xcat-dep/gpxe/ipxe-droppackets.patch

diff -urN ipxe-configged/src/net/arp.c ipxe-droppackets/src/net/arp.c
--- ipxe-configged/src/net/arp.c	2010-11-11 14:09:48.132334810 -0500
+++ ipxe-droppackets/src/net/arp.c	2010-11-12 10:33:14.482334325 -0500
@@ -234,7 +234,7 @@
 		goto done;
 	
 	/* Create new ARP table entry if necessary */
-	if ( ! merge ) {
+	if ( ( arphdr->ar_op == htons ( ARPOP_REPLY ) ) && ( ! merge ) ) {
 		arp = &arp_table[next_new_arp_entry++ % NUM_ARP_ENTRIES];
 		arp->ll_protocol = ll_protocol;
 		arp->net_protocol = net_protocol;
diff -urN ipxe-configged/src/net/arp.c.orig ipxe-droppackets/src/net/arp.c.orig
--- ipxe-configged/src/net/arp.c.orig	1969-12-31 19:00:00.000000000 -0500
+++ ipxe-droppackets/src/net/arp.c.orig	2010-11-11 14:09:48.132334810 -0500
@@ -0,0 +1,290 @@
+/*
+ * Copyright (C) 2006 Michael Brown <mbrown@fensystems.co.uk>.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+FILE_LICENCE ( GPL2_OR_LATER );
+
+#include <stdint.h>
+#include <string.h>
+#include <byteswap.h>
+#include <errno.h>
+#include <ipxe/if_ether.h>
+#include <ipxe/if_arp.h>
+#include <ipxe/iobuf.h>
+#include <ipxe/netdevice.h>
+#include <ipxe/arp.h>
+
+/** @file
+ *
+ * Address Resolution Protocol
+ *
+ * This file implements the address resolution protocol as defined in
+ * RFC826.  The implementation is media-independent and
+ * protocol-independent; it is not limited to Ethernet or to IPv4.
+ *
+ */
+
+/** An ARP cache entry */
+struct arp_entry {
+	/** Network-layer protocol */
+	struct net_protocol *net_protocol;
+	/** Link-layer protocol */
+	struct ll_protocol *ll_protocol;
+	/** Network-layer address */
+	uint8_t net_addr[MAX_NET_ADDR_LEN];
+	/** Link-layer address */
+	uint8_t ll_addr[MAX_LL_ADDR_LEN];
+};
+
+/** Number of entries in the ARP cache
+ *
+ * This is a global cache, covering all network interfaces,
+ * network-layer protocols and link-layer protocols.
+ */
+#define NUM_ARP_ENTRIES 4
+
+/** The ARP cache */
+static struct arp_entry arp_table[NUM_ARP_ENTRIES];
+#define arp_table_end &arp_table[NUM_ARP_ENTRIES]
+
+static unsigned int next_new_arp_entry = 0;
+
+struct net_protocol arp_protocol __net_protocol;
+
+/**
+ * Find entry in the ARP cache
+ *
+ * @v ll_protocol	Link-layer protocol
+ * @v net_protocol	Network-layer protocol
+ * @v net_addr		Network-layer address
+ * @ret arp		ARP cache entry, or NULL if not found
+ *
+ */
+static struct arp_entry *
+arp_find_entry ( struct ll_protocol *ll_protocol,
+		 struct net_protocol *net_protocol,
+		 const void *net_addr ) {
+	struct arp_entry *arp;
+
+	for ( arp = arp_table ; arp < arp_table_end ; arp++ ) {
+		if ( ( arp->ll_protocol == ll_protocol ) &&
+		     ( arp->net_protocol == net_protocol ) &&
+		     ( memcmp ( arp->net_addr, net_addr,
+				net_protocol->net_addr_len ) == 0 ) )
+			return arp;
+	}
+	return NULL;
+}
+
+/**
+ * Look up media-specific link-layer address in the ARP cache
+ *
+ * @v netdev		Network device
+ * @v net_protocol	Network-layer protocol
+ * @v dest_net_addr	Destination network-layer address
+ * @v source_net_addr	Source network-layer address
+ * @ret dest_ll_addr	Destination link layer address
+ * @ret rc		Return status code
+ *
+ * This function will use the ARP cache to look up the link-layer
+ * address for the link-layer protocol associated with the network
+ * device and the given network-layer protocol and addresses.  If
+ * found, the destination link-layer address will be filled in in @c
+ * dest_ll_addr.
+ *
+ * If no address is found in the ARP cache, an ARP request will be
+ * transmitted on the specified network device and -ENOENT will be
+ * returned.
+ */
+int arp_resolve ( struct net_device *netdev, struct net_protocol *net_protocol,
+		  const void *dest_net_addr, const void *source_net_addr,
+		  void *dest_ll_addr ) {
+	struct ll_protocol *ll_protocol = netdev->ll_protocol;
+	const struct arp_entry *arp;
+	struct io_buffer *iobuf;
+	struct arphdr *arphdr;
+	int rc;
+
+	/* Look for existing entry in ARP table */
+	arp = arp_find_entry ( ll_protocol, net_protocol, dest_net_addr );
+	if ( arp ) {
+		DBG ( "ARP cache hit: %s %s => %s %s\n",
+		      net_protocol->name, net_protocol->ntoa ( arp->net_addr ),
+		      ll_protocol->name, ll_protocol->ntoa ( arp->ll_addr ) );
+		memcpy ( dest_ll_addr, arp->ll_addr, ll_protocol->ll_addr_len);
+		return 0;
+	}
+	DBG ( "ARP cache miss: %s %s\n", net_protocol->name,
+	      net_protocol->ntoa ( dest_net_addr ) );
+
+	/* Allocate ARP packet */
+	iobuf = alloc_iob ( MAX_LL_HEADER_LEN + sizeof ( *arphdr ) +
+			  2 * ( MAX_LL_ADDR_LEN + MAX_NET_ADDR_LEN ) );
+	if ( ! iobuf )
+		return -ENOMEM;
+	iob_reserve ( iobuf, MAX_LL_HEADER_LEN );
+
+	/* Build up ARP request */
+	arphdr = iob_put ( iobuf, sizeof ( *arphdr ) );
+	arphdr->ar_hrd = ll_protocol->ll_proto;
+	arphdr->ar_hln = ll_protocol->ll_addr_len;
+	arphdr->ar_pro = net_protocol->net_proto;
+	arphdr->ar_pln = net_protocol->net_addr_len;
+	arphdr->ar_op = htons ( ARPOP_REQUEST );
+	memcpy ( iob_put ( iobuf, ll_protocol->ll_addr_len ),
+		 netdev->ll_addr, ll_protocol->ll_addr_len );
+	memcpy ( iob_put ( iobuf, net_protocol->net_addr_len ),
+		 source_net_addr, net_protocol->net_addr_len );
+	memset ( iob_put ( iobuf, ll_protocol->ll_addr_len ),
+		 0, ll_protocol->ll_addr_len );
+	memcpy ( iob_put ( iobuf, net_protocol->net_addr_len ),
+		 dest_net_addr, net_protocol->net_addr_len );
+
+	/* Transmit ARP request */
+	if ( ( rc = net_tx ( iobuf, netdev, &arp_protocol,
+			     netdev->ll_broadcast, netdev->ll_addr ) ) != 0 )
+		return rc;
+
+	return -ENOENT;
+}
+
+/**
+ * Identify ARP protocol
+ *
+ * @v net_proto			Network-layer protocol, in network-endian order
+ * @ret arp_net_protocol	ARP protocol, or NULL
+ *
+ */
+static struct arp_net_protocol * arp_find_protocol ( uint16_t net_proto ) {
+	struct arp_net_protocol *arp_net_protocol;
+
+	for_each_table_entry ( arp_net_protocol, ARP_NET_PROTOCOLS ) {
+		if ( arp_net_protocol->net_protocol->net_proto == net_proto ) {
+			return arp_net_protocol;
+		}
+	}
+	return NULL;
+}
+
+/**
+ * Process incoming ARP packets
+ *
+ * @v iobuf		I/O buffer
+ * @v netdev		Network device
+ * @v ll_source		Link-layer source address
+ * @ret rc		Return status code
+ *
+ * This handles ARP requests and responses as detailed in RFC826.  The
+ * method detailed within the RFC is pretty optimised, handling
+ * requests and responses with basically a single code path and
+ * avoiding the need for extraneous ARP requests; read the RFC for
+ * details.
+ */
+static int arp_rx ( struct io_buffer *iobuf, struct net_device *netdev,
+		    const void *ll_dest __unused,
+		    const void *ll_source __unused ) {
+	struct arphdr *arphdr = iobuf->data;
+	struct arp_net_protocol *arp_net_protocol;
+	struct net_protocol *net_protocol;
+	struct ll_protocol *ll_protocol;
+	struct arp_entry *arp;
+	int merge = 0;
+
+	/* Identify network-layer and link-layer protocols */
+	arp_net_protocol = arp_find_protocol ( arphdr->ar_pro );
+	if ( ! arp_net_protocol )
+		goto done;
+	net_protocol = arp_net_protocol->net_protocol;
+	ll_protocol = netdev->ll_protocol;
+
+	/* Sanity checks */
+	if ( ( arphdr->ar_hrd != ll_protocol->ll_proto ) ||
+	     ( arphdr->ar_hln != ll_protocol->ll_addr_len ) ||
+	     ( arphdr->ar_pln != net_protocol->net_addr_len ) )
+		goto done;
+
+	/* See if we have an entry for this sender, and update it if so */
+	arp = arp_find_entry ( ll_protocol, net_protocol,
+			       arp_sender_pa ( arphdr ) );
+	if ( arp ) {
+		memcpy ( arp->ll_addr, arp_sender_ha ( arphdr ),
+			 arphdr->ar_hln );
+		merge = 1;
+		DBG ( "ARP cache update: %s %s => %s %s\n",
+		      net_protocol->name, net_protocol->ntoa ( arp->net_addr ),
+		      ll_protocol->name, ll_protocol->ntoa ( arp->ll_addr ) );
+	}
+
+	/* See if we own the target protocol address */
+	if ( arp_net_protocol->check ( netdev, arp_target_pa ( arphdr ) ) != 0)
+		goto done;
+	
+	/* Create new ARP table entry if necessary */
+	if ( ! merge ) {
+		arp = &arp_table[next_new_arp_entry++ % NUM_ARP_ENTRIES];
+		arp->ll_protocol = ll_protocol;
+		arp->net_protocol = net_protocol;
+		memcpy ( arp->ll_addr, arp_sender_ha ( arphdr ),
+			 arphdr->ar_hln );
+		memcpy ( arp->net_addr, arp_sender_pa ( arphdr ),
+			 arphdr->ar_pln);
+		DBG ( "ARP cache add: %s %s => %s %s\n",
+		      net_protocol->name, net_protocol->ntoa ( arp->net_addr ),
+		      ll_protocol->name, ll_protocol->ntoa ( arp->ll_addr ) );
+	}
+
+	/* If it's not a request, there's nothing more to do */
+	if ( arphdr->ar_op != htons ( ARPOP_REQUEST ) )
+		goto done;
+
+	/* Change request to a reply */
+	DBG ( "ARP reply: %s %s => %s %s\n", net_protocol->name,
+	      net_protocol->ntoa ( arp_target_pa ( arphdr ) ),
+	      ll_protocol->name, ll_protocol->ntoa ( netdev->ll_addr ) );
+	arphdr->ar_op = htons ( ARPOP_REPLY );
+	memswap ( arp_sender_ha ( arphdr ), arp_target_ha ( arphdr ),
+		 arphdr->ar_hln + arphdr->ar_pln );
+	memcpy ( arp_sender_ha ( arphdr ), netdev->ll_addr, arphdr->ar_hln );
+
+	/* Send reply */
+	net_tx ( iob_disown ( iobuf ), netdev, &arp_protocol,
+		 arp_target_ha ( arphdr ), netdev->ll_addr );
+
+ done:
+	free_iob ( iobuf );
+	return 0;
+}
+
+/**
+ * Transcribe ARP address
+ *
+ * @v net_addr	ARP address
+ * @ret string	"<ARP>"
+ *
+ * This operation is meaningless for the ARP protocol.
+ */
+static const char * arp_ntoa ( const void *net_addr __unused ) {
+	return "<ARP>";
+}
+
+/** ARP protocol */
+struct net_protocol arp_protocol __net_protocol = {
+	.name = "ARP",
+	.net_proto = htons ( ETH_P_ARP ),
+	.rx = arp_rx,
+	.ntoa = arp_ntoa,
+};
diff -urN ipxe-configged/src/net/icmp.c ipxe-droppackets/src/net/icmp.c
--- ipxe-configged/src/net/icmp.c	2010-11-11 14:09:48.132334810 -0500
+++ ipxe-droppackets/src/net/icmp.c	2010-11-12 10:33:14.482334325 -0500
@@ -69,11 +69,14 @@
 	}
 
 	/* We respond only to pings */
-	if ( icmp->type != ICMP_ECHO_REQUEST ) {
+	/* Always discard, responding to pings is more than most systems do in this state and 
+	 * gPXE is in a relatively precarious position, resource management wise, and thus it is
+         * easier just to disable this function. */
+	/* if ( icmp->type != ICMP_ECHO_REQUEST ) { */
 		DBG ( "ICMP ignoring type %d\n", icmp->type );
 		rc = 0;
 		goto done;
-	}
+	/* } */
 
 	DBG ( "ICMP responding to ping\n" );
 
diff -urN ipxe-configged/src/net/icmp.c.orig ipxe-droppackets/src/net/icmp.c.orig
--- ipxe-configged/src/net/icmp.c.orig	1969-12-31 19:00:00.000000000 -0500
+++ ipxe-droppackets/src/net/icmp.c.orig	2010-11-11 14:09:48.132334810 -0500
@@ -0,0 +1,103 @@
+/*
+ * Copyright (C) 2009 Michael Brown <mbrown@fensystems.co.uk>.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+FILE_LICENCE ( GPL2_OR_LATER );
+
+#include <string.h>
+#include <errno.h>
+#include <ipxe/iobuf.h>
+#include <ipxe/in.h>
+#include <ipxe/tcpip.h>
+#include <ipxe/icmp.h>
+
+/** @file
+ *
+ * ICMP protocol
+ *
+ */
+
+struct tcpip_protocol icmp_protocol __tcpip_protocol;
+
+/**
+ * Process a received packet
+ *
+ * @v iobuf		I/O buffer
+ * @v st_src		Partially-filled source address
+ * @v st_dest		Partially-filled destination address
+ * @v pshdr_csum	Pseudo-header checksum
+ * @ret rc		Return status code
+ */
+static int icmp_rx ( struct io_buffer *iobuf, struct sockaddr_tcpip *st_src,
+		     struct sockaddr_tcpip *st_dest,
+		     uint16_t pshdr_csum __unused ) {
+	struct icmp_header *icmp = iobuf->data;
+	size_t len = iob_len ( iobuf );
+	unsigned int csum;
+	int rc;
+
+	/* Sanity check */
+	if ( len < sizeof ( *icmp ) ) {
+		DBG ( "ICMP packet too short at %zd bytes (min %zd bytes)\n",
+		      len, sizeof ( *icmp ) );
+		rc = -EINVAL;
+		goto done;
+	}
+
+	/* Verify checksum */
+	csum = tcpip_chksum ( icmp, len );
+	if ( csum != 0 ) {
+		DBG ( "ICMP checksum incorrect (is %04x, should be 0000)\n",
+		      csum );
+		DBG_HD ( icmp, len );
+		rc = -EINVAL;
+		goto done;
+	}
+
+	/* We respond only to pings */
+	if ( icmp->type != ICMP_ECHO_REQUEST ) {
+		DBG ( "ICMP ignoring type %d\n", icmp->type );
+		rc = 0;
+		goto done;
+	}
+
+	DBG ( "ICMP responding to ping\n" );
+
+	/* Change type to response and recalculate checksum */
+	icmp->type = ICMP_ECHO_RESPONSE;
+	icmp->chksum = 0;
+	icmp->chksum = tcpip_chksum ( icmp, len );
+
+	/* Transmit the response */
+	if ( ( rc = tcpip_tx ( iob_disown ( iobuf ), &icmp_protocol, st_dest,
+			       st_src, NULL, NULL ) ) != 0 ) {
+		DBG ( "ICMP could not transmit ping response: %s\n",
+		      strerror ( rc ) );
+		goto done;
+	}
+
+ done:
+	free_iob ( iobuf );
+	return rc;
+}
+
+/** ICMP TCP/IP protocol */
+struct tcpip_protocol icmp_protocol __tcpip_protocol = {
+	.name = "ICMP",
+	.rx = icmp_rx,
+	.tcpip_proto = IP_ICMP,
+};
diff -urN ipxe-configged/src/net/tcp.c ipxe-droppackets/src/net/tcp.c
--- ipxe-configged/src/net/tcp.c	2010-11-11 14:09:48.132334810 -0500
+++ ipxe-droppackets/src/net/tcp.c	2010-11-12 10:33:14.482334325 -0500
@@ -1119,7 +1119,11 @@
 
 	/* If no connection was found, send RST */
 	if ( ! tcp ) {
-		tcp_xmit_reset ( tcp, st_src, tcphdr );
+		/* We simply drop unrecognized TCP connections.  Hopefully, this is no worse than I typically set up a firewall.
+		 * If we try to even remotely respond to random TCP streams, it can exhaust gPXE resources easily in a very large
+		 * environment */
+		/* tcp_xmit_reset ( tcp, st_src, tcphdr ); Don't consume precious outbound resource for irrelevant communication. */
+		
 		rc = -ENOTCONN;
 		goto discard;
 	}
diff -urN ipxe-configged/src/net/tcp.c.orig ipxe-droppackets/src/net/tcp.c.orig
--- ipxe-configged/src/net/tcp.c.orig	1969-12-31 19:00:00.000000000 -0500
+++ ipxe-droppackets/src/net/tcp.c.orig	2010-11-11 14:09:48.132334810 -0500
@@ -0,0 +1,1336 @@
+#include <string.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <assert.h>
+#include <errno.h>
+#include <byteswap.h>
+#include <ipxe/timer.h>
+#include <ipxe/iobuf.h>
+#include <ipxe/malloc.h>
+#include <ipxe/retry.h>
+#include <ipxe/refcnt.h>
+#include <ipxe/xfer.h>
+#include <ipxe/open.h>
+#include <ipxe/uri.h>
+#include <ipxe/tcpip.h>
+#include <ipxe/tcp.h>
+
+/** @file
+ *
+ * TCP protocol
+ *
+ */
+
+FILE_LICENCE ( GPL2_OR_LATER );
+
+/** A TCP connection */
+struct tcp_connection {
+	/** Reference counter */
+	struct refcnt refcnt;
+	/** List of TCP connections */
+	struct list_head list;
+
+	/** Flags */
+	unsigned int flags;
+
+	/** Data transfer interface */
+	struct interface xfer;
+
+	/** Remote socket address */
+	struct sockaddr_tcpip peer;
+	/** Local port */
+	unsigned int local_port;
+
+	/** Current TCP state */
+	unsigned int tcp_state;
+	/** Previous TCP state
+	 *
+	 * Maintained only for debug messages
+	 */
+	unsigned int prev_tcp_state;
+	/** Current sequence number
+	 *
+	 * Equivalent to SND.UNA in RFC 793 terminology.
+	 */
+	uint32_t snd_seq;
+	/** Unacknowledged sequence count
+	 *
+	 * Equivalent to (SND.NXT-SND.UNA) in RFC 793 terminology.
+	 */
+	uint32_t snd_sent;
+	/** Send window
+	 *
+	 * Equivalent to SND.WND in RFC 793 terminology
+	 */
+	uint32_t snd_win;
+	/** Current acknowledgement number
+	 *
+	 * Equivalent to RCV.NXT in RFC 793 terminology.
+	 */
+	uint32_t rcv_ack;
+	/** Receive window
+	 *
+	 * Equivalent to RCV.WND in RFC 793 terminology.
+	 */
+	uint32_t rcv_win;
+	/** Most recent received timestamp
+	 *
+	 * Equivalent to TS.Recent in RFC 1323 terminology.
+	 */
+	uint32_t ts_recent;
+
+	/** Transmit queue */
+	struct list_head tx_queue;
+	/** Receive queue */
+	struct list_head rx_queue;
+	/** Retransmission timer */
+	struct retry_timer timer;
+	/** Shutdown (TIME_WAIT) timer */
+	struct retry_timer wait;
+};
+
+/** TCP flags */
+enum tcp_flags {
+	/** TCP data transfer interface has been closed */
+	TCP_XFER_CLOSED = 0x0001,
+	/** TCP timestamps are enabled */
+	TCP_TS_ENABLED = 0x0002,
+	/** TCP acknowledgement is pending */
+	TCP_ACK_PENDING = 0x0004,
+};
+
+/** TCP internal header
+ *
+ * This is the header that replaces the TCP header for packets
+ * enqueued on the receive queue.
+ */
+struct tcp_rx_queued_header {
+	/** SEQ value, in host-endian order
+	 *
+	 * This represents the SEQ value at the time the packet is
+	 * enqueued, and so excludes the SYN, if present.
+	 */
+	uint32_t seq;
+	/** Flags
+	 *
+	 * Only FIN is valid within this flags byte; all other flags
+	 * have already been processed by the time the packet is
+	 * enqueued.
+	 */
+	uint8_t flags;
+	/** Reserved */
+	uint8_t reserved[3];
+};
+
+/**
+ * List of registered TCP connections
+ */
+static LIST_HEAD ( tcp_conns );
+
+/* Forward declarations */
+static struct interface_descriptor tcp_xfer_desc;
+static void tcp_expired ( struct retry_timer *timer, int over );
+static void tcp_wait_expired ( struct retry_timer *timer, int over );
+static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
+			uint32_t win );
+
+/**
+ * Name TCP state
+ *
+ * @v state		TCP state
+ * @ret name		Name of TCP state
+ */
+static inline __attribute__ (( always_inline )) const char *
+tcp_state ( int state ) {
+	switch ( state ) {
+	case TCP_CLOSED:		return "CLOSED";
+	case TCP_LISTEN:		return "LISTEN";
+	case TCP_SYN_SENT:		return "SYN_SENT";
+	case TCP_SYN_RCVD:		return "SYN_RCVD";
+	case TCP_ESTABLISHED:		return "ESTABLISHED";
+	case TCP_FIN_WAIT_1:		return "FIN_WAIT_1";
+	case TCP_FIN_WAIT_2:		return "FIN_WAIT_2";
+	case TCP_CLOSING_OR_LAST_ACK:	return "CLOSING/LAST_ACK";
+	case TCP_TIME_WAIT:		return "TIME_WAIT";
+	case TCP_CLOSE_WAIT:		return "CLOSE_WAIT";
+	default:			return "INVALID";
+	}
+}
+
+/**
+ * Dump TCP state transition
+ *
+ * @v tcp		TCP connection
+ */
+static inline __attribute__ (( always_inline )) void
+tcp_dump_state ( struct tcp_connection *tcp ) {
+
+	if ( tcp->tcp_state != tcp->prev_tcp_state ) {
+		DBGC ( tcp, "TCP %p transitioned from %s to %s\n", tcp,
+		       tcp_state ( tcp->prev_tcp_state ),
+		       tcp_state ( tcp->tcp_state ) );
+	}
+	tcp->prev_tcp_state = tcp->tcp_state;
+}
+
+/**
+ * Dump TCP flags
+ *
+ * @v flags		TCP flags
+ */
+static inline __attribute__ (( always_inline )) void
+tcp_dump_flags ( struct tcp_connection *tcp, unsigned int flags ) {
+	if ( flags & TCP_RST )
+		DBGC2 ( tcp, " RST" );
+	if ( flags & TCP_SYN )
+		DBGC2 ( tcp, " SYN" );
+	if ( flags & TCP_PSH )
+		DBGC2 ( tcp, " PSH" );
+	if ( flags & TCP_FIN )
+		DBGC2 ( tcp, " FIN" );
+	if ( flags & TCP_ACK )
+		DBGC2 ( tcp, " ACK" );
+}
+
+/***************************************************************************
+ *
+ * Open and close
+ *
+ ***************************************************************************
+ */
+
+/**
+ * Bind TCP connection to local port
+ *
+ * @v tcp		TCP connection
+ * @v port		Local port number
+ * @ret rc		Return status code
+ *
+ * If the port is 0, the connection is assigned an available port
+ * between 1024 and 65535.
+ */
+static int tcp_bind ( struct tcp_connection *tcp, unsigned int port ) {
+	struct tcp_connection *existing;
+	uint16_t try_port;
+	unsigned int i;
+
+	/* If no port is specified, find an available port */
+	if ( ! port ) {
+		try_port = random();
+		for ( i = 0 ; i < 65536 ; i++ ) {
+			try_port++;
+			if ( try_port < 1024 )
+				continue;
+			if ( tcp_bind ( tcp, try_port ) == 0 )
+				return 0;
+		}
+		DBGC ( tcp, "TCP %p could not bind: no free ports\n", tcp );
+		return -EADDRINUSE;
+	}
+
+	/* Attempt bind to local port */
+	list_for_each_entry ( existing, &tcp_conns, list ) {
+		if ( existing->local_port == port ) {
+			DBGC ( tcp, "TCP %p could not bind: port %d in use\n",
+			       tcp, port );
+			return -EADDRINUSE;
+		}
+	}
+	tcp->local_port = port;
+
+	DBGC ( tcp, "TCP %p bound to port %d\n", tcp, port );
+	return 0;
+}
+
+/**
+ * Open a TCP connection
+ *
+ * @v xfer		Data transfer interface
+ * @v peer		Peer socket address
+ * @v local		Local socket address, or NULL
+ * @ret rc		Return status code
+ */
+static int tcp_open ( struct interface *xfer, struct sockaddr *peer,
+		      struct sockaddr *local ) {
+	struct sockaddr_tcpip *st_peer = ( struct sockaddr_tcpip * ) peer;
+	struct sockaddr_tcpip *st_local = ( struct sockaddr_tcpip * ) local;
+	struct tcp_connection *tcp;
+	unsigned int bind_port;
+	int rc;
+
+	/* Allocate and initialise structure */
+	tcp = zalloc ( sizeof ( *tcp ) );
+	if ( ! tcp )
+		return -ENOMEM;
+	DBGC ( tcp, "TCP %p allocated\n", tcp );
+	ref_init ( &tcp->refcnt, NULL );
+	intf_init ( &tcp->xfer, &tcp_xfer_desc, &tcp->refcnt );
+	timer_init ( &tcp->timer, tcp_expired, &tcp->refcnt );
+	timer_init ( &tcp->wait, tcp_wait_expired, &tcp->refcnt );
+	tcp->prev_tcp_state = TCP_CLOSED;
+	tcp->tcp_state = TCP_STATE_SENT ( TCP_SYN );
+	tcp_dump_state ( tcp );
+	tcp->snd_seq = random();
+	INIT_LIST_HEAD ( &tcp->tx_queue );
+	INIT_LIST_HEAD ( &tcp->rx_queue );
+	memcpy ( &tcp->peer, st_peer, sizeof ( tcp->peer ) );
+
+	/* Bind to local port */
+	bind_port = ( st_local ? ntohs ( st_local->st_port ) : 0 );
+	if ( ( rc = tcp_bind ( tcp, bind_port ) ) != 0 )
+		goto err;
+
+	/* Start timer to initiate SYN */
+	start_timer_nodelay ( &tcp->timer );
+
+	/* Attach parent interface, transfer reference to connection
+	 * list and return
+	 */
+	intf_plug_plug ( &tcp->xfer, xfer );
+	list_add ( &tcp->list, &tcp_conns );
+	return 0;
+
+ err:
+	ref_put ( &tcp->refcnt );
+	return rc;
+}
+
+/**
+ * Close TCP connection
+ *
+ * @v tcp		TCP connection
+ * @v rc		Reason for close
+ *
+ * Closes the data transfer interface.  If the TCP state machine is in
+ * a suitable state, the connection will be deleted.
+ */
+static void tcp_close ( struct tcp_connection *tcp, int rc ) {
+	struct io_buffer *iobuf;
+	struct io_buffer *tmp;
+
+	/* Close data transfer interface */
+	intf_shutdown ( &tcp->xfer, rc );
+	tcp->flags |= TCP_XFER_CLOSED;
+
+	/* If we are in CLOSED, or have otherwise not yet received a
+	 * SYN (i.e. we are in LISTEN or SYN_SENT), just delete the
+	 * connection.
+	 */
+	if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
+
+		/* Transition to CLOSED for the sake of debugging messages */
+		tcp->tcp_state = TCP_CLOSED;
+		tcp_dump_state ( tcp );
+
+		/* Free any unprocessed I/O buffers */
+		list_for_each_entry_safe ( iobuf, tmp, &tcp->rx_queue, list ) {
+			list_del ( &iobuf->list );
+			free_iob ( iobuf );
+		}
+
+		/* Free any unsent I/O buffers */
+		list_for_each_entry_safe ( iobuf, tmp, &tcp->tx_queue, list ) {
+			list_del ( &iobuf->list );
+			free_iob ( iobuf );
+		}
+
+		/* Remove from list and drop reference */
+		stop_timer ( &tcp->timer );
+		list_del ( &tcp->list );
+		ref_put ( &tcp->refcnt );
+		DBGC ( tcp, "TCP %p connection deleted\n", tcp );
+		return;
+	}
+
+	/* If we have not had our SYN acknowledged (i.e. we are in
+	 * SYN_RCVD), pretend that it has been acknowledged so that we
+	 * can send a FIN without breaking things.
+	 */
+	if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
+		tcp_rx_ack ( tcp, ( tcp->snd_seq + 1 ), 0 );
+
+	/* If we have no data remaining to send, start sending FIN */
+	if ( list_empty ( &tcp->tx_queue ) ) {
+		tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
+		tcp_dump_state ( tcp );
+	}
+}
+
+/***************************************************************************
+ *
+ * Transmit data path
+ *
+ ***************************************************************************
+ */
+
+/**
+ * Calculate transmission window
+ *
+ * @v tcp		TCP connection
+ * @ret len		Maximum length that can be sent in a single packet
+ */
+static size_t tcp_xmit_win ( struct tcp_connection *tcp ) {
+	size_t len;
+
+	/* Not ready if we're not in a suitable connection state */
+	if ( ! TCP_CAN_SEND_DATA ( tcp->tcp_state ) )
+		return 0;
+
+	/* Length is the minimum of the receiver's window and the path MTU */
+	len = tcp->snd_win;
+	if ( len > TCP_PATH_MTU )
+		len = TCP_PATH_MTU;
+
+	return len;
+}
+
+/**
+ * Process TCP transmit queue
+ *
+ * @v tcp		TCP connection
+ * @v max_len		Maximum length to process
+ * @v dest		I/O buffer to fill with data, or NULL
+ * @v remove		Remove data from queue
+ * @ret len		Length of data processed
+ *
+ * This processes at most @c max_len bytes from the TCP connection's
+ * transmit queue.  Data will be copied into the @c dest I/O buffer
+ * (if provided) and, if @c remove is true, removed from the transmit
+ * queue.
+ */
+static size_t tcp_process_tx_queue ( struct tcp_connection *tcp, size_t max_len,
+				     struct io_buffer *dest, int remove ) {
+	struct io_buffer *iobuf;
+	struct io_buffer *tmp;
+	size_t frag_len;
+	size_t len = 0;
+
+	list_for_each_entry_safe ( iobuf, tmp, &tcp->tx_queue, list ) {
+		frag_len = iob_len ( iobuf );
+		if ( frag_len > max_len )
+			frag_len = max_len;
+		if ( dest ) {
+			memcpy ( iob_put ( dest, frag_len ), iobuf->data,
+				 frag_len );
+		}
+		if ( remove ) {
+			iob_pull ( iobuf, frag_len );
+			if ( ! iob_len ( iobuf ) ) {
+				list_del ( &iobuf->list );
+				free_iob ( iobuf );
+			}
+		}
+		len += frag_len;
+		max_len -= frag_len;
+	}
+	return len;
+}
+
+/**
+ * Transmit any outstanding data
+ *
+ * @v tcp		TCP connection
+ * 
+ * Transmits any outstanding data on the connection.
+ *
+ * Note that even if an error is returned, the retransmission timer
+ * will have been started if necessary, and so the stack will
+ * eventually attempt to retransmit the failed packet.
+ */
+static int tcp_xmit ( struct tcp_connection *tcp ) {
+	struct io_buffer *iobuf;
+	struct tcp_header *tcphdr;
+	struct tcp_mss_option *mssopt;
+	struct tcp_timestamp_padded_option *tsopt;
+	void *payload;
+	unsigned int flags;
+	size_t len = 0;
+	uint32_t seq_len;
+	uint32_t app_win;
+	uint32_t max_rcv_win;
+	int rc;
+
+	/* If retransmission timer is already running, do nothing */
+	if ( timer_running ( &tcp->timer ) )
+		return 0;
+
+	/* Calculate both the actual (payload) and sequence space
+	 * lengths that we wish to transmit.
+	 */
+	if ( TCP_CAN_SEND_DATA ( tcp->tcp_state ) ) {
+		len = tcp_process_tx_queue ( tcp, tcp_xmit_win ( tcp ),
+					     NULL, 0 );
+	}
+	seq_len = len;
+	flags = TCP_FLAGS_SENDING ( tcp->tcp_state );
+	if ( flags & ( TCP_SYN | TCP_FIN ) ) {
+		/* SYN or FIN consume one byte, and we can never send both */
+		assert ( ! ( ( flags & TCP_SYN ) && ( flags & TCP_FIN ) ) );
+		seq_len++;
+	}
+	tcp->snd_sent = seq_len;
+
+	/* If we have nothing to transmit, stop now */
+	if ( ( seq_len == 0 ) && ! ( tcp->flags & TCP_ACK_PENDING ) )
+		return 0;
+
+	/* If we are transmitting anything that requires
+	 * acknowledgement (i.e. consumes sequence space), start the
+	 * retransmission timer.  Do this before attempting to
+	 * allocate the I/O buffer, in case allocation itself fails.
+	 */
+	if ( seq_len )
+		start_timer ( &tcp->timer );
+
+	/* Allocate I/O buffer */
+	iobuf = alloc_iob ( len + MAX_HDR_LEN );
+	if ( ! iobuf ) {
+		DBGC ( tcp, "TCP %p could not allocate iobuf for %08x..%08x "
+		       "%08x\n", tcp, tcp->snd_seq, ( tcp->snd_seq + seq_len ),
+		       tcp->rcv_ack );
+		return -ENOMEM;
+	}
+	iob_reserve ( iobuf, MAX_HDR_LEN );
+
+	/* Fill data payload from transmit queue */
+	tcp_process_tx_queue ( tcp, len, iobuf, 0 );
+
+	/* Expand receive window if possible */
+	max_rcv_win = ( ( freemem * 3 ) / 4 );
+	if ( max_rcv_win > TCP_MAX_WINDOW_SIZE )
+		max_rcv_win = TCP_MAX_WINDOW_SIZE;
+	app_win = xfer_window ( &tcp->xfer );
+	if ( max_rcv_win > app_win )
+		max_rcv_win = app_win;
+	max_rcv_win &= ~0x03; /* Keep everything dword-aligned */
+	if ( tcp->rcv_win < max_rcv_win )
+		tcp->rcv_win = max_rcv_win;
+
+	/* Fill up the TCP header */
+	payload = iobuf->data;
+	if ( flags & TCP_SYN ) {
+		mssopt = iob_push ( iobuf, sizeof ( *mssopt ) );
+		mssopt->kind = TCP_OPTION_MSS;
+		mssopt->length = sizeof ( *mssopt );
+		mssopt->mss = htons ( TCP_MSS );
+	}
+	if ( ( flags & TCP_SYN ) || ( tcp->flags & TCP_TS_ENABLED ) ) {
+		tsopt = iob_push ( iobuf, sizeof ( *tsopt ) );
+		memset ( tsopt->nop, TCP_OPTION_NOP, sizeof ( tsopt->nop ) );
+		tsopt->tsopt.kind = TCP_OPTION_TS;
+		tsopt->tsopt.length = sizeof ( tsopt->tsopt );
+		tsopt->tsopt.tsval = htonl ( currticks() );
+		tsopt->tsopt.tsecr = htonl ( tcp->ts_recent );
+	}
+	if ( len != 0 )
+		flags |= TCP_PSH;
+	tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
+	memset ( tcphdr, 0, sizeof ( *tcphdr ) );
+	tcphdr->src = htons ( tcp->local_port );
+	tcphdr->dest = tcp->peer.st_port;
+	tcphdr->seq = htonl ( tcp->snd_seq );
+	tcphdr->ack = htonl ( tcp->rcv_ack );
+	tcphdr->hlen = ( ( payload - iobuf->data ) << 2 );
+	tcphdr->flags = flags;
+	tcphdr->win = htons ( tcp->rcv_win );
+	tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
+
+	/* Dump header */
+	DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4zd",
+		tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
+		ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) + seq_len ),
+		ntohl ( tcphdr->ack ), len );
+	tcp_dump_flags ( tcp, tcphdr->flags );
+	DBGC2 ( tcp, "\n" );
+
+	/* Transmit packet */
+	if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, &tcp->peer, NULL,
+			       &tcphdr->csum ) ) != 0 ) {
+		DBGC ( tcp, "TCP %p could not transmit %08x..%08x %08x: %s\n",
+		       tcp, tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ),
+		       tcp->rcv_ack, strerror ( rc ) );
+		return rc;
+	}
+
+	/* Clear ACK-pending flag */
+	tcp->flags &= ~TCP_ACK_PENDING;
+
+	return 0;
+}
+
+/**
+ * Retransmission timer expired
+ *
+ * @v timer		Retransmission timer
+ * @v over		Failure indicator
+ */
+static void tcp_expired ( struct retry_timer *timer, int over ) {
+	struct tcp_connection *tcp =
+		container_of ( timer, struct tcp_connection, timer );
+
+	DBGC ( tcp, "TCP %p timer %s in %s for %08x..%08x %08x\n", tcp,
+	       ( over ? "expired" : "fired" ), tcp_state ( tcp->tcp_state ),
+	       tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
+
+	assert ( ( tcp->tcp_state == TCP_SYN_SENT ) ||
+		 ( tcp->tcp_state == TCP_SYN_RCVD ) ||
+		 ( tcp->tcp_state == TCP_ESTABLISHED ) ||
+		 ( tcp->tcp_state == TCP_FIN_WAIT_1 ) ||
+		 ( tcp->tcp_state == TCP_CLOSE_WAIT ) ||
+		 ( tcp->tcp_state == TCP_CLOSING_OR_LAST_ACK ) );
+
+	if ( over ) {
+		/* If we have finally timed out and given up,
+		 * terminate the connection
+		 */
+		tcp->tcp_state = TCP_CLOSED;
+		tcp_dump_state ( tcp );
+		tcp_close ( tcp, -ETIMEDOUT );
+	} else {
+		/* Otherwise, retransmit the packet */
+		tcp_xmit ( tcp );
+	}
+}
+
+/**
+ * Shutdown timer expired
+ *
+ * @v timer		Shutdown timer
+ * @v over		Failure indicator
+ */
+static void tcp_wait_expired ( struct retry_timer *timer, int over __unused ) {
+	struct tcp_connection *tcp =
+		container_of ( timer, struct tcp_connection, wait );
+
+	assert ( tcp->tcp_state == TCP_TIME_WAIT );
+
+	DBGC ( tcp, "TCP %p wait complete in %s for %08x..%08x %08x\n", tcp,
+	       tcp_state ( tcp->tcp_state ), tcp->snd_seq,
+	       ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
+
+	tcp->tcp_state = TCP_CLOSED;
+	tcp_dump_state ( tcp );
+	tcp_close ( tcp, 0 );
+}
+
+/**
+ * Send RST response to incoming packet
+ *
+ * @v in_tcphdr		TCP header of incoming packet
+ * @ret rc		Return status code
+ */
+static int tcp_xmit_reset ( struct tcp_connection *tcp,
+			    struct sockaddr_tcpip *st_dest,
+			    struct tcp_header *in_tcphdr ) {
+	struct io_buffer *iobuf;
+	struct tcp_header *tcphdr;
+	int rc;
+
+	/* Allocate space for dataless TX buffer */
+	iobuf = alloc_iob ( MAX_HDR_LEN );
+	if ( ! iobuf ) {
+		DBGC ( tcp, "TCP %p could not allocate iobuf for RST "
+		       "%08x..%08x %08x\n", tcp, ntohl ( in_tcphdr->ack ),
+		       ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ) );
+		return -ENOMEM;
+	}
+	iob_reserve ( iobuf, MAX_HDR_LEN );
+
+	/* Construct RST response */
+	tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
+	memset ( tcphdr, 0, sizeof ( *tcphdr ) );
+	tcphdr->src = in_tcphdr->dest;
+	tcphdr->dest = in_tcphdr->src;
+	tcphdr->seq = in_tcphdr->ack;
+	tcphdr->ack = in_tcphdr->seq;
+	tcphdr->hlen = ( ( sizeof ( *tcphdr ) / 4 ) << 4 );
+	tcphdr->flags = ( TCP_RST | TCP_ACK );
+	tcphdr->win = htons ( TCP_MAX_WINDOW_SIZE );
+	tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
+
+	/* Dump header */
+	DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4d",
+		tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
+		ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) ),
+		ntohl ( tcphdr->ack ), 0 );
+	tcp_dump_flags ( tcp, tcphdr->flags );
+	DBGC2 ( tcp, "\n" );
+
+	/* Transmit packet */
+	if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, st_dest,
+			       NULL, &tcphdr->csum ) ) != 0 ) {
+		DBGC ( tcp, "TCP %p could not transmit RST %08x..%08x %08x: "
+		       "%s\n", tcp, ntohl ( in_tcphdr->ack ),
+		       ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ),
+		       strerror ( rc ) );
+		return rc;
+	}
+
+	return 0;
+}
+
+/***************************************************************************
+ *
+ * Receive data path
+ *
+ ***************************************************************************
+ */
+
+/**
+ * Identify TCP connection by local port number
+ *
+ * @v local_port	Local port
+ * @ret tcp		TCP connection, or NULL
+ */
+static struct tcp_connection * tcp_demux ( unsigned int local_port ) {
+	struct tcp_connection *tcp;
+
+	list_for_each_entry ( tcp, &tcp_conns, list ) {
+		if ( tcp->local_port == local_port )
+			return tcp;
+	}
+	return NULL;
+}
+
+/**
+ * Parse TCP received options
+ *
+ * @v tcp		TCP connection
+ * @v data		Raw options data
+ * @v len		Raw options length
+ * @v options		Options structure to fill in
+ */
+static void tcp_rx_opts ( struct tcp_connection *tcp, const void *data,
+			  size_t len, struct tcp_options *options ) {
+	const void *end = ( data + len );
+	const struct tcp_option *option;
+	unsigned int kind;
+
+	memset ( options, 0, sizeof ( *options ) );
+	while ( data < end ) {
+		option = data;
+		kind = option->kind;
+		if ( kind == TCP_OPTION_END )
+			return;
+		if ( kind == TCP_OPTION_NOP ) {
+			data++;
+			continue;
+		}
+		switch ( kind ) {
+		case TCP_OPTION_MSS:
+			options->mssopt = data;
+			break;
+		case TCP_OPTION_TS:
+			options->tsopt = data;
+			break;
+		default:
+			DBGC ( tcp, "TCP %p received unknown option %d\n",
+			       tcp, kind );
+			break;
+		}
+		data += option->length;
+	}
+}
+
+/**
+ * Consume received sequence space
+ *
+ * @v tcp		TCP connection
+ * @v seq_len		Sequence space length to consume
+ */
+static void tcp_rx_seq ( struct tcp_connection *tcp, uint32_t seq_len ) {
+	tcp->rcv_ack += seq_len;
+	if ( tcp->rcv_win > seq_len ) {
+		tcp->rcv_win -= seq_len;
+	} else {
+		tcp->rcv_win = 0;
+	}
+	tcp->flags |= TCP_ACK_PENDING;
+}
+
+/**
+ * Handle TCP received SYN
+ *
+ * @v tcp		TCP connection
+ * @v seq		SEQ value (in host-endian order)
+ * @v options		TCP options
+ * @ret rc		Return status code
+ */
+static int tcp_rx_syn ( struct tcp_connection *tcp, uint32_t seq,
+			struct tcp_options *options ) {
+
+	/* Synchronise sequence numbers on first SYN */
+	if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
+		tcp->rcv_ack = seq;
+		if ( options->tsopt )
+			tcp->flags |= TCP_TS_ENABLED;
+	}
+
+	/* Ignore duplicate SYN */
+	if ( seq != tcp->rcv_ack )
+		return 0;
+
+	/* Acknowledge SYN */
+	tcp_rx_seq ( tcp, 1 );
+
+	/* Mark SYN as received and start sending ACKs with each packet */
+	tcp->tcp_state |= ( TCP_STATE_SENT ( TCP_ACK ) |
+			    TCP_STATE_RCVD ( TCP_SYN ) );
+
+	return 0;
+}
+
+/**
+ * Handle TCP received ACK
+ *
+ * @v tcp		TCP connection
+ * @v ack		ACK value (in host-endian order)
+ * @v win		WIN value (in host-endian order)
+ * @ret rc		Return status code
+ */
+static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
+			uint32_t win ) {
+	uint32_t ack_len = ( ack - tcp->snd_seq );
+	size_t len;
+	unsigned int acked_flags;
+
+	/* Check for out-of-range or old duplicate ACKs */
+	if ( ack_len > tcp->snd_sent ) {
+		DBGC ( tcp, "TCP %p received ACK for %08x..%08x, "
+		       "sent only %08x..%08x\n", tcp, tcp->snd_seq,
+		       ( tcp->snd_seq + ack_len ), tcp->snd_seq,
+		       ( tcp->snd_seq + tcp->snd_sent ) );
+
+		if ( TCP_HAS_BEEN_ESTABLISHED ( tcp->tcp_state ) ) {
+			/* Just ignore what might be old duplicate ACKs */
+			return 0;
+		} else {
+			/* Send RST if an out-of-range ACK is received
+			 * on a not-yet-established connection, as per
+			 * RFC 793.
+			 */
+			return -EINVAL;
+		}
+	}
+
+	/* Ignore ACKs that don't actually acknowledge any new data.
+	 * (In particular, do not stop the retransmission timer; this
+	 * avoids creating a sorceror's apprentice syndrome when a
+	 * duplicate ACK is received and we still have data in our
+	 * transmit queue.)
+	 */
+	if ( ack_len == 0 )
+		return 0;
+
+	/* Stop the retransmission timer */
+	stop_timer ( &tcp->timer );
+
+	/* Determine acknowledged flags and data length */
+	len = ack_len;
+	acked_flags = ( TCP_FLAGS_SENDING ( tcp->tcp_state ) &
+			( TCP_SYN | TCP_FIN ) );
+	if ( acked_flags )
+		len--;
+
+	/* Update SEQ and sent counters, and window size */
+	tcp->snd_seq = ack;
+	tcp->snd_sent = 0;
+	tcp->snd_win = win;
+
+	/* Remove any acknowledged data from transmit queue */
+	tcp_process_tx_queue ( tcp, len, NULL, 1 );
+		
+	/* Mark SYN/FIN as acknowledged if applicable. */
+	if ( acked_flags )
+		tcp->tcp_state |= TCP_STATE_ACKED ( acked_flags );
+
+	/* Start sending FIN if we've had all possible data ACKed */
+	if ( list_empty ( &tcp->tx_queue ) && ( tcp->flags & TCP_XFER_CLOSED ) )
+		tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
+
+	return 0;
+}
+
+/**
+ * Handle TCP received data
+ *
+ * @v tcp		TCP connection
+ * @v seq		SEQ value (in host-endian order)
+ * @v iobuf		I/O buffer
+ * @ret rc		Return status code
+ *
+ * This function takes ownership of the I/O buffer.
+ */
+static int tcp_rx_data ( struct tcp_connection *tcp, uint32_t seq,
+			 struct io_buffer *iobuf ) {
+	uint32_t already_rcvd;
+	uint32_t len;
+	int rc;
+
+	/* Ignore duplicate or out-of-order data */
+	already_rcvd = ( tcp->rcv_ack - seq );
+	len = iob_len ( iobuf );
+	if ( already_rcvd >= len ) {
+		free_iob ( iobuf );
+		return 0;
+	}
+	iob_pull ( iobuf, already_rcvd );
+	len -= already_rcvd;
+
+	/* Acknowledge new data */
+	tcp_rx_seq ( tcp, len );
+
+	/* Deliver data to application */
+	if ( ( rc = xfer_deliver_iob ( &tcp->xfer, iobuf ) ) != 0 ) {
+		DBGC ( tcp, "TCP %p could not deliver %08x..%08x: %s\n",
+		       tcp, seq, ( seq + len ), strerror ( rc ) );
+		return rc;
+	}
+
+	return 0;
+}
+
+/**
+ * Handle TCP received FIN
+ *
+ * @v tcp		TCP connection
+ * @v seq		SEQ value (in host-endian order)
+ * @ret rc		Return status code
+ */
+static int tcp_rx_fin ( struct tcp_connection *tcp, uint32_t seq ) {
+
+	/* Ignore duplicate or out-of-order FIN */
+	if ( seq != tcp->rcv_ack )
+		return 0;
+
+	/* Acknowledge FIN */
+	tcp_rx_seq ( tcp, 1 );
+
+	/* Mark FIN as received */
+	tcp->tcp_state |= TCP_STATE_RCVD ( TCP_FIN );
+
+	/* Close connection */
+	tcp_close ( tcp, 0 );
+
+	return 0;
+}
+
+/**
+ * Handle TCP received RST
+ *
+ * @v tcp		TCP connection
+ * @v seq		SEQ value (in host-endian order)
+ * @ret rc		Return status code
+ */
+static int tcp_rx_rst ( struct tcp_connection *tcp, uint32_t seq ) {
+
+	/* Accept RST only if it falls within the window.  If we have
+	 * not yet received a SYN, then we have no window to test
+	 * against, so fall back to checking that our SYN has been
+	 * ACKed.
+	 */
+	if ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) {
+		if ( ! tcp_in_window ( seq, tcp->rcv_ack, tcp->rcv_win ) )
+			return 0;
+	} else {
+		if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
+			return 0;
+	}
+
+	/* Abort connection */
+	tcp->tcp_state = TCP_CLOSED;
+	tcp_dump_state ( tcp );
+	tcp_close ( tcp, -ECONNRESET );
+
+	DBGC ( tcp, "TCP %p connection reset by peer\n", tcp );
+	return -ECONNRESET;
+}
+
+/**
+ * Enqueue received TCP packet
+ *
+ * @v tcp		TCP connection
+ * @v seq		SEQ value (in host-endian order)
+ * @v flags		TCP flags
+ * @v iobuf		I/O buffer
+ */
+static void tcp_rx_enqueue ( struct tcp_connection *tcp, uint32_t seq,
+			     uint8_t flags, struct io_buffer *iobuf ) {
+	struct tcp_rx_queued_header *tcpqhdr;
+	struct io_buffer *queued;
+	size_t len;
+	uint32_t seq_len;
+
+	/* Calculate remaining flags and sequence length.  Note that
+	 * SYN, if present, has already been processed by this point.
+	 */
+	flags &= TCP_FIN;
+	len = iob_len ( iobuf );
+	seq_len = ( len + ( flags ? 1 : 0 ) );
+
+	/* Discard immediately (to save memory) if:
+	 *
+	 * a) we have not yet received a SYN (and so have no defined
+	 *    receive window), or
+	 * b) the packet lies entirely outside the receive window, or
+	 * c) there is no further content to process.
+	 */
+	if ( ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) ||
+	     ( tcp_cmp ( seq, tcp->rcv_ack + tcp->rcv_win ) >= 0 ) ||
+	     ( tcp_cmp ( seq + seq_len, tcp->rcv_ack ) < 0 ) ||
+	     ( seq_len == 0 ) ) {
+		free_iob ( iobuf );
+		return;
+	}
+
+	/* Add internal header */
+	tcpqhdr = iob_push ( iobuf, sizeof ( *tcpqhdr ) );
+	tcpqhdr->seq = seq;
+	tcpqhdr->flags = flags;
+
+	/* Add to RX queue */
+	list_for_each_entry ( queued, &tcp->rx_queue, list ) {
+		tcpqhdr = queued->data;
+		if ( tcp_cmp ( seq, tcpqhdr->seq ) < 0 )
+			break;
+	}
+	list_add_tail ( &iobuf->list, &queued->list );
+}
+
+/**
+ * Process receive queue
+ *
+ * @v tcp		TCP connection
+ */
+static void tcp_process_rx_queue ( struct tcp_connection *tcp ) {
+	struct io_buffer *iobuf;
+	struct tcp_rx_queued_header *tcpqhdr;
+	uint32_t seq;
+	unsigned int flags;
+	size_t len;
+
+	/* Process all applicable received buffers.  Note that we
+	 * cannot use list_for_each_entry() to iterate over the RX
+	 * queue, since tcp_discard() may remove packets from the RX
+	 * queue while we are processing.
+	 */
+	while ( ( iobuf = list_first_entry ( &tcp->rx_queue, struct io_buffer,
+					     list ) ) ) {
+
+		/* Stop processing when we hit the first gap */
+		tcpqhdr = iobuf->data;
+		if ( tcp_cmp ( tcpqhdr->seq, tcp->rcv_ack ) > 0 )
+			break;
+
+		/* Strip internal header and remove from RX queue */
+		list_del ( &iobuf->list );
+		seq = tcpqhdr->seq;
+		flags = tcpqhdr->flags;
+		iob_pull ( iobuf, sizeof ( *tcpqhdr ) );
+		len = iob_len ( iobuf );
+
+		/* Handle new data, if any */
+		tcp_rx_data ( tcp, seq, iob_disown ( iobuf ) );
+		seq += len;
+
+		/* Handle FIN, if present */
+		if ( flags & TCP_FIN ) {
+			tcp_rx_fin ( tcp, seq );
+			seq++;
+		}
+	}
+}
+
+/**
+ * Process received packet
+ *
+ * @v iobuf		I/O buffer
+ * @v st_src		Partially-filled source address
+ * @v st_dest		Partially-filled destination address
+ * @v pshdr_csum	Pseudo-header checksum
+ * @ret rc		Return status code
+  */
+static int tcp_rx ( struct io_buffer *iobuf,
+		    struct sockaddr_tcpip *st_src,
+		    struct sockaddr_tcpip *st_dest __unused,
+		    uint16_t pshdr_csum ) {
+	struct tcp_header *tcphdr = iobuf->data;
+	struct tcp_connection *tcp;
+	struct tcp_options options;
+	size_t hlen;
+	uint16_t csum;
+	uint32_t seq;
+	uint32_t ack;
+	uint32_t win;
+	unsigned int flags;
+	size_t len;
+	uint32_t seq_len;
+	int rc;
+
+	/* Sanity check packet */
+	if ( iob_len ( iobuf ) < sizeof ( *tcphdr ) ) {
+		DBG ( "TCP packet too short at %zd bytes (min %zd bytes)\n",
+		      iob_len ( iobuf ), sizeof ( *tcphdr ) );
+		rc = -EINVAL;
+		goto discard;
+	}
+	hlen = ( ( tcphdr->hlen & TCP_MASK_HLEN ) / 16 ) * 4;
+	if ( hlen < sizeof ( *tcphdr ) ) {
+		DBG ( "TCP header too short at %zd bytes (min %zd bytes)\n",
+		      hlen, sizeof ( *tcphdr ) );
+		rc = -EINVAL;
+		goto discard;
+	}
+	if ( hlen > iob_len ( iobuf ) ) {
+		DBG ( "TCP header too long at %zd bytes (max %zd bytes)\n",
+		      hlen, iob_len ( iobuf ) );
+		rc = -EINVAL;
+		goto discard;
+	}
+	csum = tcpip_continue_chksum ( pshdr_csum, iobuf->data,
+				       iob_len ( iobuf ) );
+	if ( csum != 0 ) {
+		DBG ( "TCP checksum incorrect (is %04x including checksum "
+		      "field, should be 0000)\n", csum );
+		rc = -EINVAL;
+		goto discard;
+	}
+	
+	/* Parse parameters from header and strip header */
+	tcp = tcp_demux ( ntohs ( tcphdr->dest ) );
+	seq = ntohl ( tcphdr->seq );
+	ack = ntohl ( tcphdr->ack );
+	win = ntohs ( tcphdr->win );
+	flags = tcphdr->flags;
+	tcp_rx_opts ( tcp, ( ( ( void * ) tcphdr ) + sizeof ( *tcphdr ) ),
+		      ( hlen - sizeof ( *tcphdr ) ), &options );
+	iob_pull ( iobuf, hlen );
+	len = iob_len ( iobuf );
+	seq_len = ( len + ( ( flags & TCP_SYN ) ? 1 : 0 ) +
+		    ( ( flags & TCP_FIN ) ? 1 : 0 ) );
+
+	/* Dump header */
+	DBGC2 ( tcp, "TCP %p RX %d<-%d           %08x %08x..%08x %4zd",
+		tcp, ntohs ( tcphdr->dest ), ntohs ( tcphdr->src ),
+		ntohl ( tcphdr->ack ), ntohl ( tcphdr->seq ),
+		( ntohl ( tcphdr->seq ) + seq_len ), len );
+	tcp_dump_flags ( tcp, tcphdr->flags );
+	DBGC2 ( tcp, "\n" );
+
+	/* If no connection was found, send RST */
+	if ( ! tcp ) {
+		tcp_xmit_reset ( tcp, st_src, tcphdr );
+		rc = -ENOTCONN;
+		goto discard;
+	}
+
+	/* Update timestamp, if applicable */
+	if ( options.tsopt && tcp_in_window ( tcp->rcv_ack, seq, seq_len ) )
+		tcp->ts_recent = ntohl ( options.tsopt->tsval );
+
+	/* Handle ACK, if present */
+	if ( flags & TCP_ACK ) {
+		if ( ( rc = tcp_rx_ack ( tcp, ack, win ) ) != 0 ) {
+			tcp_xmit_reset ( tcp, st_src, tcphdr );
+			goto discard;
+		}
+	}
+
+	/* Force an ACK if this packet is out of order */
+	if ( ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) &&
+	     ( seq != tcp->rcv_ack ) ) {
+		tcp->flags |= TCP_ACK_PENDING;
+	}
+
+	/* Handle SYN, if present */
+	if ( flags & TCP_SYN ) {
+		tcp_rx_syn ( tcp, seq, &options );
+		seq++;
+	}
+
+	/* Handle RST, if present */
+	if ( flags & TCP_RST ) {
+		if ( ( rc = tcp_rx_rst ( tcp, seq ) ) != 0 )
+			goto discard;
+	}
+
+	/* Enqueue received data */
+	tcp_rx_enqueue ( tcp, seq, flags, iob_disown ( iobuf ) );
+
+	/* Process receive queue */
+	tcp_process_rx_queue ( tcp );
+
+	/* Dump out any state change as a result of the received packet */
+	tcp_dump_state ( tcp );
+
+	/* Send out any pending data */
+	tcp_xmit ( tcp );
+
+	/* If this packet was the last we expect to receive, set up
+	 * timer to expire and cause the connection to be freed.
+	 */
+	if ( TCP_CLOSED_GRACEFULLY ( tcp->tcp_state ) ) {
+		stop_timer ( &tcp->wait );
+		start_timer_fixed ( &tcp->wait, ( 2 * TCP_MSL ) );
+	}
+
+	return 0;
+
+ discard:
+	/* Free received packet */
+	free_iob ( iobuf );
+	return rc;
+}
+
+/** TCP protocol */
+struct tcpip_protocol tcp_protocol __tcpip_protocol = {
+	.name = "TCP",
+	.rx = tcp_rx,
+	.tcpip_proto = IP_TCP,
+};
+
+/**
+ * Discard some cached TCP data
+ *
+ * @ret discarded	Number of cached items discarded
+ */
+static unsigned int tcp_discard ( void ) {
+	struct tcp_connection *tcp;
+	struct io_buffer *iobuf;
+	unsigned int discarded = 0;
+
+	/* Try to drop one queued RX packet from each connection */
+	list_for_each_entry ( tcp, &tcp_conns, list ) {
+		list_for_each_entry_reverse ( iobuf, &tcp->rx_queue, list ) {
+			list_del ( &iobuf->list );
+			free_iob ( iobuf );
+			discarded++;
+			break;
+		}
+	}
+
+	return discarded;
+}
+
+/** TCP cache discarder */
+struct cache_discarder tcp_cache_discarder __cache_discarder = {
+	.discard = tcp_discard,
+};
+
+/***************************************************************************
+ *
+ * Data transfer interface
+ *
+ ***************************************************************************
+ */
+
+/**
+ * Close interface
+ *
+ * @v tcp		TCP connection
+ * @v rc		Reason for close
+ */
+static void tcp_xfer_close ( struct tcp_connection *tcp, int rc ) {
+
+	/* Close data transfer interface */
+	tcp_close ( tcp, rc );
+
+	/* Transmit FIN, if possible */
+	tcp_xmit ( tcp );
+}
+
+/**
+ * Check flow control window
+ *
+ * @v tcp		TCP connection
+ * @ret len		Length of window
+ */
+static size_t tcp_xfer_window ( struct tcp_connection *tcp ) {
+
+	/* Not ready if data queue is non-empty.  This imposes a limit
+	 * of only one unACKed packet in the TX queue at any time; we
+	 * do this to conserve memory usage.
+	 */
+	if ( ! list_empty ( &tcp->tx_queue ) )
+		return 0;
+
+	/* Return TCP window length */
+	return tcp_xmit_win ( tcp );
+}
+
+/**
+ * Deliver datagram as I/O buffer
+ *
+ * @v tcp		TCP connection
+ * @v iobuf		Datagram I/O buffer
+ * @v meta		Data transfer metadata
+ * @ret rc		Return status code
+ */
+static int tcp_xfer_deliver ( struct tcp_connection *tcp,
+			      struct io_buffer *iobuf,
+			      struct xfer_metadata *meta __unused ) {
+
+	/* Enqueue packet */
+	list_add_tail ( &iobuf->list, &tcp->tx_queue );
+
+	/* Transmit data, if possible */
+	tcp_xmit ( tcp );
+
+	return 0;
+}
+
+/** TCP data transfer interface operations */
+static struct interface_operation tcp_xfer_operations[] = {
+	INTF_OP ( xfer_deliver, struct tcp_connection *, tcp_xfer_deliver ),
+	INTF_OP ( xfer_window, struct tcp_connection *, tcp_xfer_window ),
+	INTF_OP ( intf_close, struct tcp_connection *, tcp_xfer_close ),
+};
+
+/** TCP data transfer interface descriptor */
+static struct interface_descriptor tcp_xfer_desc =
+	INTF_DESC ( struct tcp_connection, xfer, tcp_xfer_operations );
+
+/***************************************************************************
+ *
+ * Openers
+ *
+ ***************************************************************************
+ */
+
+/** TCP socket opener */
+struct socket_opener tcp_socket_opener __socket_opener = {
+	.semantics	= TCP_SOCK_STREAM,
+	.family		= AF_INET,
+	.open		= tcp_open,
+};
+
+/** Linkage hack */
+int tcp_sock_stream = TCP_SOCK_STREAM;
+
+/**
+ * Open TCP URI
+ *
+ * @v xfer		Data transfer interface
+ * @v uri		URI
+ * @ret rc		Return status code
+ */
+static int tcp_open_uri ( struct interface *xfer, struct uri *uri ) {
+	struct sockaddr_tcpip peer;
+
+	/* Sanity check */
+	if ( ! uri->host )
+		return -EINVAL;
+
+	memset ( &peer, 0, sizeof ( peer ) );
+	peer.st_port = htons ( uri_port ( uri, 0 ) );
+	return xfer_open_named_socket ( xfer, SOCK_STREAM,
+					( struct sockaddr * ) &peer,
+					uri->host, NULL );
+}
+
+/** TCP URI opener */
+struct uri_opener tcp_uri_opener __uri_opener = {
+	.scheme		= "tcp",
+	.open		= tcp_open_uri,
+};
+