First draft of iSCSI protocol support. Is capable of retrieving disk

blocks.

src/include/gpxe/iscsi.h

+#ifndef _ISCSI_H
+#define _ISCSI_H
+
+/** @file
+ *
+ * iSCSI protocol
+ *
+ */
+
+#include <stdint.h>
+#include <gpxe/tcp.h>
+#include <gpxe/scsi.h>
+
+/**
+ * iSCSI segment lengths
+ *
+ * iSCSI uses an icky structure with one one-byte field (a dword
+ * count) and one three-byte field (a byte count).  This structure,
+ * and the accompanying macros, relieve some of the pain.
+ */
+union iscsi_segment_lengths {
+	struct {
+		/** The AHS length (measured in dwords) */
+		uint8_t ahs_len;
+		/** The data length (measured in bytes), in network
+		 * byte order
+		 */
+		uint8_t data_len[3];
+	} bytes;
+	/** Ths data length (measured in bytes), in network byte
+	 * order, with ahs_len as the first byte.
+	 */
+	uint32_t ahs_and_data_len;
+};
+
+/** The length of the additional header segment, in dwords */
+#define ISCSI_AHS_LEN( segment_lengths ) \
+	( (segment_lengths).bytes.ahs_len )
+
+/** The length of the data segment, in bytes, excluding any padding */
+#define ISCSI_DATA_LEN( segment_lengths ) \
+	( ntohl ( (segment_lengths).ahs_and_data_len ) & 0xffffff )
+
+/** The padding of the data segment, in bytes */
+#define ISCSI_DATA_PAD_LEN( segment_lengths ) \
+	( ( 0 - (segment_lengths).bytes.data_len[2] ) & 0x03 )
+
+/** Set additional header and data segment lengths */
+#define ISCSI_SET_LENGTHS( segment_lengths, ahs_len, data_len ) do {	\
+	(segment_lengths).ahs_and_data_len =				\
+		htonl ( data_len | ( ahs_len << 24 ) );			\
+	} while ( 0 )
+
+/**
+ * iSCSI basic header segment common fields
+ *
+ */
+struct iscsi_bhs_common {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Fields specific to the PDU type */
+	uint8_t other_a[2];
+	/** Segment lengths */
+	union iscsi_segment_lengths lengths;
+	/** Fields specific to the PDU type */
+	uint8_t other_b[8];
+	/** Initiator Task Tag */
+	uint32_t itt;
+	/** Fields specific to the PDU type */
+	uint8_t other_c[28];
+};
+
+/** Opcode mask */
+#define ISCSI_OPCODE_MASK 0x3f
+
+/** Immediate delivery */
+#define ISCSI_FLAG_IMMEDIATE 0x40
+
+/** Final PDU of a sequence */
+#define ISCSI_FLAG_FINAL 0x80
+
+/**
+ * iSCSI login request basic header segment
+ *
+ */
+struct iscsi_bhs_login_request {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Maximum supported version number */
+	uint8_t version_max;
+	/** Minimum supported version number */
+	uint8_t version_min;
+	/** Segment lengths */
+	union iscsi_segment_lengths lengths;
+	/** Initiator session ID (IANA format) enterprise number and flags */
+	uint32_t isid_iana_en;
+	/** Initiator session ID (IANA format) qualifier */
+	uint16_t isid_iana_qual;
+	/** Target session identifying handle */
+	uint16_t tsih;
+	/** Initiator Task Tag */
+	uint32_t itt;
+	/** Connection ID */
+	uint16_t cid;
+	/** Reserved */
+	uint16_t reserved_a;
+	/** Command sequence number */
+	uint32_t cmdsn;
+	/** Expected status sequence number */
+	uint32_t expstatsn;
+	/** Reserved */
+	uint8_t reserved_b[16];
+};
+
+/** Login request opcode */
+#define ISCSI_OPCODE_LOGIN_REQUEST 0x03
+
+/** Willingness to transition to next stage */
+#define ISCSI_LOGIN_FLAG_TRANSITION 0x80
+
+/** Key=value pairs continued in subsequent request */
+#define ISCSI_LOGIN_FLAG_CONTINUE 0x40
+
+/* Current stage values and mask */
+#define ISCSI_LOGIN_CSG_MASK 0x0c
+#define ISCSI_LOGIN_CSG_SECURITY_NEGOTIATION 0x00
+#define ISCSI_LOGIN_CSG_OPERATIONAL_NEGOTIATION 0x04
+#define ISCSI_LOGIN_CSG_FULL_FEATURE_PHASE 0x0c
+
+/* Next stage values and mask */
+#define ISCSI_LOGIN_NSG_MASK 0x03
+#define ISCSI_LOGIN_NSG_SECURITY_NEGOTIATION 0x00
+#define ISCSI_LOGIN_NSG_OPERATIONAL_NEGOTIATION 0x01
+#define ISCSI_LOGIN_NSG_FULL_FEATURE_PHASE 0x03
+
+/** ISID IANA format marker */
+#define ISCSI_ISID_IANA 0x40000000
+
+/** Fen Systems Ltd. IANA enterprise number
+ *
+ * Permission is hereby granted to use Fen Systems Ltd.'s IANA
+ * enterprise number with this iSCSI implementation.
+ */
+#define IANA_EN_FEN_SYSTEMS 10019
+
+/**
+ * iSCSI login response basic header segment
+ *
+ */
+struct iscsi_bhs_login_response {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Maximum supported version number */
+	uint8_t version_max;
+	/** Minimum supported version number */
+	uint8_t version_min;
+	/** Segment lengths */
+	union iscsi_segment_lengths lengths;
+	/** Initiator session ID (IANA format) enterprise number and flags */
+	uint32_t isid_iana_en;
+	/** Initiator session ID (IANA format) qualifier */
+	uint16_t isid_iana_qual;
+	/** Target session identifying handle */
+	uint16_t tsih;
+	/** Initiator Task Tag */
+	uint32_t itt;
+	/** Reserved */
+	uint32_t reserved_a;
+	/** Status sequence number */
+	uint32_t statsn;
+	/** Expected command sequence number */
+	uint32_t expcmdsn;
+	/** Maximum command sequence number */
+	uint32_t maxcmdsn;
+	/** Status class */
+	uint8_t status_class;
+	/** Status detail */
+	uint8_t status_detail;
+	/** Reserved */
+	uint8_t reserved_b[10];
+};
+
+/** Login response opcode */
+#define ISCSI_OPCODE_LOGIN_RESPONSE 0x23
+
+/**
+ * iSCSI SCSI command basic header segment
+ *
+ */
+struct iscsi_bhs_scsi_command {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Reserved */
+	uint16_t reserved_a;
+	/** Segment lengths */
+	union iscsi_segment_lengths lengths;
+	/** SCSI Logical Unit Number */
+	uint8_t lun[8];
+	/** Initiator Task Tag */
+	uint32_t itt;
+	/** Expected data transfer length */
+	uint32_t exp_len;
+	/** Command sequence number */
+	uint32_t cmdsn;
+	/** Expected status sequence number */
+	uint32_t expstatsn;
+	/** SCSI Command Descriptor Block (CDB) */
+	union scsi_cdb cdb;
+};
+
+/** SCSI command opcode */
+#define ISCSI_OPCODE_SCSI_COMMAND 0x01
+
+/** Command will read data */
+#define ISCSI_COMMAND_FLAG_READ 0x40
+
+/** Command will write data */
+#define ISCSI_COMMAND_FLAG_WRITE 0x20
+
+/* Task attributes */
+#define ISCSI_COMMAND_ATTR_UNTAGGED 0x00
+#define ISCSI_COMMAND_ATTR_SIMPLE 0x01
+#define ISCSI_COMMAND_ATTR_ORDERED 0x02
+#define ISCSI_COMMAND_ATTR_HEAD_OF_QUEUE 0x03
+#define ISCSI_COMMAND_ATTR_ACA 0x04
+
+/**
+ * iSCSI SCSI response basic header segment
+ *
+ */
+struct iscsi_bhs_scsi_response {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Response code */
+	uint8_t response;
+	/** SCSI status code */
+	uint8_t status;
+	/** Segment lengths */
+	union iscsi_segment_lengths lengths;
+	/** Reserved */
+	uint8_t reserved_a[8];
+	/** Initiator Task Tag */
+	uint32_t itt;
+	/** SNACK tag */
+	uint32_t snack;
+	/** Status sequence number */
+	uint32_t statsn;
+	/** Expected command sequence number */
+	uint32_t expcmdsn;
+	/** Maximum command sequence number */
+	uint32_t maxcmdsn;
+	/** Expected data sequence number */
+	uint32_t expdatasn;
+	/** Reserved */
+	uint8_t reserved_b[8];
+};
+
+/** SCSI response opcode */
+#define ISCSI_OPCODE_SCSI_RESPONSE 0x21
+
+/** SCSI command completed at target */
+#define ISCSI_RESPONSE_COMMAND_COMPLETE 0x00
+
+/** SCSI target failure */
+#define ISCSI_RESPONSE_TARGET_FAILURE 0x01
+
+/**
+ * iSCSI data in basic header segment
+ *
+ */
+struct iscsi_bhs_data_in {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Reserved */
+	uint8_t reserved_a;
+	/** SCSI status code */
+	uint8_t status;
+	/** Segment lengths */
+	union iscsi_segment_lengths lengths;
+	/** Logical Unit Number */
+	uint8_t lun[8];
+	/** Initiator Task Tag */
+	uint32_t itt;
+	/** Target Transfer Tag */
+	uint32_t ttt;
+	/** Status sequence number */
+	uint32_t statsn;
+	/** Expected command sequence number */
+	uint32_t expcmdsn;
+	/** Maximum command sequence number */
+	uint32_t maxcmdsn;
+	/** Data sequence number */
+	uint32_t datasn;
+	/** Buffer offset */
+	uint32_t offset;
+	/** Residual count */
+	uint32_t residual_count;
+};
+
+/** Data in opcode */
+#define ISCSI_OPCODE_DATA_IN 0x25
+
+/** Data requires acknowledgement */
+#define ISCSI_DATA_FLAG_ACKNOWLEDGE 0x40
+
+/** Data overflow occurred */
+#define ISCSI_DATA_FLAG_OVERFLOW 0x04
+
+/** Data underflow occurred */
+#define ISCSI_DATA_FLAG_UNDERFLOW 0x02
+
+/** SCSI status code and verflow/underflow flags are valid */
+#define ISCSI_DATA_FLAG_STATUS 0x01
+
+/**
+ * An iSCSI basic header segment
+ */
+union iscsi_bhs {
+	struct iscsi_bhs_common common;
+	struct iscsi_bhs_login_request login_request;
+	struct iscsi_bhs_login_response login_response;
+	struct iscsi_bhs_scsi_command scsi_command;
+	struct iscsi_bhs_scsi_response scsi_response;
+	struct iscsi_bhs_data_in data_in;
+	unsigned char bytes[ sizeof ( struct iscsi_bhs_common ) ];
+};
+
+/** State */
+enum iscsi_state {
+	/** In the process of logging in */
+	ISCSI_STATE_FAILED = -1,
+	ISCSI_STATE_NOT_CONNECTED = 0,
+	ISCSI_STATE_IDLE,
+	ISCSI_STATE_LOGGING_IN,
+	ISCSI_STATE_READING_DATA,
+};
+
+/** State of an iSCSI TX engine */
+enum iscsi_tx_state {
+	/** Nothing to send */
+	ISCSI_TX_IDLE = 0,
+	/** Sending the basic header segment */
+	ISCSI_TX_BHS,
+	/** Sending the additional header segment */
+	ISCSI_TX_AHS,
+	/** Sending the data segment */
+	ISCSI_TX_DATA,
+	/** Sending the data segment padding */
+	ISCSI_TX_DATA_PADDING,
+};
+
+/** State of an iSCSI RX engine */
+enum iscsi_rx_state {
+	/** Receiving the basic header segment */
+	ISCSI_RX_BHS = 0,
+	/** Receiving the additional header segment */
+	ISCSI_RX_AHS,
+	/** Receiving the data segment */
+	ISCSI_RX_DATA,
+	/** Receiving the data segment padding */
+	ISCSI_RX_DATA_PADDING,
+};
+
+/** An iSCSI session */
+struct iscsi_session {
+	/** TCP connection for this session */
+	struct tcp_connection tcp;
+
+	/** Initiator IQN */
+	const char *initiator;
+	/** Target IQN */
+	const char *target;
+
+	/** Block size in bytes */
+	size_t block_size;
+	/** Starting block number of the current data transfer */
+	unsigned long block_start;
+	/** Block count of the current data transfer */
+	unsigned long block_count;
+	/** Block read callback function
+	 *
+	 * Note that this may be called several times, since it is
+	 * called per-packet rather than per-block.
+	 */
+	void ( * block_read_callback ) ( void *private, const void *data,
+					 unsigned long offset, size_t len );
+	/** Block read callback private data
+	 *
+	 * This is passed to block_read_callback()
+	 */
+	void *block_read_private;
+
+	/** State of the session */
+	enum iscsi_state state;
+	/** Target session identifying handle
+	 *
+	 * This is assigned by the target when we first log in, and
+	 * must be reused on subsequent login attempts.
+	 */
+	uint16_t tsih;
+
+	/** Initiator task tag
+	 *
+	 * This is the tag of the current command.  It is incremented
+	 * whenever a final response PDU is received.
+	 */
+	uint32_t itt;
+	/** Command sequence number
+	 *
+	 * This is the sequence number of the current command, used to
+	 * fill out the CmdSN field in iSCSI request PDUs.  It is
+	 * updated with the value of the ExpCmdSN field whenever we
+	 * receive an iSCSI response PDU containing such a field.
+	 */
+	uint32_t cmdsn;
+	/** Status sequence number
+	 *
+	 * This is the most recent status sequence number present in
+	 * the StatSN field of an iSCSI response PDU containing such a
+	 * field.  Whenever we send an iSCSI request PDU, we fill out
+	 * the ExpStatSN field with this value plus one.
+	 */
+	uint32_t statsn;
+	
+	/** Basic header segment for current TX PDU */
+	union iscsi_bhs tx_bhs;
+	/** State of the TX engine */
+	enum iscsi_tx_state tx_state;
+	/** Byte offset within the current TX state */
+	size_t tx_offset;
+
+	/** Basic header segment for current RX PDU */
+	union iscsi_bhs rx_bhs;
+	/** State of the RX engine */
+	enum iscsi_rx_state rx_state;
+	/** Byte offset within the current RX state */
+	size_t rx_offset;
+};
+
+static inline int iscsi_busy ( struct iscsi_session *iscsi ) {
+	return ( iscsi->state > ISCSI_STATE_IDLE );
+}
+
+static inline int iscsi_error ( struct iscsi_session *iscsi ) {
+	return ( iscsi->state == ISCSI_STATE_FAILED );
+}
+
+#endif /* _ISCSI_H */

src/include/gpxe/scsi.h

+#ifndef _SCSI_H
+#define _SCSI_H
+
+#include <stdint.h>
+
+struct scsi_cdb_read_10 {
+	/** Opcode */
+	uint8_t opcode;
+	/** Flags */
+	uint8_t flags;
+	/** Start address
+	 *
+	 * This is a logical block number, in big-endian order.
+	 */
+	uint32_t lba;
+	/** Group number */
+	uint8_t group;
+	/** Transfer length
+	 *
+	 * This is a logical block count.
+	 */
+	uint16_t len;
+	/** Control byte */
+	uint8_t control;
+} __attribute__ (( packed ));
+
+#define SCSI_OPCODE_READ_10 0x28
+
+union scsi_cdb {
+	struct scsi_cdb_read_10 read_10;
+	char bytes[16];
+};
+
+#endif /* _SCSI_H */

src/proto/iscsi.c

+#include <stddef.h>
+#include <string.h>
+#include <vsprintf.h>
+#include <assert.h>
+#include <byteswap.h>
+#include <gpxe/iscsi.h>
+
+/** @file
+ *
+ * iSCSI protocol
+ *
+ */
+
+/****************************************************************************
+ *
+ * Utility functions
+ *
+ */
+
+/**
+ * Start up a new TX PDU
+ *
+ * @v iscsi		iSCSI session
+ *
+ * This initiates the process of sending a new PDU.  Only one PDU may
+ * be in transit at any one time.
+ */
+static void iscsi_start_tx ( struct iscsi_session *iscsi ) {
+	assert ( iscsi->tx_state == ISCSI_TX_IDLE );
+	iscsi->tx_state = ISCSI_TX_BHS;
+	iscsi->tx_offset = 0;
+}
+
+/**
+ * Mark session as failed
+ *
+ * @v iscsi		iSCSI session
+ *
+ * This marks the session as permanently failed.  The session will not
+ * be automatically logged back in.
+ */
+static void iscsi_fail ( struct iscsi_session *iscsi ) {
+	iscsi->state = ISCSI_STATE_FAILED;
+	tcp_close ( &iscsi->tcp );
+}
+
+/****************************************************************************
+ *
+ * iSCSI SCSI command issuing
+ *
+ */
+
+/**
+ * Start up a block read
+ *
+ * @v iscsi		iSCSI session
+ *
+ */
+static void iscsi_start_read_block ( struct iscsi_session *iscsi ) {
+	struct iscsi_bhs_scsi_command *command = &iscsi->tx_bhs.scsi_command;
+	struct scsi_cdb_read_10 *read = &command->cdb.read_10;
+
+	assert ( iscsi->block_size != 0 );
+	assert ( iscsi->block_count != 0 );
+	assert ( iscsi->block_read_callback != NULL );
+
+	/* Construct BHS */
+	memset ( command, 0, sizeof ( *command ) );
+	command->opcode = ISCSI_OPCODE_SCSI_COMMAND;
+	command->flags = ( ISCSI_FLAG_FINAL |
+			   ISCSI_COMMAND_FLAG_READ |
+			   ISCSI_COMMAND_ATTR_SIMPLE );
+	/* lengths left as zero */
+	/* lun left as zero, on the assumption that no-one uses LUNs > 0 */
+	command->itt = htonl ( iscsi->itt );
+	command->exp_len = htonl ( iscsi->block_count * iscsi->block_size );
+	command->cmdsn = htonl ( iscsi->cmdsn );
+	command->expstatsn = htonl ( iscsi->statsn + 1 );
+	read->opcode = SCSI_OPCODE_READ_10;
+	read->lba = htonl ( iscsi->block_start );
+	read->len = htons ( iscsi->block_count );
+
+	iscsi->state = ISCSI_STATE_READING_DATA;
+	iscsi_start_tx ( iscsi );
+}
+
+/**
+ * Receive data segment of an iSCSI data-in PDU
+ *
+ * @v iscsi		iSCSI session
+ * @v data		Received data
+ * @v len		Length of received data
+ * @v remaining		Data remaining after this data
+ * 
+ */
+static void iscsi_rx_data_in ( struct iscsi_session *iscsi, void *data,
+			       size_t len, size_t remaining ) {
+	struct iscsi_bhs_data_in *data_in = &iscsi->rx_bhs.data_in;
+	unsigned long offset;
+
+	/* Update cmdsn and statsn */
+	iscsi->cmdsn = ntohl ( data_in->expcmdsn );
+	iscsi->statsn = ntohl ( data_in->statsn );
+
+	/* Process data via callback */
+	offset = ntohl ( data_in->offset ) + iscsi->rx_offset;
+	iscsi->block_read_callback ( iscsi->block_read_private,
+				     data, offset, len );
+
+	/* If this is the end, mark state as idle */
+	if ( ( data_in->flags & ISCSI_FLAG_FINAL ) && ( remaining == 0 ) )
+		iscsi->state = ISCSI_STATE_IDLE;
+}
+
+/****************************************************************************
+ *
+ * iSCSI login
+ *
+ */
+
+/**
+ * Build iSCSI login request strings
+ *
+ * @v iscsi		iSCSI session
+ *
+ * These are the initial set of strings sent in the first login
+ * request PDU.
+ */
+static int iscsi_build_login_request_strings ( struct iscsi_session *iscsi,
+					       void *data, size_t len ) {
+	return snprintf ( data, len,
+			  "InitiatorName=%s:initiator%c"
+			  "TargetName=%s%c"
+			  "MaxRecvDataSegmentLength=512%c"
+			  "SessionType=Normal%c"
+			  "DataDigest=None%c"
+			  "HeaderDigest=None%c",
+			  iscsi->initiator, 0, iscsi->target, 0,
+			  0, 0, 0, 0 );
+}
+
+/**
+ * Transmit data segment of an iSCSI login request PDU
+ *
+ * @v iscsi		iSCSI session
+ *
+ * For login requests, the data segment consists of the login strings.
+ */
+static void iscsi_tx_login_request ( struct iscsi_session *iscsi ) {
+	int len;
+
+	len = iscsi_build_login_request_strings ( iscsi, tcp_buffer,
+						  tcp_buflen );
+	tcp_send ( &iscsi->tcp, tcp_buffer + iscsi->tx_offset,
+		   len - iscsi->tx_offset );
+}
+
+/**
+ * Start up a login request
+ *
+ * @v iscsi		iSCSI session
+ *
+ */
+static void iscsi_start_login ( struct iscsi_session *iscsi ) {
+	struct iscsi_bhs_login_request *request = &iscsi->tx_bhs.login_request;
+	int len;
+
+	/* Construct login request BHS */
+	memset ( request, 0, sizeof ( *request ) );
+	request->opcode = ( ISCSI_OPCODE_LOGIN_REQUEST |
+			    ISCSI_FLAG_IMMEDIATE );
+	request->flags = ( ISCSI_LOGIN_FLAG_TRANSITION |
+			   ISCSI_LOGIN_CSG_OPERATIONAL_NEGOTIATION |
+			   ISCSI_LOGIN_NSG_FULL_FEATURE_PHASE );
+	/* version_max and version_min left as zero */
+	len = iscsi_build_login_request_strings ( iscsi, NULL, 0 );
+	ISCSI_SET_LENGTHS ( request->lengths, 0, len );
+	request->isid_iana_en = htonl ( ISCSI_ISID_IANA |
+					IANA_EN_FEN_SYSTEMS );
+	/* isid_iana_qual left as zero */
+	request->tsih = htons ( iscsi->tsih );
+	/* itt left as zero */
+	/* cid left as zero */
+	request->cmdsn = htonl ( iscsi->cmdsn );
+	request->expstatsn = htonl ( iscsi->statsn + 1 );
+
+	iscsi->state = ISCSI_STATE_LOGGING_IN;
+	iscsi_start_tx ( iscsi );
+}
+
+/**
+ * Receive data segment of an iSCSI login response PDU
+ *
+ * @v iscsi		iSCSI session
+ * @v data		Received data
+ * @v len		Length of received data
+ * @v remaining		Data remaining after this data
+ * 
+ */
+static void iscsi_rx_login_response ( struct iscsi_session *iscsi,
+				      void *data __unused,
+				      size_t len __unused,
+				      size_t remaining __unused ) {
+	struct iscsi_bhs_login_request *request = &iscsi->tx_bhs.login_request;
+	struct iscsi_bhs_login_response *response
+		= &iscsi->rx_bhs.login_response;
+
+	/* Sanity check */
+	if ( iscsi->state != ISCSI_STATE_LOGGING_IN ) {
+		printf ( "Spurious iSCSI login response\n" );
+		iscsi_fail ( iscsi );
+		return;
+	}
+
+	/* Check for fatal errors */
+	if ( response->status_class != 0 ) {
+		printf ( "iSCSI login failure: class %02x detail %02x\n",
+			 response->status_class, response->status_detail );
+		iscsi_fail ( iscsi );
+		return;
+	}
+
+	/* Update cmdsn and statsn */
+	iscsi->cmdsn = ntohl ( response->expcmdsn );
+	iscsi->statsn = ntohl ( response->statsn );
+
+	/* If server did not transition, we send it another login
+	 * request with empty strings.
+	 */
+	if ( ! ( response->flags & ISCSI_LOGIN_FLAG_TRANSITION ) ) {
+		ISCSI_SET_LENGTHS ( request->lengths, 0, 0 );
+		iscsi_start_tx ( iscsi );
+		return;
+	}
+
+	/* Record TSIH  for future reference */
+	iscsi->tsih = ntohl ( response->tsih );
+
+	/* Start reading data */
+	iscsi_start_read_block ( iscsi );
+}
+
+/****************************************************************************
+ *
+ * iSCSI to TCP interface
+ *
+ */
+
+static inline struct iscsi_session *
+tcp_to_iscsi ( struct tcp_connection *conn ) {
+	return container_of ( conn, struct iscsi_session, tcp );
+}
+
+static void iscsi_aborted ( struct tcp_connection *conn ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+
+}
+
+static void iscsi_timedout ( struct tcp_connection *conn ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+
+}
+
+static void iscsi_closed ( struct tcp_connection *conn ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+
+}
+
+static void iscsi_connected ( struct tcp_connection *conn ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+
+	/* Prepare to receive PDUs. */
+	iscsi->rx_state = ISCSI_RX_BHS;
+	iscsi->rx_offset = 0;
+
+	/* TX state should already have been set up */
+	assert ( iscsi->tx_state != ISCSI_TX_IDLE );
+	assert ( iscsi->tx_offset == 0 );
+}
+
+/**
+ * Transmit data segment of an iSCSI PDU
+ *
+ * @v iscsi		iSCSI session
+ * 
+ * Handle transmission of part of a PDU data segment.  iscsi::tx_bhs
+ * will be valid when this is called.
+ */
+static void iscsi_tx_data ( struct iscsi_session *iscsi ) {
+	struct iscsi_bhs_common *common = &iscsi->tx_bhs.common;
+
+	switch ( common->opcode & ISCSI_OPCODE_MASK ) {
+	case ISCSI_OPCODE_LOGIN_REQUEST:
+		iscsi_tx_login_request ( iscsi );
+		break;
+	default:
+		assert ( 0 );
+		break;
+	}
+}
+
+/**
+ * Handle TCP ACKs
+ *
+ * @v iscsi		iSCSI session
+ * 
+ * Updates iscsi->tx_offset and, if applicable, transitions to the
+ * next TX state.
+ */
+static void iscsi_acked ( struct tcp_connection *conn, size_t len ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+	struct iscsi_bhs_common *common = &iscsi->tx_bhs.common;
+	size_t max_tx_offset;
+	enum iscsi_tx_state next_state;
+	
+	iscsi->tx_offset += len;
+	while ( 1 ) {
+		switch ( iscsi->tx_state ) {
+		case ISCSI_TX_BHS:
+			max_tx_offset = sizeof ( iscsi->tx_bhs );
+			next_state = ISCSI_TX_AHS;
+			break;
+		case ISCSI_TX_AHS:
+			max_tx_offset = 4 * ISCSI_AHS_LEN ( common->lengths );
+			next_state = ISCSI_TX_DATA;
+			break;
+		case ISCSI_TX_DATA:
+			max_tx_offset = ISCSI_DATA_LEN ( common->lengths );
+			next_state = ISCSI_TX_DATA_PADDING;
+			break;
+		case ISCSI_TX_DATA_PADDING:
+			max_tx_offset = ISCSI_DATA_PAD_LEN ( common->lengths );
+			next_state = ISCSI_TX_IDLE;
+			break;
+		case ISCSI_TX_IDLE:
+			return;
+		default:
+			assert ( 0 );
+			return;
+		}
+		assert ( iscsi->tx_offset <= max_tx_offset );
+
+		/* If the whole of the current portion has not yet
+		 * been acked, stay in this state for now.
+		 */
+		if ( iscsi->tx_offset != max_tx_offset )
+			return;
+		
+		iscsi->tx_state = next_state;
+		iscsi->tx_offset = 0;
+	}
+}
+
+/**
+ * Transmit iSCSI PDU
+ *
+ * @v iscsi		iSCSI session
+ * 
+ * Constructs data to be sent for the current TX state
+ */
+static void iscsi_senddata ( struct tcp_connection *conn ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+	struct iscsi_bhs_common *common = &iscsi->tx_bhs.common;
+	static const char pad[] = { '\0', '\0', '\0' };
+
+	switch ( iscsi->tx_state ) {
+	case ISCSI_TX_IDLE:
+		/* Do nothing */
+		break;
+	case ISCSI_TX_BHS:
+		tcp_send ( conn, &iscsi->tx_bhs.bytes[iscsi->tx_offset],
+			   ( sizeof ( iscsi->tx_bhs ) - iscsi->tx_offset ) );
+		break;
+	case ISCSI_TX_AHS:
+		/* We don't yet have an AHS transmission mechanism */
+		assert ( 0 );
+		break;
+	case ISCSI_TX_DATA:
+		iscsi_tx_data ( iscsi );
+		break;
+	case ISCSI_TX_DATA_PADDING:
+		tcp_send ( conn, pad, ( ISCSI_DATA_PAD_LEN ( common->lengths )
+					- iscsi->tx_offset ) );
+		break;
+	default:
+		assert ( 0 );
+		break;
+	}
+}
+
+/**
+ * Receive data segment of an iSCSI PDU
+ *
+ * @v iscsi		iSCSI session
+ * @v data		Received data
+ * @v len		Length of received data
+ * @v remaining		Data remaining after this data
+ *
+ * Handle processing of part of a PDU data segment.  iscsi::rx_bhs
+ * will be valid when this is called.
+ */
+static void iscsi_rx_data ( struct iscsi_session *iscsi, void *data,
+			    size_t len, size_t remaining ) {
+	struct iscsi_bhs_common *common = &iscsi->rx_bhs.common;
+
+	switch ( common->opcode & ISCSI_OPCODE_MASK ) {
+	case ISCSI_OPCODE_LOGIN_RESPONSE:
+		iscsi_rx_login_response ( iscsi, data, len, remaining );
+		break;
+	case ISCSI_OPCODE_DATA_IN:
+		iscsi_rx_data_in ( iscsi, data, len, remaining );
+		break;
+	default:
+		printf ( "Unknown iSCSI opcode %02x\n", common->opcode );
+		break;
+	}
+}
+
+/**
+ * Discard portion of an iSCSI PDU.
+ *
+ * @v iscsi		iSCSI session
+ * @v data		Received data
+ * @v len		Length of received data
+ * @v remaining		Data remaining after this data
+ *
+ * This discards data from a portion of a received PDU.
+ */
+static void iscsi_rx_discard ( struct iscsi_session *iscsi __unused,
+			       void *data __unused, size_t len __unused,
+			       size_t remaining __unused ) {
+	/* Do nothing */
+}
+
+/**
+ * Receive basic header segment of an iSCSI PDU
+ *
+ * @v iscsi		iSCSI session
+ * @v data		Received data
+ * @v len		Length of received data
+ * @v remaining		Data remaining after this data
+ *
+ * This fills in iscsi::rx_bhs with the data from the BHS portion of
+ * the received PDU.
+ */
+static void iscsi_rx_bhs ( struct iscsi_session *iscsi, void *data,
+			   size_t len, size_t remaining __unused ) {
+	memcpy ( &iscsi->rx_bhs.bytes[iscsi->rx_offset], data, len );
+}
+
+/**
+ * Receive new data
+ *
+ * @v tcp		TCP connection
+ * @v data		Received data
+ * @v len		Length of received data
+ *
+ * This handles received PDUs.  The receive strategy is to fill in
+ * iscsi::rx_bhs with the contents of the BHS portion of the PDU,
+ * throw away any AHS portion, and then process each part of the data
+ * portion as it arrives.  The data processing routine therefore
+ * always has a full copy of the BHS available, even for portions of
+ * the data in different packets to the BHS.
+ */
+static void iscsi_newdata ( struct tcp_connection *conn, void *data,
+			    size_t len ) {
+	struct iscsi_session *iscsi = tcp_to_iscsi ( conn );
+	struct iscsi_bhs_common *common = &iscsi->rx_bhs.common;
+	void ( *process ) ( struct iscsi_session *iscsi, void *data,
+			    size_t len, size_t remaining );
+	size_t max_rx_offset;
+	enum iscsi_rx_state next_state;
+	size_t frag_len;
+	size_t remaining;
+
+	while ( 1 ) {
+		switch ( iscsi->rx_state ) {
+		case ISCSI_RX_BHS:
+			process = iscsi_rx_bhs;
+			max_rx_offset = sizeof ( iscsi->rx_bhs );
+			next_state = ISCSI_RX_AHS;			
+			break;
+		case ISCSI_RX_AHS:
+			process = iscsi_rx_discard;
+			max_rx_offset = 4 * ISCSI_AHS_LEN ( common->lengths );
+			next_state = ISCSI_RX_DATA;
+			break;
+		case ISCSI_RX_DATA:
+			process = iscsi_rx_data;
+			max_rx_offset = ISCSI_DATA_LEN ( common->lengths );
+			next_state = ISCSI_RX_DATA_PADDING;
+			break;
+		case ISCSI_RX_DATA_PADDING:
+			process = iscsi_rx_discard;
+			max_rx_offset = ISCSI_DATA_PAD_LEN ( common->lengths );
+			next_state = ISCSI_RX_BHS;
+			break;
+		default:
+			assert ( 0 );
+			return;
+		}
+
+		frag_len = max_rx_offset - iscsi->rx_offset;
+		if ( frag_len > len )
+			frag_len = len;
+		remaining = max_rx_offset - iscsi->rx_offset - frag_len;
+		process ( iscsi, data, frag_len, remaining );
+
+		iscsi->rx_offset += frag_len;
+		data += frag_len;
+		len -= frag_len;
+
+		/* If all the data for this state has not yet been
+		 * received, stay in this state for now.
+		 */
+		if ( iscsi->rx_offset != max_rx_offset )
+			return;
+
+		iscsi->rx_state = next_state;
+		iscsi->rx_offset = 0;
+	}
+}
+
+/** iSCSI TCP operations */
+static struct tcp_operations iscsi_tcp_operations = {
+	.aborted	= iscsi_aborted,
+	.timedout	= iscsi_timedout,
+	.closed		= iscsi_closed,
+	.connected	= iscsi_connected,
+	.acked		= iscsi_acked,
+	.newdata	= iscsi_newdata,
+	.senddata	= iscsi_senddata,
+};
+
+/**
+ * Wake up session
+ *
+ * @v iscsi		iSCSI session
+ *
+ */
+void iscsi_wakeup ( struct iscsi_session *iscsi ) {
+	iscsi->tcp.tcp_op = &iscsi_tcp_operations;
+
+	switch ( iscsi->state ) {
+	case ISCSI_STATE_NOT_CONNECTED:
+	case ISCSI_STATE_FAILED:
+		if ( tcp_connect ( &iscsi->tcp ) != 0 )
+			iscsi_fail ( iscsi );
+		iscsi_start_login ( iscsi );
+		break;
+	case ISCSI_STATE_IDLE:
+		iscsi_start_read_block ( iscsi );
+		break;
+	default:
+		/* Stay in same state */
+		break;
+	}
+}