gpxe/src/net/tcp.c

   1 #include <string.h>
   2 #include <stdlib.h>
   3 #include <stdio.h>
   4 #include <assert.h>
   5 #include <errno.h>
   6 #include <byteswap.h>
   7 #include <gpxe/timer.h>
   8 #include <gpxe/iobuf.h>
   9 #include <gpxe/malloc.h>
  10 #include <gpxe/retry.h>
  11 #include <gpxe/refcnt.h>
  12 #include <gpxe/xfer.h>
  13 #include <gpxe/open.h>
  14 #include <gpxe/uri.h>
  15 #include <gpxe/tcpip.h>
  16 #include <gpxe/tcp.h>
  17
  18 /** @file
  19  *
  20  * TCP protocol
  21  *
  22  */
  23
  24 /** A TCP connection */
  25 struct tcp_connection {
  26         /** Reference counter */
  27         struct refcnt refcnt;
  28         /** List of TCP connections */
  29         struct list_head list;
  30
  31         /** Data transfer interface */
  32         struct xfer_interface xfer;
  33         /** Data transfer interface closed flag */
  34         int xfer_closed;
  35
  36         /** Remote socket address */
  37         struct sockaddr_tcpip peer;
  38         /** Local port, in network byte order */
  39         unsigned int local_port;
  40
  41         /** Current TCP state */
  42         unsigned int tcp_state;
  43         /** Previous TCP state
  44          *
  45          * Maintained only for debug messages
  46          */
  47         unsigned int prev_tcp_state;
  48         /** Current sequence number
  49          *
  50          * Equivalent to SND.UNA in RFC 793 terminology.
  51          */
  52         uint32_t snd_seq;
  53         /** Unacknowledged sequence count
  54          *
  55          * Equivalent to (SND.NXT-SND.UNA) in RFC 793 terminology.
  56          */
  57         uint32_t snd_sent;
  58         /** Send window
  59          *
  60          * Equivalent to SND.WND in RFC 793 terminology
  61          */
  62         uint32_t snd_win;
  63         /** Current acknowledgement number
  64          *
  65          * Equivalent to RCV.NXT in RFC 793 terminology.
  66          */
  67         uint32_t rcv_ack;
  68         /** Most recent received timestamp
  69          *
  70          * Equivalent to TS.Recent in RFC 1323 terminology.
  71          */
  72         uint32_t ts_recent;
  73         /** Timestamps enabled */
  74         int timestamps;
  75
  76         /** Transmit queue */
  77         struct list_head queue;
  78         /** Retransmission timer */
  79         struct retry_timer timer;
  80 };
  81
  82 /**
  83  * List of registered TCP connections
  84  */
  85 static LIST_HEAD ( tcp_conns );
  86
  87 /* Forward declarations */
  88 static struct xfer_interface_operations tcp_xfer_operations;
  89 static void tcp_expired ( struct retry_timer *timer, int over );
  90 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
  91                         uint32_t win );
  92
  93 /**
  94  * Name TCP state
  95  *
  96  * @v state             TCP state
  97  * @ret name            Name of TCP state
  98  */
  99 static inline __attribute__ (( always_inline )) const char *
 100 tcp_state ( int state ) {
 101         switch ( state ) {
 102         case TCP_CLOSED:                return "CLOSED";
 103         case TCP_LISTEN:                return "LISTEN";
 104         case TCP_SYN_SENT:              return "SYN_SENT";
 105         case TCP_SYN_RCVD:              return "SYN_RCVD";
 106         case TCP_ESTABLISHED:           return "ESTABLISHED";
 107         case TCP_FIN_WAIT_1:            return "FIN_WAIT_1";
 108         case TCP_FIN_WAIT_2:            return "FIN_WAIT_2";
 109         case TCP_CLOSING_OR_LAST_ACK:   return "CLOSING/LAST_ACK";
 110         case TCP_TIME_WAIT:             return "TIME_WAIT";
 111         case TCP_CLOSE_WAIT:            return "CLOSE_WAIT";
 112         default:                        return "INVALID";
 113         }
 114 }
 115
 116 /**
 117  * Dump TCP state transition
 118  *
 119  * @v tcp               TCP connection
 120  */
 121 static inline __attribute__ (( always_inline )) void
 122 tcp_dump_state ( struct tcp_connection *tcp ) {
 123
 124         if ( tcp->tcp_state != tcp->prev_tcp_state ) {
 125                 DBGC ( tcp, "TCP %p transitioned from %s to %s\n", tcp,
 126                        tcp_state ( tcp->prev_tcp_state ),
 127                        tcp_state ( tcp->tcp_state ) );
 128         }
 129         tcp->prev_tcp_state = tcp->tcp_state;
 130 }
 131
 132 /**
 133  * Dump TCP flags
 134  *
 135  * @v flags             TCP flags
 136  */
 137 static inline __attribute__ (( always_inline )) void
 138 tcp_dump_flags ( struct tcp_connection *tcp, unsigned int flags ) {
 139         if ( flags & TCP_RST )
 140                 DBGC ( tcp, " RST" );
 141         if ( flags & TCP_SYN )
 142                 DBGC ( tcp, " SYN" );
 143         if ( flags & TCP_PSH )
 144                 DBGC ( tcp, " PSH" );
 145         if ( flags & TCP_FIN )
 146                 DBGC ( tcp, " FIN" );
 147         if ( flags & TCP_ACK )
 148                 DBGC ( tcp, " ACK" );
 149 }
 150
 151 /***************************************************************************
 152  *
 153  * Open and close
 154  *
 155  ***************************************************************************
 156  */
 157
 158 /**
 159  * Bind TCP connection to local port
 160  *
 161  * @v tcp               TCP connection
 162  * @v port              Local port number, in network-endian order
 163  * @ret rc              Return status code
 164  *
 165  * If the port is 0, the connection is assigned an available port
 166  * between 1024 and 65535.
 167  */
 168 static int tcp_bind ( struct tcp_connection *tcp, unsigned int port ) {
 169         struct tcp_connection *existing;
 170         static uint16_t try_port = 1023;
 171
 172         /* If no port specified, find the first available port */
 173         if ( ! port ) {
 174                 while ( try_port ) {
 175                         try_port++;
 176                         if ( try_port < 1024 )
 177                                 continue;
 178                         if ( tcp_bind ( tcp, htons ( try_port ) ) == 0 )
 179                                 return 0;
 180                 }
 181                 DBGC ( tcp, "TCP %p could not bind: no free ports\n", tcp );
 182                 return -EADDRINUSE;
 183         }
 184
 185         /* Attempt bind to local port */
 186         list_for_each_entry ( existing, &tcp_conns, list ) {
 187                 if ( existing->local_port == port ) {
 188                         DBGC ( tcp, "TCP %p could not bind: port %d in use\n",
 189                                tcp, ntohs ( port ) );
 190                         return -EADDRINUSE;
 191                 }
 192         }
 193         tcp->local_port = port;
 194
 195         DBGC ( tcp, "TCP %p bound to port %d\n", tcp, ntohs ( port ) );
 196         return 0;
 197 }
 198
 199 /**
 200  * Open a TCP connection
 201  *
 202  * @v xfer              Data transfer interface
 203  * @v peer              Peer socket address
 204  * @v local             Local socket address, or NULL
 205  * @ret rc              Return status code
 206  */
 207 static int tcp_open ( struct xfer_interface *xfer, struct sockaddr *peer,
 208                       struct sockaddr *local ) {
 209         struct sockaddr_tcpip *st_peer = ( struct sockaddr_tcpip * ) peer;
 210         struct sockaddr_tcpip *st_local = ( struct sockaddr_tcpip * ) local;
 211         struct tcp_connection *tcp;
 212         unsigned int bind_port;
 213         int rc;
 214
 215         /* Allocate and initialise structure */
 216         tcp = zalloc ( sizeof ( *tcp ) );
 217         if ( ! tcp )
 218                 return -ENOMEM;
 219         DBGC ( tcp, "TCP %p allocated\n", tcp );
 220         xfer_init ( &tcp->xfer, &tcp_xfer_operations, &tcp->refcnt );
 221         tcp->prev_tcp_state = TCP_CLOSED;
 222         tcp->tcp_state = TCP_STATE_SENT ( TCP_SYN );
 223         tcp_dump_state ( tcp );
 224         tcp->snd_seq = random();
 225         INIT_LIST_HEAD ( &tcp->queue );
 226         tcp->timer.expired = tcp_expired;
 227         memcpy ( &tcp->peer, st_peer, sizeof ( tcp->peer ) );
 228
 229         /* Bind to local port */
 230         bind_port = ( st_local ? st_local->st_port : 0 );
 231         if ( ( rc = tcp_bind ( tcp, bind_port ) ) != 0 )
 232                 goto err;
 233
 234         /* Start timer to initiate SYN */
 235         start_timer_nodelay ( &tcp->timer );
 236
 237         /* Attach parent interface, transfer reference to connection
 238          * list and return
 239          */
 240         xfer_plug_plug ( &tcp->xfer, xfer );
 241         list_add ( &tcp->list, &tcp_conns );
 242         return 0;
 243
 244  err:
 245         ref_put ( &tcp->refcnt );
 246         return rc;
 247 }
 248
 249 /**
 250  * Close TCP connection
 251  *
 252  * @v tcp               TCP connection
 253  * @v rc                Reason for close
 254  *
 255  * Closes the data transfer interface.  If the TCP state machine is in
 256  * a suitable state, the connection will be deleted.
 257  */
 258 static void tcp_close ( struct tcp_connection *tcp, int rc ) {
 259         struct io_buffer *iobuf;
 260         struct io_buffer *tmp;
 261
 262         /* Close data transfer interface */
 263         xfer_nullify ( &tcp->xfer );
 264         xfer_close ( &tcp->xfer, rc );
 265         tcp->xfer_closed = 1;
 266
 267         /* If we are in CLOSED, or have otherwise not yet received a
 268          * SYN (i.e. we are in LISTEN or SYN_SENT), just delete the
 269          * connection.
 270          */
 271         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 272
 273                 /* Transition to CLOSED for the sake of debugging messages */
 274                 tcp->tcp_state = TCP_CLOSED;
 275                 tcp_dump_state ( tcp );
 276
 277                 /* Free any unsent I/O buffers */
 278                 list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 279                         list_del ( &iobuf->list );
 280                         free_iob ( iobuf );
 281                 }
 282
 283                 /* Remove from list and drop reference */
 284                 stop_timer ( &tcp->timer );
 285                 list_del ( &tcp->list );
 286                 ref_put ( &tcp->refcnt );
 287                 DBGC ( tcp, "TCP %p connection deleted\n", tcp );
 288                 return;
 289         }
 290
 291         /* If we have not had our SYN acknowledged (i.e. we are in
 292          * SYN_RCVD), pretend that it has been acknowledged so that we
 293          * can send a FIN without breaking things.
 294          */
 295         if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 296                 tcp_rx_ack ( tcp, ( tcp->snd_seq + 1 ), 0 );
 297
 298         /* If we have no data remaining to send, start sending FIN */
 299         if ( list_empty ( &tcp->queue ) ) {
 300                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 301                 tcp_dump_state ( tcp );
 302         }
 303 }
 304
 305 /***************************************************************************
 306  *
 307  * Transmit data path
 308  *
 309  ***************************************************************************
 310  */
 311
 312 /**
 313  * Calculate transmission window
 314  *
 315  * @v tcp               TCP connection
 316  * @ret len             Maximum length that can be sent in a single packet
 317  */
 318 static size_t tcp_xmit_win ( struct tcp_connection *tcp ) {
 319         size_t len;
 320
 321         /* Not ready if we're not in a suitable connection state */
 322         if ( ! TCP_CAN_SEND_DATA ( tcp->tcp_state ) )
 323                 return 0;
 324
 325         /* Length is the minimum of the receiver's window and the path MTU */
 326         len = tcp->snd_win;
 327         if ( len > TCP_PATH_MTU )
 328                 len = TCP_PATH_MTU;
 329
 330         return len;
 331 }
 332
 333 /**
 334  * Process TCP transmit queue
 335  *
 336  * @v tcp               TCP connection
 337  * @v max_len           Maximum length to process
 338  * @v dest              I/O buffer to fill with data, or NULL
 339  * @v remove            Remove data from queue
 340  * @ret len             Length of data processed
 341  *
 342  * This processes at most @c max_len bytes from the TCP connection's
 343  * transmit queue.  Data will be copied into the @c dest I/O buffer
 344  * (if provided) and, if @c remove is true, removed from the transmit
 345  * queue.
 346  */
 347 static size_t tcp_process_queue ( struct tcp_connection *tcp, size_t max_len,
 348                                   struct io_buffer *dest, int remove ) {
 349         struct io_buffer *iobuf;
 350         struct io_buffer *tmp;
 351         size_t frag_len;
 352         size_t len = 0;
 353
 354         list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 355                 frag_len = iob_len ( iobuf );
 356                 if ( frag_len > max_len )
 357                         frag_len = max_len;
 358                 if ( dest ) {
 359                         memcpy ( iob_put ( dest, frag_len ), iobuf->data,
 360                                  frag_len );
 361                 }
 362                 if ( remove ) {
 363                         iob_pull ( iobuf, frag_len );
 364                         if ( ! iob_len ( iobuf ) ) {
 365                                 list_del ( &iobuf->list );
 366                                 free_iob ( iobuf );
 367                         }
 368                 }
 369                 len += frag_len;
 370                 max_len -= frag_len;
 371         }
 372         return len;
 373 }
 374
 375 /**
 376  * Transmit any outstanding data
 377  *
 378  * @v tcp               TCP connection
 379  * @v force_send        Force sending of packet
 380  *
 381  * Transmits any outstanding data on the connection.
 382  *
 383  * Note that even if an error is returned, the retransmission timer
 384  * will have been started if necessary, and so the stack will
 385  * eventually attempt to retransmit the failed packet.
 386  */
 387 static int tcp_xmit ( struct tcp_connection *tcp, int force_send ) {
 388         struct io_buffer *iobuf;
 389         struct tcp_header *tcphdr;
 390         struct tcp_mss_option *mssopt;
 391         struct tcp_timestamp_padded_option *tsopt;
 392         void *payload;
 393         unsigned int flags;
 394         size_t len = 0;
 395         size_t seq_len;
 396         size_t app_win;
 397         size_t rcv_win;
 398
 399         /* If retransmission timer is already running, do nothing */
 400         if ( timer_running ( &tcp->timer ) )
 401                 return 0;
 402
 403         /* Calculate both the actual (payload) and sequence space
 404          * lengths that we wish to transmit.
 405          */
 406         if ( TCP_CAN_SEND_DATA ( tcp->tcp_state ) ) {
 407                 len = tcp_process_queue ( tcp, tcp_xmit_win ( tcp ),
 408                                           NULL, 0 );
 409         }
 410         seq_len = len;
 411         flags = TCP_FLAGS_SENDING ( tcp->tcp_state );
 412         if ( flags & ( TCP_SYN | TCP_FIN ) ) {
 413                 /* SYN or FIN consume one byte, and we can never send both */
 414                 assert ( ! ( ( flags & TCP_SYN ) && ( flags & TCP_FIN ) ) );
 415                 seq_len++;
 416         }
 417         tcp->snd_sent = seq_len;
 418
 419         /* If we have nothing to transmit, stop now */
 420         if ( ( seq_len == 0 ) && ! force_send )
 421                 return 0;
 422
 423         /* If we are transmitting anything that requires
 424          * acknowledgement (i.e. consumes sequence space), start the
 425          * retransmission timer.  Do this before attempting to
 426          * allocate the I/O buffer, in case allocation itself fails.
 427          */
 428         if ( seq_len )
 429                 start_timer ( &tcp->timer );
 430
 431         /* Allocate I/O buffer */
 432         iobuf = alloc_iob ( len + MAX_HDR_LEN );
 433         if ( ! iobuf ) {
 434                 DBGC ( tcp, "TCP %p could not allocate data buffer\n", tcp );
 435                 return -ENOMEM;
 436         }
 437         iob_reserve ( iobuf, MAX_HDR_LEN );
 438
 439         /* Fill data payload from transmit queue */
 440         tcp_process_queue ( tcp, len, iobuf, 0 );
 441
 442         /* Estimate window size */
 443         rcv_win = ( ( freemem * 3 ) / 4 );
 444         if ( rcv_win > TCP_MAX_WINDOW_SIZE )
 445                 rcv_win = TCP_MAX_WINDOW_SIZE;
 446         app_win = xfer_window ( &tcp->xfer );
 447         if ( rcv_win > app_win )
 448                 rcv_win = app_win;
 449         rcv_win &= ~0x03; /* Keep everything dword-aligned */
 450
 451         /* Fill up the TCP header */
 452         payload = iobuf->data;
 453         if ( flags & TCP_SYN ) {
 454                 mssopt = iob_push ( iobuf, sizeof ( *mssopt ) );
 455                 mssopt->kind = TCP_OPTION_MSS;
 456                 mssopt->length = sizeof ( *mssopt );
 457                 mssopt->mss = htons ( TCP_MSS );
 458         }
 459         if ( ( flags & TCP_SYN ) || tcp->timestamps ) {
 460                 tsopt = iob_push ( iobuf, sizeof ( *tsopt ) );
 461                 memset ( tsopt->nop, TCP_OPTION_NOP, sizeof ( tsopt->nop ) );
 462                 tsopt->tsopt.kind = TCP_OPTION_TS;
 463                 tsopt->tsopt.length = sizeof ( tsopt->tsopt );
 464                 tsopt->tsopt.tsval = ntohl ( currticks() );
 465                 tsopt->tsopt.tsecr = ntohl ( tcp->ts_recent );
 466         }
 467         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 468         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 469         tcphdr->src = tcp->local_port;
 470         tcphdr->dest = tcp->peer.st_port;
 471         tcphdr->seq = htonl ( tcp->snd_seq );
 472         tcphdr->ack = htonl ( tcp->rcv_ack );
 473         tcphdr->hlen = ( ( payload - iobuf->data ) << 2 );
 474         tcphdr->flags = flags;
 475         tcphdr->win = htons ( rcv_win );
 476         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 477
 478         /* Dump header */
 479         DBGC ( tcp, "TCP %p TX %d->%d %08lx..%08lx           %08lx %4zd",
 480                tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 481                ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) + seq_len ),
 482                ntohl ( tcphdr->ack ), len );
 483         tcp_dump_flags ( tcp, tcphdr->flags );
 484         DBGC ( tcp, "\n" );
 485
 486         /* Transmit packet */
 487         return tcpip_tx ( iobuf, &tcp_protocol, &tcp->peer, NULL,
 488                           &tcphdr->csum );
 489 }
 490
 491 /**
 492  * Retransmission timer expired
 493  *
 494  * @v timer     Retry timer
 495  * @v over      Failure indicator
 496  */
 497 static void tcp_expired ( struct retry_timer *timer, int over ) {
 498         struct tcp_connection *tcp =
 499                 container_of ( timer, struct tcp_connection, timer );
 500         int graceful_close = TCP_CLOSED_GRACEFULLY ( tcp->tcp_state );
 501
 502         DBGC ( tcp, "TCP %p timer %s in %s\n", tcp,
 503                ( over ? "expired" : "fired" ), tcp_state ( tcp->tcp_state ) );
 504
 505         assert ( ( tcp->tcp_state == TCP_SYN_SENT ) ||
 506                  ( tcp->tcp_state == TCP_SYN_RCVD ) ||
 507                  ( tcp->tcp_state == TCP_ESTABLISHED ) ||
 508                  ( tcp->tcp_state == TCP_FIN_WAIT_1 ) ||
 509                  ( tcp->tcp_state == TCP_TIME_WAIT ) ||
 510                  ( tcp->tcp_state == TCP_CLOSE_WAIT ) ||
 511                  ( tcp->tcp_state == TCP_CLOSING_OR_LAST_ACK ) );
 512
 513         if ( over || graceful_close ) {
 514                 /* If we have finally timed out and given up, or if
 515                  * this is the result of a graceful close, terminate
 516                  * the connection
 517                  */
 518                 tcp->tcp_state = TCP_CLOSED;
 519                 tcp_dump_state ( tcp );
 520                 tcp_close ( tcp, -ETIMEDOUT );
 521         } else {
 522                 /* Otherwise, retransmit the packet */
 523                 tcp_xmit ( tcp, 0 );
 524         }
 525 }
 526
 527 /**
 528  * Send RST response to incoming packet
 529  *
 530  * @v in_tcphdr         TCP header of incoming packet
 531  * @ret rc              Return status code
 532  */
 533 static int tcp_xmit_reset ( struct tcp_connection *tcp,
 534                             struct sockaddr_tcpip *st_dest,
 535                             struct tcp_header *in_tcphdr ) {
 536         struct io_buffer *iobuf;
 537         struct tcp_header *tcphdr;
 538
 539         /* Allocate space for dataless TX buffer */
 540         iobuf = alloc_iob ( MAX_HDR_LEN );
 541         if ( ! iobuf ) {
 542                 DBGC ( tcp, "TCP %p could not allocate data buffer\n", tcp );
 543                 return -ENOMEM;
 544         }
 545         iob_reserve ( iobuf, MAX_HDR_LEN );
 546
 547         /* Construct RST response */
 548         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 549         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 550         tcphdr->src = in_tcphdr->dest;
 551         tcphdr->dest = in_tcphdr->src;
 552         tcphdr->seq = in_tcphdr->ack;
 553         tcphdr->ack = in_tcphdr->seq;
 554         tcphdr->hlen = ( ( sizeof ( *tcphdr ) / 4 ) << 4 );
 555         tcphdr->flags = ( TCP_RST | TCP_ACK );
 556         tcphdr->win = htons ( TCP_MAX_WINDOW_SIZE );
 557         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 558
 559         /* Dump header */
 560         DBGC ( tcp, "TCP %p TX %d->%d %08lx..%08lx           %08lx %4d",
 561                tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 562                ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) ),
 563                ntohl ( tcphdr->ack ), 0 );
 564         tcp_dump_flags ( tcp, tcphdr->flags );
 565         DBGC ( tcp, "\n" );
 566
 567         /* Transmit packet */
 568         return tcpip_tx ( iobuf, &tcp_protocol, st_dest,
 569                           NULL, &tcphdr->csum );
 570 }
 571
 572 /***************************************************************************
 573  *
 574  * Receive data path
 575  *
 576  ***************************************************************************
 577  */
 578
 579 /**
 580  * Identify TCP connection by local port number
 581  *
 582  * @v local_port        Local port (in network-endian order)
 583  * @ret tcp             TCP connection, or NULL
 584  */
 585 static struct tcp_connection * tcp_demux ( unsigned int local_port ) {
 586         struct tcp_connection *tcp;
 587
 588         list_for_each_entry ( tcp, &tcp_conns, list ) {
 589                 if ( tcp->local_port == local_port )
 590                         return tcp;
 591         }
 592         return NULL;
 593 }
 594
 595 /**
 596  * Parse TCP received options
 597  *
 598  * @v tcp               TCP connection
 599  * @v data              Raw options data
 600  * @v len               Raw options length
 601  * @v options           Options structure to fill in
 602  */
 603 static void tcp_rx_opts ( struct tcp_connection *tcp, const void *data,
 604                           size_t len, struct tcp_options *options ) {
 605         const void *end = ( data + len );
 606         const struct tcp_option *option;
 607         unsigned int kind;
 608
 609         memset ( options, 0, sizeof ( *options ) );
 610         while ( data < end ) {
 611                 option = data;
 612                 kind = option->kind;
 613                 if ( kind == TCP_OPTION_END )
 614                         return;
 615                 if ( kind == TCP_OPTION_NOP ) {
 616                         data++;
 617                         continue;
 618                 }
 619                 switch ( kind ) {
 620                 case TCP_OPTION_MSS:
 621                         options->mssopt = data;
 622                         break;
 623                 case TCP_OPTION_TS:
 624                         options->tsopt = data;
 625                         break;
 626                 default:
 627                         DBGC ( tcp, "TCP %p received unknown option %d\n",
 628                                tcp, kind );
 629                         break;
 630                 }
 631                 data += option->length;
 632         }
 633 }
 634
 635 /**
 636  * Handle TCP received SYN
 637  *
 638  * @v tcp               TCP connection
 639  * @v seq               SEQ value (in host-endian order)
 640  * @v options           TCP options
 641  * @ret rc              Return status code
 642  */
 643 static int tcp_rx_syn ( struct tcp_connection *tcp, uint32_t seq,
 644                         struct tcp_options *options ) {
 645
 646         /* Synchronise sequence numbers on first SYN */
 647         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 648                 tcp->rcv_ack = seq;
 649                 if ( options->tsopt )
 650                         tcp->timestamps = 1;
 651         }
 652
 653         /* Ignore duplicate SYN */
 654         if ( ( tcp->rcv_ack - seq ) > 0 )
 655                 return 0;
 656
 657         /* Mark SYN as received and start sending ACKs with each packet */
 658         tcp->tcp_state |= ( TCP_STATE_SENT ( TCP_ACK ) |
 659                             TCP_STATE_RCVD ( TCP_SYN ) );
 660
 661         /* Acknowledge SYN */
 662         tcp->rcv_ack++;
 663
 664         return 0;
 665 }
 666
 667 /**
 668  * Handle TCP received ACK
 669  *
 670  * @v tcp               TCP connection
 671  * @v ack               ACK value (in host-endian order)
 672  * @v win               WIN value (in host-endian order)
 673  * @ret rc              Return status code
 674  */
 675 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
 676                         uint32_t win ) {
 677         size_t ack_len = ( ack - tcp->snd_seq );
 678         size_t len;
 679         unsigned int acked_flags;
 680
 681         /* Ignore duplicate or out-of-range ACK */
 682         if ( ack_len > tcp->snd_sent ) {
 683                 DBGC ( tcp, "TCP %p received ACK for [%08lx,%08lx), "
 684                        "sent only [%08lx,%08lx)\n", tcp, tcp->snd_seq,
 685                        ( tcp->snd_seq + ack_len ), tcp->snd_seq,
 686                        ( tcp->snd_seq + tcp->snd_sent ) );
 687                 return -EINVAL;
 688         }
 689
 690         /* Acknowledge any flags being sent */
 691         len = ack_len;
 692         acked_flags = ( TCP_FLAGS_SENDING ( tcp->tcp_state ) &
 693                         ( TCP_SYN | TCP_FIN ) );
 694         if ( acked_flags )
 695                 len--;
 696
 697         /* Update SEQ and sent counters, and window size */
 698         tcp->snd_seq = ack;
 699         tcp->snd_sent = 0;
 700         tcp->snd_win = win;
 701
 702         /* Stop the retransmission timer */
 703         stop_timer ( &tcp->timer );
 704
 705         /* Remove any acknowledged data from transmit queue */
 706         tcp_process_queue ( tcp, len, NULL, 1 );
 707
 708         /* Mark SYN/FIN as acknowledged if applicable. */
 709         if ( acked_flags )
 710                 tcp->tcp_state |= TCP_STATE_ACKED ( acked_flags );
 711
 712         /* Start sending FIN if we've had all possible data ACKed */
 713         if ( list_empty ( &tcp->queue ) && tcp->xfer_closed )
 714                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 715
 716         return 0;
 717 }
 718
 719 /**
 720  * Handle TCP received data
 721  *
 722  * @v tcp               TCP connection
 723  * @v seq               SEQ value (in host-endian order)
 724  * @v iobuf             I/O buffer
 725  * @ret rc              Return status code
 726  *
 727  * This function takes ownership of the I/O buffer.
 728  */
 729 static int tcp_rx_data ( struct tcp_connection *tcp, uint32_t seq,
 730                          struct io_buffer *iobuf ) {
 731         size_t already_rcvd;
 732         size_t len;
 733         int rc;
 734
 735         /* Ignore duplicate or out-of-order data */
 736         already_rcvd = ( tcp->rcv_ack - seq );
 737         len = iob_len ( iobuf );
 738         if ( already_rcvd >= len ) {
 739                 free_iob ( iobuf );
 740                 return 0;
 741         }
 742         iob_pull ( iobuf, already_rcvd );
 743         len -= already_rcvd;
 744
 745         /* Deliver data to application */
 746         if ( ( rc = xfer_deliver_iob ( &tcp->xfer, iobuf ) ) != 0 )
 747                 return rc;
 748
 749         /* Acknowledge new data */
 750         tcp->rcv_ack += len;
 751         return 0;
 752 }
 753
 754 /**
 755  * Handle TCP received FIN
 756  *
 757  * @v tcp               TCP connection
 758  * @v seq               SEQ value (in host-endian order)
 759  * @ret rc              Return status code
 760  */
 761 static int tcp_rx_fin ( struct tcp_connection *tcp, uint32_t seq ) {
 762
 763         /* Ignore duplicate or out-of-order FIN */
 764         if ( ( tcp->rcv_ack - seq ) > 0 )
 765                 return 0;
 766
 767         /* Mark FIN as received and acknowledge it */
 768         tcp->tcp_state |= TCP_STATE_RCVD ( TCP_FIN );
 769         tcp->rcv_ack++;
 770
 771         /* Close connection */
 772         tcp_close ( tcp, 0 );
 773
 774         return 0;
 775 }
 776
 777 /**
 778  * Handle TCP received RST
 779  *
 780  * @v tcp               TCP connection
 781  * @v seq               SEQ value (in host-endian order)
 782  * @ret rc              Return status code
 783  */
 784 static int tcp_rx_rst ( struct tcp_connection *tcp, uint32_t seq ) {
 785
 786         /* Accept RST only if it falls within the window.  If we have
 787          * not yet received a SYN, then we have no window to test
 788          * against, so fall back to checking that our SYN has been
 789          * ACKed.
 790          */
 791         if ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) {
 792                 if ( ( tcp->rcv_ack - seq ) > 0 )
 793                         return 0;
 794         } else {
 795                 if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 796                         return 0;
 797         }
 798
 799         /* Abort connection */
 800         tcp->tcp_state = TCP_CLOSED;
 801         tcp_dump_state ( tcp );
 802         tcp_close ( tcp, -ECONNRESET );
 803
 804         return -ECONNRESET;
 805 }
 806
 807 /**
 808  * Process received packet
 809  *
 810  * @v iobuf             I/O buffer
 811  * @v st_src            Partially-filled source address
 812  * @v st_dest           Partially-filled destination address
 813  * @v pshdr_csum        Pseudo-header checksum
 814  * @ret rc              Return status code
 815   */
 816 static int tcp_rx ( struct io_buffer *iobuf,
 817                     struct sockaddr_tcpip *st_src,
 818                     struct sockaddr_tcpip *st_dest __unused,
 819                     uint16_t pshdr_csum ) {
 820         struct tcp_header *tcphdr = iobuf->data;
 821         struct tcp_connection *tcp;
 822         struct tcp_options options;
 823         size_t hlen;
 824         uint16_t csum;
 825         uint32_t start_seq;
 826         uint32_t seq;
 827         uint32_t ack;
 828         uint32_t win;
 829         unsigned int flags;
 830         size_t len;
 831         int rc;
 832
 833         /* Sanity check packet */
 834         if ( iob_len ( iobuf ) < sizeof ( *tcphdr ) ) {
 835                 DBG ( "TCP packet too short at %zd bytes (min %zd bytes)\n",
 836                       iob_len ( iobuf ), sizeof ( *tcphdr ) );
 837                 rc = -EINVAL;
 838                 goto discard;
 839         }
 840         hlen = ( ( tcphdr->hlen & TCP_MASK_HLEN ) / 16 ) * 4;
 841         if ( hlen < sizeof ( *tcphdr ) ) {
 842                 DBG ( "TCP header too short at %zd bytes (min %zd bytes)\n",
 843                       hlen, sizeof ( *tcphdr ) );
 844                 rc = -EINVAL;
 845                 goto discard;
 846         }
 847         if ( hlen > iob_len ( iobuf ) ) {
 848                 DBG ( "TCP header too long at %zd bytes (max %zd bytes)\n",
 849                       hlen, iob_len ( iobuf ) );
 850                 rc = -EINVAL;
 851                 goto discard;
 852         }
 853         csum = tcpip_continue_chksum ( pshdr_csum, iobuf->data, iob_len ( iobuf ));
 854         if ( csum != 0 ) {
 855                 DBG ( "TCP checksum incorrect (is %04x including checksum "
 856                       "field, should be 0000)\n", csum );
 857                 rc = -EINVAL;
 858                 goto discard;
 859         }
 860
 861         /* Parse parameters from header and strip header */
 862         tcp = tcp_demux ( tcphdr->dest );
 863         start_seq = seq = ntohl ( tcphdr->seq );
 864         ack = ntohl ( tcphdr->ack );
 865         win = ntohs ( tcphdr->win );
 866         flags = tcphdr->flags;
 867         tcp_rx_opts ( tcp, ( ( ( void * ) tcphdr ) + sizeof ( *tcphdr ) ),
 868                       ( hlen - sizeof ( *tcphdr ) ), &options );
 869         iob_pull ( iobuf, hlen );
 870         len = iob_len ( iobuf );
 871
 872         /* Dump header */
 873         DBGC ( tcp, "TCP %p RX %d<-%d           %08lx %08lx..%08lx %4zd",
 874                tcp, ntohs ( tcphdr->dest ), ntohs ( tcphdr->src ),
 875                ntohl ( tcphdr->ack ), ntohl ( tcphdr->seq ),
 876                ( ntohl ( tcphdr->seq ) + len +
 877                  ( ( tcphdr->flags & ( TCP_SYN | TCP_FIN ) ) ? 1 : 0 ) ), len);
 878         tcp_dump_flags ( tcp, tcphdr->flags );
 879         DBGC ( tcp, "\n" );
 880
 881         /* If no connection was found, send RST */
 882         if ( ! tcp ) {
 883                 tcp_xmit_reset ( tcp, st_src, tcphdr );
 884                 rc = -ENOTCONN;
 885                 goto discard;
 886         }
 887
 888         /* Handle ACK, if present */
 889         if ( flags & TCP_ACK ) {
 890                 if ( ( rc = tcp_rx_ack ( tcp, ack, win ) ) != 0 ) {
 891                         tcp_xmit_reset ( tcp, st_src, tcphdr );
 892                         goto discard;
 893                 }
 894         }
 895
 896         /* Handle SYN, if present */
 897         if ( flags & TCP_SYN ) {
 898                 tcp_rx_syn ( tcp, seq, &options );
 899                 seq++;
 900         }
 901
 902         /* Handle RST, if present */
 903         if ( flags & TCP_RST ) {
 904                 if ( ( rc = tcp_rx_rst ( tcp, seq ) ) != 0 )
 905                         goto discard;
 906         }
 907
 908         /* Handle new data, if any */
 909         tcp_rx_data ( tcp, seq, iobuf );
 910         seq += len;
 911
 912         /* Handle FIN, if present */
 913         if ( flags & TCP_FIN ) {
 914                 tcp_rx_fin ( tcp, seq );
 915                 seq++;
 916         }
 917
 918         /* Update timestamp, if present and applicable */
 919         if ( ( seq == tcp->rcv_ack ) && options.tsopt )
 920                 tcp->ts_recent = ntohl ( options.tsopt->tsval );
 921
 922         /* Dump out any state change as a result of the received packet */
 923         tcp_dump_state ( tcp );
 924
 925         /* Send out any pending data.  If peer is expecting an ACK for
 926          * this packet then force sending a reply.
 927          */
 928         tcp_xmit ( tcp, ( start_seq != seq ) );
 929
 930         /* If this packet was the last we expect to receive, set up
 931          * timer to expire and cause the connection to be freed.
 932          */
 933         if ( TCP_CLOSED_GRACEFULLY ( tcp->tcp_state ) ) {
 934                 tcp->timer.timeout = ( 2 * TCP_MSL );
 935                 start_timer ( &tcp->timer );
 936         }
 937
 938         return 0;
 939
 940  discard:
 941         /* Free received packet */
 942         free_iob ( iobuf );
 943         return rc;
 944 }
 945
 946 /** TCP protocol */
 947 struct tcpip_protocol tcp_protocol __tcpip_protocol = {
 948         .name = "TCP",
 949         .rx = tcp_rx,
 950         .tcpip_proto = IP_TCP,
 951 };
 952
 953 /***************************************************************************
 954  *
 955  * Data transfer interface
 956  *
 957  ***************************************************************************
 958  */
 959
 960 /**
 961  * Close interface
 962  *
 963  * @v xfer              Data transfer interface
 964  * @v rc                Reason for close
 965  */
 966 static void tcp_xfer_close ( struct xfer_interface *xfer, int rc ) {
 967         struct tcp_connection *tcp =
 968                 container_of ( xfer, struct tcp_connection, xfer );
 969
 970         /* Close data transfer interface */
 971         tcp_close ( tcp, rc );
 972
 973         /* Transmit FIN, if possible */
 974         tcp_xmit ( tcp, 0 );
 975 }
 976
 977 /**
 978  * Check flow control window
 979  *
 980  * @v xfer              Data transfer interface
 981  * @ret len             Length of window
 982  */
 983 static size_t tcp_xfer_window ( struct xfer_interface *xfer ) {
 984         struct tcp_connection *tcp =
 985                 container_of ( xfer, struct tcp_connection, xfer );
 986
 987         /* Not ready if data queue is non-empty.  This imposes a limit
 988          * of only one unACKed packet in the TX queue at any time; we
 989          * do this to conserve memory usage.
 990          */
 991         if ( ! list_empty ( &tcp->queue ) )
 992                 return 0;
 993
 994         /* Return TCP window length */
 995         return tcp_xmit_win ( tcp );
 996 }
 997
 998 /**
 999  * Deliver datagram as I/O buffer
1000  *
1001  * @v xfer              Data transfer interface
1002  * @v iobuf             Datagram I/O buffer
1003  * @v meta              Data transfer metadata, or NULL
1004  * @ret rc              Return status code
1005  */
1006 static int tcp_xfer_deliver_iob ( struct xfer_interface *xfer,
1007                                   struct io_buffer *iobuf,
1008                                   struct xfer_metadata *meta __unused ) {
1009         struct tcp_connection *tcp =
1010                 container_of ( xfer, struct tcp_connection, xfer );
1011
1012         /* Enqueue packet */
1013         list_add_tail ( &iobuf->list, &tcp->queue );
1014
1015         /* Transmit data, if possible */
1016         tcp_xmit ( tcp, 0 );
1017
1018         return 0;
1019 }
1020
1021 /** TCP data transfer interface operations */
1022 static struct xfer_interface_operations tcp_xfer_operations = {
1023         .close          = tcp_xfer_close,
1024         .vredirect      = ignore_xfer_vredirect,
1025         .window         = tcp_xfer_window,
1026         .alloc_iob      = default_xfer_alloc_iob,
1027         .deliver_iob    = tcp_xfer_deliver_iob,
1028         .deliver_raw    = xfer_deliver_as_iob,
1029 };
1030
1031 /***************************************************************************
1032  *
1033  * Openers
1034  *
1035  ***************************************************************************
1036  */
1037
1038 /** TCP socket opener */
1039 struct socket_opener tcp_socket_opener __socket_opener = {
1040         .semantics      = SOCK_STREAM,
1041         .family         = AF_INET,
1042         .open           = tcp_open,
1043 };
1044
1045 char TCP_SOCK_STREAM[1];
1046
1047 /**
1048  * Open TCP URI
1049  *
1050  * @v xfer              Data transfer interface
1051  * @v uri               URI
1052  * @ret rc              Return status code
1053  */
1054 static int tcp_open_uri ( struct xfer_interface *xfer, struct uri *uri ) {
1055         struct sockaddr_tcpip peer;
1056
1057         /* Sanity check */
1058         if ( ! uri->host )
1059                 return -EINVAL;
1060
1061         memset ( &peer, 0, sizeof ( peer ) );
1062         peer.st_port = htons ( uri_port ( uri, 0 ) );
1063         return xfer_open_named_socket ( xfer, SOCK_STREAM,
1064                                         ( struct sockaddr * ) &peer,
1065                                         uri->host, NULL );
1066 }
1067
1068 /** TCP URI opener */
1069 struct uri_opener tcp_uri_opener __uri_opener = {
1070         .scheme         = "tcp",
1071         .open           = tcp_open_uri,
1072 };
1073