch_tcp(src_ip: [u8; 4], dst_ip: [u8; 4], pkt: &PacketBuf, checksum_rx:
         }
     }
 
-    let result = tcp::tcp_input(src_ip, dst_ip, &hdr, options, payload, now_ms);
+    let actions = tcp::input(src_ip, dst_ip, &hdr, options, payload, now_ms);
 
-    if let Some(seg) = result.response {
-        let _ = socket::socket_send_tcp_segment(&seg, &[]);
+    for seg in actions.segments() {
+        let _ = socket::socket_send_tcp_segment(seg, &[]);
     }
-    socket::socket_notify_tcp_activity(&result);
+    socket::socket_notify_tcp_activity(&actions);
 }
 
 /// Dispatch a UDP datagram to the socket layer, with DNS interception.

net/src/lib.rs

@@ -32,15 +32,6 @@ pub mod route;
 pub mod socket;
 pub mod socket_file_ops;
 pub mod tcp;
-/// Compatibility shim for the former top-level `net::tcp_socket` module.
-///
-/// The real implementation moved to `tcp::listener` under P2.4 of the TCP
-/// overhaul.  Existing call sites (socket.rs, tests) keep working through
-/// this glob re-export; future work will update them to import from
-/// `tcp::listener` directly.
-pub mod tcp_socket {
-    pub use crate::tcp::listener::*;
-}
 pub mod timer;
 pub mod udp;
 pub mod unix_socket;