Network API now allows for multiple network devices (although the

implementation allows for only one, and does so without compromising on
the efficiency of static allocation).

Link-layer protocols are cleanly separated from the device drivers.

Network-layer protocols are cleanly separated from individual network
devices.

Link-layer and network-layer protocols are cleanly separated from each
other.

src/drivers/net/pnic.c

@@ -12,21 +12,21 @@ Bochs Pseudo NIC driver for Etherboot
  * See pnic_api.h for an explanation of the Bochs Pseudo NIC.
  */
 
-/* to get some global routines like printf */
-#include "etherboot.h"
-/* to get the interface to the body of the program */
-#include "nic.h"
-/* to get the PCI support functions, if this is a PCI NIC */
-#include "pci.h"
-
-/* PNIC API */
-#include "pnic_api.h"
+#include <stdint.h>
+#include <io.h>
+#include <vsprintf.h>
+#include <errno.h>
+#include <gpxe/pci.h>
+#include <gpxe/if_ether.h>
+#include <gpxe/ethernet.h>
+#include <gpxe/pkbuff.h>
+#include <gpxe/netdevice.h>
 
-/* Function prototypes */
-static int pnic_api_check ( uint16_t api_version );
+#include "pnic_api.h"
 
-/* NIC specific static variables go here */
-uint8_t tx_buffer[ETH_FRAME_LEN] __shared;
+struct pnic {
+	unsigned short ioaddr;
+};
 
 /* 
  * Utility functions: issue a PNIC command, retrieve result.  Use
@@ -40,7 +40,7 @@ uint8_t tx_buffer[ETH_FRAME_LEN] __shared;
  * of data).
  */
 
-static uint16_t pnic_command_quiet ( struct nic *nic, uint16_t command,
+static uint16_t pnic_command_quiet ( struct pnic *pnic, uint16_t command,
 				     void *input, uint16_t input_length,
 				     void *output, uint16_t output_max_length,
 				     uint16_t *output_length ) {
@@ -50,18 +50,19 @@ static uint16_t pnic_command_quiet ( struct nic *nic, uint16_t command,
 
 	if ( input != NULL ) {
 		/* Write input length */
-		outw ( input_length, nic->ioaddr + PNIC_REG_LEN );
+		outw ( input_length, pnic->ioaddr + PNIC_REG_LEN );
 		/* Write input data */
 		for ( i = 0; i < input_length; i++ ) {
-			outb( ((char*)input)[i], nic->ioaddr + PNIC_REG_DATA );
+			outb( ((char*)input)[i],
+			      pnic->ioaddr + PNIC_REG_DATA );
 		}
 	}
 	/* Write command */
-	outw ( command, nic->ioaddr + PNIC_REG_CMD );
+	outw ( command, pnic->ioaddr + PNIC_REG_CMD );
 	/* Retrieve status */
-	status = inw ( nic->ioaddr + PNIC_REG_STAT );
+	status = inw ( pnic->ioaddr + PNIC_REG_STAT );
 	/* Retrieve output length */
-	_output_length = inw ( nic->ioaddr + PNIC_REG_LEN );
+	_output_length = inw ( pnic->ioaddr + PNIC_REG_LEN );
 	if ( output_length == NULL ) {
 		if ( _output_length != output_max_length ) {
 			printf ( "pnic_command %#hx: wrong data length "
@@ -81,17 +82,17 @@ static uint16_t pnic_command_quiet ( struct nic *nic, uint16_t command,
 		/* Retrieve output data */
 		for ( i = 0; i < _output_length; i++ ) {
 			((char*)output)[i] =
-				inb ( nic->ioaddr + PNIC_REG_DATA );
+				inb ( pnic->ioaddr + PNIC_REG_DATA );
 		}
 	}
 	return status;
 }
 
-static uint16_t pnic_command ( struct nic *nic, uint16_t command,
+static uint16_t pnic_command ( struct pnic *pnic, uint16_t command,
 			       void *input, uint16_t input_length,
 			       void *output, uint16_t output_max_length,
 			       uint16_t *output_length ) {
-	uint16_t status = pnic_command_quiet ( nic, command,
+	uint16_t status = pnic_command_quiet ( pnic, command,
 					       input, input_length,
 					       output, output_max_length,
 					       output_length );
@@ -110,135 +111,134 @@ static int pnic_api_check ( uint16_t api_version ) {
 			 PNIC_API_VERSION >> 8, PNIC_API_VERSION & 0xff );
 	}
 	if ( api_version < PNIC_API_VERSION ) {
-		printf ( "*** You may need to update your copy of Bochs ***\n" );
+		printf ( "** You may need to update your copy of Bochs **\n" );
 	}
 	return ( api_version == PNIC_API_VERSION );
 }
 
-/**************************************************************************
-CONNECT - connect adapter to the network
-***************************************************************************/
-static int pnic_connect ( struct nic *nic __unused ) {
-	/* Nothing to do */
-	return 1;
-}
-
 /**************************************************************************
 POLL - Wait for a frame
 ***************************************************************************/
-static int pnic_poll ( struct nic *nic, int retrieve ) {
+static void pnic_poll ( struct net_device *netdev ) {
+	struct pnic *pnic = netdev->priv;
+	struct pk_buff *pkb;
 	uint16_t length;
 	uint16_t qlen;
 
-	/* Check receive queue length to see if there's anything to
-	 * get.  Necessary since once we've called PNIC_CMD_RECV we
-	 * have to read out the packet, otherwise it's lost forever.
-	 */
-	if ( pnic_command ( nic, PNIC_CMD_RECV_QLEN, NULL, 0,
-			    &qlen, sizeof(qlen), NULL )
-	     != PNIC_STATUS_OK ) return ( 0 );
-	if ( qlen == 0 ) return ( 0 );
-
-	/* There is a packet ready.  Return 1 if we're only checking. */
-	if ( ! retrieve ) return ( 1 );
-
-	/* Retrieve the packet */
-	if ( pnic_command ( nic, PNIC_CMD_RECV, NULL, 0,
-			    nic->packet, ETH_FRAME_LEN, &length )
-	     != PNIC_STATUS_OK ) return ( 0 );
-	nic->packetlen = length;
-	return ( 1 );
+	/* Fetch all available packets */
+	while ( 1 ) {
+		if ( pnic_command ( pnic, PNIC_CMD_RECV_QLEN, NULL, 0,
+				    &qlen, sizeof ( qlen ), NULL )
+		     != PNIC_STATUS_OK )
+			break;
+		if ( qlen == 0 )
+			break;
+		pkb = alloc_pkb ( ETH_FRAME_LEN );
+		if ( ! pkb )
+			break;
+		if ( pnic_command ( pnic, PNIC_CMD_RECV, NULL, 0,
+				    pkb->data, ETH_FRAME_LEN, &length )
+		     != PNIC_STATUS_OK ) {
+			free_pkb ( pkb );
+			break;
+		}
+		pkb_put ( pkb, length );
+		netdev_rx ( netdev, pkb );
+	}
 }
 
 /**************************************************************************
 TRANSMIT - Transmit a frame
 ***************************************************************************/
-static void pnic_transmit ( struct nic *nic, const char *dest,
-			    unsigned int type, unsigned int size,
-			    const char *data ) {
-	unsigned int nstype = htons ( type );
-
-	if ( ( ETH_HLEN + size ) >= ETH_FRAME_LEN ) {
-		printf ( "pnic_transmit: packet too large\n" );
-		return;
-	}
+static int pnic_transmit ( struct net_device *netdev, struct pk_buff *pkb ) {
+	struct pnic *pnic = netdev->priv;
 
-	/* Assemble packet */
-	memcpy ( tx_buffer, dest, ETH_ALEN );
-	memcpy ( tx_buffer + ETH_ALEN, nic->node_addr, ETH_ALEN );
-	memcpy ( tx_buffer + 2 * ETH_ALEN, &nstype, 2 );
-	memcpy ( tx_buffer + ETH_HLEN, data, size );
-
-	pnic_command ( nic, PNIC_CMD_XMIT, tx_buffer, ETH_HLEN + size,
+	pnic_command ( pnic, PNIC_CMD_XMIT, pkb, pkb_len ( pkb ),
 		       NULL, 0, NULL );
-}
-
-/**************************************************************************
-DISABLE - Turn off ethernet interface
-***************************************************************************/
-static void pnic_disable ( struct nic *nic, struct pci_device *pci __unused ) {
-	nic_disable ( nic );
-	pnic_command ( nic, PNIC_CMD_RESET, NULL, 0, NULL, 0, NULL );
+	free_pkb ( pkb );
+	return 0;
 }
 
 /**************************************************************************
 IRQ - Handle card interrupt status
 ***************************************************************************/
-static void pnic_irq ( struct nic *nic, irq_action_t action ) {
+#if 0
+static void pnic_irq ( struct net_device *netdev, irq_action_t action ) {
+	struct pnic *pnic = netdev->priv;
 	uint8_t enabled;
 
 	switch ( action ) {
 	case DISABLE :
 	case ENABLE :
 		enabled = ( action == ENABLE ? 1 : 0 );
-		pnic_command ( nic, PNIC_CMD_MASK_IRQ,
-			       &enabled, sizeof(enabled), NULL, 0, NULL );
+		pnic_command ( pnic, PNIC_CMD_MASK_IRQ,
+			       &enabled, sizeof ( enabled ), NULL, 0, NULL );
 		break;
 	case FORCE :
-		pnic_command ( nic, PNIC_CMD_FORCE_IRQ,
+		pnic_command ( pnic, PNIC_CMD_FORCE_IRQ,
 			       NULL, 0, NULL, 0, NULL );
 		break;
 	}
 }
+#endif
 
 /**************************************************************************
-NIC operations table
+DISABLE - Turn off ethernet interface
 ***************************************************************************/
-static struct nic_operations pnic_operations = {
-	.connect	= pnic_connect,
-	.poll		= pnic_poll,
-	.transmit	= pnic_transmit,
-	.irq		= pnic_irq,
-};
+static void pnic_remove ( struct pci_device *pci ) {
+	struct net_device *netdev = pci_get_drvdata ( pci );
+	struct pnic *pnic = netdev->priv;
+
+	unregister_netdev ( netdev );
+	pnic_command ( pnic, PNIC_CMD_RESET, NULL, 0, NULL, 0, NULL );
+}
 
 /**************************************************************************
 PROBE - Look for an adapter, this routine's visible to the outside
 ***************************************************************************/
-static int pnic_probe ( struct nic *nic, struct pci_device *pci ) {
+static int pnic_probe ( struct pci_device *pci ) {
+	struct net_device *netdev;
+	struct pnic *pnic;
 	uint16_t api_version;
 	uint16_t status;
+	int rc;
 
-	/* Retrieve relevant information about PCI device */
-	pci_fill_nic ( nic, pci );
+	/* Allocate net device */
+	netdev = alloc_etherdev ( sizeof ( *pnic ) );
+	if ( ! netdev ) {
+		rc = -ENOMEM;
+		goto err;
+	}
+	pnic = netdev->priv;
+	pci_set_drvdata ( pci, netdev );
+	pnic->ioaddr = pci->ioaddr;
 
 	/* API version check */
-	status = pnic_command_quiet( nic, PNIC_CMD_API_VER, NULL, 0,
-				     &api_version,
-				     sizeof(api_version), NULL );
+	status = pnic_command_quiet ( pnic, PNIC_CMD_API_VER, NULL, 0,
+				      &api_version,
+				      sizeof ( api_version ), NULL );
 	if ( status != PNIC_STATUS_OK ) {
 		printf ( "PNIC failed installation check, code %#hx\n",
 			 status );
-		return 0;
+		rc = -EIO;
+		goto err;
 	}
-	pnic_api_check(api_version);
+	pnic_api_check ( api_version );
 
 	/* Get MAC address */
-	status = pnic_command ( nic, PNIC_CMD_READ_MAC, NULL, 0,
-				nic->node_addr, ETH_ALEN, NULL );
+	status = pnic_command ( pnic, PNIC_CMD_READ_MAC, NULL, 0,
+				netdev->ll_addr, ETH_ALEN, NULL );
+
+	/* Point to NIC specific routines */
+	netdev->poll	 = pnic_poll;
+	netdev->transmit = pnic_transmit;
+
+	return 0;
 
-	/* point to NIC specific routines */
-	nic->nic_op	= &pnic_operations;
-	return 1;
+ err:
+	/* Free net device */
+	free_netdev ( netdev );
+	return rc;
 }
 
 static struct pci_id pnic_nics[] = {
@@ -246,7 +246,18 @@ static struct pci_id pnic_nics[] = {
 PCI_ROM ( 0xfefe, 0xefef, "pnic", "Bochs Pseudo NIC Adaptor" ),
 };
 
-PCI_DRIVER ( pnic_driver, pnic_nics, PCI_NO_CLASS );
+static struct pci_driver pnic_driver = {
+	.ids = pnic_nics,
+	.id_count = ( sizeof ( pnic_nics ) / sizeof ( pnic_nics[0] ) ),
+	.class = PCI_NO_CLASS,
+	//	.probe = pnic_probe,
+	//	.remove = pnic_remove,
+};
+
+// PCI_DRIVER ( pnic_driver );
+
+#include "dev.h"
+extern struct type_driver test_driver;
 
-DRIVER ( "PNIC", nic_driver, pci_driver, pnic_driver,
-	 pnic_probe, pnic_disable );
+DRIVER ( "PNIC", test_driver, pci_driver, pnic_driver,
+	 pnic_probe, pnic_remove );

src/include/gpxe/arp.h

@@ -1,5 +1,5 @@
-#ifndef _ARP_H
-#define _ARP_H
+#ifndef _GPXE_ARP_H
+#define _GPXE_ARP_H
 
 /** @file
  *
@@ -7,17 +7,10 @@
  *
  */
 
-struct net_device;
-struct net_interface;
-struct pk_buff;
+struct net_header;
+struct ll_header;
 
-extern int arp_resolve ( struct net_device *netdev, struct pk_buff *pkb,
-			 const void **ll_addr );
+extern int arp_resolve ( const struct net_header *nethdr,
+			 struct ll_header *llhdr );
 
-extern int arp_process ( struct net_interface *arp_netif,
-			 struct pk_buff *pkb );
-
-extern int arp_add_generic_header ( struct net_interface *arp_netif,
-				    struct pk_buff *pkb );
-
-#endif /* _ARP_H */
+#endif /* _GPXE_ARP_H */

src/include/gpxe/netdevice.h

@@ -1,203 +1,326 @@
-#ifndef _NETDEVICE_H
-#define _NETDEVICE_H
+#ifndef _GPXE_NETDEVICE_H
+#define _GPXE_NETDEVICE_H
 
 /** @file
  *
- * Network device and network interface
+ * Network device management
  *
  */
 
 #include <stdint.h>
-#include <gpxe/list.h>
+#include <gpxe/tables.h>
 
-struct net_device;
-struct net_interface;
 struct pk_buff;
+struct net_protocol;
+struct ll_protocol;
 
 /** Maximum length of a link-layer address */
-#define MAX_LLH_ADDR_LEN 6
+#define MAX_LL_ADDR_LEN 6
 
 /** Maximum length of a network-layer address */
 #define MAX_NET_ADDR_LEN 4
 
+/* Network-layer address may be required to hold a link-layer address
+ * (if NETADDR_FL_RAW is set
+ */
+#if MAX_NET_ADDR_LEN < MAX_LL_ADDR_LEN
+#undef MAX_NET_ADDR_LEN
+#define MAX_NET_ADDR_LEN MAX_LL_ADDR_LEN
+#endif
+
+/** A generic network-layer header */
+struct net_header {
+	/** Network-layer protocol */
+	struct net_protocol *net_protocol;
+	/** Destination address flags
+	 *
+	 * This is the bitwise OR of zero or more NETADDR_FL_XXX
+	 * values.
+	 */
+	int dest_flags;
+	/** Network-layer destination address */
+	uint8_t dest_net_addr[MAX_NET_ADDR_LEN];
+	/** Network-layer source address */
+	uint8_t source_net_addr[MAX_NET_ADDR_LEN];
+};
+
+/** Address is a broadcast address */
+#define NETADDR_FL_BROADCAST 0x01
+
+/** Address is a multicast address */
+#define NETADDR_FL_MULTICAST 0x02
+
+/** Address is a raw hardware address */
+#define NETADDR_FL_RAW 0x04
+
+/** A generic link-layer header */
+struct ll_header {
+	/** Link-layer protocol */
+	struct ll_protocol *ll_protocol;
+	/** Destination address flags
+	 *
+	 * This is the bitwise OR of zero or more NETADDR_FL_XXX
+	 * values.
+	 */
+	int dest_flags;
+	/** Link-layer destination address */
+	uint8_t dest_ll_addr[MAX_LL_ADDR_LEN];
+	/** Link-layer source address */
+	uint8_t source_ll_addr[MAX_LL_ADDR_LEN];
+	/** Network-layer protocol
+	 *
+	 *
+	 * This is an ETH_P_XXX constant, in network-byte order
+	 */
+	uint16_t net_proto;
+};
+
 /**
- * A network device
- *
- * This structure represents a piece of networking hardware.  It has
- * properties such as a link-layer address and methods for
- * transmitting and receiving raw packets.  It does not know anything
- * about network-layer protocols (e.g. IP) or their addresses; these
- * are handled by struct @c net_interface instead.
+ * A network-layer protocol
  *
- * Note that this structure must represent a generic network device,
- * not just an Ethernet device.
  */
-struct net_device {
-	/** Transmit packet
+struct net_protocol {
+	/**
+	 * Perform network-layer routing
 	 *
-	 * @v netdev	Network device
 	 * @v pkb	Packet buffer
+	 * @ret source	Network-layer source address
+	 * @ret dest	Network-layer destination address
 	 * @ret rc	Return status code
 	 *
-	 * This method should cause the hardware to initiate
-	 * transmission of the packet buffer.  The buffer may be
-	 * reused immediately after the method returns, and so the
-	 * method should either wait for packet transmission to
-	 * complete, or take a copy of the buffer contents.
+	 * This method should fill in the source and destination
+	 * addresses with enough information to allow the link layer
+	 * to route the packet.
+	 *
+	 * For example, in the case of IPv4, this method should fill
+	 * in @c source with the IP addresses of the local adapter and
+	 * @c dest with the next hop destination (e.g. the gateway).
 	 */
-	int ( * transmit ) ( struct net_device *netdev,
-			     struct pk_buff *pkb );
-	/** Poll for received packet
+	int ( * route ) ( const struct pk_buff *pkb,
+			  struct net_header *nethdr );
+	/**
+	 * Handle received packets
 	 *
-	 * @v netdev	Network device
-	 * @v pkb	Packet buffer to contain received packet
+	 * @v pkb	Packet buffer
 	 * @ret rc	Return status code
 	 *
-	 * This method should cause the hardware to check for a
-	 * received packet.  If no packet is available, the method
-	 * should return -EAGAIN (i.e. this method is *always*
-	 * considered to be a non-blocking read).  If a packet is
-	 * available, the method should fill the packet buffer and
-	 * return zero for success.
+	 * If this method returns success, it has taken ownership of
+	 * the packet buffer.
 	 */
-	int ( * poll ) ( struct net_device *netdev, struct pk_buff *pkb );
-	/** Build link-layer header
+	int ( * rx ) ( struct pk_buff *pkb );
+	/** Network-layer protocol
 	 *
-	 * @v netdev	Network device
-	 * @v pkb	Packet buffer
+	 * This is an ETH_P_XXX constant, in network-byte order
+	 */
+	uint16_t net_proto;
+	/** Network-layer address length */
+	uint8_t net_addr_len;
+};
+
+/**
+ * A link-layer protocol
+ *
+ */
+struct ll_protocol {
+	/**
+	 * Perform link-layer routing
+	 *
+	 * @v nethdr	Generic network-layer header
+	 * @ret llhdr	Generic link-layer header
 	 * @ret rc	Return status code
 	 *
-	 * This method should fill in the link-layer header based on
-	 * the metadata contained in @c pkb.
+	 * This method should construct the generic link-layer header
+	 * based on the generic network-layer header.
 	 *
 	 * If a link-layer header cannot be constructed (e.g. because
 	 * of a missing ARP cache entry), then this method should
 	 * return an error (after transmitting an ARP request, if
 	 * applicable).
 	 */
-	int ( * build_llh ) ( struct net_device *netdev, struct pk_buff *pkb );
-	/** Parse link-layer header
+	int ( * route ) ( const struct net_header *nethdr,
+			  struct ll_header *llhdr );
+	/**
+	 * Fill media-specific link-layer header
 	 *
-	 * @v netdev	Network device
+	 * @v llhdr	Generic link-layer header
 	 * @v pkb	Packet buffer
-	 * @ret rc	Return status code
 	 *
-	 * This method should parse the link-layer header and fill in
-	 * the metadata in @c pkb.
+	 * This method should fill in the link-layer header in the
+	 * packet buffer based on information in the generic
+	 * link-layer header.
+	 */
+	void ( * fill_llh ) ( const struct ll_header *llhdr,
+			      struct pk_buff *pkb );
+	/**
+	 * Parse media-specific link-layer header
+	 *
+	 * @v pkb	Packet buffer
+	 * @v llhdr	Generic link-layer header
+	 *
+	 * This method should fill in the generic link-layer header
+	 * based on information in the link-layer header in the packet
+	 * buffer.
 	 */
-	int ( * parse_llh ) ( struct net_device *netdev, struct pk_buff *pkb );
+	void ( * parse_llh ) ( const struct pk_buff *pkb,
+			       struct ll_header *llhdr );
+
 	/** Link-layer protocol
 	 *
 	 * This is an ARPHRD_XXX constant, in network byte order.
 	 */
 	uint16_t ll_proto;
-	/** Link-layer header length */
-	uint8_t ll_hlen;
 	/** Link-layer address length */
 	uint8_t ll_addr_len;
-	/** Link-layer address
-	 *
-	 * For Ethernet, this is the MAC address.
-	 */
-	uint8_t ll_addr[MAX_LLH_ADDR_LEN];
-	/** Linked list of network devices */
-	struct list_head devices;
-	/** List of network interfaces */
-	struct list_head interfaces;
-	/** Driver private data */
-	void *priv;
+	/** Link-layer header length */
+	uint8_t ll_header_len;
 };
 
 /**
- * A network interface
- *
- * This structure represents a particular network layer protocol's
- * interface to a piece of network hardware (a struct @c net_device).
+ * A network-layer address assigned to a network device
  *
  */
-struct net_interface {
-	/** Underlying net device */
-	struct net_device *netdev;
-	/** Linked list of interfaces for this device */
-	struct list_head interfaces;
-	/** Network-layer protocol
-	 *
-	 * This is an ETH_P_XXX constant, in network byte order.
-	 */
-	uint16_t net_proto;
-	/** Network-layer address length */
-	uint8_t net_addr_len;
+struct net_address {
+	/** Network-layer protocol */
+	struct net_protocol *net_protocol;
 	/** Network-layer address */
 	uint8_t net_addr[MAX_NET_ADDR_LEN];
-	/** Fill in packet metadata
+};
+
+/**
+ * A network device
+ *
+ * This structure represents a piece of networking hardware.  It has
+ * properties such as a link-layer address and methods for
+ * transmitting and receiving raw packets.
+ *
+ * Note that this structure must represent a generic network device,
+ * not just an Ethernet device.
+ */
+struct net_device {
+	/** Transmit packet
 	 *
-	 * @v netif	Network interface
+	 * @v netdev	Network device
 	 * @v pkb	Packet buffer
 	 * @ret rc	Return status code
 	 *
-	 * This method should fill in the @c pkb metadata with enough
-	 * information to enable net_device::build_llh to construct
-	 * the link-layer header.
+	 * This method should cause the hardware to initiate
+	 * transmission of the packet buffer.
+	 *
+	 * If the method returns success, ownership of the packet
+	 * buffer is transferred to the @c net_device, which must
+	 * eventually call free_pkb() to release the buffer.
 	 */
-	int ( * add_llh_metadata ) ( struct net_interface *netif,
-				     struct pk_buff *pkb );
-	/** Received packet processor
+	int ( * transmit ) ( struct net_device *netdev, struct pk_buff *pkb );
+	/** Poll for received packet
 	 *
-	 * @v netif	Network interface
-	 * @v pkb	Packet buffer
-	 * @ret rc	Return status code
+	 * @v netdev	Network device
 	 *
-	 * This method is called for packets arriving on the
-	 * associated network device that match this interface's
-	 * network-layer protocol.
+	 * This method should cause the hardware to check for received
+	 * packets.  Any received packets should be delivered via
+	 * netdev_rx().
+	 */
+	void ( * poll ) ( struct net_device *netdev );
+
+	/** Link-layer protocol */
+	struct ll_protocol *ll_protocol;
+	/** Link-layer address
 	 *
-	 * When this method is called, the link-layer header will
-	 * already have been stripped from the packet.
+	 * For Ethernet, this is the MAC address.
 	 */
-	int ( * rx_packet ) ( struct net_interface *netif,
-			      struct pk_buff *pkb );
+	uint8_t ll_addr[MAX_LL_ADDR_LEN];
+
+	/** Driver private data */
+	void *priv;
 };
 
+extern struct net_device static_single_netdev;
+
+/**
+ * Allocate network device
+ *
+ * @v priv_size		Size of private data area (net_device::priv)
+ * @ret netdev		Network device, or NULL
+ *
+ * Allocates space for a network device and its private data area.
+ *
+ * This macro allows for a very efficient implementation in the case
+ * of a single static network device; it neatly avoids dynamic
+ * allocation and can never return failure, meaning that the failure
+ * path will be optimised away.  However, driver writers should not
+ * rely on this feature; the drivers should be written to allow for
+ * multiple instances of network devices.
+ */
+#define alloc_netdev( priv_size ) ( {		\
+	static char priv_data[priv_size];	\
+	static_single_netdev.priv = priv_data;	\
+	&static_single_netdev; } )
+
 /**
- * Find interface for a specific protocol
+ * Register network device
  *
- * @v netdev	Network device
- * @v net_proto	Network-layer protocol, in network byte order
- * @ret netif	Network interface, or NULL if none found
+ * @v netdev		Network device
+ * @ret rc		Return status code
  *
+ * Adds the network device to the list of network devices.
  */
-static inline struct net_interface *
-netdev_find_netif ( const struct net_device *netdev, uint16_t net_proto ) {
-	struct net_interface *netif;
-
-	list_for_each_entry ( netif, &netdev->interfaces, interfaces ) {
-		if ( netif->net_proto == net_proto )
-			return netif;
-	}
-	return NULL;
+static inline int
+register_netdev ( struct net_device *netdev __attribute__ (( unused )) ) {
+	return 0;
 }
 
-extern int register_netdevice ( struct net_device *netdev );
-extern void unregister_netdevice ( struct net_device *netdev );
-extern int netdev_send ( struct net_device *netdev, struct pk_buff *pkb );
-extern int netdev_poll ( struct net_device *netdev, struct pk_buff *pkb );
-extern int netif_send ( struct net_interface *netif, struct pk_buff *pkb );
-extern int netdev_rx_packet ( struct net_device *netdev, struct pk_buff *pkb );
-extern int net_poll ( struct pk_buff *pkb, struct net_device **netdev );
+/**
+ * Unregister network device
+ *
+ * @v netdev		Network device
+ *
+ * Removes the network device from the list of network devices.
+ */
+static inline void 
+unregister_netdev ( struct net_device *netdev __attribute__ (( unused )) ) {
+	/* Nothing to do */
+}
 
+/**
+ * Free network device
+ *
+ * @v netdev		Network device
+ */
+static inline void
+free_netdev ( struct net_device *netdev __attribute__ (( unused )) ) {
+	/* Nothing to do */
+}
 
+/**
+ * Register a link-layer protocol
+ *
+ * @v protocol		Link-layer protocol
+ */
+#define LL_PROTOCOL( protocol ) \
+	struct ll_protocol protocol __table ( ll_protocols, 00 )
 
-extern struct net_device static_single_netdev;
+/**
+ * Register a network-layer protocol
+ *
+ * @v protocol		Network-layer protocol
+ */
+#define NET_PROTOCOL( protocol ) \
+	struct net_protocol protocol __table ( net_protocols, 00 )
 
-/* Must be a macro because priv_data[] is of variable size */
-#define alloc_netdevice( priv_size ) ( {		\
-	static char priv_data[priv_size];		\
-	static_single_netdev.priv = priv_data;	\
-	&static_single_netdev; } )
+/**
+ * Register a network-layer address for the static single network device
+ *
+ * @v net_address	Network-layer address
+ */
+#define STATIC_SINGLE_NETDEV_ADDRESS( address ) \
+	struct net_address address __table ( sgl_netdev_addresses, 00 )
 
+extern struct net_protocol *net_find_protocol ( uint16_t net_proto );
+extern struct net_device * net_find_address ( struct net_protocol *net_proto,
+					      void *net_addr );
 
-static inline void free_netdevice ( struct net_device *netdev __unused ) {
-	/* Do nothing */
-}
+extern int net_transmit ( struct pk_buff *pkb );
+extern int net_poll ( void );
+extern void netdev_rx ( struct net_device *netdev, struct pk_buff *pkb );
+extern struct pk_buff * net_rx_dequeue ( void );
 
-#endif /* _NETDEVICE_H */
+#endif /* _GPXE_NETDEVICE_H */

src/include/gpxe/pkbuff.h

@@ -1,5 +1,5 @@
-#ifndef _PKBUFF_H
-#define _PKBUFF_H
+#ifndef _GPXE_PKBUFF_H
+#define _GPXE_PKBUFF_H
 
 /** @file
  *
@@ -12,6 +12,10 @@
 
 #include <stdint.h>
 #include <assert.h>
+#include <gpxe/list.h>
+
+struct net_protocol;
+struct ll_protocol;
 
 /** A packet buffer
  *
@@ -27,38 +31,14 @@ struct pk_buff {
 	/** End of the buffer */
         void *end;
 
-	/** The network-layer protocol
-	 *
-	 * This is the network-layer protocol expressed as an
-	 * ETH_P_XXX constant, in network-byte order.
-	 */
-	uint16_t net_proto;
-	/** Flags
-	 *
-	 * Filled in only on outgoing packets.  Value is the
-	 * bitwise-OR of zero or more PKB_FL_XXX constants.
-	 */
-	uint8_t flags;
-	/** Network-layer address length 
-	 *
-	 * Filled in only on outgoing packets.
-	 */
-	uint8_t net_addr_len;
-	/** Network-layer address
-	 *
-	 * Filled in only on outgoing packets.
-	 */
-	void *net_addr;
-};
-
-/** Packet is a broadcast packet */
-#define PKB_FL_BROADCAST 0x01
+	/** List of which this buffer is a member */
+	struct list_head list;
 
-/** Packet is a multicast packet */
-#define PKB_FL_MULTICAST 0x02
-
-/** Network-layer address is a raw hardware address */
-#define PKB_FL_RAW_NET_ADDR 0x04
+	/** The network-layer protocol */
+	struct net_protocol *net_protocol;
+	/** The link-layer protocol */
+	struct ll_protocol *ll_protocol;
+};
 
 /**
  * Add data to start of packet buffer
@@ -130,4 +110,7 @@ static inline size_t pkb_len ( struct pk_buff *pkb ) {
 	return ( pkb->tail - pkb->data );
 }
 
-#endif /* _PKBUFF_H */
+extern struct pk_buff * alloc_pkb ( size_t len );
+extern void free_pkb ( struct pk_buff *pkb );
+
+#endif /* _GPXE_PKBUFF_H */

src/net/arp.c

@@ -39,13 +39,13 @@
 /** An ARP cache entry */
 struct arp_entry {
 	/** Network-layer protocol */
-	uint16_t net_proto;
+	struct net_protocol *net_protocol;
 	/** Link-layer protocol */
-	uint16_t ll_proto;
+	struct ll_protocol *ll_protocol;
 	/** Network-layer address */
 	uint8_t net_addr[MAX_NET_ADDR_LEN];
 	/** Link-layer address */
-	uint8_t ll_addr[MAX_LLH_ADDR_LEN];
+	uint8_t ll_addr[MAX_LL_ADDR_LEN];
 };
 
 /** Number of entries in the ARP cache
@@ -61,25 +61,28 @@ static struct arp_entry arp_table[NUM_ARP_ENTRIES];
 
 static unsigned int next_new_arp_entry = 0;
 
+struct net_protocol arp_protocol;
+
 /**
  * Find entry in the ARP cache
  *
- * @v ll_proto		Link-layer protocol
- * @v net_proto		Network-layer protocol
+ * @v ll_protocol	Link-layer protocol
+ * @v net_protocol	Network-layer protocol
  * @v net_addr		Network-layer address
- * @v net_addr_len	Network-layer address length
  * @ret arp		ARP cache entry, or NULL if not found
  *
  */
 static struct arp_entry *
-arp_find_entry ( uint16_t ll_proto, uint16_t net_proto, const void *net_addr,
-		 size_t net_addr_len ) {
+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_proto == ll_proto ) &&
-		     ( arp->net_proto == net_proto ) &&
-		     ( memcmp ( arp->net_addr, net_addr, net_addr_len ) == 0 ))
+		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;
@@ -88,59 +91,64 @@ arp_find_entry ( uint16_t ll_proto, uint16_t net_proto, const void *net_addr,
 /**
  * Look up media-specific link-layer address in the ARP cache
  *
- * @v netdev		Network device
- * @v pkb		Packet buffer
- * @ret ll_addr		Pointer to link-layer address
+ * @v nethdr		Generic network-layer header
+ * @ret llhdr		Generic link-layer header
  * @ret rc		Return status code
  *
  * This function will use the ARP cache to look up the link-layer
- * address for the media corresponding to @c netdev and the
- * network-layer address as specified in the @c pkb metadata.
+ * address for the link-layer protocol specified in @c llhdr and the
+ * network-layer protocol and address as specified in @c nethdr.  If
+ * found, the destination link-layer address will be filled in in @c
+ * llhdr.
  *
  * If no address is found in the ARP cache, an ARP request will be
- * transmitted, -ENOENT will be returned, and the packet buffer
- * contents will be undefined.
+ * transmitted and -ENOENT will be returned.
  */
-int arp_resolve ( struct net_device *netdev, struct pk_buff *pkb,
-		  const void **ll_addr ) {
+int arp_resolve ( const struct net_header *nethdr, struct ll_header *llhdr ) {
+	struct net_protocol *net_protocol = nethdr->net_protocol;
+	struct ll_protocol *ll_protocol = llhdr->ll_protocol;
 	const struct arp_entry *arp;
-	struct net_interface *netif;
+	struct pk_buff *pkb;
 	struct arphdr *arphdr;
+	int rc;
 
 	/* Look for existing entry in ARP table */
-	arp = arp_find_entry ( netdev->ll_proto, pkb->net_proto,
-			       pkb->net_addr, pkb->net_addr_len );
+	arp = arp_find_entry ( ll_protocol, net_protocol,
+			       nethdr->dest_net_addr );
 	if ( arp ) {
-		*ll_addr = arp->ll_addr;
+		memcpy ( llhdr->dest_ll_addr, arp->ll_addr,
+			 sizeof ( llhdr->dest_ll_addr ) );
 		return 0;
 	}
 
-	/* Find interface for this protocol */
-	netif = netdev_find_netif ( netdev, pkb->net_proto );
-	if ( ! netif )
-		return -EAFNOSUPPORT;
+	/* Allocate ARP packet */
+	pkb = alloc_pkb ( sizeof ( *arphdr ) +
+			  2 * ( MAX_LL_ADDR_LEN + MAX_NET_ADDR_LEN ) );
+	if ( ! pkb )
+		return -ENOMEM;
+	pkb->net_protocol = &arp_protocol;
 
 	/* Build up ARP request */
-	pkb_empty ( pkb );
 	arphdr = pkb_put ( pkb, sizeof ( *arphdr ) );
-	arphdr->ar_hrd = netdev->ll_proto;
-	arphdr->ar_hln = netdev->ll_addr_len;
-	arphdr->ar_pro = pkb->net_proto;
-	arphdr->ar_pln = pkb->net_addr_len;
+	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 ( pkb_put ( pkb, netdev->ll_addr_len ),
-		 netdev->ll_addr, netdev->ll_addr_len );
-	memcpy ( pkb_put ( pkb, netif->net_addr_len ),
-		 netif->net_addr, netif->net_addr_len );
-	memset ( pkb_put ( pkb, netdev->ll_addr_len ),
-		 0xff, netdev->ll_addr_len );
-	memcpy ( pkb_put ( pkb, netif->net_addr_len ),
-		 pkb->net_addr, netif->net_addr_len );
-
-	/* Locate ARP interface and send ARP request */
-	netif = netdev_find_netif ( netdev, htons ( ETH_P_ARP ) );
-	assert ( netif != NULL );
-	netif_send ( netif, pkb );
+	memcpy ( pkb_put ( pkb, ll_protocol->ll_addr_len ),
+		 llhdr->source_ll_addr, ll_protocol->ll_addr_len );
+	memcpy ( pkb_put ( pkb, net_protocol->net_addr_len ),
+		 nethdr->source_net_addr, net_protocol->net_addr_len );
+	memset ( pkb_put ( pkb, ll_protocol->ll_addr_len ),
+		 0, ll_protocol->ll_addr_len );
+	memcpy ( pkb_put ( pkb, net_protocol->net_addr_len ),
+		 nethdr->dest_net_addr, net_protocol->net_addr_len );
+
+	/* Transmit ARP request */
+	if ( ( rc = net_transmit ( pkb ) ) != 0 ) {
+		free_pkb ( pkb );
+		return rc;
+	}
 
 	return -ENOENT;
 }
@@ -148,7 +156,6 @@ int arp_resolve ( struct net_device *netdev, struct pk_buff *pkb,
 /**
  * Process incoming ARP packets
  *
- * @v arp_netif		Network interface for ARP packets
  * @v pkb		Packet buffer
  * @ret rc		Return status code
  *
@@ -158,80 +165,96 @@ int arp_resolve ( struct net_device *netdev, struct pk_buff *pkb,
  * avoiding the need for extraneous ARP requests; read the RFC for
  * details.
  */
-int arp_process ( struct net_interface *arp_netif, struct pk_buff *pkb ) {
+static int arp_rx ( struct pk_buff *pkb ) {
 	struct arphdr *arphdr = pkb->data;
-	struct net_device *netdev = arp_netif->netdev;
-	struct net_interface *netif;
+	struct ll_protocol *ll_protocol;
+	struct net_protocol *net_protocol;
 	struct arp_entry *arp;
+	struct net_device *netdev;
 	int merge = 0;
 
-	/* Check for correct link-layer protocol and length */
-	if ( ( arphdr->ar_hrd != netdev->ll_proto ) ||
-	     ( arphdr->ar_hln != netdev->ll_addr_len ) )
-		return 0;
+	/* Identify link-layer and network-layer protocols */
+	ll_protocol = pkb->ll_protocol;
+	net_protocol = net_find_protocol ( arphdr->ar_pro );
+	if ( ! net_protocol )
+		goto done;
 
-	/* See if we have an interface for this network-layer protocol */
-	netif = netdev_find_netif ( netdev, arphdr->ar_pro );
-	if ( ! netif )
-		return 0;
-	if ( arphdr->ar_pln != netif->net_addr_len )
-		return 0;
+	/* 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 ( arphdr->ar_hrd, arphdr->ar_pro,
-			       arp_sender_pa ( arphdr ), arphdr->ar_pln );
+	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;
 	}
 
-	/* See if we are the target protocol address */
-	if ( memcmp ( arp_target_pa ( arphdr ), netif->net_addr,
-		      arphdr->ar_pln ) != 0 )
-		return 0;
-
+	/* See if we own the target protocol address */
+	netdev = net_find_address ( net_protocol, arp_target_pa ( arphdr ) );
+	if ( ! netdev )
+		goto done;
+	
 	/* Create new ARP table entry if necessary */
 	if ( ! merge ) {
 		arp = &arp_table[next_new_arp_entry++ % NUM_ARP_ENTRIES];
-		arp->ll_proto = arphdr->ar_hrd;
-		arp->net_proto = arphdr->ar_pro;
+		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 );
+			 arphdr->ar_pln);
 	}
 
 	/* If it's not a request, there's nothing more to do */
 	if ( arphdr->ar_op != htons ( ARPOP_REQUEST ) )
-		return 0;
+		goto done;
 
 	/* Change request to a reply, and send it */
 	arphdr->ar_op = htons ( ARPOP_REPLY );
-	memcpy ( arp_sender_ha ( arphdr ), arp_target_ha ( arphdr ),
+	memswap ( arp_sender_ha ( arphdr ), arp_target_ha ( arphdr ),
 		 arphdr->ar_hln + arphdr->ar_pln );
 	memcpy ( arp_target_ha ( arphdr ), netdev->ll_addr, arphdr->ar_hln );
-	memcpy ( arp_target_pa ( arphdr ), netif->net_addr, arphdr->ar_pln );
-	netif_send ( arp_netif, pkb );
+	if ( net_transmit ( pkb ) == 0 )
+		pkb = NULL;
 
+ done:
+	free_pkb ( pkb );
 	return 0;
 }
 
 /**
- * Add media-independent link-layer header
+ * Perform ARP network-layer routing
  *
- * @v arp_netif		Network interface for ARP packets
- * @v pkb		Packet buffer
- * @ret rc		Return status code
+ * @v pkb	Packet buffer
+ * @ret source	Network-layer source address
+ * @ret dest	Network-layer destination address
+ * @ret rc	Return status code
  */
-int arp_add_llh_metadata ( struct net_interface *arp_netif __unused,
-			   struct pk_buff *pkb ) {
+static int arp_route ( const struct pk_buff *pkb,
+		       struct net_header *nethdr ) {
 	struct arphdr *arphdr = pkb->data;
 
-	pkb->net_proto = htons ( ETH_P_ARP );
-	pkb->flags = PKB_FL_RAW_NET_ADDR;
-	pkb->net_addr_len = arphdr->ar_hln;
-	pkb->net_addr = arp_target_ha ( arphdr );
+	memcpy ( nethdr->source_net_addr, arp_sender_ha ( arphdr ),
+		 arphdr->ar_hln );
+	memcpy ( nethdr->dest_net_addr, arp_target_ha ( arphdr ),
+		 arphdr->ar_hln );
+	nethdr->dest_flags = NETADDR_FL_RAW;
+	if ( arphdr->ar_op == htons ( ARPOP_REQUEST ) )
+		nethdr->dest_flags |= NETADDR_FL_BROADCAST;
 	
 	return 0;
 }
+
+/** ARP protocol */
+struct net_protocol arp_protocol = {
+	.net_proto = ETH_P_ARP,
+	.rx = arp_rx,
+	.route = arp_route,
+};
+
+NET_PROTOCOL ( arp_protocol );

src/net/ethernet.c

@@ -20,10 +20,12 @@
 #include <string.h>
 #include <byteswap.h>
 #include <assert.h>
+#include <gpxe/if_arp.h>
 #include <gpxe/if_ether.h>
 #include <gpxe/netdevice.h>
 #include <gpxe/pkbuff.h>
 #include <gpxe/arp.h>
+#include <gpxe/ethernet.h>
 
 /** @file
  *
@@ -32,85 +34,104 @@
  */
 
 /** Ethernet broadcast MAC address */
-static uint8_t eth_broadcast[] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+static uint8_t eth_broadcast[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 
 /**
- * Build Ethernet link-layer header
+ * Perform Ethernet routing
  *
- * @v netdev	Network device
- * @v pkb	Packet buffer
+ * @v nethdr	Generic network-layer header
+ * @ret llhdr	Generic link-layer header
  * @ret rc	Return status code
  *
- * This constructs the Ethernet link-layer header (destination MAC,
- * source MAC, network-layer protocol) based on the metadata found in
- * @c pkb.
+ * Constructs the generic link-layer header based on the generic
+ * network-layer header, i.e. maps network-layer addresses (e.g. IPv4
+ * addresses) to MAC addresses.
  *
  * If the destination MAC address cannot be determined, an ARP request
- * is sent for the requested network-layer address instead.
+ * is sent for the requested network-layer address and -ENOENT is
+ * returned.
  */
-int eth_build_llh ( struct net_device *netdev, struct pk_buff *pkb ) {
-	struct ethhdr *ethhdr = pkb->data;
-	const void *eth_dest;
+static int eth_route ( const struct net_header *nethdr,
+		       struct ll_header *llhdr ) {
 	int rc;
 
-	/* Do the easy bits */
-	ethhdr->h_protocol = pkb->net_proto;
-	memcpy ( ethhdr->h_source, netdev->ll_addr,
-		 sizeof ( ethhdr->h_source ) );
-
 	/* Work out the destination MAC address */
-	if ( pkb->flags & PKB_FL_RAW_NET_ADDR ) {
-		eth_dest = pkb->net_addr;
-	} else if ( pkb->flags & PKB_FL_BROADCAST ) {
-		eth_dest = eth_broadcast;
-	} else if ( pkb->flags & PKB_FL_MULTICAST ) {
+	if ( nethdr->dest_flags & NETADDR_FL_RAW ) {
+		memcpy ( llhdr->dest_ll_addr, nethdr->dest_net_addr, ETH_ALEN);
+	} else if ( nethdr->dest_flags & NETADDR_FL_BROADCAST ) {
+		memcpy ( llhdr->dest_ll_addr, eth_broadcast, ETH_ALEN );
+	} else if ( nethdr->dest_flags & NETADDR_FL_MULTICAST ) {
 		/* IP multicast is a special case; there exists a
 		 * direct mapping from IP address to MAC address
 		 */
-		assert ( pkb->net_proto == htons ( ETH_P_IP ) );
-		ethhdr->h_dest[0] = 0x01;
-		ethhdr->h_dest[1] = 0x00;
-		ethhdr->h_dest[2] = 0x5e;
-		ethhdr->h_dest[3] = *( ( char * ) pkb->net_addr + 1 ) & 0x7f;
-		ethhdr->h_dest[4] = *( ( char * ) pkb->net_addr + 2 );
-		ethhdr->h_dest[5] = *( ( char * ) pkb->net_addr + 3 );
-		eth_dest = ethhdr->h_dest;
+		assert ( nethdr->net_protocol->net_proto == htons(ETH_P_IP) );
+		llhdr->dest_ll_addr[0] = 0x01;
+		llhdr->dest_ll_addr[1] = 0x00;
+		llhdr->dest_ll_addr[2] = 0x5e;
+		llhdr->dest_ll_addr[3] = nethdr->dest_net_addr[1] & 0x7f;
+		llhdr->dest_ll_addr[4] = nethdr->dest_net_addr[2];
+		llhdr->dest_ll_addr[5] = nethdr->dest_net_addr[3];
 	} else {
 		/* Otherwise, look up the address using ARP */
-		if ( ( rc = arp_resolve ( netdev, pkb, &eth_dest ) ) != 0 )
+		if ( ( rc = arp_resolve ( nethdr, llhdr ) ) != 0 )
 			return rc;
 	}
 
-	/* Fill in destination MAC address */
-	memcpy ( ethhdr->h_dest, eth_dest, sizeof ( ethhdr->h_dest ) );
-
 	return 0;
 }
 
+/**
+ * Fill in Ethernet link-layer header
+ *
+ * @v pkb	Packet buffer
+ * @v llhdr	Generic link-layer header
+ *
+ * Fills in the Ethernet link-layer header in the packet buffer based
+ * on information in the generic link-layer header.
+ */
+static void eth_fill_llh ( const struct ll_header *llhdr,
+			   struct pk_buff *pkb ) {
+	struct ethhdr *ethhdr = pkb->data;
+
+	memcpy ( ethhdr->h_dest, llhdr->dest_ll_addr, ETH_ALEN );
+	memcpy ( ethhdr->h_source, llhdr->source_ll_addr, ETH_ALEN );
+	ethhdr->h_protocol = llhdr->net_proto;
+}
+
 /**
  * Parse Ethernet link-layer header
  *
- * @v netdev	Network device
  * @v pkb	Packet buffer
- * @ret rc	Return status code
+ * @v llhdr	Generic link-layer header
  *
- * This parses the Ethernet link-layer header (destination MAC, source
- * MAC, network-layer protocol) and fills in the metadata in @c pkb.
+ * Fills in the generic link-layer header based on information in the
+ * Ethernet link-layer header in the packet buffer.
  */
-int eth_parse_llh ( struct net_device *netdev __unused, struct pk_buff *pkb ) {
+static void eth_parse_llh ( const struct pk_buff *pkb,
+			    struct ll_header *llhdr ) {
 	struct ethhdr *ethhdr = pkb->data;
 
-	pkb->net_proto = ethhdr->h_protocol;
-	pkb->flags = PKB_FL_RAW_NET_ADDR;
-	pkb->net_addr_len = sizeof ( ethhdr->h_dest );
-	pkb->net_addr = ethhdr->h_dest;
+	memcpy ( llhdr->dest_ll_addr, ethhdr->h_dest, ETH_ALEN );
+	memcpy ( llhdr->source_ll_addr, ethhdr->h_source, ETH_ALEN );
+	llhdr->net_proto = ethhdr->h_protocol;
 
-	if ( memcmp ( ethhdr->h_dest, eth_broadcast,
-		      sizeof ( ethhdr->h_dest ) ) == 0 ) {
-		pkb->flags |= PKB_FL_BROADCAST;
+	if ( memcmp ( ethhdr->h_dest, eth_broadcast, ETH_ALEN ) == 0 ) {
+		llhdr->dest_flags = NETADDR_FL_BROADCAST;
 	} else if ( ethhdr->h_dest[0] & 0x01 ) {
-		pkb->flags |= PKB_FL_MULTICAST;
+		llhdr->dest_flags = NETADDR_FL_MULTICAST;
+	} else {
+		llhdr->dest_flags = 0;
 	}
-
-	return 0;
 }
+
+/** Ethernet protocol */
+struct ll_protocol ethernet_protocol = {
+	.ll_proto = htons ( ARPHRD_ETHER ),
+	.ll_addr_len = ETH_ALEN,
+	.ll_header_len = ETH_HLEN,
+	.route = eth_route,
+	.fill_llh = eth_fill_llh,
+	.parse_llh = eth_parse_llh,
+};
+
+LL_PROTOCOL ( ethernet_protocol );

src/net/netdevice.c

@@ -18,146 +18,231 @@
 
 #include <stdint.h>
 #include <byteswap.h>
+#include <string.h>
 #include <errno.h>
 #include <gpxe/if_ether.h>
 #include <gpxe/pkbuff.h>
+#include <gpxe/tables.h>
 #include <gpxe/netdevice.h>
 
 /** @file
  *
- * Network devices and network interfaces
+ * Network device management
  *
  */
 
-/** List of all registered network devices */
-static LIST_HEAD ( net_devices );
-
 /**
- * Register network device
+ * Static single instance of a network device
+ *
+ * The gPXE API is designed to accommodate multiple network devices.
+ * However, in the interests of code size, the implementation behind
+ * the API supports only a single instance of a network device.
  *
- * @v netdev	Network device
- * @ret rc	Return status code
+ * No code outside of netdevice.c should ever refer directly to @c
+ * static_single_netdev.
  *
- * Adds the network device to the list of network devices.
+ * Callers should always check the return status of alloc_netdev(),
+ * register_netdev() etc.  In the current implementation this code
+ * will be optimised out by the compiler, so there is no penalty.
  */
-int register_netdevice ( struct net_device *netdev ) {
-	list_add ( &netdev->devices, &net_devices );
-	return 0;
-}
+struct net_device static_single_netdev;
+
+/** Registered network-layer protocols */
+static struct net_protocol net_protocols[0] __table_start ( net_protocols );
+static struct net_protocol net_protocols_end[0] __table_end ( net_protocols );
+
+/** Network-layer addresses for @c static_single_netdev */
+static struct net_address static_single_netdev_addresses[0]
+	__table_start ( sgl_netdev_addresses );
+static struct net_address static_single_netdev_addresses_end[0]
+	__table_end ( sgl_netdev_addresses );
+
+/** Recevied packet queue */
+static LIST_HEAD ( rx_queue );
 
 /**
- * Unregister network device
+ * Identify network protocol
  *
- * @v netdev	Network device
+ * @v net_proto		Network-layer protocol, in network-byte order
+ * @ret net_protocol	Network-layer protocol, or NULL
  *
- * Removes the network device from the list of network devices.
+ * Identify a network-layer protocol from a protocol number, which
+ * must be an ETH_P_XXX constant in network-byte order.
  */
-void unregister_netdevice ( struct net_device *netdev ) {
-	list_del ( &netdev->devices );
+struct net_protocol * net_find_protocol ( uint16_t net_proto ) {
+	struct net_protocol *net_protocol;
+
+	for ( net_protocol = net_protocols ; net_protocol < net_protocols_end ;
+	      net_protocol++ ) {
+		if ( net_protocol->net_proto == net_proto )
+			return net_protocol;
+	}
+	return NULL;
 }
 
 /**
- * Transmit packet via network device
+ * Identify network device by network-layer address
  *
- * @v netdev	Network device
- * @v pkb	Packet buffer
- * @ret rc	Return status code
+ * @v net_protocol	Network-layer protocol
+ * @v net_addr		Network-layer address
+ * @ret netdev		Network device, or NULL
  *
- * Transmits the packet via the network device.  The @c pkb link-layer
- * metadata must already have been filled in, and space for the
- * link-layer header must already be present in the packet buffer.
+ * Searches through all network devices to find the device with the
+ * specified network-layer address.
+ *
+ * Note that even with a static single network device, this function
+ * can still return NULL.
  */
-int netdev_send ( struct net_device *netdev, struct pk_buff *pkb ) {
-	int rc;
-
-	if ( pkb->net_proto != ETH_P_RAW ) {
-		if ( ( rc = netdev->build_llh ( netdev, pkb ) ) != 0 )
-			return rc;
+struct net_device * net_find_address ( struct net_protocol *net_protocol,
+				       void *net_addr ) {
+	struct net_address *net_address;
+	struct net_device *netdev = &static_single_netdev;
+	
+	for ( net_address = static_single_netdev_addresses ;
+	      net_address < static_single_netdev_addresses_end ;
+	      net_address ++ ) {
+		if ( ( net_address->net_protocol == net_protocol ) &&
+		     ( memcmp ( net_address->net_addr, net_addr,
+				net_protocol->net_addr_len ) == 0 ) )
+			return netdev;
 	}
-	return netdev->transmit ( netdev, pkb );
+	return NULL;
 }
 
 /**
- * Poll for packet on network device
+ * Transmit packet
  *
- * @v netdev	Network device
- * @v pkb	Packet buffer
- * @ret rc	Return status code
+ * @v pkb		Packet buffer
+ * @ret rc		Return status code
  *
- * Polls the network device for a packet.  If a packet is available,
- * it will be added to the packet buffer, and the link-layer metadata
- * fields in @c pkb will be filled in.
+ * Transmits the packet via the appropriate network device.  If this
+ * function returns success, it has taken ownership of the packet
+ * buffer.
  */
-int netdev_poll ( struct net_device *netdev, struct pk_buff *pkb ) {
+int net_transmit ( struct pk_buff *pkb ) {
+	struct net_protocol *net_protocol;
+	struct net_header nethdr;
+	struct ll_protocol *ll_protocol;
+	struct ll_header llhdr;
+	struct net_device *netdev;
 	int rc;
 
-	if ( ( rc = netdev->poll ( netdev, pkb ) ) != 0 )
-		return rc;
-	return netdev->parse_llh ( netdev, pkb );
+	/* Perform network-layer routing */
+	net_protocol = pkb->net_protocol;
+	nethdr.net_protocol = net_protocol;
+	if ( ( rc = net_protocol->route ( pkb, &nethdr ) ) != 0 )
+		goto err;
+
+	/* Identify transmitting network device */
+	netdev = net_find_address ( net_protocol, nethdr.source_net_addr );
+	if ( ! netdev )
+		goto err;
+
+	/* Perform link-layer routing */
+	ll_protocol = netdev->ll_protocol;
+	llhdr.ll_protocol = ll_protocol;
+	llhdr.net_proto = net_protocol->net_proto;
+	memcpy ( llhdr.source_ll_addr, netdev->ll_addr,
+		 ll_protocol->ll_addr_len);
+	if ( ( rc = ll_protocol->route ( &nethdr, &llhdr ) ) != 0 )
+		goto err;
+
+	/* Prepend link-layer header */
+	pkb_push ( pkb, ll_protocol->ll_header_len );
+	ll_protocol->fill_llh ( &llhdr, pkb );
+
+	/* Transmit packet */
+	if ( ( rc = netdev->transmit ( netdev, pkb ) ) != 0 )
+		goto err;
+
+	return 0;
+
+ err:
+	free_pkb ( pkb );
+	return rc;
 }
 
 /**
- * Transmit packet via network interface
+ * Poll for packet on all network devices
  *
- * @v netif	Network interface
- * @v pkb	Packet buffer
- * @ret rc	Return status code
+ * @ret True		There are packets present in the receive queue
+ * @ret False		There are no packets present in the receive queue
  *
- * Transmits the packet via the network interface.  The packet must
- * start with a network-layer header (e.g. an IP header, for an IP
- * interface).  The packet contents are undefined on return.
+ * Polls all network devices for received packets.  Any received
+ * packets will be added to the RX packet queue via netdev_rx().
  */
-int netif_send ( struct net_interface *netif, struct pk_buff *pkb ) {
-	struct net_device *netdev = netif->netdev;
-	int rc;
+int net_poll ( void ) {
+	struct net_device *netdev = &static_single_netdev;
+
+	netdev->poll ( netdev );
 
-	if ( ( rc = netif->add_llh_metadata ( netif, pkb ) ) != 0 )
-		return rc;
-	pkb_push ( pkb, netdev->ll_hlen );
-	return netdev_send ( netdev, pkb );
+	return ( ! list_empty ( &rx_queue ) );
 }
 
 /**
- * Process received packet
+ * Add packet to receive queue
  *
- * @v netif	Network interface
- * @v pkb	Packet buffer
- * @ret rc	Return status code
+ * @v netdev		Network device
+ * @v pkb		Packet buffer
  *
- * Processes a packet received via netdev_poll().  The interface
- * corresponding to the network-layer protocol is identified, the
- * link-layer header is stripped from the packet and the packet is
- * passed to the net_interface::rx_packet() method.
+ * The packet is added to the RX queue.  Ownership of the packet is
+ * transferred to the RX queue; the caller must not touch the packet
+ * buffer after calling netdev_rx().
  */
-int netdev_rx_packet ( struct net_device *netdev, struct pk_buff *pkb ) {
-	struct net_interface *netif;
-
-	netif = netdev_find_netif ( netdev, pkb->net_proto );
-	if ( ! netif )
-		return -EAFNOSUPPORT;
-
-	pkb_pull ( pkb, netdev->ll_hlen );
-	return netif->rx_packet ( netif, pkb );
+void netdev_rx ( struct net_device *netdev, struct pk_buff *pkb ) {
+	pkb->ll_protocol = netdev->ll_protocol;
+	list_add_tail ( &pkb->list, &rx_queue );
 }
 
 /**
- * Poll for packet on all network devices
+ * Remove packet from receive queue
  *
- * @v pkb	Packet buffer
- * @ret netdev	Network device
- * @ret rc	Return status code
+ * @ret pkb		Packet buffer, or NULL
  *
- * Polls all network devices for a packet.  If a packet is available
- * on any interface, @c netdev will be filled in and the packet will
- * be received as per netdev_poll().
+ * Removes the first packet from the RX queue and returns it.
+ * Ownership of the packet is transferred to the caller.
  */
-int net_poll ( struct pk_buff *pkb, struct net_device **netdev ) {
-	int rc;
+struct pk_buff * net_rx_dequeue ( void ) {
+	struct pk_buff *pkb;
 
-	list_for_each_entry ( (*netdev), &net_devices, devices ) {
-		if ( ( rc = netdev_poll ( *netdev, pkb ) ) == 0 )
-			return rc;
+	list_for_each_entry ( pkb, &rx_queue, list ) {
+		list_del ( &pkb->list );
+		return pkb;
 	}
+	return NULL;
+}
 
-	return -EAGAIN;
+void net_run ( void ) {
+	struct pk_buff *pkb;
+	struct ll_protocol *ll_protocol;
+	struct ll_header llhdr;
+	struct net_protocol *net_protocol;
+
+	while ( ( pkb = net_rx_dequeue () ) ) {
+
+		/* Parse link-layer header */
+		ll_protocol = pkb->ll_protocol;
+		ll_protocol->parse_llh ( pkb, &llhdr );
+
+		/* Identify network-layer protocol */
+		net_protocol = net_find_protocol ( llhdr.net_proto );
+		if ( ! net_protocol ) {
+			DBG ( "Unknown network-layer protocol %02x\n",
+			      ntohs ( llhdr.net_proto ) );
+			free_pkb ( pkb );
+			continue;
+		}
+
+		/* Strip off link-layer header */
+		pkb_pull ( pkb, ll_protocol->ll_header_len );
+
+		/* Hand off to network layer */
+		if ( net_protocol->rx ( pkb ) != 0 ) {
+			free_pkb ( pkb );
+			continue;
+		}
+	}
 }
+
+
+

src/proto/uip/uipopt.h

@@ -464,7 +464,7 @@ void uip_log(char *msg);
  *
  * \hideinitializer
  */
-#define UIP_LLH_LEN     14
+#define UIP_LLH_LEN     0
 
 
 /** @} */