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 FILE_LICENCE ( GPL2_OR_LATER );
  25
  26 /** A TCP connection */
  27 struct tcp_connection {
  28         /** Reference counter */
  29         struct refcnt refcnt;
  30         /** List of TCP connections */
  31         struct list_head list;
  32
  33         /** Data transfer interface */
  34         struct xfer_interface xfer;
  35         /** Data transfer interface closed flag */
  36         int xfer_closed;
  37
  38         /** Remote socket address */
  39         struct sockaddr_tcpip peer;
  40         /** Local port, in network byte order */
  41         unsigned int local_port;
  42
  43         /** Current TCP state */
  44         unsigned int tcp_state;
  45         /** Previous TCP state
  46          *
  47          * Maintained only for debug messages
  48          */
  49         unsigned int prev_tcp_state;
  50         /** Current sequence number
  51          *
  52          * Equivalent to SND.UNA in RFC 793 terminology.
  53          */
  54         uint32_t snd_seq;
  55         /** Unacknowledged sequence count
  56          *
  57          * Equivalent to (SND.NXT-SND.UNA) in RFC 793 terminology.
  58          */
  59         uint32_t snd_sent;
  60         /** Send window
  61          *
  62          * Equivalent to SND.WND in RFC 793 terminology
  63          */
  64         uint32_t snd_win;
  65         /** Current acknowledgement number
  66          *
  67          * Equivalent to RCV.NXT in RFC 793 terminology.
  68          */
  69         uint32_t rcv_ack;
  70         /** Receive window
  71          *
  72          * Equivalent to RCV.WND in RFC 793 terminology.
  73          */
  74         uint32_t rcv_win;
  75         /** Most recent received timestamp
  76          *
  77          * Equivalent to TS.Recent in RFC 1323 terminology.
  78          */
  79         uint32_t ts_recent;
  80         /** Timestamps enabled */
  81         int timestamps;
  82
  83         /** Transmit queue */
  84         struct list_head queue;
  85         /** Retransmission timer */
  86         struct retry_timer timer;
  87 };
  88
  89 /**
  90  * List of registered TCP connections
  91  */
  92 static LIST_HEAD ( tcp_conns );
  93
  94 /* Forward declarations */
  95 static struct xfer_interface_operations tcp_xfer_operations;
  96 static void tcp_expired ( struct retry_timer *timer, int over );
  97 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
  98                         uint32_t win );
  99
 100 /**
 101  * Name TCP state
 102  *
 103  * @v state             TCP state
 104  * @ret name            Name of TCP state
 105  */
 106 static inline __attribute__ (( always_inline )) const char *
 107 tcp_state ( int state ) {
 108         switch ( state ) {
 109         case TCP_CLOSED:                return "CLOSED";
 110         case TCP_LISTEN:                return "LISTEN";
 111         case TCP_SYN_SENT:              return "SYN_SENT";
 112         case TCP_SYN_RCVD:              return "SYN_RCVD";
 113         case TCP_ESTABLISHED:           return "ESTABLISHED";
 114         case TCP_FIN_WAIT_1:            return "FIN_WAIT_1";
 115         case TCP_FIN_WAIT_2:            return "FIN_WAIT_2";
 116         case TCP_CLOSING_OR_LAST_ACK:   return "CLOSING/LAST_ACK";
 117         case TCP_TIME_WAIT:             return "TIME_WAIT";
 118         case TCP_CLOSE_WAIT:            return "CLOSE_WAIT";
 119         default:                        return "INVALID";
 120         }
 121 }
 122
 123 /**
 124  * Dump TCP state transition
 125  *
 126  * @v tcp               TCP connection
 127  */
 128 static inline __attribute__ (( always_inline )) void
 129 tcp_dump_state ( struct tcp_connection *tcp ) {
 130
 131         if ( tcp->tcp_state != tcp->prev_tcp_state ) {
 132                 DBGC ( tcp, "TCP %p transitioned from %s to %s\n", tcp,
 133                        tcp_state ( tcp->prev_tcp_state ),
 134                        tcp_state ( tcp->tcp_state ) );
 135         }
 136         tcp->prev_tcp_state = tcp->tcp_state;
 137 }
 138
 139 /**
 140  * Dump TCP flags
 141  *
 142  * @v flags             TCP flags
 143  */
 144 static inline __attribute__ (( always_inline )) void
 145 tcp_dump_flags ( struct tcp_connection *tcp, unsigned int flags ) {
 146         if ( flags & TCP_RST )
 147                 DBGC2 ( tcp, " RST" );
 148         if ( flags & TCP_SYN )
 149                 DBGC2 ( tcp, " SYN" );
 150         if ( flags & TCP_PSH )
 151                 DBGC2 ( tcp, " PSH" );
 152         if ( flags & TCP_FIN )
 153                 DBGC2 ( tcp, " FIN" );
 154         if ( flags & TCP_ACK )
 155                 DBGC2 ( tcp, " ACK" );
 156 }
 157
 158 /***************************************************************************
 159  *
 160  * Open and close
 161  *
 162  ***************************************************************************
 163  */
 164
 165 /**
 166  * Bind TCP connection to local port
 167  *
 168  * @v tcp               TCP connection
 169  * @v port              Local port number, in network-endian order
 170  * @ret rc              Return status code
 171  *
 172  * If the port is 0, the connection is assigned an available port
 173  * between 1024 and 65535.
 174  */
 175 static int tcp_bind ( struct tcp_connection *tcp, unsigned int port ) {
 176         struct tcp_connection *existing;
 177         static uint16_t try_port = 1023;
 178
 179         /* If no port specified, find the first available port */
 180         if ( ! port ) {
 181                 while ( try_port ) {
 182                         try_port++;
 183                         if ( try_port < 1024 )
 184                                 continue;
 185                         if ( tcp_bind ( tcp, htons ( try_port ) ) == 0 )
 186                                 return 0;
 187                 }
 188                 DBGC ( tcp, "TCP %p could not bind: no free ports\n", tcp );
 189                 return -EADDRINUSE;
 190         }
 191
 192         /* Attempt bind to local port */
 193         list_for_each_entry ( existing, &tcp_conns, list ) {
 194                 if ( existing->local_port == port ) {
 195                         DBGC ( tcp, "TCP %p could not bind: port %d in use\n",
 196                                tcp, ntohs ( port ) );
 197                         return -EADDRINUSE;
 198                 }
 199         }
 200         tcp->local_port = port;
 201
 202         DBGC ( tcp, "TCP %p bound to port %d\n", tcp, ntohs ( port ) );
 203         return 0;
 204 }
 205
 206 /**
 207  * Open a TCP connection
 208  *
 209  * @v xfer              Data transfer interface
 210  * @v peer              Peer socket address
 211  * @v local             Local socket address, or NULL
 212  * @ret rc              Return status code
 213  */
 214 static int tcp_open ( struct xfer_interface *xfer, struct sockaddr *peer,
 215                       struct sockaddr *local ) {
 216         struct sockaddr_tcpip *st_peer = ( struct sockaddr_tcpip * ) peer;
 217         struct sockaddr_tcpip *st_local = ( struct sockaddr_tcpip * ) local;
 218         struct tcp_connection *tcp;
 219         unsigned int bind_port;
 220         int rc;
 221
 222         /* Allocate and initialise structure */
 223         tcp = zalloc ( sizeof ( *tcp ) );
 224         if ( ! tcp )
 225                 return -ENOMEM;
 226         DBGC ( tcp, "TCP %p allocated\n", tcp );
 227         xfer_init ( &tcp->xfer, &tcp_xfer_operations, &tcp->refcnt );
 228         tcp->prev_tcp_state = TCP_CLOSED;
 229         tcp->tcp_state = TCP_STATE_SENT ( TCP_SYN );
 230         tcp_dump_state ( tcp );
 231         tcp->snd_seq = random();
 232         INIT_LIST_HEAD ( &tcp->queue );
 233         tcp->timer.expired = tcp_expired;
 234         memcpy ( &tcp->peer, st_peer, sizeof ( tcp->peer ) );
 235
 236         /* Bind to local port */
 237         bind_port = ( st_local ? st_local->st_port : 0 );
 238         if ( ( rc = tcp_bind ( tcp, bind_port ) ) != 0 )
 239                 goto err;
 240
 241         /* Start timer to initiate SYN */
 242         start_timer_nodelay ( &tcp->timer );
 243
 244         /* Attach parent interface, transfer reference to connection
 245          * list and return
 246          */
 247         xfer_plug_plug ( &tcp->xfer, xfer );
 248         list_add ( &tcp->list, &tcp_conns );
 249         return 0;
 250
 251  err:
 252         ref_put ( &tcp->refcnt );
 253         return rc;
 254 }
 255
 256 /**
 257  * Close TCP connection
 258  *
 259  * @v tcp               TCP connection
 260  * @v rc                Reason for close
 261  *
 262  * Closes the data transfer interface.  If the TCP state machine is in
 263  * a suitable state, the connection will be deleted.
 264  */
 265 static void tcp_close ( struct tcp_connection *tcp, int rc ) {
 266         struct io_buffer *iobuf;
 267         struct io_buffer *tmp;
 268
 269         /* Close data transfer interface */
 270         xfer_nullify ( &tcp->xfer );
 271         xfer_close ( &tcp->xfer, rc );
 272         tcp->xfer_closed = 1;
 273
 274         /* If we are in CLOSED, or have otherwise not yet received a
 275          * SYN (i.e. we are in LISTEN or SYN_SENT), just delete the
 276          * connection.
 277          */
 278         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 279
 280                 /* Transition to CLOSED for the sake of debugging messages */
 281                 tcp->tcp_state = TCP_CLOSED;
 282                 tcp_dump_state ( tcp );
 283
 284                 /* Free any unsent I/O buffers */
 285                 list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 286                         list_del ( &iobuf->list );
 287                         free_iob ( iobuf );
 288                 }
 289
 290                 /* Remove from list and drop reference */
 291                 stop_timer ( &tcp->timer );
 292                 list_del ( &tcp->list );
 293                 ref_put ( &tcp->refcnt );
 294                 DBGC ( tcp, "TCP %p connection deleted\n", tcp );
 295                 return;
 296         }
 297
 298         /* If we have not had our SYN acknowledged (i.e. we are in
 299          * SYN_RCVD), pretend that it has been acknowledged so that we
 300          * can send a FIN without breaking things.
 301          */
 302         if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 303                 tcp_rx_ack ( tcp, ( tcp->snd_seq + 1 ), 0 );
 304
 305         /* If we have no data remaining to send, start sending FIN */
 306         if ( list_empty ( &tcp->queue ) ) {
 307                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 308                 tcp_dump_state ( tcp );
 309         }
 310 }
 311
 312 /***************************************************************************
 313  *
 314  * Transmit data path
 315  *
 316  ***************************************************************************
 317  */
 318
 319 /**
 320  * Calculate transmission window
 321  *
 322  * @v tcp               TCP connection
 323  * @ret len             Maximum length that can be sent in a single packet
 324  */
 325 static size_t tcp_xmit_win ( struct tcp_connection *tcp ) {
 326         size_t len;
 327
 328         /* Not ready if we're not in a suitable connection state */
 329         if ( ! TCP_CAN_SEND_DATA ( tcp->tcp_state ) )
 330                 return 0;
 331
 332         /* Length is the minimum of the receiver's window and the path MTU */
 333         len = tcp->snd_win;
 334         if ( len > TCP_PATH_MTU )
 335                 len = TCP_PATH_MTU;
 336
 337         return len;
 338 }
 339
 340 /**
 341  * Process TCP transmit queue
 342  *
 343  * @v tcp               TCP connection
 344  * @v max_len           Maximum length to process
 345  * @v dest              I/O buffer to fill with data, or NULL
 346  * @v remove            Remove data from queue
 347  * @ret len             Length of data processed
 348  *
 349  * This processes at most @c max_len bytes from the TCP connection's
 350  * transmit queue.  Data will be copied into the @c dest I/O buffer
 351  * (if provided) and, if @c remove is true, removed from the transmit
 352  * queue.
 353  */
 354 static size_t tcp_process_queue ( struct tcp_connection *tcp, size_t max_len,
 355                                   struct io_buffer *dest, int remove ) {
 356         struct io_buffer *iobuf;
 357         struct io_buffer *tmp;
 358         size_t frag_len;
 359         size_t len = 0;
 360
 361         list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 362                 frag_len = iob_len ( iobuf );
 363                 if ( frag_len > max_len )
 364                         frag_len = max_len;
 365                 if ( dest ) {
 366                         memcpy ( iob_put ( dest, frag_len ), iobuf->data,
 367                                  frag_len );
 368                 }
 369                 if ( remove ) {
 370                         iob_pull ( iobuf, frag_len );
 371                         if ( ! iob_len ( iobuf ) ) {
 372                                 list_del ( &iobuf->list );
 373                                 free_iob ( iobuf );
 374                         }
 375                 }
 376                 len += frag_len;
 377                 max_len -= frag_len;
 378         }
 379         return len;
 380 }
 381
 382 /**
 383  * Transmit any outstanding data
 384  *
 385  * @v tcp               TCP connection
 386  * @v force_send        Force sending of packet
 387  *
 388  * Transmits any outstanding data on the connection.
 389  *
 390  * Note that even if an error is returned, the retransmission timer
 391  * will have been started if necessary, and so the stack will
 392  * eventually attempt to retransmit the failed packet.
 393  */
 394 static int tcp_xmit ( struct tcp_connection *tcp, int force_send ) {
 395         struct io_buffer *iobuf;
 396         struct tcp_header *tcphdr;
 397         struct tcp_mss_option *mssopt;
 398         struct tcp_timestamp_padded_option *tsopt;
 399         void *payload;
 400         unsigned int flags;
 401         size_t len = 0;
 402         uint32_t seq_len;
 403         uint32_t app_win;
 404         uint32_t max_rcv_win;
 405         int rc;
 406
 407         /* If retransmission timer is already running, do nothing */
 408         if ( timer_running ( &tcp->timer ) )
 409                 return 0;
 410
 411         /* Calculate both the actual (payload) and sequence space
 412          * lengths that we wish to transmit.
 413          */
 414         if ( TCP_CAN_SEND_DATA ( tcp->tcp_state ) ) {
 415                 len = tcp_process_queue ( tcp, tcp_xmit_win ( tcp ),
 416                                           NULL, 0 );
 417         }
 418         seq_len = len;
 419         flags = TCP_FLAGS_SENDING ( tcp->tcp_state );
 420         if ( flags & ( TCP_SYN | TCP_FIN ) ) {
 421                 /* SYN or FIN consume one byte, and we can never send both */
 422                 assert ( ! ( ( flags & TCP_SYN ) && ( flags & TCP_FIN ) ) );
 423                 seq_len++;
 424         }
 425         tcp->snd_sent = seq_len;
 426
 427         /* If we have nothing to transmit, stop now */
 428         if ( ( seq_len == 0 ) && ! force_send )
 429                 return 0;
 430
 431         /* If we are transmitting anything that requires
 432          * acknowledgement (i.e. consumes sequence space), start the
 433          * retransmission timer.  Do this before attempting to
 434          * allocate the I/O buffer, in case allocation itself fails.
 435          */
 436         if ( seq_len )
 437                 start_timer ( &tcp->timer );
 438
 439         /* Allocate I/O buffer */
 440         iobuf = alloc_iob ( len + MAX_HDR_LEN );
 441         if ( ! iobuf ) {
 442                 DBGC ( tcp, "TCP %p could not allocate iobuf for %08x..%08x "
 443                        "%08x\n", tcp, tcp->snd_seq, ( tcp->snd_seq + seq_len ),
 444                        tcp->rcv_ack );
 445                 return -ENOMEM;
 446         }
 447         iob_reserve ( iobuf, MAX_HDR_LEN );
 448
 449         /* Fill data payload from transmit queue */
 450         tcp_process_queue ( tcp, len, iobuf, 0 );
 451
 452         /* Expand receive window if possible */
 453         max_rcv_win = ( ( freemem * 3 ) / 4 );
 454         if ( max_rcv_win > TCP_MAX_WINDOW_SIZE )
 455                 max_rcv_win = TCP_MAX_WINDOW_SIZE;
 456         app_win = xfer_window ( &tcp->xfer );
 457         if ( max_rcv_win > app_win )
 458                 max_rcv_win = app_win;
 459         max_rcv_win &= ~0x03; /* Keep everything dword-aligned */
 460         if ( tcp->rcv_win < max_rcv_win )
 461                 tcp->rcv_win = max_rcv_win;
 462
 463         /* Fill up the TCP header */
 464         payload = iobuf->data;
 465         if ( flags & TCP_SYN ) {
 466                 mssopt = iob_push ( iobuf, sizeof ( *mssopt ) );
 467                 mssopt->kind = TCP_OPTION_MSS;
 468                 mssopt->length = sizeof ( *mssopt );
 469                 mssopt->mss = htons ( TCP_MSS );
 470         }
 471         if ( ( flags & TCP_SYN ) || tcp->timestamps ) {
 472                 tsopt = iob_push ( iobuf, sizeof ( *tsopt ) );
 473                 memset ( tsopt->nop, TCP_OPTION_NOP, sizeof ( tsopt->nop ) );
 474                 tsopt->tsopt.kind = TCP_OPTION_TS;
 475                 tsopt->tsopt.length = sizeof ( tsopt->tsopt );
 476                 tsopt->tsopt.tsval = ntohl ( currticks() );
 477                 tsopt->tsopt.tsecr = ntohl ( tcp->ts_recent );
 478         }
 479         if ( ! ( flags & TCP_SYN ) )
 480                 flags |= TCP_PSH;
 481         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 482         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 483         tcphdr->src = tcp->local_port;
 484         tcphdr->dest = tcp->peer.st_port;
 485         tcphdr->seq = htonl ( tcp->snd_seq );
 486         tcphdr->ack = htonl ( tcp->rcv_ack );
 487         tcphdr->hlen = ( ( payload - iobuf->data ) << 2 );
 488         tcphdr->flags = flags;
 489         tcphdr->win = htons ( tcp->rcv_win );
 490         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 491
 492         /* Dump header */
 493         DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4zd",
 494                 tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 495                 ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) + seq_len ),
 496                 ntohl ( tcphdr->ack ), len );
 497         tcp_dump_flags ( tcp, tcphdr->flags );
 498         DBGC2 ( tcp, "\n" );
 499
 500         /* Transmit packet */
 501         if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, &tcp->peer, NULL,
 502                                &tcphdr->csum ) ) != 0 ) {
 503                 DBGC ( tcp, "TCP %p could not transmit %08x..%08x %08x: %s\n",
 504                        tcp, tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ),
 505                        tcp->rcv_ack, strerror ( rc ) );
 506                 return rc;
 507         }
 508
 509         return 0;
 510 }
 511
 512 /**
 513  * Retransmission timer expired
 514  *
 515  * @v timer     Retry timer
 516  * @v over      Failure indicator
 517  */
 518 static void tcp_expired ( struct retry_timer *timer, int over ) {
 519         struct tcp_connection *tcp =
 520                 container_of ( timer, struct tcp_connection, timer );
 521         int graceful_close = TCP_CLOSED_GRACEFULLY ( tcp->tcp_state );
 522
 523         DBGC ( tcp, "TCP %p timer %s in %s for %08x..%08x %08x\n", tcp,
 524                ( over ? "expired" : "fired" ), tcp_state ( tcp->tcp_state ),
 525                tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
 526
 527         assert ( ( tcp->tcp_state == TCP_SYN_SENT ) ||
 528                  ( tcp->tcp_state == TCP_SYN_RCVD ) ||
 529                  ( tcp->tcp_state == TCP_ESTABLISHED ) ||
 530                  ( tcp->tcp_state == TCP_FIN_WAIT_1 ) ||
 531                  ( tcp->tcp_state == TCP_TIME_WAIT ) ||
 532                  ( tcp->tcp_state == TCP_CLOSE_WAIT ) ||
 533                  ( tcp->tcp_state == TCP_CLOSING_OR_LAST_ACK ) );
 534
 535         if ( over || graceful_close ) {
 536                 /* If we have finally timed out and given up, or if
 537                  * this is the result of a graceful close, terminate
 538                  * the connection
 539                  */
 540                 tcp->tcp_state = TCP_CLOSED;
 541                 tcp_dump_state ( tcp );
 542                 tcp_close ( tcp, -ETIMEDOUT );
 543         } else {
 544                 /* Otherwise, retransmit the packet */
 545                 tcp_xmit ( tcp, 0 );
 546         }
 547 }
 548
 549 /**
 550  * Send RST response to incoming packet
 551  *
 552  * @v in_tcphdr         TCP header of incoming packet
 553  * @ret rc              Return status code
 554  */
 555 static int tcp_xmit_reset ( struct tcp_connection *tcp,
 556                             struct sockaddr_tcpip *st_dest,
 557                             struct tcp_header *in_tcphdr ) {
 558         struct io_buffer *iobuf;
 559         struct tcp_header *tcphdr;
 560         int rc;
 561
 562         /* Allocate space for dataless TX buffer */
 563         iobuf = alloc_iob ( MAX_HDR_LEN );
 564         if ( ! iobuf ) {
 565                 DBGC ( tcp, "TCP %p could not allocate iobuf for RST "
 566                        "%08x..%08x %08x\n", tcp, ntohl ( in_tcphdr->ack ),
 567                        ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ) );
 568                 return -ENOMEM;
 569         }
 570         iob_reserve ( iobuf, MAX_HDR_LEN );
 571
 572         /* Construct RST response */
 573         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 574         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 575         tcphdr->src = in_tcphdr->dest;
 576         tcphdr->dest = in_tcphdr->src;
 577         tcphdr->seq = in_tcphdr->ack;
 578         tcphdr->ack = in_tcphdr->seq;
 579         tcphdr->hlen = ( ( sizeof ( *tcphdr ) / 4 ) << 4 );
 580         tcphdr->flags = ( TCP_RST | TCP_ACK );
 581         tcphdr->win = htons ( TCP_MAX_WINDOW_SIZE );
 582         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 583
 584         /* Dump header */
 585         DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4d",
 586                 tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 587                 ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) ),
 588                 ntohl ( tcphdr->ack ), 0 );
 589         tcp_dump_flags ( tcp, tcphdr->flags );
 590         DBGC2 ( tcp, "\n" );
 591
 592         /* Transmit packet */
 593         if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, st_dest,
 594                                NULL, &tcphdr->csum ) ) != 0 ) {
 595                 DBGC ( tcp, "TCP %p could not transmit RST %08x..%08x %08x: "
 596                        "%s\n", tcp, ntohl ( in_tcphdr->ack ),
 597                        ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ),
 598                        strerror ( rc ) );
 599                 return rc;
 600         }
 601
 602         return 0;
 603 }
 604
 605 /***************************************************************************
 606  *
 607  * Receive data path
 608  *
 609  ***************************************************************************
 610  */
 611
 612 /**
 613  * Identify TCP connection by local port number
 614  *
 615  * @v local_port        Local port (in network-endian order)
 616  * @ret tcp             TCP connection, or NULL
 617  */
 618 static struct tcp_connection * tcp_demux ( unsigned int local_port ) {
 619         struct tcp_connection *tcp;
 620
 621         list_for_each_entry ( tcp, &tcp_conns, list ) {
 622                 if ( tcp->local_port == local_port )
 623                         return tcp;
 624         }
 625         return NULL;
 626 }
 627
 628 /**
 629  * Parse TCP received options
 630  *
 631  * @v tcp               TCP connection
 632  * @v data              Raw options data
 633  * @v len               Raw options length
 634  * @v options           Options structure to fill in
 635  */
 636 static void tcp_rx_opts ( struct tcp_connection *tcp, const void *data,
 637                           size_t len, struct tcp_options *options ) {
 638         const void *end = ( data + len );
 639         const struct tcp_option *option;
 640         unsigned int kind;
 641
 642         memset ( options, 0, sizeof ( *options ) );
 643         while ( data < end ) {
 644                 option = data;
 645                 kind = option->kind;
 646                 if ( kind == TCP_OPTION_END )
 647                         return;
 648                 if ( kind == TCP_OPTION_NOP ) {
 649                         data++;
 650                         continue;
 651                 }
 652                 switch ( kind ) {
 653                 case TCP_OPTION_MSS:
 654                         options->mssopt = data;
 655                         break;
 656                 case TCP_OPTION_TS:
 657                         options->tsopt = data;
 658                         break;
 659                 default:
 660                         DBGC ( tcp, "TCP %p received unknown option %d\n",
 661                                tcp, kind );
 662                         break;
 663                 }
 664                 data += option->length;
 665         }
 666 }
 667
 668 /**
 669  * Consume received sequence space
 670  *
 671  * @v tcp               TCP connection
 672  * @v seq_len           Sequence space length to consume
 673  */
 674 static void tcp_rx_seq ( struct tcp_connection *tcp, uint32_t seq_len ) {
 675         tcp->rcv_ack += seq_len;
 676         if ( tcp->rcv_win > seq_len ) {
 677                 tcp->rcv_win -= seq_len;
 678         } else {
 679                 tcp->rcv_win = 0;
 680         }
 681 }
 682
 683 /**
 684  * Handle TCP received SYN
 685  *
 686  * @v tcp               TCP connection
 687  * @v seq               SEQ value (in host-endian order)
 688  * @v options           TCP options
 689  * @ret rc              Return status code
 690  */
 691 static int tcp_rx_syn ( struct tcp_connection *tcp, uint32_t seq,
 692                         struct tcp_options *options ) {
 693
 694         /* Synchronise sequence numbers on first SYN */
 695         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 696                 tcp->rcv_ack = seq;
 697                 if ( options->tsopt )
 698                         tcp->timestamps = 1;
 699         }
 700
 701         /* Ignore duplicate SYN */
 702         if ( ( tcp->rcv_ack - seq ) > 0 )
 703                 return 0;
 704
 705         /* Mark SYN as received and start sending ACKs with each packet */
 706         tcp->tcp_state |= ( TCP_STATE_SENT ( TCP_ACK ) |
 707                             TCP_STATE_RCVD ( TCP_SYN ) );
 708
 709         /* Acknowledge SYN */
 710         tcp_rx_seq ( tcp, 1 );
 711
 712         return 0;
 713 }
 714
 715 /**
 716  * Handle TCP received ACK
 717  *
 718  * @v tcp               TCP connection
 719  * @v ack               ACK value (in host-endian order)
 720  * @v win               WIN value (in host-endian order)
 721  * @ret rc              Return status code
 722  */
 723 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
 724                         uint32_t win ) {
 725         uint32_t ack_len = ( ack - tcp->snd_seq );
 726         size_t len;
 727         unsigned int acked_flags;
 728
 729         /* Check for out-of-range or old duplicate ACKs */
 730         if ( ack_len > tcp->snd_sent ) {
 731                 DBGC ( tcp, "TCP %p received ACK for %08x..%08x, "
 732                        "sent only %08x..%08x\n", tcp, tcp->snd_seq,
 733                        ( tcp->snd_seq + ack_len ), tcp->snd_seq,
 734                        ( tcp->snd_seq + tcp->snd_sent ) );
 735
 736                 if ( TCP_HAS_BEEN_ESTABLISHED ( tcp->tcp_state ) ) {
 737                         /* Just ignore what might be old duplicate ACKs */
 738                         return 0;
 739                 } else {
 740                         /* Send RST if an out-of-range ACK is received
 741                          * on a not-yet-established connection, as per
 742                          * RFC 793.
 743                          */
 744                         return -EINVAL;
 745                 }
 746         }
 747
 748         /* Ignore ACKs that don't actually acknowledge any new data.
 749          * (In particular, do not stop the retransmission timer; this
 750          * avoids creating a sorceror's apprentice syndrome when a
 751          * duplicate ACK is received and we still have data in our
 752          * transmit queue.)
 753          */
 754         if ( ack_len == 0 )
 755                 return 0;
 756
 757         /* Stop the retransmission timer */
 758         stop_timer ( &tcp->timer );
 759
 760         /* Determine acknowledged flags and data length */
 761         len = ack_len;
 762         acked_flags = ( TCP_FLAGS_SENDING ( tcp->tcp_state ) &
 763                         ( TCP_SYN | TCP_FIN ) );
 764         if ( acked_flags )
 765                 len--;
 766
 767         /* Update SEQ and sent counters, and window size */
 768         tcp->snd_seq = ack;
 769         tcp->snd_sent = 0;
 770         tcp->snd_win = win;
 771
 772         /* Remove any acknowledged data from transmit queue */
 773         tcp_process_queue ( tcp, len, NULL, 1 );
 774
 775         /* Mark SYN/FIN as acknowledged if applicable. */
 776         if ( acked_flags )
 777                 tcp->tcp_state |= TCP_STATE_ACKED ( acked_flags );
 778
 779         /* Start sending FIN if we've had all possible data ACKed */
 780         if ( list_empty ( &tcp->queue ) && tcp->xfer_closed )
 781                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 782
 783         return 0;
 784 }
 785
 786 /**
 787  * Handle TCP received data
 788  *
 789  * @v tcp               TCP connection
 790  * @v seq               SEQ value (in host-endian order)
 791  * @v iobuf             I/O buffer
 792  * @ret rc              Return status code
 793  *
 794  * This function takes ownership of the I/O buffer.
 795  */
 796 static int tcp_rx_data ( struct tcp_connection *tcp, uint32_t seq,
 797                          struct io_buffer *iobuf ) {
 798         uint32_t already_rcvd;
 799         uint32_t len;
 800         int rc;
 801
 802         /* Ignore duplicate or out-of-order data */
 803         already_rcvd = ( tcp->rcv_ack - seq );
 804         len = iob_len ( iobuf );
 805         if ( already_rcvd >= len ) {
 806                 free_iob ( iobuf );
 807                 return 0;
 808         }
 809         iob_pull ( iobuf, already_rcvd );
 810         len -= already_rcvd;
 811
 812         /* Deliver data to application */
 813         if ( ( rc = xfer_deliver_iob ( &tcp->xfer, iobuf ) ) != 0 ) {
 814                 DBGC ( tcp, "TCP %p could not deliver %08x..%08x: %s\n",
 815                        tcp, seq, ( seq + len ), strerror ( rc ) );
 816                 return rc;
 817         }
 818
 819         /* Acknowledge new data */
 820         tcp_rx_seq ( tcp, len );
 821
 822         return 0;
 823 }
 824
 825 /**
 826  * Handle TCP received FIN
 827  *
 828  * @v tcp               TCP connection
 829  * @v seq               SEQ value (in host-endian order)
 830  * @ret rc              Return status code
 831  */
 832 static int tcp_rx_fin ( struct tcp_connection *tcp, uint32_t seq ) {
 833
 834         /* Ignore duplicate or out-of-order FIN */
 835         if ( ( tcp->rcv_ack - seq ) > 0 )
 836                 return 0;
 837
 838         /* Mark FIN as received and acknowledge it */
 839         tcp->tcp_state |= TCP_STATE_RCVD ( TCP_FIN );
 840         tcp_rx_seq ( tcp, 1 );
 841
 842         /* Close connection */
 843         tcp_close ( tcp, 0 );
 844
 845         return 0;
 846 }
 847
 848 /**
 849  * Handle TCP received RST
 850  *
 851  * @v tcp               TCP connection
 852  * @v seq               SEQ value (in host-endian order)
 853  * @ret rc              Return status code
 854  */
 855 static int tcp_rx_rst ( struct tcp_connection *tcp, uint32_t seq ) {
 856
 857         /* Accept RST only if it falls within the window.  If we have
 858          * not yet received a SYN, then we have no window to test
 859          * against, so fall back to checking that our SYN has been
 860          * ACKed.
 861          */
 862         if ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) {
 863                 if ( ( seq - tcp->rcv_ack ) >= tcp->rcv_win )
 864                         return 0;
 865         } else {
 866                 if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 867                         return 0;
 868         }
 869
 870         /* Abort connection */
 871         tcp->tcp_state = TCP_CLOSED;
 872         tcp_dump_state ( tcp );
 873         tcp_close ( tcp, -ECONNRESET );
 874
 875         DBGC ( tcp, "TCP %p connection reset by peer\n", tcp );
 876         return -ECONNRESET;
 877 }
 878
 879 /**
 880  * Process received packet
 881  *
 882  * @v iobuf             I/O buffer
 883  * @v st_src            Partially-filled source address
 884  * @v st_dest           Partially-filled destination address
 885  * @v pshdr_csum        Pseudo-header checksum
 886  * @ret rc              Return status code
 887   */
 888 static int tcp_rx ( struct io_buffer *iobuf,
 889                     struct sockaddr_tcpip *st_src,
 890                     struct sockaddr_tcpip *st_dest __unused,
 891                     uint16_t pshdr_csum ) {
 892         struct tcp_header *tcphdr = iobuf->data;
 893         struct tcp_connection *tcp;
 894         struct tcp_options options;
 895         size_t hlen;
 896         uint16_t csum;
 897         uint32_t start_seq;
 898         uint32_t seq;
 899         uint32_t ack;
 900         uint32_t win;
 901         unsigned int flags;
 902         size_t len;
 903         int rc;
 904
 905         /* Sanity check packet */
 906         if ( iob_len ( iobuf ) < sizeof ( *tcphdr ) ) {
 907                 DBG ( "TCP packet too short at %zd bytes (min %zd bytes)\n",
 908                       iob_len ( iobuf ), sizeof ( *tcphdr ) );
 909                 rc = -EINVAL;
 910                 goto discard;
 911         }
 912         hlen = ( ( tcphdr->hlen & TCP_MASK_HLEN ) / 16 ) * 4;
 913         if ( hlen < sizeof ( *tcphdr ) ) {
 914                 DBG ( "TCP header too short at %zd bytes (min %zd bytes)\n",
 915                       hlen, sizeof ( *tcphdr ) );
 916                 rc = -EINVAL;
 917                 goto discard;
 918         }
 919         if ( hlen > iob_len ( iobuf ) ) {
 920                 DBG ( "TCP header too long at %zd bytes (max %zd bytes)\n",
 921                       hlen, iob_len ( iobuf ) );
 922                 rc = -EINVAL;
 923                 goto discard;
 924         }
 925         csum = tcpip_continue_chksum ( pshdr_csum, iobuf->data,
 926                                        iob_len ( iobuf ) );
 927         if ( csum != 0 ) {
 928                 DBG ( "TCP checksum incorrect (is %04x including checksum "
 929                       "field, should be 0000)\n", csum );
 930                 rc = -EINVAL;
 931                 goto discard;
 932         }
 933
 934         /* Parse parameters from header and strip header */
 935         tcp = tcp_demux ( tcphdr->dest );
 936         start_seq = seq = ntohl ( tcphdr->seq );
 937         ack = ntohl ( tcphdr->ack );
 938         win = ntohs ( tcphdr->win );
 939         flags = tcphdr->flags;
 940         tcp_rx_opts ( tcp, ( ( ( void * ) tcphdr ) + sizeof ( *tcphdr ) ),
 941                       ( hlen - sizeof ( *tcphdr ) ), &options );
 942         iob_pull ( iobuf, hlen );
 943         len = iob_len ( iobuf );
 944
 945         /* Dump header */
 946         DBGC2 ( tcp, "TCP %p RX %d<-%d           %08x %08x..%08zx %4zd",
 947                 tcp, ntohs ( tcphdr->dest ), ntohs ( tcphdr->src ),
 948                 ntohl ( tcphdr->ack ), ntohl ( tcphdr->seq ),
 949                 ( ntohl ( tcphdr->seq ) + len +
 950                   ( ( tcphdr->flags & ( TCP_SYN | TCP_FIN ) ) ? 1 : 0 )), len);
 951         tcp_dump_flags ( tcp, tcphdr->flags );
 952         DBGC2 ( tcp, "\n" );
 953
 954         /* If no connection was found, send RST */
 955         if ( ! tcp ) {
 956                 tcp_xmit_reset ( tcp, st_src, tcphdr );
 957                 rc = -ENOTCONN;
 958                 goto discard;
 959         }
 960
 961         /* Handle ACK, if present */
 962         if ( flags & TCP_ACK ) {
 963                 if ( ( rc = tcp_rx_ack ( tcp, ack, win ) ) != 0 ) {
 964                         tcp_xmit_reset ( tcp, st_src, tcphdr );
 965                         goto discard;
 966                 }
 967         }
 968
 969         /* Handle SYN, if present */
 970         if ( flags & TCP_SYN ) {
 971                 tcp_rx_syn ( tcp, seq, &options );
 972                 seq++;
 973         }
 974
 975         /* Handle RST, if present */
 976         if ( flags & TCP_RST ) {
 977                 if ( ( rc = tcp_rx_rst ( tcp, seq ) ) != 0 )
 978                         goto discard;
 979         }
 980
 981         /* Handle new data, if any */
 982         tcp_rx_data ( tcp, seq, iobuf );
 983         seq += len;
 984
 985         /* Handle FIN, if present */
 986         if ( flags & TCP_FIN ) {
 987                 tcp_rx_fin ( tcp, seq );
 988                 seq++;
 989         }
 990
 991         /* Update timestamp, if present and applicable */
 992         if ( ( seq == tcp->rcv_ack ) && options.tsopt )
 993                 tcp->ts_recent = ntohl ( options.tsopt->tsval );
 994
 995         /* Dump out any state change as a result of the received packet */
 996         tcp_dump_state ( tcp );
 997
 998         /* Send out any pending data.  We force sending a reply if either
 999          *
1000          *  a) the peer is expecting an ACK (i.e. consumed sequence space), or
1001          *  b) either end of the packet was outside the receive window
1002          *
1003          * Case (b) enables us to support TCP keepalives using
1004          * zero-length packets, which we would otherwise ignore.  Note
1005          * that for case (b), we need *only* consider zero-length
1006          * packets, since non-zero-length packets will already be
1007          * caught by case (a).
1008          */
1009         tcp_xmit ( tcp, ( ( start_seq != seq ) ||
1010                           ( ( seq - tcp->rcv_ack ) > tcp->rcv_win ) ) );
1011
1012         /* If this packet was the last we expect to receive, set up
1013          * timer to expire and cause the connection to be freed.
1014          */
1015         if ( TCP_CLOSED_GRACEFULLY ( tcp->tcp_state ) ) {
1016                 tcp->timer.timeout = ( 2 * TCP_MSL );
1017                 start_timer ( &tcp->timer );
1018         }
1019
1020         return 0;
1021
1022  discard:
1023         /* Free received packet */
1024         free_iob ( iobuf );
1025         return rc;
1026 }
1027
1028 /** TCP protocol */
1029 struct tcpip_protocol tcp_protocol __tcpip_protocol = {
1030         .name = "TCP",
1031         .rx = tcp_rx,
1032         .tcpip_proto = IP_TCP,
1033 };
1034
1035 /***************************************************************************
1036  *
1037  * Data transfer interface
1038  *
1039  ***************************************************************************
1040  */
1041
1042 /**
1043  * Close interface
1044  *
1045  * @v xfer              Data transfer interface
1046  * @v rc                Reason for close
1047  */
1048 static void tcp_xfer_close ( struct xfer_interface *xfer, int rc ) {
1049         struct tcp_connection *tcp =
1050                 container_of ( xfer, struct tcp_connection, xfer );
1051
1052         /* Close data transfer interface */
1053         tcp_close ( tcp, rc );
1054
1055         /* Transmit FIN, if possible */
1056         tcp_xmit ( tcp, 0 );
1057 }
1058
1059 /**
1060  * Check flow control window
1061  *
1062  * @v xfer              Data transfer interface
1063  * @ret len             Length of window
1064  */
1065 static size_t tcp_xfer_window ( struct xfer_interface *xfer ) {
1066         struct tcp_connection *tcp =
1067                 container_of ( xfer, struct tcp_connection, xfer );
1068
1069         /* Not ready if data queue is non-empty.  This imposes a limit
1070          * of only one unACKed packet in the TX queue at any time; we
1071          * do this to conserve memory usage.
1072          */
1073         if ( ! list_empty ( &tcp->queue ) )
1074                 return 0;
1075
1076         /* Return TCP window length */
1077         return tcp_xmit_win ( tcp );
1078 }
1079
1080 /**
1081  * Deliver datagram as I/O buffer
1082  *
1083  * @v xfer              Data transfer interface
1084  * @v iobuf             Datagram I/O buffer
1085  * @v meta              Data transfer metadata
1086  * @ret rc              Return status code
1087  */
1088 static int tcp_xfer_deliver_iob ( struct xfer_interface *xfer,
1089                                   struct io_buffer *iobuf,
1090                                   struct xfer_metadata *meta __unused ) {
1091         struct tcp_connection *tcp =
1092                 container_of ( xfer, struct tcp_connection, xfer );
1093
1094         /* Enqueue packet */
1095         list_add_tail ( &iobuf->list, &tcp->queue );
1096
1097         /* Transmit data, if possible */
1098         tcp_xmit ( tcp, 0 );
1099
1100         return 0;
1101 }
1102
1103 /** TCP data transfer interface operations */
1104 static struct xfer_interface_operations tcp_xfer_operations = {
1105         .close          = tcp_xfer_close,
1106         .vredirect      = ignore_xfer_vredirect,
1107         .window         = tcp_xfer_window,
1108         .alloc_iob      = default_xfer_alloc_iob,
1109         .deliver_iob    = tcp_xfer_deliver_iob,
1110         .deliver_raw    = xfer_deliver_as_iob,
1111 };
1112
1113 /***************************************************************************
1114  *
1115  * Openers
1116  *
1117  ***************************************************************************
1118  */
1119
1120 /** TCP socket opener */
1121 struct socket_opener tcp_socket_opener __socket_opener = {
1122         .semantics      = TCP_SOCK_STREAM,
1123         .family         = AF_INET,
1124         .open           = tcp_open,
1125 };
1126
1127 /** Linkage hack */
1128 int tcp_sock_stream = TCP_SOCK_STREAM;
1129
1130 /**
1131  * Open TCP URI
1132  *
1133  * @v xfer              Data transfer interface
1134  * @v uri               URI
1135  * @ret rc              Return status code
1136  */
1137 static int tcp_open_uri ( struct xfer_interface *xfer, struct uri *uri ) {
1138         struct sockaddr_tcpip peer;
1139
1140         /* Sanity check */
1141         if ( ! uri->host )
1142                 return -EINVAL;
1143
1144         memset ( &peer, 0, sizeof ( peer ) );
1145         peer.st_port = htons ( uri_port ( uri, 0 ) );
1146         return xfer_open_named_socket ( xfer, SOCK_STREAM,
1147                                         ( struct sockaddr * ) &peer,
1148                                         uri->host, NULL );
1149 }
1150
1151 /** TCP URI opener */
1152 struct uri_opener tcp_uri_opener __uri_opener = {
1153         .scheme         = "tcp",
1154         .open           = tcp_open_uri,
1155 };
1156