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 */
  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         /** Shutdown (TIME_WAIT) timer */
  88         struct retry_timer wait;
  89 };
  90
  91 /**
  92  * List of registered TCP connections
  93  */
  94 static LIST_HEAD ( tcp_conns );
  95
  96 /* Forward declarations */
  97 static struct xfer_interface_operations tcp_xfer_operations;
  98 static void tcp_expired ( struct retry_timer *timer, int over );
  99 static void tcp_wait_expired ( struct retry_timer *timer, int over );
 100 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
 101                         uint32_t win );
 102
 103 /**
 104  * Name TCP state
 105  *
 106  * @v state             TCP state
 107  * @ret name            Name of TCP state
 108  */
 109 static inline __attribute__ (( always_inline )) const char *
 110 tcp_state ( int state ) {
 111         switch ( state ) {
 112         case TCP_CLOSED:                return "CLOSED";
 113         case TCP_LISTEN:                return "LISTEN";
 114         case TCP_SYN_SENT:              return "SYN_SENT";
 115         case TCP_SYN_RCVD:              return "SYN_RCVD";
 116         case TCP_ESTABLISHED:           return "ESTABLISHED";
 117         case TCP_FIN_WAIT_1:            return "FIN_WAIT_1";
 118         case TCP_FIN_WAIT_2:            return "FIN_WAIT_2";
 119         case TCP_CLOSING_OR_LAST_ACK:   return "CLOSING/LAST_ACK";
 120         case TCP_TIME_WAIT:             return "TIME_WAIT";
 121         case TCP_CLOSE_WAIT:            return "CLOSE_WAIT";
 122         default:                        return "INVALID";
 123         }
 124 }
 125
 126 /**
 127  * Dump TCP state transition
 128  *
 129  * @v tcp               TCP connection
 130  */
 131 static inline __attribute__ (( always_inline )) void
 132 tcp_dump_state ( struct tcp_connection *tcp ) {
 133
 134         if ( tcp->tcp_state != tcp->prev_tcp_state ) {
 135                 DBGC ( tcp, "TCP %p transitioned from %s to %s\n", tcp,
 136                        tcp_state ( tcp->prev_tcp_state ),
 137                        tcp_state ( tcp->tcp_state ) );
 138         }
 139         tcp->prev_tcp_state = tcp->tcp_state;
 140 }
 141
 142 /**
 143  * Dump TCP flags
 144  *
 145  * @v flags             TCP flags
 146  */
 147 static inline __attribute__ (( always_inline )) void
 148 tcp_dump_flags ( struct tcp_connection *tcp, unsigned int flags ) {
 149         if ( flags & TCP_RST )
 150                 DBGC2 ( tcp, " RST" );
 151         if ( flags & TCP_SYN )
 152                 DBGC2 ( tcp, " SYN" );
 153         if ( flags & TCP_PSH )
 154                 DBGC2 ( tcp, " PSH" );
 155         if ( flags & TCP_FIN )
 156                 DBGC2 ( tcp, " FIN" );
 157         if ( flags & TCP_ACK )
 158                 DBGC2 ( tcp, " ACK" );
 159 }
 160
 161 /***************************************************************************
 162  *
 163  * Open and close
 164  *
 165  ***************************************************************************
 166  */
 167
 168 /**
 169  * Bind TCP connection to local port
 170  *
 171  * @v tcp               TCP connection
 172  * @v port              Local port number
 173  * @ret rc              Return status code
 174  *
 175  * If the port is 0, the connection is assigned an available port
 176  * between 1024 and 65535.
 177  */
 178 static int tcp_bind ( struct tcp_connection *tcp, unsigned int port ) {
 179         struct tcp_connection *existing;
 180         uint16_t try_port;
 181         int i;
 182
 183         /* If no port specified, find an available port */
 184         if ( ! port ) {
 185                 try_port = ( random() % 64512 ) + 1023;
 186                 for ( i = 0 ; i < 65536 ; ++i ) {
 187                         if ( ++try_port < 1024 )
 188                                 continue;
 189                         if ( tcp_bind ( tcp, try_port ) == 0 )
 190                                 return 0;
 191                 }
 192                 DBGC ( tcp, "TCP %p could not bind: no free ports\n", tcp );
 193                 return -EADDRINUSE;
 194         }
 195
 196         /* Attempt bind to local port */
 197         list_for_each_entry ( existing, &tcp_conns, list ) {
 198                 if ( existing->local_port == port ) {
 199                         DBGC ( tcp, "TCP %p could not bind: port %d in use\n",
 200                                tcp, port );
 201                         return -EADDRINUSE;
 202                 }
 203         }
 204         tcp->local_port = port;
 205
 206         DBGC ( tcp, "TCP %p bound to port %d\n", tcp, port );
 207         return 0;
 208 }
 209
 210 /**
 211  * Open a TCP connection
 212  *
 213  * @v xfer              Data transfer interface
 214  * @v peer              Peer socket address
 215  * @v local             Local socket address, or NULL
 216  * @ret rc              Return status code
 217  */
 218 static int tcp_open ( struct xfer_interface *xfer, struct sockaddr *peer,
 219                       struct sockaddr *local ) {
 220         struct sockaddr_tcpip *st_peer = ( struct sockaddr_tcpip * ) peer;
 221         struct sockaddr_tcpip *st_local = ( struct sockaddr_tcpip * ) local;
 222         struct tcp_connection *tcp;
 223         unsigned int bind_port;
 224         int rc;
 225
 226         /* Allocate and initialise structure */
 227         tcp = zalloc ( sizeof ( *tcp ) );
 228         if ( ! tcp )
 229                 return -ENOMEM;
 230         DBGC ( tcp, "TCP %p allocated\n", tcp );
 231         ref_init ( &tcp->refcnt, NULL );
 232         xfer_init ( &tcp->xfer, &tcp_xfer_operations, &tcp->refcnt );
 233         timer_init ( &tcp->timer, tcp_expired );
 234         timer_init ( &tcp->wait, tcp_wait_expired );
 235         tcp->prev_tcp_state = TCP_CLOSED;
 236         tcp->tcp_state = TCP_STATE_SENT ( TCP_SYN );
 237         tcp_dump_state ( tcp );
 238         tcp->snd_seq = random();
 239         INIT_LIST_HEAD ( &tcp->queue );
 240         memcpy ( &tcp->peer, st_peer, sizeof ( tcp->peer ) );
 241
 242         /* Bind to local port */
 243         bind_port = ( st_local ? ntohs ( st_local->st_port ) : 0 );
 244         if ( ( rc = tcp_bind ( tcp, bind_port ) ) != 0 )
 245                 goto err;
 246
 247         /* Start timer to initiate SYN */
 248         start_timer_nodelay ( &tcp->timer );
 249
 250         /* Attach parent interface, transfer reference to connection
 251          * list and return
 252          */
 253         xfer_plug_plug ( &tcp->xfer, xfer );
 254         list_add ( &tcp->list, &tcp_conns );
 255         return 0;
 256
 257  err:
 258         ref_put ( &tcp->refcnt );
 259         return rc;
 260 }
 261
 262 /**
 263  * Close TCP connection
 264  *
 265  * @v tcp               TCP connection
 266  * @v rc                Reason for close
 267  *
 268  * Closes the data transfer interface.  If the TCP state machine is in
 269  * a suitable state, the connection will be deleted.
 270  */
 271 static void tcp_close ( struct tcp_connection *tcp, int rc ) {
 272         struct io_buffer *iobuf;
 273         struct io_buffer *tmp;
 274
 275         /* Close data transfer interface */
 276         xfer_nullify ( &tcp->xfer );
 277         xfer_close ( &tcp->xfer, rc );
 278         tcp->xfer_closed = 1;
 279
 280         /* If we are in CLOSED, or have otherwise not yet received a
 281          * SYN (i.e. we are in LISTEN or SYN_SENT), just delete the
 282          * connection.
 283          */
 284         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 285
 286                 /* Transition to CLOSED for the sake of debugging messages */
 287                 tcp->tcp_state = TCP_CLOSED;
 288                 tcp_dump_state ( tcp );
 289
 290                 /* Free any unsent I/O buffers */
 291                 list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 292                         list_del ( &iobuf->list );
 293                         free_iob ( iobuf );
 294                 }
 295
 296                 /* Remove from list and drop reference */
 297                 stop_timer ( &tcp->timer );
 298                 list_del ( &tcp->list );
 299                 ref_put ( &tcp->refcnt );
 300                 DBGC ( tcp, "TCP %p connection deleted\n", tcp );
 301                 return;
 302         }
 303
 304         /* If we have not had our SYN acknowledged (i.e. we are in
 305          * SYN_RCVD), pretend that it has been acknowledged so that we
 306          * can send a FIN without breaking things.
 307          */
 308         if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 309                 tcp_rx_ack ( tcp, ( tcp->snd_seq + 1 ), 0 );
 310
 311         /* If we have no data remaining to send, start sending FIN */
 312         if ( list_empty ( &tcp->queue ) ) {
 313                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 314                 tcp_dump_state ( tcp );
 315         }
 316 }
 317
 318 /***************************************************************************
 319  *
 320  * Transmit data path
 321  *
 322  ***************************************************************************
 323  */
 324
 325 /**
 326  * Calculate transmission window
 327  *
 328  * @v tcp               TCP connection
 329  * @ret len             Maximum length that can be sent in a single packet
 330  */
 331 static size_t tcp_xmit_win ( struct tcp_connection *tcp ) {
 332         size_t len;
 333
 334         /* Not ready if we're not in a suitable connection state */
 335         if ( ! TCP_CAN_SEND_DATA ( tcp->tcp_state ) )
 336                 return 0;
 337
 338         /* Length is the minimum of the receiver's window and the path MTU */
 339         len = tcp->snd_win;
 340         if ( len > TCP_PATH_MTU )
 341                 len = TCP_PATH_MTU;
 342
 343         return len;
 344 }
 345
 346 /**
 347  * Process TCP transmit queue
 348  *
 349  * @v tcp               TCP connection
 350  * @v max_len           Maximum length to process
 351  * @v dest              I/O buffer to fill with data, or NULL
 352  * @v remove            Remove data from queue
 353  * @ret len             Length of data processed
 354  *
 355  * This processes at most @c max_len bytes from the TCP connection's
 356  * transmit queue.  Data will be copied into the @c dest I/O buffer
 357  * (if provided) and, if @c remove is true, removed from the transmit
 358  * queue.
 359  */
 360 static size_t tcp_process_queue ( struct tcp_connection *tcp, size_t max_len,
 361                                   struct io_buffer *dest, int remove ) {
 362         struct io_buffer *iobuf;
 363         struct io_buffer *tmp;
 364         size_t frag_len;
 365         size_t len = 0;
 366
 367         list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 368                 frag_len = iob_len ( iobuf );
 369                 if ( frag_len > max_len )
 370                         frag_len = max_len;
 371                 if ( dest ) {
 372                         memcpy ( iob_put ( dest, frag_len ), iobuf->data,
 373                                  frag_len );
 374                 }
 375                 if ( remove ) {
 376                         iob_pull ( iobuf, frag_len );
 377                         if ( ! iob_len ( iobuf ) ) {
 378                                 list_del ( &iobuf->list );
 379                                 free_iob ( iobuf );
 380                         }
 381                 }
 382                 len += frag_len;
 383                 max_len -= frag_len;
 384         }
 385         return len;
 386 }
 387
 388 /**
 389  * Transmit any outstanding data
 390  *
 391  * @v tcp               TCP connection
 392  * @v force_send        Force sending of packet
 393  *
 394  * Transmits any outstanding data on the connection.
 395  *
 396  * Note that even if an error is returned, the retransmission timer
 397  * will have been started if necessary, and so the stack will
 398  * eventually attempt to retransmit the failed packet.
 399  */
 400 static int tcp_xmit ( struct tcp_connection *tcp, int force_send ) {
 401         struct io_buffer *iobuf;
 402         struct tcp_header *tcphdr;
 403         struct tcp_mss_option *mssopt;
 404         struct tcp_timestamp_padded_option *tsopt;
 405         void *payload;
 406         unsigned int flags;
 407         size_t len = 0;
 408         uint32_t seq_len;
 409         uint32_t app_win;
 410         uint32_t max_rcv_win;
 411         int rc;
 412
 413         /* If retransmission timer is already running, do nothing */
 414         if ( timer_running ( &tcp->timer ) )
 415                 return 0;
 416
 417         /* Calculate both the actual (payload) and sequence space
 418          * lengths that we wish to transmit.
 419          */
 420         if ( TCP_CAN_SEND_DATA ( tcp->tcp_state ) ) {
 421                 len = tcp_process_queue ( tcp, tcp_xmit_win ( tcp ),
 422                                           NULL, 0 );
 423         }
 424         seq_len = len;
 425         flags = TCP_FLAGS_SENDING ( tcp->tcp_state );
 426         if ( flags & ( TCP_SYN | TCP_FIN ) ) {
 427                 /* SYN or FIN consume one byte, and we can never send both */
 428                 assert ( ! ( ( flags & TCP_SYN ) && ( flags & TCP_FIN ) ) );
 429                 seq_len++;
 430         }
 431         tcp->snd_sent = seq_len;
 432
 433         /* If we have nothing to transmit, stop now */
 434         if ( ( seq_len == 0 ) && ! force_send )
 435                 return 0;
 436
 437         /* If we are transmitting anything that requires
 438          * acknowledgement (i.e. consumes sequence space), start the
 439          * retransmission timer.  Do this before attempting to
 440          * allocate the I/O buffer, in case allocation itself fails.
 441          */
 442         if ( seq_len )
 443                 start_timer ( &tcp->timer );
 444
 445         /* Allocate I/O buffer */
 446         iobuf = alloc_iob ( len + MAX_HDR_LEN );
 447         if ( ! iobuf ) {
 448                 DBGC ( tcp, "TCP %p could not allocate iobuf for %08x..%08x "
 449                        "%08x\n", tcp, tcp->snd_seq, ( tcp->snd_seq + seq_len ),
 450                        tcp->rcv_ack );
 451                 return -ENOMEM;
 452         }
 453         iob_reserve ( iobuf, MAX_HDR_LEN );
 454
 455         /* Fill data payload from transmit queue */
 456         tcp_process_queue ( tcp, len, iobuf, 0 );
 457
 458         /* Expand receive window if possible */
 459         max_rcv_win = ( ( freemem * 3 ) / 4 );
 460         if ( max_rcv_win > TCP_MAX_WINDOW_SIZE )
 461                 max_rcv_win = TCP_MAX_WINDOW_SIZE;
 462         app_win = xfer_window ( &tcp->xfer );
 463         if ( max_rcv_win > app_win )
 464                 max_rcv_win = app_win;
 465         max_rcv_win &= ~0x03; /* Keep everything dword-aligned */
 466         if ( tcp->rcv_win < max_rcv_win )
 467                 tcp->rcv_win = max_rcv_win;
 468
 469         /* Fill up the TCP header */
 470         payload = iobuf->data;
 471         if ( flags & TCP_SYN ) {
 472                 mssopt = iob_push ( iobuf, sizeof ( *mssopt ) );
 473                 mssopt->kind = TCP_OPTION_MSS;
 474                 mssopt->length = sizeof ( *mssopt );
 475                 mssopt->mss = htons ( TCP_MSS );
 476         }
 477         if ( ( flags & TCP_SYN ) || tcp->timestamps ) {
 478                 tsopt = iob_push ( iobuf, sizeof ( *tsopt ) );
 479                 memset ( tsopt->nop, TCP_OPTION_NOP, sizeof ( tsopt->nop ) );
 480                 tsopt->tsopt.kind = TCP_OPTION_TS;
 481                 tsopt->tsopt.length = sizeof ( tsopt->tsopt );
 482                 tsopt->tsopt.tsval = htonl ( currticks() );
 483                 tsopt->tsopt.tsecr = htonl ( tcp->ts_recent );
 484         }
 485         if ( ! ( flags & TCP_SYN ) )
 486                 flags |= TCP_PSH;
 487         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 488         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 489         tcphdr->src = htons ( tcp->local_port );
 490         tcphdr->dest = tcp->peer.st_port;
 491         tcphdr->seq = htonl ( tcp->snd_seq );
 492         tcphdr->ack = htonl ( tcp->rcv_ack );
 493         tcphdr->hlen = ( ( payload - iobuf->data ) << 2 );
 494         tcphdr->flags = flags;
 495         tcphdr->win = htons ( tcp->rcv_win );
 496         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 497
 498         /* Dump header */
 499         DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4zd",
 500                 tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 501                 ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) + seq_len ),
 502                 ntohl ( tcphdr->ack ), len );
 503         tcp_dump_flags ( tcp, tcphdr->flags );
 504         DBGC2 ( tcp, "\n" );
 505
 506         /* Transmit packet */
 507         if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, &tcp->peer, NULL,
 508                                &tcphdr->csum ) ) != 0 ) {
 509                 DBGC ( tcp, "TCP %p could not transmit %08x..%08x %08x: %s\n",
 510                        tcp, tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ),
 511                        tcp->rcv_ack, strerror ( rc ) );
 512                 return rc;
 513         }
 514
 515         return 0;
 516 }
 517
 518 /**
 519  * Retransmission timer expired
 520  *
 521  * @v timer             Retransmission timer
 522  * @v over              Failure indicator
 523  */
 524 static void tcp_expired ( struct retry_timer *timer, int over ) {
 525         struct tcp_connection *tcp =
 526                 container_of ( timer, struct tcp_connection, timer );
 527
 528         DBGC ( tcp, "TCP %p timer %s in %s for %08x..%08x %08x\n", tcp,
 529                ( over ? "expired" : "fired" ), tcp_state ( tcp->tcp_state ),
 530                tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
 531
 532         assert ( ( tcp->tcp_state == TCP_SYN_SENT ) ||
 533                  ( tcp->tcp_state == TCP_SYN_RCVD ) ||
 534                  ( tcp->tcp_state == TCP_ESTABLISHED ) ||
 535                  ( tcp->tcp_state == TCP_FIN_WAIT_1 ) ||
 536                  ( tcp->tcp_state == TCP_CLOSE_WAIT ) ||
 537                  ( tcp->tcp_state == TCP_CLOSING_OR_LAST_ACK ) );
 538
 539         if ( over ) {
 540                 /* If we have finally timed out and given up,
 541                  * terminate the connection
 542                  */
 543                 tcp->tcp_state = TCP_CLOSED;
 544                 tcp_dump_state ( tcp );
 545                 tcp_close ( tcp, -ETIMEDOUT );
 546         } else {
 547                 /* Otherwise, retransmit the packet */
 548                 tcp_xmit ( tcp, 0 );
 549         }
 550 }
 551
 552 /**
 553  * Shutdown timer expired
 554  *
 555  * @v timer             Shutdown timer
 556  * @v over              Failure indicator
 557  */
 558 static void tcp_wait_expired ( struct retry_timer *timer, int over __unused ) {
 559         struct tcp_connection *tcp =
 560                 container_of ( timer, struct tcp_connection, wait );
 561
 562         assert ( tcp->tcp_state == TCP_TIME_WAIT );
 563
 564         DBGC ( tcp, "TCP %p wait complete in %s for %08x..%08x %08x\n", tcp,
 565                tcp_state ( tcp->tcp_state ), tcp->snd_seq,
 566                ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
 567
 568         tcp->tcp_state = TCP_CLOSED;
 569         tcp_dump_state ( tcp );
 570         tcp_close ( tcp, 0 );
 571 }
 572
 573 /**
 574  * Send RST response to incoming packet
 575  *
 576  * @v in_tcphdr         TCP header of incoming packet
 577  * @ret rc              Return status code
 578  */
 579 static int tcp_xmit_reset ( struct tcp_connection *tcp,
 580                             struct sockaddr_tcpip *st_dest,
 581                             struct tcp_header *in_tcphdr ) {
 582         struct io_buffer *iobuf;
 583         struct tcp_header *tcphdr;
 584         int rc;
 585
 586         /* Allocate space for dataless TX buffer */
 587         iobuf = alloc_iob ( MAX_HDR_LEN );
 588         if ( ! iobuf ) {
 589                 DBGC ( tcp, "TCP %p could not allocate iobuf for RST "
 590                        "%08x..%08x %08x\n", tcp, ntohl ( in_tcphdr->ack ),
 591                        ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ) );
 592                 return -ENOMEM;
 593         }
 594         iob_reserve ( iobuf, MAX_HDR_LEN );
 595
 596         /* Construct RST response */
 597         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 598         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 599         tcphdr->src = in_tcphdr->dest;
 600         tcphdr->dest = in_tcphdr->src;
 601         tcphdr->seq = in_tcphdr->ack;
 602         tcphdr->ack = in_tcphdr->seq;
 603         tcphdr->hlen = ( ( sizeof ( *tcphdr ) / 4 ) << 4 );
 604         tcphdr->flags = ( TCP_RST | TCP_ACK );
 605         tcphdr->win = htons ( TCP_MAX_WINDOW_SIZE );
 606         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 607
 608         /* Dump header */
 609         DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4d",
 610                 tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 611                 ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) ),
 612                 ntohl ( tcphdr->ack ), 0 );
 613         tcp_dump_flags ( tcp, tcphdr->flags );
 614         DBGC2 ( tcp, "\n" );
 615
 616         /* Transmit packet */
 617         if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, st_dest,
 618                                NULL, &tcphdr->csum ) ) != 0 ) {
 619                 DBGC ( tcp, "TCP %p could not transmit RST %08x..%08x %08x: "
 620                        "%s\n", tcp, ntohl ( in_tcphdr->ack ),
 621                        ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ),
 622                        strerror ( rc ) );
 623                 return rc;
 624         }
 625
 626         return 0;
 627 }
 628
 629 /***************************************************************************
 630  *
 631  * Receive data path
 632  *
 633  ***************************************************************************
 634  */
 635
 636 /**
 637  * Identify TCP connection by local port number
 638  *
 639  * @v local_port        Local port
 640  * @ret tcp             TCP connection, or NULL
 641  */
 642 static struct tcp_connection * tcp_demux ( unsigned int local_port ) {
 643         struct tcp_connection *tcp;
 644
 645         list_for_each_entry ( tcp, &tcp_conns, list ) {
 646                 if ( tcp->local_port == local_port )
 647                         return tcp;
 648         }
 649         return NULL;
 650 }
 651
 652 /**
 653  * Parse TCP received options
 654  *
 655  * @v tcp               TCP connection
 656  * @v data              Raw options data
 657  * @v len               Raw options length
 658  * @v options           Options structure to fill in
 659  */
 660 static void tcp_rx_opts ( struct tcp_connection *tcp, const void *data,
 661                           size_t len, struct tcp_options *options ) {
 662         const void *end = ( data + len );
 663         const struct tcp_option *option;
 664         unsigned int kind;
 665
 666         memset ( options, 0, sizeof ( *options ) );
 667         while ( data < end ) {
 668                 option = data;
 669                 kind = option->kind;
 670                 if ( kind == TCP_OPTION_END )
 671                         return;
 672                 if ( kind == TCP_OPTION_NOP ) {
 673                         data++;
 674                         continue;
 675                 }
 676                 switch ( kind ) {
 677                 case TCP_OPTION_MSS:
 678                         options->mssopt = data;
 679                         break;
 680                 case TCP_OPTION_TS:
 681                         options->tsopt = data;
 682                         break;
 683                 default:
 684                         DBGC ( tcp, "TCP %p received unknown option %d\n",
 685                                tcp, kind );
 686                         break;
 687                 }
 688                 data += option->length;
 689         }
 690 }
 691
 692 /**
 693  * Consume received sequence space
 694  *
 695  * @v tcp               TCP connection
 696  * @v seq_len           Sequence space length to consume
 697  */
 698 static void tcp_rx_seq ( struct tcp_connection *tcp, uint32_t seq_len ) {
 699         tcp->rcv_ack += seq_len;
 700         if ( tcp->rcv_win > seq_len ) {
 701                 tcp->rcv_win -= seq_len;
 702         } else {
 703                 tcp->rcv_win = 0;
 704         }
 705 }
 706
 707 /**
 708  * Handle TCP received SYN
 709  *
 710  * @v tcp               TCP connection
 711  * @v seq               SEQ value (in host-endian order)
 712  * @v options           TCP options
 713  * @ret rc              Return status code
 714  */
 715 static int tcp_rx_syn ( struct tcp_connection *tcp, uint32_t seq,
 716                         struct tcp_options *options ) {
 717
 718         /* Synchronise sequence numbers on first SYN */
 719         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 720                 tcp->rcv_ack = seq;
 721                 if ( options->tsopt )
 722                         tcp->timestamps = 1;
 723         }
 724
 725         /* Ignore duplicate SYN */
 726         if ( ( tcp->rcv_ack - seq ) > 0 )
 727                 return 0;
 728
 729         /* Acknowledge SYN */
 730         tcp_rx_seq ( tcp, 1 );
 731
 732         /* Mark SYN as received and start sending ACKs with each packet */
 733         tcp->tcp_state |= ( TCP_STATE_SENT ( TCP_ACK ) |
 734                             TCP_STATE_RCVD ( TCP_SYN ) );
 735
 736         return 0;
 737 }
 738
 739 /**
 740  * Handle TCP received ACK
 741  *
 742  * @v tcp               TCP connection
 743  * @v ack               ACK value (in host-endian order)
 744  * @v win               WIN value (in host-endian order)
 745  * @ret rc              Return status code
 746  */
 747 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
 748                         uint32_t win ) {
 749         uint32_t ack_len = ( ack - tcp->snd_seq );
 750         size_t len;
 751         unsigned int acked_flags;
 752
 753         /* Check for out-of-range or old duplicate ACKs */
 754         if ( ack_len > tcp->snd_sent ) {
 755                 DBGC ( tcp, "TCP %p received ACK for %08x..%08x, "
 756                        "sent only %08x..%08x\n", tcp, tcp->snd_seq,
 757                        ( tcp->snd_seq + ack_len ), tcp->snd_seq,
 758                        ( tcp->snd_seq + tcp->snd_sent ) );
 759
 760                 if ( TCP_HAS_BEEN_ESTABLISHED ( tcp->tcp_state ) ) {
 761                         /* Just ignore what might be old duplicate ACKs */
 762                         return 0;
 763                 } else {
 764                         /* Send RST if an out-of-range ACK is received
 765                          * on a not-yet-established connection, as per
 766                          * RFC 793.
 767                          */
 768                         return -EINVAL;
 769                 }
 770         }
 771
 772         /* Ignore ACKs that don't actually acknowledge any new data.
 773          * (In particular, do not stop the retransmission timer; this
 774          * avoids creating a sorceror's apprentice syndrome when a
 775          * duplicate ACK is received and we still have data in our
 776          * transmit queue.)
 777          */
 778         if ( ack_len == 0 )
 779                 return 0;
 780
 781         /* Stop the retransmission timer */
 782         stop_timer ( &tcp->timer );
 783
 784         /* Determine acknowledged flags and data length */
 785         len = ack_len;
 786         acked_flags = ( TCP_FLAGS_SENDING ( tcp->tcp_state ) &
 787                         ( TCP_SYN | TCP_FIN ) );
 788         if ( acked_flags )
 789                 len--;
 790
 791         /* Update SEQ and sent counters, and window size */
 792         tcp->snd_seq = ack;
 793         tcp->snd_sent = 0;
 794         tcp->snd_win = win;
 795
 796         /* Remove any acknowledged data from transmit queue */
 797         tcp_process_queue ( tcp, len, NULL, 1 );
 798
 799         /* Mark SYN/FIN as acknowledged if applicable. */
 800         if ( acked_flags )
 801                 tcp->tcp_state |= TCP_STATE_ACKED ( acked_flags );
 802
 803         /* Start sending FIN if we've had all possible data ACKed */
 804         if ( list_empty ( &tcp->queue ) && tcp->xfer_closed )
 805                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 806
 807         return 0;
 808 }
 809
 810 /**
 811  * Handle TCP received data
 812  *
 813  * @v tcp               TCP connection
 814  * @v seq               SEQ value (in host-endian order)
 815  * @v iobuf             I/O buffer
 816  * @ret rc              Return status code
 817  *
 818  * This function takes ownership of the I/O buffer.
 819  */
 820 static int tcp_rx_data ( struct tcp_connection *tcp, uint32_t seq,
 821                          struct io_buffer *iobuf ) {
 822         uint32_t already_rcvd;
 823         uint32_t len;
 824         int rc;
 825
 826         /* Ignore duplicate or out-of-order data */
 827         already_rcvd = ( tcp->rcv_ack - seq );
 828         len = iob_len ( iobuf );
 829         if ( already_rcvd >= len ) {
 830                 free_iob ( iobuf );
 831                 return 0;
 832         }
 833         iob_pull ( iobuf, already_rcvd );
 834         len -= already_rcvd;
 835
 836         /* Acknowledge new data */
 837         tcp_rx_seq ( tcp, len );
 838
 839         /* Deliver data to application */
 840         if ( ( rc = xfer_deliver_iob ( &tcp->xfer, iobuf ) ) != 0 ) {
 841                 DBGC ( tcp, "TCP %p could not deliver %08x..%08x: %s\n",
 842                        tcp, seq, ( seq + len ), strerror ( rc ) );
 843                 return rc;
 844         }
 845
 846         return 0;
 847 }
 848
 849 /**
 850  * Handle TCP received FIN
 851  *
 852  * @v tcp               TCP connection
 853  * @v seq               SEQ value (in host-endian order)
 854  * @ret rc              Return status code
 855  */
 856 static int tcp_rx_fin ( struct tcp_connection *tcp, uint32_t seq ) {
 857
 858         /* Ignore duplicate or out-of-order FIN */
 859         if ( ( tcp->rcv_ack - seq ) > 0 )
 860                 return 0;
 861
 862         /* Acknowledge FIN */
 863         tcp_rx_seq ( tcp, 1 );
 864
 865         /* Mark FIN as received */
 866         tcp->tcp_state |= TCP_STATE_RCVD ( TCP_FIN );
 867
 868         /* Close connection */
 869         tcp_close ( tcp, 0 );
 870
 871         return 0;
 872 }
 873
 874 /**
 875  * Handle TCP received RST
 876  *
 877  * @v tcp               TCP connection
 878  * @v seq               SEQ value (in host-endian order)
 879  * @ret rc              Return status code
 880  */
 881 static int tcp_rx_rst ( struct tcp_connection *tcp, uint32_t seq ) {
 882
 883         /* Accept RST only if it falls within the window.  If we have
 884          * not yet received a SYN, then we have no window to test
 885          * against, so fall back to checking that our SYN has been
 886          * ACKed.
 887          */
 888         if ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) {
 889                 if ( ( seq - tcp->rcv_ack ) >= tcp->rcv_win )
 890                         return 0;
 891         } else {
 892                 if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 893                         return 0;
 894         }
 895
 896         /* Abort connection */
 897         tcp->tcp_state = TCP_CLOSED;
 898         tcp_dump_state ( tcp );
 899         tcp_close ( tcp, -ECONNRESET );
 900
 901         DBGC ( tcp, "TCP %p connection reset by peer\n", tcp );
 902         return -ECONNRESET;
 903 }
 904
 905 /**
 906  * Process received packet
 907  *
 908  * @v iobuf             I/O buffer
 909  * @v st_src            Partially-filled source address
 910  * @v st_dest           Partially-filled destination address
 911  * @v pshdr_csum        Pseudo-header checksum
 912  * @ret rc              Return status code
 913   */
 914 static int tcp_rx ( struct io_buffer *iobuf,
 915                     struct sockaddr_tcpip *st_src,
 916                     struct sockaddr_tcpip *st_dest __unused,
 917                     uint16_t pshdr_csum ) {
 918         struct tcp_header *tcphdr = iobuf->data;
 919         struct tcp_connection *tcp;
 920         struct tcp_options options;
 921         size_t hlen;
 922         uint16_t csum;
 923         uint32_t start_seq;
 924         uint32_t seq;
 925         uint32_t ack;
 926         uint32_t win;
 927         uint32_t ts_recent;
 928         unsigned int flags;
 929         size_t len;
 930         int rc;
 931
 932         /* Sanity check packet */
 933         if ( iob_len ( iobuf ) < sizeof ( *tcphdr ) ) {
 934                 DBG ( "TCP packet too short at %zd bytes (min %zd bytes)\n",
 935                       iob_len ( iobuf ), sizeof ( *tcphdr ) );
 936                 rc = -EINVAL;
 937                 goto discard;
 938         }
 939         hlen = ( ( tcphdr->hlen & TCP_MASK_HLEN ) / 16 ) * 4;
 940         if ( hlen < sizeof ( *tcphdr ) ) {
 941                 DBG ( "TCP header too short at %zd bytes (min %zd bytes)\n",
 942                       hlen, sizeof ( *tcphdr ) );
 943                 rc = -EINVAL;
 944                 goto discard;
 945         }
 946         if ( hlen > iob_len ( iobuf ) ) {
 947                 DBG ( "TCP header too long at %zd bytes (max %zd bytes)\n",
 948                       hlen, iob_len ( iobuf ) );
 949                 rc = -EINVAL;
 950                 goto discard;
 951         }
 952         csum = tcpip_continue_chksum ( pshdr_csum, iobuf->data,
 953                                        iob_len ( iobuf ) );
 954         if ( csum != 0 ) {
 955                 DBG ( "TCP checksum incorrect (is %04x including checksum "
 956                       "field, should be 0000)\n", csum );
 957                 rc = -EINVAL;
 958                 goto discard;
 959         }
 960
 961         /* Parse parameters from header and strip header */
 962         tcp = tcp_demux ( ntohs ( tcphdr->dest ) );
 963         start_seq = seq = ntohl ( tcphdr->seq );
 964         ack = ntohl ( tcphdr->ack );
 965         win = ntohs ( tcphdr->win );
 966         flags = tcphdr->flags;
 967         tcp_rx_opts ( tcp, ( ( ( void * ) tcphdr ) + sizeof ( *tcphdr ) ),
 968                       ( hlen - sizeof ( *tcphdr ) ), &options );
 969         ts_recent = ( options.tsopt ?
 970                       ntohl ( options.tsopt->tsval ) : tcp->ts_recent );
 971         iob_pull ( iobuf, hlen );
 972         len = iob_len ( iobuf );
 973
 974         /* Dump header */
 975         DBGC2 ( tcp, "TCP %p RX %d<-%d           %08x %08x..%08zx %4zd",
 976                 tcp, ntohs ( tcphdr->dest ), ntohs ( tcphdr->src ),
 977                 ntohl ( tcphdr->ack ), ntohl ( tcphdr->seq ),
 978                 ( ntohl ( tcphdr->seq ) + len +
 979                   ( ( tcphdr->flags & ( TCP_SYN | TCP_FIN ) ) ? 1 : 0 )), len);
 980         tcp_dump_flags ( tcp, tcphdr->flags );
 981         DBGC2 ( tcp, "\n" );
 982
 983         /* If no connection was found, send RST */
 984         if ( ! tcp ) {
 985                 tcp_xmit_reset ( tcp, st_src, tcphdr );
 986                 rc = -ENOTCONN;
 987                 goto discard;
 988         }
 989
 990         /* Handle ACK, if present */
 991         if ( flags & TCP_ACK ) {
 992                 if ( ( rc = tcp_rx_ack ( tcp, ack, win ) ) != 0 ) {
 993                         tcp_xmit_reset ( tcp, st_src, tcphdr );
 994                         goto discard;
 995                 }
 996         }
 997
 998         /* Handle SYN, if present */
 999         if ( flags & TCP_SYN ) {
1000                 tcp_rx_syn ( tcp, seq, &options );
1001                 seq++;
1002         }
1003
1004         /* Handle RST, if present */
1005         if ( flags & TCP_RST ) {
1006                 if ( ( rc = tcp_rx_rst ( tcp, seq ) ) != 0 )
1007                         goto discard;
1008         }
1009
1010         /* Handle new data, if any */
1011         tcp_rx_data ( tcp, seq, iob_disown ( iobuf ) );
1012         seq += len;
1013
1014         /* Handle FIN, if present */
1015         if ( flags & TCP_FIN ) {
1016                 tcp_rx_fin ( tcp, seq );
1017                 seq++;
1018         }
1019
1020         /* Update timestamp, if applicable */
1021         if ( seq == tcp->rcv_ack )
1022                 tcp->ts_recent = ts_recent;
1023
1024         /* Dump out any state change as a result of the received packet */
1025         tcp_dump_state ( tcp );
1026
1027         /* Send out any pending data.  We force sending a reply if either
1028          *
1029          *  a) the peer is expecting an ACK (i.e. consumed sequence space), or
1030          *  b) either end of the packet was outside the receive window
1031          *
1032          * Case (b) enables us to support TCP keepalives using
1033          * zero-length packets, which we would otherwise ignore.  Note
1034          * that for case (b), we need *only* consider zero-length
1035          * packets, since non-zero-length packets will already be
1036          * caught by case (a).
1037          */
1038         tcp_xmit ( tcp, ( ( start_seq != seq ) ||
1039                           ( ( seq - tcp->rcv_ack ) > tcp->rcv_win ) ) );
1040
1041         /* If this packet was the last we expect to receive, set up
1042          * timer to expire and cause the connection to be freed.
1043          */
1044         if ( TCP_CLOSED_GRACEFULLY ( tcp->tcp_state ) ) {
1045                 stop_timer ( &tcp->wait );
1046                 start_timer_fixed ( &tcp->wait, ( 2 * TCP_MSL ) );
1047         }
1048
1049         return 0;
1050
1051  discard:
1052         /* Free received packet */
1053         free_iob ( iobuf );
1054         return rc;
1055 }
1056
1057 /** TCP protocol */
1058 struct tcpip_protocol tcp_protocol __tcpip_protocol = {
1059         .name = "TCP",
1060         .rx = tcp_rx,
1061         .tcpip_proto = IP_TCP,
1062 };
1063
1064 /***************************************************************************
1065  *
1066  * Data transfer interface
1067  *
1068  ***************************************************************************
1069  */
1070
1071 /**
1072  * Close interface
1073  *
1074  * @v xfer              Data transfer interface
1075  * @v rc                Reason for close
1076  */
1077 static void tcp_xfer_close ( struct xfer_interface *xfer, int rc ) {
1078         struct tcp_connection *tcp =
1079                 container_of ( xfer, struct tcp_connection, xfer );
1080
1081         /* Close data transfer interface */
1082         tcp_close ( tcp, rc );
1083
1084         /* Transmit FIN, if possible */
1085         tcp_xmit ( tcp, 0 );
1086 }
1087
1088 /**
1089  * Check flow control window
1090  *
1091  * @v xfer              Data transfer interface
1092  * @ret len             Length of window
1093  */
1094 static size_t tcp_xfer_window ( struct xfer_interface *xfer ) {
1095         struct tcp_connection *tcp =
1096                 container_of ( xfer, struct tcp_connection, xfer );
1097
1098         /* Not ready if data queue is non-empty.  This imposes a limit
1099          * of only one unACKed packet in the TX queue at any time; we
1100          * do this to conserve memory usage.
1101          */
1102         if ( ! list_empty ( &tcp->queue ) )
1103                 return 0;
1104
1105         /* Return TCP window length */
1106         return tcp_xmit_win ( tcp );
1107 }
1108
1109 /**
1110  * Deliver datagram as I/O buffer
1111  *
1112  * @v xfer              Data transfer interface
1113  * @v iobuf             Datagram I/O buffer
1114  * @v meta              Data transfer metadata
1115  * @ret rc              Return status code
1116  */
1117 static int tcp_xfer_deliver_iob ( struct xfer_interface *xfer,
1118                                   struct io_buffer *iobuf,
1119                                   struct xfer_metadata *meta __unused ) {
1120         struct tcp_connection *tcp =
1121                 container_of ( xfer, struct tcp_connection, xfer );
1122
1123         /* Enqueue packet */
1124         list_add_tail ( &iobuf->list, &tcp->queue );
1125
1126         /* Transmit data, if possible */
1127         tcp_xmit ( tcp, 0 );
1128
1129         return 0;
1130 }
1131
1132 /** TCP data transfer interface operations */
1133 static struct xfer_interface_operations tcp_xfer_operations = {
1134         .close          = tcp_xfer_close,
1135         .vredirect      = ignore_xfer_vredirect,
1136         .window         = tcp_xfer_window,
1137         .alloc_iob      = default_xfer_alloc_iob,
1138         .deliver_iob    = tcp_xfer_deliver_iob,
1139         .deliver_raw    = xfer_deliver_as_iob,
1140 };
1141
1142 /***************************************************************************
1143  *
1144  * Openers
1145  *
1146  ***************************************************************************
1147  */
1148
1149 /** TCP socket opener */
1150 struct socket_opener tcp_socket_opener __socket_opener = {
1151         .semantics      = TCP_SOCK_STREAM,
1152         .family         = AF_INET,
1153         .open           = tcp_open,
1154 };
1155
1156 /** Linkage hack */
1157 int tcp_sock_stream = TCP_SOCK_STREAM;
1158
1159 /**
1160  * Open TCP URI
1161  *
1162  * @v xfer              Data transfer interface
1163  * @v uri               URI
1164  * @ret rc              Return status code
1165  */
1166 static int tcp_open_uri ( struct xfer_interface *xfer, struct uri *uri ) {
1167         struct sockaddr_tcpip peer;
1168
1169         /* Sanity check */
1170         if ( ! uri->host )
1171                 return -EINVAL;
1172
1173         memset ( &peer, 0, sizeof ( peer ) );
1174         peer.st_port = htons ( uri_port ( uri, 0 ) );
1175         return xfer_open_named_socket ( xfer, SOCK_STREAM,
1176                                         ( struct sockaddr * ) &peer,
1177                                         uri->host, NULL );
1178 }
1179
1180 /** TCP URI opener */
1181 struct uri_opener tcp_uri_opener __uri_opener = {
1182         .scheme         = "tcp",
1183         .open           = tcp_open_uri,
1184 };
1185