tcp.c File Reference

TCP protocol. More...

#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/init.h>
#include <ipxe/retry.h>
#include <ipxe/refcnt.h>
#include <ipxe/pending.h>
#include <ipxe/xfer.h>
#include <ipxe/open.h>
#include <ipxe/uri.h>
#include <ipxe/netdevice.h>
#include <ipxe/profile.h>
#include <ipxe/process.h>
#include <ipxe/job.h>
#include <ipxe/tcpip.h>
#include <ipxe/tcp.h>

Go to the source code of this file.

Data Structures
struct	tcp_connection
	A TCP connection. More...
struct	tcp_rx_queued_header
	TCP internal header. More...

Enumerations
enum	tcp_flags { TCP_XFER_CLOSED = 0x0001, TCP_TS_ENABLED = 0x0002, TCP_ACK_PENDING = 0x0004, TCP_SACK_ENABLED = 0x0008 }
	TCP flags. More...

Functions
	FILE_LICENCE (GPL2_OR_LATER_OR_UBDL)
	FILE_SECBOOT (PERMITTED)
static	LIST_HEAD (tcp_conns)
	List of registered TCP connections. More...
static void	tcp_expired (struct retry_timer *timer, int over)
	Retransmission timer expired. More...
static void	tcp_keepalive_expired (struct retry_timer *timer, int over __unused)
	Keepalive timer expired. More...
static void	tcp_wait_expired (struct retry_timer *timer, int over __unused)
	Shutdown timer expired. More...
static struct tcp_connection *	tcp_demux (unsigned int local_port)
	Identify TCP connection by local port number. More...
static int	tcp_rx_ack (struct tcp_connection *tcp, uint32_t ack, uint32_t win)
	Handle TCP received ACK. More...
static const char *	tcp_state (int state)
	Name TCP state. More...
static void	tcp_dump_state (struct tcp_connection *tcp)
	Dump TCP state transition. More...
static void	tcp_dump_flags (struct tcp_connection *tcp, unsigned int flags)
	Dump TCP flags. More...
static int	tcp_port_available (int port)
	Check if local TCP port is available. More...
static int	tcp_open (struct interface *xfer, struct sockaddr *peer, struct sockaddr *local)
	Open a TCP connection. More...
static void	tcp_close (struct tcp_connection *tcp, int rc)
	Close TCP connection. More...
static size_t	tcp_xmit_win (struct tcp_connection *tcp)
	Calculate transmission window. More...
static size_t	tcp_xfer_window (struct tcp_connection *tcp)
	Check data-transfer flow control window. More...
static uint32_t	tcp_sack_block (struct tcp_connection *tcp, uint32_t seq, struct tcp_sack_block *sack)
	Find selective acknowledgement block. More...
static unsigned int	tcp_sack (struct tcp_connection *tcp, uint32_t seq)
	Update TCP selective acknowledgement list. More...
static size_t	tcp_process_tx_queue (struct tcp_connection *tcp, size_t max_len, struct io_buffer *dest, int remove)
	Process TCP transmit queue. More...
static void	tcp_xmit_sack (struct tcp_connection *tcp, uint32_t sack_seq)
	Transmit any outstanding data (with selective acknowledgement) More...
static void	tcp_xmit (struct tcp_connection *tcp)
	Transmit any outstanding data. More...
static int	tcp_xmit_reset (struct tcp_connection *tcp, struct sockaddr_tcpip *st_dest, struct tcp_header *in_tcphdr)
	Send RST response to incoming packet. More...
static int	tcp_rx_opts (struct tcp_connection *tcp, const struct tcp_header *tcphdr, size_t hlen, struct tcp_options *options)
	Parse TCP received options. More...
static void	tcp_rx_seq (struct tcp_connection *tcp, uint32_t seq_len)
	Consume received sequence space. More...
static int	tcp_rx_syn (struct tcp_connection *tcp, uint32_t seq, struct tcp_options *options)
	Handle TCP received SYN. More...
static int	tcp_rx_data (struct tcp_connection *tcp, uint32_t seq, struct io_buffer *iobuf)
	Handle TCP received data. More...
static int	tcp_rx_fin (struct tcp_connection *tcp, uint32_t seq)
	Handle TCP received FIN. More...
static int	tcp_rx_rst (struct tcp_connection *tcp, uint32_t seq)
	Handle TCP received RST. More...
static void	tcp_rx_enqueue (struct tcp_connection *tcp, uint32_t seq, uint8_t flags, struct io_buffer *iobuf)
	Enqueue received TCP packet. More...
static void	tcp_process_rx_queue (struct tcp_connection *tcp)
	Process receive queue. More...
static int	tcp_rx (struct io_buffer *iobuf, struct net_device *netdev __unused, struct sockaddr_tcpip *st_src, struct sockaddr_tcpip *st_dest __unused, uint16_t pshdr_csum)
	Process received packet. More...
static unsigned int	tcp_discard (void)
	Discard some cached TCP data. More...
struct cache_discarder tcp_discarder	__cache_discarder (CACHE_NORMAL)
	TCP cache discarder. More...
static struct tcp_connection *	tcp_first_unclosed (void)
	Find first TCP connection that has not yet been closed. More...
static struct tcp_connection *	tcp_first_unfinished (void)
	Find first TCP connection that has not yet finished all operations. More...
static void	tcp_shutdown (int booting __unused)
	Shut down all TCP connections. More...
struct startup_fn tcp_startup_fn	__startup_fn (STARTUP_LATE)
	TCP shutdown function. More...
static void	tcp_xfer_close (struct tcp_connection *tcp, int rc)
	Close interface. More...
static int	tcp_xfer_deliver (struct tcp_connection *tcp, struct io_buffer *iobuf, struct xfer_metadata *meta __unused)
	Deliver datagram as I/O buffer. More...
static int	tcp_progress (struct tcp_connection *tcp, struct job_progress *progress)
	Report job progress. More...
static int	tcp_open_uri (struct interface *xfer, struct uri *uri)
	Open TCP URI. More...

Variables
struct tcp_statistics	tcp_stats
	TCP statistics. More...
static struct profiler tcp_tx_profiler	__profiler = { .name = "tcp.tx" }
	Transmit profiler. More...
static struct process_descriptor	tcp_process_desc
	TCP process descriptor. More...
static struct interface_descriptor	tcp_xfer_desc
	TCP data transfer interface descriptor. More...
struct tcpip_protocol tcp_protocol	__tcpip_protocol
	TCP protocol. More...
static struct interface_operation	tcp_xfer_operations []
	TCP data transfer interface operations. More...
struct socket_opener tcp_socket_opener	__socket_opener
	TCP socket opener. More...
int	tcp_sock_stream = TCP_SOCK_STREAM
	Linkage hack. More...
struct uri_opener tcp_uri_opener	__uri_opener
	TCP URI opener. More...

Detailed Description

TCP protocol.

Definition in file tcp.c.

Enumeration Type Documentation

tcp_flags

enum tcp_flags

TCP flags.

Enumerator
TCP_XFER_CLOSED	TCP data transfer interface has been closed.
TCP_TS_ENABLED	TCP timestamps are enabled.
TCP_ACK_PENDING	TCP acknowledgement is pending.
TCP_SACK_ENABLED	TCP selective acknowledgement is enabled.

Definition at line 130 of file tcp.c.

               {
        /** 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 selective acknowledgement is enabled */
        TCP_SACK_ENABLED = 0x0008,
};

Function Documentation

FILE_LICENCE()

FILE_LICENCE

(

GPL2_OR_LATER_OR_UBDL

)

FILE_SECBOOT()

FILE_SECBOOT

(

PERMITTED

)

LIST_HEAD()

static LIST_HEAD ( tcp_conns  )

static

List of registered TCP connections.

tcp_expired()

static void tcp_expired ( struct retry_timer *  timer, int  over  )

static

Retransmission timer expired.

Parameters

timer	Retransmission timer
over	Failure indicator

Definition at line 789 of file tcp.c.

                                                                {
        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 );
        }
}

References assert(), container_of, DBGC, ETIMEDOUT, tcp_connection::rcv_ack, tcp_connection::snd_sent, tcp_connection::snd_seq, tcp_close(), TCP_CLOSE_WAIT, TCP_CLOSED, TCP_CLOSING_OR_LAST_ACK, tcp_dump_state(), TCP_ESTABLISHED, TCP_FIN_WAIT_1, tcp_connection::tcp_state, tcp_state(), TCP_SYN_RCVD, TCP_SYN_SENT, and tcp_xmit().

Referenced by tcp_open().

tcp_keepalive_expired()

static void tcp_keepalive_expired ( struct retry_timer *  timer, int over  __unused  )

static

Keepalive timer expired.

Parameters

timer	Keepalive timer
over	Failure indicator

Definition at line 823 of file tcp.c.

                                                        {
        struct tcp_connection *tcp =
                container_of ( timer, struct tcp_connection, keepalive );

        DBGC ( tcp, "TCP %p sending keepalive\n", tcp );

        /* Reset keepalive timer */
        start_timer_fixed ( &tcp->keepalive, TCP_KEEPALIVE_DELAY );

        /* Send keepalive.  We do this only to preserve or restore
         * state in intermediate devices (e.g. firewall NAT tables);
         * we don't actually care about eliciting a response to verify
         * that the peer is still alive.  We therefore send just a
         * pure ACK, to keep our transmit path simple.
         */
        tcp->flags |= TCP_ACK_PENDING;
        tcp_xmit ( tcp );
}

References container_of, DBGC, tcp_connection::flags, tcp_connection::keepalive, start_timer_fixed(), TCP_ACK_PENDING, TCP_KEEPALIVE_DELAY, and tcp_xmit().

Referenced by tcp_open().

tcp_wait_expired()

static void tcp_wait_expired ( struct retry_timer *  timer, int over  __unused  )

static

Shutdown timer expired.

Parameters

timer	Shutdown timer
over	Failure indicator

Definition at line 849 of file tcp.c.

                                                                              {
        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 );
}

References assert(), container_of, DBGC, tcp_connection::rcv_ack, tcp_connection::snd_sent, tcp_connection::snd_seq, tcp_close(), TCP_CLOSED, tcp_dump_state(), tcp_connection::tcp_state, tcp_state(), TCP_TIME_WAIT, and tcp_connection::wait.

Referenced by tcp_open().

tcp_demux()

static struct tcp_connection * tcp_demux ( unsigned int  local_port )

static

Identify TCP connection by local port number.

Parameters

local_port

Local port

Return values

tcp	TCP connection, or NULL

Definition at line 933 of file tcp.c.

                                                                     {
        struct tcp_connection *tcp;

        list_for_each_entry ( tcp, &tcp_conns, list ) {
                if ( tcp->local_port == local_port )
                        return tcp;
        }
        return NULL;
}

References tcp_connection::list, list_for_each_entry, tcp_connection::local_port, and NULL.

Referenced by tcp_port_available(), and tcp_rx().

tcp_rx_ack()

static int tcp_rx_ack ( struct tcp_connection *  tcp, uint32_t  ack, uint32_t  win  )

static

Handle TCP received ACK.

Parameters

tcp	TCP connection
ack	ACK value (in host-endian order)
win	WIN value (in host-endian order)

Return values

rc	Return status code

Definition at line 1119 of file tcp.c.

                                       {
        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;
                }
        }

        /* Update window size */
        tcp->snd_win = win;

        /* Hold off (or start) the keepalive timer, if applicable */
        if ( ! ( tcp->tcp_state & TCP_STATE_SENT ( TCP_FIN ) ) )
                start_timer_fixed ( &tcp->keepalive, TCP_KEEPALIVE_DELAY );

        /* 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--;
                pending_put ( &tcp->pending_flags );
        }

        /* Update SEQ and sent counters */
        tcp->snd_seq = ack;
        tcp->snd_sent = 0;

        /* 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 ) ) ) {
                tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
                pending_get ( &tcp->pending_flags );
        }

        return 0;
}

References DBGC, EINVAL, tcp_connection::flags, tcp_connection::keepalive, len, list_empty, NULL, tcp_connection::pending_flags, pending_get(), pending_put(), tcp_connection::snd_sent, tcp_connection::snd_seq, tcp_connection::snd_win, start_timer_fixed(), stop_timer(), TCP_FIN, TCP_FLAGS_SENDING, TCP_HAS_BEEN_ESTABLISHED, TCP_KEEPALIVE_DELAY, tcp_process_tx_queue(), tcp_connection::tcp_state, TCP_STATE_ACKED, TCP_STATE_SENT, TCP_SYN, TCP_XFER_CLOSED, tcp_connection::timer, and tcp_connection::tx_queue.

Referenced by tcp_close(), and tcp_rx().

tcp_state()

static const char* tcp_state ( int  state )

inlinestatic

Name TCP state.

Parameters

state

TCP state

Return values

name	Name of TCP state

Definition at line 200 of file tcp.c.

                        {
        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";
        }
}

References state, TCP_CLOSE_WAIT, TCP_CLOSED, TCP_CLOSING_OR_LAST_ACK, TCP_ESTABLISHED, TCP_FIN_WAIT_1, TCP_FIN_WAIT_2, TCP_LISTEN, TCP_SYN_RCVD, TCP_SYN_SENT, and TCP_TIME_WAIT.

Referenced by tcp_dump_state(), tcp_expired(), and tcp_wait_expired().

tcp_dump_state()

static void tcp_dump_state ( struct tcp_connection *  tcp )

inlinestatic

Dump TCP state transition.

Parameters

tcp	TCP connection

Definition at line 222 of file tcp.c.

                                              {

        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;
}

References DBGC, tcp_connection::prev_tcp_state, tcp_connection::tcp_state, and tcp_state().

Referenced by tcp_close(), tcp_expired(), tcp_open(), tcp_rx(), tcp_rx_rst(), tcp_shutdown(), and tcp_wait_expired().

tcp_dump_flags()

static void tcp_dump_flags ( struct tcp_connection *  tcp, unsigned int  flags  )

inlinestatic

Dump TCP flags.

Parameters

flags

TCP flags

Definition at line 238 of file tcp.c.

                                                                  {
        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" );
}

References DBGC2, flags, TCP_ACK, TCP_FIN, TCP_PSH, TCP_RST, and TCP_SYN.

Referenced by tcp_rx(), tcp_xmit_reset(), and tcp_xmit_sack().

tcp_port_available()

static int tcp_port_available ( int  port )

static

Check if local TCP port is available.

Parameters

port	Local port number

Return values

port	Local port number, or negative error

Definition at line 264 of file tcp.c.

                                           {

        return ( tcp_demux ( port ) ? -EADDRINUSE : port );
}

References EADDRINUSE, port, and tcp_demux().

Referenced by tcp_open().

tcp_open()

static int tcp_open ( struct interface *  xfer, struct sockaddr *  peer, struct sockaddr *  local  )

static

Open a TCP connection.

Parameters

xfer	Data transfer interface
peer	Peer socket address
local	Local socket address, or NULL

Return values

rc	Return status code

Definition at line 277 of file tcp.c.

                                               {
        struct sockaddr_tcpip *st_peer = ( struct sockaddr_tcpip * ) peer;
        struct sockaddr_tcpip *st_local = ( struct sockaddr_tcpip * ) local;
        struct tcp_connection *tcp;
        size_t mtu;
        int 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 );
        process_init_stopped ( &tcp->process, &tcp_process_desc, &tcp->refcnt );
        timer_init ( &tcp->timer, tcp_expired, &tcp->refcnt );
        timer_init ( &tcp->keepalive, tcp_keepalive_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 ) );

        /* Calculate MSS */
        mtu = tcpip_mtu ( &tcp->peer );
        if ( ! mtu ) {
                DBGC ( tcp, "TCP %p has no route to %s\n",
                       tcp, sock_ntoa ( peer ) );
                rc = -ENETUNREACH;
                goto err;
        }
        tcp->mss = ( mtu - sizeof ( struct tcp_header ) );

        /* Bind to local port */
        port = tcpip_bind ( st_local, tcp_port_available );
        if ( port < 0 ) {
                rc = port;
                DBGC ( tcp, "TCP %p could not bind: %s\n",
                       tcp, strerror ( rc ) );
                goto err;
        }
        tcp->local_port = port;
        DBGC ( tcp, "TCP %p bound to port %d\n", tcp, tcp->local_port );

        /* Start timer to initiate SYN */
        start_timer_nodelay ( &tcp->timer );

        /* Add a pending operation for the SYN */
        pending_get ( &tcp->pending_flags );

        /* 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;
}

References DBGC, ENETUNREACH, ENOMEM, INIT_LIST_HEAD, intf_init(), intf_plug_plug(), tcp_connection::keepalive, tcp_connection::list, list_add, tcp_connection::local_port, memcpy(), tcp_connection::mss, mtu, NULL, peer, tcp_connection::peer, tcp_connection::pending_flags, pending_get(), port, tcp_connection::prev_tcp_state, tcp_connection::process, process_init_stopped(), random(), rc, ref_init, ref_put, tcp_connection::refcnt, tcp_connection::rx_queue, tcp_connection::snd_seq, sock_ntoa(), start_timer_nodelay(), strerror(), TCP_CLOSED, tcp_dump_state(), tcp_expired(), tcp_keepalive_expired(), tcp_port_available(), tcp_process_desc, tcp_connection::tcp_state, TCP_STATE_SENT, TCP_SYN, tcp_wait_expired(), tcp_xfer_desc, tcpip_bind(), tcpip_mtu(), tcp_connection::timer, tcp_connection::tx_queue, tcp_connection::wait, tcp_connection::xfer, and zalloc().

tcp_close()

static void tcp_close ( struct tcp_connection *  tcp, int  rc  )

static

Close TCP connection.

Parameters

tcp	TCP connection
rc	Reason for close

Closes the data transfer interface. If the TCP state machine is in a suitable state, the connection will be deleted.

Definition at line 353 of file tcp.c.

                                                             {
        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 );
                        pending_put ( &tcp->pending_data );
                }
                assert ( ! is_pending ( &tcp->pending_data ) );

                /* Remove pending operations for SYN and FIN, if applicable */
                pending_put ( &tcp->pending_flags );
                pending_put ( &tcp->pending_flags );

                /* Remove from list and drop reference */
                process_del ( &tcp->process );
                stop_timer ( &tcp->timer );
                stop_timer ( &tcp->keepalive );
                stop_timer ( &tcp->wait );
                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 );

        /* Stop keepalive timer */
        stop_timer ( &tcp->keepalive );

        /* 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->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
                tcp_dump_state ( tcp );
                process_add ( &tcp->process );

                /* Add a pending operation for the FIN */
                pending_get ( &tcp->pending_flags );
        }
}

References assert(), DBGC, tcp_connection::flags, free_iob(), intf_shutdown(), is_pending(), tcp_connection::keepalive, io_buffer::list, tcp_connection::list, list_del, list_empty, list_for_each_entry_safe, tcp_connection::pending_data, tcp_connection::pending_flags, pending_get(), pending_put(), tcp_connection::process, process_add(), process_del(), rc, ref_put, tcp_connection::refcnt, tcp_connection::rx_queue, tcp_connection::snd_seq, stop_timer(), TCP_CLOSED, tcp_dump_state(), TCP_FIN, tcp_rx_ack(), tcp_connection::tcp_state, TCP_STATE_ACKED, TCP_STATE_RCVD, TCP_STATE_SENT, TCP_SYN, TCP_XFER_CLOSED, tcp_connection::timer, tmp, tcp_connection::tx_queue, tcp_connection::wait, and tcp_connection::xfer.

Referenced by tcp_expired(), tcp_rx_fin(), tcp_rx_rst(), tcp_shutdown(), tcp_wait_expired(), and tcp_xfer_close().

tcp_xmit_win()

static size_t tcp_xmit_win ( struct tcp_connection *  tcp )

static

Calculate transmission window.

Parameters

tcp	TCP connection

Return values

len	Maximum length that can be sent in a single packet

Definition at line 436 of file tcp.c.

                                                          {
        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;
}

References len, tcp_connection::snd_win, TCP_CAN_SEND_DATA, TCP_PATH_MTU, and tcp_connection::tcp_state.

Referenced by tcp_xfer_window(), and tcp_xmit_sack().

tcp_xfer_window()

static size_t tcp_xfer_window ( struct tcp_connection *  tcp )

static

Check data-transfer flow control window.

Parameters

tcp	TCP connection

Return values

len	Length of window

Definition at line 457 of file tcp.c.

                                                             {

        /* 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 );
}

References list_empty, tcp_xmit_win(), and tcp_connection::tx_queue.

Referenced by tcp_rx().

tcp_sack_block()

static uint32_t tcp_sack_block ( struct tcp_connection *  tcp, uint32_t  seq, struct tcp_sack_block *  sack  )

static

Find selective acknowledgement block.

Parameters

tcp	TCP connection
seq	SEQ value in SACK block (in host-endian order)
sack	SACK block to fill in (in host-endian order)

Return values

len	Length of SACK block

Definition at line 478 of file tcp.c.

                                                               {
        struct io_buffer *iobuf;
        struct tcp_rx_queued_header *tcpqhdr;
        uint32_t left = tcp->rcv_ack;
        uint32_t right = left;

        /* Find highest block which does not start after SEQ */
        list_for_each_entry ( iobuf, &tcp->rx_queue, list ) {
                tcpqhdr = iobuf->data;
                if ( tcp_cmp ( tcpqhdr->seq, right ) > 0 ) {
                        if ( tcp_cmp ( tcpqhdr->seq, seq ) > 0 )
                                break;
                        left = tcpqhdr->seq;
                }
                if ( tcp_cmp ( tcpqhdr->nxt, right ) > 0 )
                        right = tcpqhdr->nxt;
        }

        /* Fail if this block does not contain SEQ */
        if ( tcp_cmp ( right, seq ) < 0 )
                return 0;

        /* Populate SACK block */
        sack->left = left;
        sack->right = right;
        return ( right - left );
}

References io_buffer::data, tcp_sack_block::left, list_for_each_entry, tcp_rx_queued_header::nxt, tcp_connection::rcv_ack, tcp_sack_block::right, tcp_connection::rx_queue, seq, tcp_rx_queued_header::seq, and tcp_cmp().

Referenced by tcp_sack().

tcp_sack()

static unsigned int tcp_sack ( struct tcp_connection *  tcp, uint32_t  seq  )

static

Update TCP selective acknowledgement list.

Parameters

tcp	TCP connection
seq	SEQ value in first SACK block (in host-endian order)

Return values

count

Number of SACK blocks

Definition at line 514 of file tcp.c.

                                                                          {
        struct tcp_sack_block sack[TCP_SACK_MAX];
        unsigned int old = 0;
        unsigned int new = 0;
        unsigned int i;
        uint32_t len;

        /* Populate first new SACK block */
        len = tcp_sack_block ( tcp, seq, &sack[0] );
        if ( len )
                new++;

        /* Populate remaining new SACK blocks based on old SACK blocks */
        for ( old = 0 ; old < TCP_SACK_MAX ; old++ ) {

                /* Stop if we run out of space in the new list */
                if ( new == TCP_SACK_MAX )
                        break;

                /* Skip empty old SACK blocks */
                if ( tcp->sack[old].left == tcp->sack[old].right )
                        continue;

                /* Populate new SACK block */
                len = tcp_sack_block ( tcp, tcp->sack[old].left, &sack[new] );
                if ( len == 0 )
                        continue;

                /* Eliminate duplicates */
                for ( i = 0 ; i < new ; i++ ) {
                        if ( sack[i].left == sack[new].left ) {
                                new--;
                                break;
                        }
                }
                new++;
        }

        /* Update SACK list */
        memset ( tcp->sack, 0, sizeof ( tcp->sack ) );
        memcpy ( tcp->sack, sack, ( new * sizeof ( tcp->sack[0] ) ) );
        return new;
}

References tcp_sack_block::left, len, memcpy(), memset(), old, tcp_sack_block::right, tcp_connection::sack, seq, tcp_sack_block(), and TCP_SACK_MAX.

Referenced by tcp_xmit_sack().

tcp_process_tx_queue()

static size_t tcp_process_tx_queue ( struct tcp_connection *  tcp, size_t  max_len, struct io_buffer *  dest, int  remove  )

static

Process TCP transmit queue.

Parameters

tcp	TCP connection
max_len	Maximum length to process
dest	I/O buffer to fill with data, or NULL
remove	Remove data from queue

Return values

len	Length of data processed

This processes at most max_len bytes from the TCP connection's transmit queue. Data will be copied into the dest I/O buffer (if provided) and, if remove is true, removed from the transmit queue.

Definition at line 572 of file tcp.c.

                                                                          {
        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 );
                                pending_put ( &tcp->pending_data );
                        }
                }
                len += frag_len;
                max_len -= frag_len;
        }
        return len;
}

References io_buffer::data, dest, free_iob(), iob_len(), iob_pull, iob_put, len, io_buffer::list, list_del, list_for_each_entry_safe, memcpy(), tcp_connection::pending_data, pending_put(), remove, tmp, and tcp_connection::tx_queue.

Referenced by tcp_rx_ack(), and tcp_xmit_sack().

tcp_xmit_sack()

static void tcp_xmit_sack ( struct tcp_connection *  tcp, uint32_t  sack_seq  )

static

Transmit any outstanding data (with selective acknowledgement)

Parameters

tcp	TCP connection
sack_seq	SEQ for first selective acknowledgement (if any)

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.

Definition at line 613 of file tcp.c.

                                                                            {
        struct io_buffer *iobuf;
        struct tcp_header *tcphdr;
        struct tcp_mss_option *mssopt;
        struct tcp_window_scale_padded_option *wsopt;
        struct tcp_timestamp_padded_option *tsopt;
        struct tcp_sack_permitted_padded_option *spopt;
        struct tcp_sack_padded_option *sackopt;
        struct tcp_sack_block *sack;
        void *payload;
        unsigned int flags;
        unsigned int sack_count;
        unsigned int i;
        size_t len = 0;
        size_t sack_len;
        uint32_t seq_len;
        uint32_t max_rcv_win;
        uint32_t max_representable_win;
        int rc;

        /* Start profiling */
        profile_start ( &tcp_tx_profiler );

        /* If retransmission timer is already running, do nothing */
        if ( timer_running ( &tcp->timer ) )
                return;

        /* 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;

        /* 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 + TCP_MAX_HEADER_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;
        }
        iob_reserve ( iobuf, TCP_MAX_HEADER_LEN );

        /* Fill data payload from transmit queue */
        tcp_process_tx_queue ( tcp, len, iobuf, 0 );

        /* Expand receive window if possible */
        max_rcv_win = xfer_window ( &tcp->xfer );
        if ( max_rcv_win > TCP_MAX_WINDOW_SIZE )
                max_rcv_win = TCP_MAX_WINDOW_SIZE;
        max_representable_win = ( 0xffff << tcp->rcv_win_scale );
        if ( max_rcv_win > max_representable_win )
                max_rcv_win = max_representable_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 );
                wsopt = iob_push ( iobuf, sizeof ( *wsopt ) );
                wsopt->nop = TCP_OPTION_NOP;
                wsopt->wsopt.kind = TCP_OPTION_WS;
                wsopt->wsopt.length = sizeof ( wsopt->wsopt );
                wsopt->wsopt.scale = TCP_RX_WINDOW_SCALE;
                spopt = iob_push ( iobuf, sizeof ( *spopt ) );
                memset ( spopt->nop, TCP_OPTION_NOP, sizeof ( spopt->nop ) );
                spopt->spopt.kind = TCP_OPTION_SACK_PERMITTED;
                spopt->spopt.length = sizeof ( spopt->spopt );
        }
        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 ( ( tcp->flags & TCP_SACK_ENABLED ) &&
             ( ! list_empty ( &tcp->rx_queue ) ) &&
             ( ( sack_count = tcp_sack ( tcp, sack_seq ) ) != 0 ) ) {
                sack_len = ( sack_count * sizeof ( *sack ) );
                sackopt = iob_push ( iobuf, ( sizeof ( *sackopt ) + sack_len ));
                memset ( sackopt->nop, TCP_OPTION_NOP, sizeof ( sackopt->nop ));
                sackopt->sackopt.kind = TCP_OPTION_SACK;
                sackopt->sackopt.length =
                        ( sizeof ( sackopt->sackopt ) + sack_len );
                sack = ( ( ( void * ) sackopt ) + sizeof ( *sackopt ) );
                for ( i = 0 ; i < sack_count ; i++, sack++ ) {
                        sack->left = htonl ( tcp->sack[i].left );
                        sack->right = htonl ( tcp->sack[i].right );
                }
        }
        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 >> tcp->rcv_win_scale );
        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;
        }

        /* Clear ACK-pending flag */
        tcp->flags &= ~TCP_ACK_PENDING;

        profile_stop ( &tcp_tx_profiler );
}

References tcp_header::ack, alloc_iob(), assert(), tcp_header::csum, currticks(), io_buffer::data, DBGC, DBGC2, tcp_header::dest, flags, tcp_header::flags, tcp_connection::flags, tcp_header::hlen, htonl, htons, iob_len(), iob_push, iob_reserve, tcp_mss_option::kind, tcp_window_scale_option::kind, tcp_sack_permitted_option::kind, tcp_sack_option::kind, tcp_timestamp_option::kind, tcp_sack_block::left, len, tcp_mss_option::length, tcp_window_scale_option::length, tcp_sack_permitted_option::length, tcp_sack_option::length, tcp_timestamp_option::length, list_empty, tcp_connection::local_port, memset(), tcp_connection::mss, tcp_mss_option::mss, tcp_sack_permitted_padded_option::nop, tcp_sack_padded_option::nop, tcp_timestamp_padded_option::nop, ntohl, ntohs, NULL, tcp_connection::peer, profile_start(), profile_stop(), rc, tcp_connection::rcv_ack, tcp_connection::rcv_win, tcp_connection::rcv_win_scale, tcp_sack_block::right, tcp_connection::rx_queue, tcp_connection::sack, tcp_sack_padded_option::sackopt, tcp_window_scale_option::scale, tcp_header::seq, tcp_connection::snd_sent, tcp_connection::snd_seq, tcp_sack_permitted_padded_option::spopt, tcp_header::src, sockaddr_tcpip::st_port, start_timer(), strerror(), TCP_ACK_PENDING, TCP_CAN_SEND_DATA, tcp_dump_flags(), TCP_FIN, TCP_FLAGS_SENDING, TCP_MAX_HEADER_LEN, TCP_MAX_WINDOW_SIZE, TCP_OPTION_MSS, TCP_OPTION_NOP, TCP_OPTION_SACK, TCP_OPTION_SACK_PERMITTED, TCP_OPTION_TS, TCP_OPTION_WS, tcp_process_tx_queue(), TCP_PSH, TCP_RX_WINDOW_SCALE, tcp_sack(), TCP_SACK_ENABLED, tcp_connection::tcp_state, TCP_SYN, TCP_TS_ENABLED, tcp_xmit_win(), tcpip_chksum(), tcpip_tx(), tcp_connection::timer, tcp_connection::ts_recent, tcp_timestamp_option::tsecr, tcp_timestamp_padded_option::tsopt, tcp_timestamp_option::tsval, tcp_header::win, wsopt, tcp_connection::xfer, and xfer_window().

Referenced by tcp_rx(), and tcp_xmit().

tcp_xmit()

static void tcp_xmit ( struct tcp_connection *  tcp )

static

Transmit any outstanding data.

Parameters

tcp	TCP connection

Definition at line 773 of file tcp.c.

                                                    {

        /* Transmit without an explicit first SACK */
        tcp_xmit_sack ( tcp, tcp->rcv_ack );
}

References tcp_connection::rcv_ack, and tcp_xmit_sack().

Referenced by tcp_expired(), tcp_keepalive_expired(), tcp_xfer_close(), and tcp_xfer_deliver().

tcp_xmit_reset()

static int tcp_xmit_reset ( struct tcp_connection *  tcp, struct sockaddr_tcpip *  st_dest, struct tcp_header *  in_tcphdr  )

static

Send RST response to incoming packet.

Parameters

in_tcphdr

TCP header of incoming packet

Return values

rc	Return status code

Definition at line 870 of file tcp.c.

                                                           {
        struct io_buffer *iobuf;
        struct tcp_header *tcphdr;
        int rc;

        /* Allocate space for dataless TX buffer */
        iobuf = alloc_iob ( TCP_MAX_HEADER_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, TCP_MAX_HEADER_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 ( 0 );
        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;
}

References tcp_header::ack, alloc_iob(), tcp_header::csum, io_buffer::data, DBGC, DBGC2, tcp_header::dest, ENOMEM, tcp_header::flags, tcp_header::hlen, htons, iob_len(), iob_push, iob_reserve, memset(), ntohl, ntohs, NULL, rc, tcp_header::seq, tcp_header::src, strerror(), TCP_ACK, tcp_dump_flags(), TCP_MAX_HEADER_LEN, TCP_RST, tcpip_chksum(), tcpip_tx(), and tcp_header::win.

Referenced by tcp_rx().

tcp_rx_opts()

static int tcp_rx_opts ( struct tcp_connection *  tcp, const struct tcp_header *  tcphdr, size_t  hlen, struct tcp_options *  options  )

static

Parse TCP received options.

Parameters

tcp	TCP connection (may be NULL)
tcphdr	TCP header
hlen	TCP header length
options	Options structure to fill in

Return values

rc	Return status code

Definition at line 952 of file tcp.c.

                                                       {
        const void *data = ( ( ( void * ) tcphdr ) + sizeof ( *tcphdr ) );
        const void *end = ( ( ( void * ) tcphdr ) + hlen );
        const struct tcp_option *option;
        unsigned int kind;
        size_t remaining;
        size_t min;

        /* Sanity check */
        assert ( hlen >= sizeof ( *tcphdr ) );

        /* Parse options */
        memset ( options, 0, sizeof ( *options ) );
        while ( ( remaining = ( end - data ) ) ) {

                /* Extract option code */
                option = data;
                kind = option->kind;

                /* Handle single-byte options */
                if ( kind == TCP_OPTION_END )
                        break;
                if ( kind == TCP_OPTION_NOP ) {
                        data++;
                        continue;
                }

                /* Handle multi-byte options */
                min = sizeof ( *option );
                switch ( kind ) {
                case TCP_OPTION_MSS:
                        /* Ignore received MSS */
                        break;
                case TCP_OPTION_WS:
                        options->wsopt = data;
                        min = sizeof ( *options->wsopt );
                        break;
                case TCP_OPTION_SACK_PERMITTED:
                        options->spopt = data;
                        min = sizeof ( *options->spopt );
                        break;
                case TCP_OPTION_SACK:
                        /* Ignore received SACKs */
                        break;
                case TCP_OPTION_TS:
                        options->tsopt = data;
                        min = sizeof ( *options->tsopt );
                        break;
                default:
                        DBGC ( tcp, "TCP %p received unknown option %d\n",
                               tcp, kind );
                        break;
                }
                if ( remaining < min ) {
                        DBGC ( tcp, "TCP %p received truncated option %d\n",
                               tcp, kind );
                        return -EINVAL;
                }
                if ( option->length < min ) {
                        DBGC ( tcp, "TCP %p received underlength option %d\n",
                               tcp, kind );
                        return -EINVAL;
                }
                if ( option->length > remaining ) {
                        DBGC ( tcp, "TCP %p received overlength option %d\n",
                               tcp, kind );
                        return -EINVAL;
                }
                data += option->length;
        }

        return 0;
}

References assert(), data, DBGC, EINVAL, end, tcp_option::kind, memset(), min, options, TCP_OPTION_END, TCP_OPTION_MSS, TCP_OPTION_NOP, TCP_OPTION_SACK, TCP_OPTION_SACK_PERMITTED, TCP_OPTION_TS, and TCP_OPTION_WS.

Referenced by tcp_rx().

tcp_rx_seq()

static void tcp_rx_seq ( struct tcp_connection *  tcp, uint32_t  seq_len  )

static

Consume received sequence space.

Parameters

tcp	TCP connection
seq_len	Sequence space length to consume

Definition at line 1034 of file tcp.c.

                                                                        {
        unsigned int sack;

        /* Sanity check */
        assert ( seq_len > 0 );

        /* Update acknowledgement number */
        tcp->rcv_ack += seq_len;

        /* Update window */
        if ( tcp->rcv_win > seq_len ) {
                tcp->rcv_win -= seq_len;
        } else {
                tcp->rcv_win = 0;
        }

        /* Update timestamp */
        tcp->ts_recent = tcp->ts_val;

        /* Update SACK list */
        for ( sack = 0 ; sack < TCP_SACK_MAX ; sack++ ) {
                if ( tcp->sack[sack].left == tcp->sack[sack].right )
                        continue;
                if ( tcp_cmp ( tcp->sack[sack].left, tcp->rcv_ack ) < 0 )
                        tcp->sack[sack].left = tcp->rcv_ack;
                if ( tcp_cmp ( tcp->sack[sack].right, tcp->rcv_ack ) < 0 )
                        tcp->sack[sack].right = tcp->rcv_ack;
        }

        /* Mark ACK as pending */
        tcp->flags |= TCP_ACK_PENDING;
}

References assert(), tcp_connection::flags, tcp_sack_block::left, tcp_connection::rcv_ack, tcp_connection::rcv_win, tcp_sack_block::right, tcp_connection::sack, TCP_ACK_PENDING, tcp_cmp(), TCP_SACK_MAX, tcp_connection::ts_recent, and tcp_connection::ts_val.

Referenced by tcp_rx_data(), tcp_rx_fin(), and tcp_rx_syn().

tcp_rx_syn()

static int tcp_rx_syn ( struct tcp_connection *  tcp, uint32_t  seq, struct tcp_options *  options  )

static

Handle TCP received SYN.

Parameters

tcp	TCP connection
seq	SEQ value (in host-endian order)
options	TCP options

Return values

rc	Return status code

Definition at line 1075 of file tcp.c.

                                                      {

        /* 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;
                if ( options->spopt )
                        tcp->flags |= TCP_SACK_ENABLED;
                if ( options->wsopt ) {
                        tcp->snd_win_scale = options->wsopt->scale;
                        tcp->rcv_win_scale = TCP_RX_WINDOW_SCALE;
                }
                DBGC ( tcp, "TCP %p using %stimestamps, %sSACK, TX window "
                       "x%d, RX window x%d\n", tcp,
                       ( ( tcp->flags & TCP_TS_ENABLED ) ? "" : "no " ),
                       ( ( tcp->flags & TCP_SACK_ENABLED ) ? "" : "no " ),
                       ( 1 << tcp->snd_win_scale ),
                       ( 1 << tcp->rcv_win_scale ) );
        }

        /* 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;
}

References DBGC, tcp_connection::flags, options, tcp_connection::rcv_ack, tcp_connection::rcv_win_scale, seq, tcp_connection::snd_win_scale, TCP_ACK, tcp_rx_seq(), TCP_RX_WINDOW_SCALE, TCP_SACK_ENABLED, tcp_connection::tcp_state, TCP_STATE_RCVD, TCP_STATE_SENT, TCP_SYN, and TCP_TS_ENABLED.

Referenced by tcp_rx().

tcp_rx_data()

static int tcp_rx_data ( struct tcp_connection *  tcp, uint32_t  seq, struct io_buffer *  iobuf  )

static

Handle TCP received data.

Parameters

tcp	TCP connection
seq	SEQ value (in host-endian order)
iobuf	I/O buffer

Return values

rc	Return status code

This function takes ownership of the I/O buffer.

Definition at line 1204 of file tcp.c.

                                                   {
        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 );

        /* Update statistics */
        tcp_stats.in_octets_good += len;

        /* Deliver data to application */
        profile_start ( &tcp_xfer_profiler );
        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;
        }
        profile_stop ( &tcp_xfer_profiler );

        return 0;
}

References DBGC, free_iob(), tcp_statistics::in_octets_good, iob_len(), iob_pull, len, profile_start(), profile_stop(), rc, tcp_connection::rcv_ack, seq, strerror(), tcp_rx_seq(), tcp_stats, tcp_connection::xfer, and xfer_deliver_iob().

Referenced by tcp_process_rx_queue().

tcp_rx_fin()

static int tcp_rx_fin ( struct tcp_connection *  tcp, uint32_t  seq  )

static

Handle TCP received FIN.

Parameters

tcp	TCP connection
seq	SEQ value (in host-endian order)

Return values

rc	Return status code

Definition at line 1245 of file tcp.c.

                                                                   {

        /* 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;
}

References tcp_connection::rcv_ack, seq, tcp_close(), TCP_FIN, tcp_rx_seq(), tcp_connection::tcp_state, and TCP_STATE_RCVD.

Referenced by tcp_process_rx_queue().

tcp_rx_rst()

static int tcp_rx_rst ( struct tcp_connection *  tcp, uint32_t  seq  )

static

Handle TCP received RST.

Parameters

tcp	TCP connection
seq	SEQ value (in host-endian order)

Return values

rc	Return status code

Definition at line 1270 of file tcp.c.

                                                                   {

        /* 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;
}

References DBGC, ECONNRESET, tcp_connection::rcv_ack, tcp_connection::rcv_win, seq, tcp_close(), TCP_CLOSED, tcp_dump_state(), tcp_in_window(), tcp_connection::tcp_state, TCP_STATE_ACKED, TCP_STATE_RCVD, and TCP_SYN.

Referenced by tcp_rx().

tcp_rx_enqueue()

static void tcp_rx_enqueue ( struct tcp_connection *  tcp, uint32_t  seq, uint8_t  flags, struct io_buffer *  iobuf  )

static

Enqueue received TCP packet.

Parameters

tcp	TCP connection
seq	SEQ value (in host-endian order)
flags	TCP flags
iobuf	I/O buffer

Definition at line 1302 of file tcp.c.

                                                                      {
        struct tcp_rx_queued_header *tcpqhdr;
        struct io_buffer *queued;
        size_t len;
        uint32_t seq_len;
        uint32_t nxt;
        uint32_t gap;

        /* 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 ) );
        nxt = ( seq + seq_len );

        /* 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 ( nxt, tcp->rcv_ack ) <= 0 ) ||
             ( seq_len == 0 ) ) {
                free_iob ( iobuf );
                return;
        }

        /* Add internal header */
        tcpqhdr = iob_push ( iobuf, sizeof ( *tcpqhdr ) );
        tcpqhdr->seq = seq;
        tcpqhdr->nxt = nxt;
        tcpqhdr->flags = flags;

        /* Add to RX queue */
        gap = tcp->rcv_ack;
        list_for_each_entry ( queued, &tcp->rx_queue, list ) {
                tcpqhdr = queued->data;
                if ( tcp_cmp ( seq, tcpqhdr->seq ) < 0 )
                        break;
                gap = tcpqhdr->nxt;
        }
        list_add_tail ( &iobuf->list, &queued->list );

        /* Update statistics */
        if ( seq != gap )
                tcp_stats.in_out_of_order++;
}

References io_buffer::data, flags, tcp_rx_queued_header::flags, free_iob(), tcp_statistics::in_out_of_order, iob_len(), iob_push, len, io_buffer::list, list_add_tail, list_for_each_entry, tcp_rx_queued_header::nxt, tcp_connection::rcv_ack, tcp_connection::rcv_win, tcp_connection::rx_queue, seq, tcp_rx_queued_header::seq, tcp_cmp(), TCP_FIN, tcp_connection::tcp_state, TCP_STATE_RCVD, tcp_stats, and TCP_SYN.

Referenced by tcp_rx().

tcp_process_rx_queue()

static void tcp_process_rx_queue ( struct tcp_connection *  tcp )

static

Process receive queue.

Parameters

tcp	TCP connection

Definition at line 1360 of file tcp.c.

                                                                {
        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++;
                }
        }
}

References io_buffer::data, flags, tcp_rx_queued_header::flags, iob_disown, iob_len(), iob_pull, len, io_buffer::list, list_del, list_first_entry, tcp_connection::rcv_ack, tcp_connection::rx_queue, seq, tcp_rx_queued_header::seq, tcp_cmp(), TCP_FIN, tcp_rx_data(), and tcp_rx_fin().

Referenced by tcp_rx().

tcp_rx()

static int tcp_rx ( struct io_buffer *  iobuf, struct net_device *netdev  __unused, struct sockaddr_tcpip *  st_src, struct sockaddr_tcpip *st_dest  __unused, uint16_t  pshdr_csum  )

static

Process received packet.

Parameters

iobuf	I/O buffer
netdev	Network device
st_src	Partially-filled source address
st_dest	Partially-filled destination address
pshdr_csum	Pseudo-header checksum

Return values

rc	Return status code

Definition at line 1409 of file tcp.c.

                                          {
        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;
        uint16_t raw_win;
        uint32_t win;
        unsigned int flags;
        size_t len;
        uint32_t seq_len;
        size_t old_xfer_window;
        int rc;

        /* Start profiling */
        profile_start ( &tcp_rx_profiler );

        /* 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 );
        raw_win = ntohs ( tcphdr->win );
        flags = tcphdr->flags;
        if ( ( rc = tcp_rx_opts ( tcp, tcphdr, hlen, &options ) ) != 0 )
                goto discard;
        if ( tcp && options.tsopt )
                tcp->ts_val = ntohl ( options.tsopt->tsval );
        iob_pull ( iobuf, hlen );
        len = iob_len ( iobuf );
        seq_len = ( len + ( ( flags & TCP_SYN ) ? 1 : 0 ) +
                    ( ( flags & TCP_FIN ) ? 1 : 0 ) );

        /* Update statistics */
        tcp_stats.in_segs++;
        tcp_stats.in_octets += len;

        /* 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, silently drop packet */
        if ( ! tcp ) {
                rc = -ENOTCONN;
                goto discard;
        }

        /* Record old data-transfer window */
        old_xfer_window = tcp_xfer_window ( tcp );

        /* Handle ACK, if present */
        if ( flags & TCP_ACK ) {
                win = ( raw_win << tcp->snd_win_scale );
                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 );

        /* Schedule transmission of ACK (and any pending data).  If we
         * have received any out-of-order packets (i.e. if the receive
         * queue remains non-empty after processing) then send the ACK
         * immediately in order to trigger Fast Retransmission.
         */
        if ( list_empty ( &tcp->rx_queue ) ) {
                process_add ( &tcp->process );
        } else {
                tcp_xmit_sack ( tcp, seq );
        }

        /* 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 ) );
        }

        /* Notify application if window has changed */
        if ( tcp_xfer_window ( tcp ) != old_xfer_window )
                xfer_window_changed ( &tcp->xfer );

        profile_stop ( &tcp_rx_profiler );
        return 0;

 discard:
        /* Free received packet */
        free_iob ( iobuf );
        return rc;
}

References tcp_header::ack, io_buffer::data, DBG, DBGC2, tcp_header::dest, EINVAL, ENOTCONN, flags, tcp_header::flags, tcp_connection::flags, free_iob(), tcp_header::hlen, tcp_statistics::in_octets, tcp_statistics::in_segs, iob_disown, iob_len(), iob_pull, len, list_empty, ntohl, ntohs, options, tcp_connection::process, process_add(), profile_start(), profile_stop(), rc, tcp_connection::rcv_ack, tcp_connection::rx_queue, seq, tcp_header::seq, tcp_connection::snd_win_scale, tcp_header::src, start_timer_fixed(), stop_timer(), TCP_ACK, TCP_ACK_PENDING, TCP_CLOSED_GRACEFULLY, tcp_demux(), tcp_dump_flags(), tcp_dump_state(), TCP_FIN, TCP_MASK_HLEN, TCP_MSL, tcp_process_rx_queue(), TCP_RST, tcp_rx_ack(), tcp_rx_enqueue(), tcp_rx_opts(), tcp_rx_rst(), tcp_rx_syn(), tcp_connection::tcp_state, TCP_STATE_RCVD, tcp_stats, TCP_SYN, tcp_xfer_window(), tcp_xmit_reset(), tcp_xmit_sack(), tcpip_continue_chksum(), tcp_connection::ts_val, tcp_connection::wait, tcp_header::win, tcp_connection::xfer, and xfer_window_changed().

tcp_discard()

static unsigned int tcp_discard ( void  )

static

Discard some cached TCP data.

Return values

discarded

Number of cached items discarded

Definition at line 1577 of file tcp.c.

                                         {
        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 ) {

                        /* Remove packet from queue */
                        list_del ( &iobuf->list );
                        free_iob ( iobuf );

                        /* Update statistics */
                        tcp_stats.in_discards++;

                        /* Report discard */
                        discarded++;
                        break;
                }
        }

        return discarded;
}

References free_iob(), tcp_statistics::in_discards, io_buffer::list, list_del, list_for_each_entry, list_for_each_entry_reverse, tcp_connection::rx_queue, and tcp_stats.

__cache_discarder()

struct cache_discarder tcp_discarder __cache_discarder

(

CACHE_NORMAL

)

TCP cache discarder.

tcp_first_unclosed()

static struct tcp_connection* tcp_first_unclosed ( void  )

static

Find first TCP connection that has not yet been closed.

Return values

tcp	First unclosed connection, or NULL

Definition at line 1612 of file tcp.c.

                                                           {
        struct tcp_connection *tcp;

        /* Find first connection which has not yet been closed */
        list_for_each_entry ( tcp, &tcp_conns, list ) {
                if ( ! ( tcp->flags & TCP_XFER_CLOSED ) )
                        return tcp;
        }
        return NULL;
}

References tcp_connection::flags, tcp_connection::list, list_for_each_entry, NULL, and TCP_XFER_CLOSED.

Referenced by tcp_shutdown().

tcp_first_unfinished()

static struct tcp_connection* tcp_first_unfinished ( void  )

static

Find first TCP connection that has not yet finished all operations.

Return values

tcp	First unfinished connection, or NULL

Definition at line 1628 of file tcp.c.

                                                             {
        struct tcp_connection *tcp;

        /* Find first connection which has not yet closed gracefully,
         * or which still has a pending transmission (e.g. to ACK the
         * received FIN).
         */
        list_for_each_entry ( tcp, &tcp_conns, list ) {
                if ( ( ! TCP_CLOSED_GRACEFULLY ( tcp->tcp_state ) ) ||
                     process_running ( &tcp->process ) ) {
                        return tcp;
                }
        }
        return NULL;
}

References tcp_connection::list, list_for_each_entry, NULL, tcp_connection::process, process_running(), TCP_CLOSED_GRACEFULLY, and tcp_connection::tcp_state.

Referenced by tcp_shutdown().

tcp_shutdown()

static void tcp_shutdown ( int booting  __unused )

static

Shut down all TCP connections.

Definition at line 1648 of file tcp.c.

                                                  {
        struct tcp_connection *tcp;
        unsigned long start;
        unsigned long elapsed;

        /* Initiate a graceful close of all connections, allowing for
         * the fact that the connection list may change as we do so.
         */
        while ( ( tcp = tcp_first_unclosed() ) ) {
                DBGC ( tcp, "TCP %p closing for shutdown\n", tcp );
                tcp_close ( tcp, -ECANCELED );
        }

        /* Wait for all connections to finish closing gracefully */
        start = currticks();
        while ( ( tcp = tcp_first_unfinished() ) &&
                ( ( elapsed = ( currticks() - start ) ) < TCP_FINISH_TIMEOUT )){
                step();
        }

        /* Forcibly close any remaining connections */
        while ( ( tcp = list_first_entry ( &tcp_conns, struct tcp_connection,
                                           list ) ) != NULL ) {
                tcp->tcp_state = TCP_CLOSED;
                tcp_dump_state ( tcp );
                tcp_close ( tcp, -ECANCELED );
        }
}

References currticks(), DBGC, ECANCELED, tcp_connection::list, list_first_entry, NULL, start, step(), tcp_close(), TCP_CLOSED, tcp_dump_state(), TCP_FINISH_TIMEOUT, tcp_first_unclosed(), tcp_first_unfinished(), and tcp_connection::tcp_state.

__startup_fn()

struct startup_fn tcp_startup_fn __startup_fn

(

STARTUP_LATE

)

TCP shutdown function.

tcp_xfer_close()

static void tcp_xfer_close ( struct tcp_connection *  tcp, int  rc  )

static

Close interface.

Parameters

tcp	TCP connection
rc	Reason for close

Definition at line 1696 of file tcp.c.

                                                                  {

        /* Close data transfer interface */
        tcp_close ( tcp, rc );

        /* Transmit FIN, if possible */
        tcp_xmit ( tcp );
}

References rc, tcp_close(), and tcp_xmit().

tcp_xfer_deliver()

static int tcp_xfer_deliver ( struct tcp_connection *  tcp, struct io_buffer *  iobuf, struct xfer_metadata *meta  __unused  )

static

Deliver datagram as I/O buffer.

Parameters

tcp	TCP connection
iobuf	Datagram I/O buffer
meta	Data transfer metadata

Return values

rc	Return status code

Definition at line 1713 of file tcp.c.

                                                                    {

        /* Enqueue packet */
        list_add_tail ( &iobuf->list, &tcp->tx_queue );

        /* Each enqueued packet is a pending operation */
        pending_get ( &tcp->pending_data );

        /* Transmit data, if possible */
        tcp_xmit ( tcp );

        return 0;
}

References io_buffer::list, list_add_tail, tcp_connection::pending_data, pending_get(), tcp_xmit(), and tcp_connection::tx_queue.

tcp_progress()

static int tcp_progress ( struct tcp_connection *  tcp, struct job_progress *  progress  )

static

Report job progress.

Parameters

tcp	TCP connection
progress	Progress report to fill in

Return values

ongoing_rc

Ongoing job status code (if known)

Definition at line 1736 of file tcp.c.

                                                          {

        /* Report connection in progress if applicable */
        if ( ! TCP_HAS_BEEN_ESTABLISHED ( tcp->tcp_state ) ) {
                snprintf ( progress->message, sizeof ( progress->message ),
                           "connecting" );
        }

        return 0;
}

References job_progress::message, snprintf(), TCP_HAS_BEEN_ESTABLISHED, and tcp_connection::tcp_state.

tcp_open_uri()

static int tcp_open_uri ( struct interface *  xfer, struct uri *  uri  )

static

Open TCP URI.

Parameters

xfer	Data transfer interface
uri	URI

Return values

rc	Return status code

Definition at line 1783 of file tcp.c.

                                                                    {
        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 );
}

References EINVAL, uri::host, htons, memset(), NULL, peer, SOCK_STREAM, uri_port(), and xfer_open_named_socket().

Variable Documentation

tcp_stats

struct tcp_statistics tcp_stats

TCP statistics.

Definition at line 172 of file tcp.c.

Referenced by ipstat(), tcp_discard(), tcp_rx(), tcp_rx_data(), and tcp_rx_enqueue().

__profiler

struct profiler tcp_xfer_profiler __profiler = { .name = "tcp.tx" }

static

Transmit profiler.

Data transfer profiler.

Receive profiler.

Definition at line 175 of file tcp.c.

tcp_process_desc

static struct process_descriptor tcp_process_desc

static

Initial value:

=
        PROC_DESC_ONCE ( struct tcp_connection, process, tcp_xmit )

TCP process descriptor.

Definition at line 184 of file tcp.c.

Referenced by tcp_open().

tcp_xfer_desc

static struct interface_descriptor tcp_xfer_desc

static

Initial value:

=
        INTF_DESC ( struct tcp_connection, xfer, tcp_xfer_operations )

TCP data transfer interface descriptor.

Definition at line 185 of file tcp.c.

Referenced by tcp_open().

__tcpip_protocol

struct tcpip_protocol tcp_protocol __tcpip_protocol

Initial value:

= {
        .name = "TCP",
        .rx = tcp_rx,
        .tcpip_proto = IP_TCP,
}

TCP protocol.

ICMPv4 TCP/IP protocol.

Definition at line 1566 of file tcp.c.

tcp_xfer_operations

struct interface_operation tcp_xfer_operations[]

static

Initial value:

= {
        INTF_OP ( xfer_deliver, struct tcp_connection *, tcp_xfer_deliver ),
        INTF_OP ( xfer_window, struct tcp_connection *, tcp_xfer_window ),
        INTF_OP ( job_progress, struct tcp_connection *, tcp_progress ),
        INTF_OP ( intf_close, struct tcp_connection *, tcp_xfer_close ),
}

TCP data transfer interface operations.

Definition at line 1749 of file tcp.c.

__socket_opener

struct socket_opener tcp_socket_opener __socket_opener

Initial value:

= {
        .semantics      = TCP_SOCK_STREAM,
        .open           = tcp_open,
}

TCP socket opener.

Definition at line 1768 of file tcp.c.

__uri_opener

struct uri_opener tcp_uri_opener __uri_opener

Initial value:

= {
        .scheme         = "tcp",
        .open           = tcp_open_uri,
}

TCP URI opener.

Definition at line 1798 of file tcp.c.