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         /** Flags */
  34         unsigned int flags;
  35
  36         /** Data transfer interface */
  37         struct xfer_interface xfer;
  38
  39         /** Remote socket address */
  40         struct sockaddr_tcpip peer;
  41         /** Local port */
  42         unsigned int local_port;
  43
  44         /** Current TCP state */
  45         unsigned int tcp_state;
  46         /** Previous TCP state
  47          *
  48          * Maintained only for debug messages
  49          */
  50         unsigned int prev_tcp_state;
  51         /** Current sequence number
  52          *
  53          * Equivalent to SND.UNA in RFC 793 terminology.
  54          */
  55         uint32_t snd_seq;
  56         /** Unacknowledged sequence count
  57          *
  58          * Equivalent to (SND.NXT-SND.UNA) in RFC 793 terminology.
  59          */
  60         uint32_t snd_sent;
  61         /** Send window
  62          *
  63          * Equivalent to SND.WND in RFC 793 terminology
  64          */
  65         uint32_t snd_win;
  66         /** Current acknowledgement number
  67          *
  68          * Equivalent to RCV.NXT in RFC 793 terminology.
  69          */
  70         uint32_t rcv_ack;
  71         /** Receive window
  72          *
  73          * Equivalent to RCV.WND in RFC 793 terminology.
  74          */
  75         uint32_t rcv_win;
  76         /** Most recent received timestamp
  77          *
  78          * Equivalent to TS.Recent in RFC 1323 terminology.
  79          */
  80         uint32_t ts_recent;
  81
  82         /** Transmit queue */
  83         struct list_head queue;
  84         /** Retransmission timer */
  85         struct retry_timer timer;
  86         /** Shutdown (TIME_WAIT) timer */
  87         struct retry_timer wait;
  88 };
  89
  90 /** TCP flags */
  91 enum tcp_flags {
  92         /** TCP data transfer interface has been closed */
  93         TCP_XFER_CLOSED = 0x0001,
  94         /** TCP timestamps are enabled */
  95         TCP_TS_ENABLED = 0x0002,
  96         /** TCP acknowledgement is pending */
  97         TCP_ACK_PENDING = 0x0004,
  98 };
  99
 100 /**
 101  * List of registered TCP connections
 102  */
 103 static LIST_HEAD ( tcp_conns );
 104
 105 /* Forward declarations */
 106 static struct xfer_interface_operations tcp_xfer_operations;
 107 static void tcp_expired ( struct retry_timer *timer, int over );
 108 static void tcp_wait_expired ( struct retry_timer *timer, int over );
 109 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
 110                         uint32_t win );
 111
 112 /**
 113  * Name TCP state
 114  *
 115  * @v state             TCP state
 116  * @ret name            Name of TCP state
 117  */
 118 static inline __attribute__ (( always_inline )) const char *
 119 tcp_state ( int state ) {
 120         switch ( state ) {
 121         case TCP_CLOSED:                return "CLOSED";
 122         case TCP_LISTEN:                return "LISTEN";
 123         case TCP_SYN_SENT:              return "SYN_SENT";
 124         case TCP_SYN_RCVD:              return "SYN_RCVD";
 125         case TCP_ESTABLISHED:           return "ESTABLISHED";
 126         case TCP_FIN_WAIT_1:            return "FIN_WAIT_1";
 127         case TCP_FIN_WAIT_2:            return "FIN_WAIT_2";
 128         case TCP_CLOSING_OR_LAST_ACK:   return "CLOSING/LAST_ACK";
 129         case TCP_TIME_WAIT:             return "TIME_WAIT";
 130         case TCP_CLOSE_WAIT:            return "CLOSE_WAIT";
 131         default:                        return "INVALID";
 132         }
 133 }
 134
 135 /**
 136  * Dump TCP state transition
 137  *
 138  * @v tcp               TCP connection
 139  */
 140 static inline __attribute__ (( always_inline )) void
 141 tcp_dump_state ( struct tcp_connection *tcp ) {
 142
 143         if ( tcp->tcp_state != tcp->prev_tcp_state ) {
 144                 DBGC ( tcp, "TCP %p transitioned from %s to %s\n", tcp,
 145                        tcp_state ( tcp->prev_tcp_state ),
 146                        tcp_state ( tcp->tcp_state ) );
 147         }
 148         tcp->prev_tcp_state = tcp->tcp_state;
 149 }
 150
 151 /**
 152  * Dump TCP flags
 153  *
 154  * @v flags             TCP flags
 155  */
 156 static inline __attribute__ (( always_inline )) void
 157 tcp_dump_flags ( struct tcp_connection *tcp, unsigned int flags ) {
 158         if ( flags & TCP_RST )
 159                 DBGC2 ( tcp, " RST" );
 160         if ( flags & TCP_SYN )
 161                 DBGC2 ( tcp, " SYN" );
 162         if ( flags & TCP_PSH )
 163                 DBGC2 ( tcp, " PSH" );
 164         if ( flags & TCP_FIN )
 165                 DBGC2 ( tcp, " FIN" );
 166         if ( flags & TCP_ACK )
 167                 DBGC2 ( tcp, " ACK" );
 168 }
 169
 170 /***************************************************************************
 171  *
 172  * Open and close
 173  *
 174  ***************************************************************************
 175  */
 176
 177 /**
 178  * Bind TCP connection to local port
 179  *
 180  * @v tcp               TCP connection
 181  * @v port              Local port number
 182  * @ret rc              Return status code
 183  *
 184  * If the port is 0, the connection is assigned an available port
 185  * between 1024 and 65535.
 186  */
 187 static int tcp_bind ( struct tcp_connection *tcp, unsigned int port ) {
 188         struct tcp_connection *existing;
 189         uint16_t try_port;
 190         int i;
 191
 192         /* If no port specified, find an available port */
 193         if ( ! port ) {
 194                 try_port = ( random() % 64512 ) + 1023;
 195                 for ( i = 0 ; i < 65536 ; ++i ) {
 196                         if ( ++try_port < 1024 )
 197                                 continue;
 198                         if ( tcp_bind ( tcp, try_port ) == 0 )
 199                                 return 0;
 200                 }
 201                 DBGC ( tcp, "TCP %p could not bind: no free ports\n", tcp );
 202                 return -EADDRINUSE;
 203         }
 204
 205         /* Attempt bind to local port */
 206         list_for_each_entry ( existing, &tcp_conns, list ) {
 207                 if ( existing->local_port == port ) {
 208                         DBGC ( tcp, "TCP %p could not bind: port %d in use\n",
 209                                tcp, port );
 210                         return -EADDRINUSE;
 211                 }
 212         }
 213         tcp->local_port = port;
 214
 215         DBGC ( tcp, "TCP %p bound to port %d\n", tcp, port );
 216         return 0;
 217 }
 218
 219 /**
 220  * Open a TCP connection
 221  *
 222  * @v xfer              Data transfer interface
 223  * @v peer              Peer socket address
 224  * @v local             Local socket address, or NULL
 225  * @ret rc              Return status code
 226  */
 227 static int tcp_open ( struct xfer_interface *xfer, struct sockaddr *peer,
 228                       struct sockaddr *local ) {
 229         struct sockaddr_tcpip *st_peer = ( struct sockaddr_tcpip * ) peer;
 230         struct sockaddr_tcpip *st_local = ( struct sockaddr_tcpip * ) local;
 231         struct tcp_connection *tcp;
 232         unsigned int bind_port;
 233         int rc;
 234
 235         /* Allocate and initialise structure */
 236         tcp = zalloc ( sizeof ( *tcp ) );
 237         if ( ! tcp )
 238                 return -ENOMEM;
 239         DBGC ( tcp, "TCP %p allocated\n", tcp );
 240         ref_init ( &tcp->refcnt, NULL );
 241         xfer_init ( &tcp->xfer, &tcp_xfer_operations, &tcp->refcnt );
 242         timer_init ( &tcp->timer, tcp_expired );
 243         timer_init ( &tcp->wait, tcp_wait_expired );
 244         tcp->prev_tcp_state = TCP_CLOSED;
 245         tcp->tcp_state = TCP_STATE_SENT ( TCP_SYN );
 246         tcp_dump_state ( tcp );
 247         tcp->snd_seq = random();
 248         INIT_LIST_HEAD ( &tcp->queue );
 249         memcpy ( &tcp->peer, st_peer, sizeof ( tcp->peer ) );
 250
 251         /* Bind to local port */
 252         bind_port = ( st_local ? ntohs ( st_local->st_port ) : 0 );
 253         if ( ( rc = tcp_bind ( tcp, bind_port ) ) != 0 )
 254                 goto err;
 255
 256         /* Start timer to initiate SYN */
 257         start_timer_nodelay ( &tcp->timer );
 258
 259         /* Attach parent interface, transfer reference to connection
 260          * list and return
 261          */
 262         xfer_plug_plug ( &tcp->xfer, xfer );
 263         list_add ( &tcp->list, &tcp_conns );
 264         return 0;
 265
 266  err:
 267         ref_put ( &tcp->refcnt );
 268         return rc;
 269 }
 270
 271 /**
 272  * Close TCP connection
 273  *
 274  * @v tcp               TCP connection
 275  * @v rc                Reason for close
 276  *
 277  * Closes the data transfer interface.  If the TCP state machine is in
 278  * a suitable state, the connection will be deleted.
 279  */
 280 static void tcp_close ( struct tcp_connection *tcp, int rc ) {
 281         struct io_buffer *iobuf;
 282         struct io_buffer *tmp;
 283
 284         /* Close data transfer interface */
 285         xfer_nullify ( &tcp->xfer );
 286         xfer_close ( &tcp->xfer, rc );
 287         tcp->flags |= TCP_XFER_CLOSED;
 288
 289         /* If we are in CLOSED, or have otherwise not yet received a
 290          * SYN (i.e. we are in LISTEN or SYN_SENT), just delete the
 291          * connection.
 292          */
 293         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 294
 295                 /* Transition to CLOSED for the sake of debugging messages */
 296                 tcp->tcp_state = TCP_CLOSED;
 297                 tcp_dump_state ( tcp );
 298
 299                 /* Free any unsent I/O buffers */
 300                 list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 301                         list_del ( &iobuf->list );
 302                         free_iob ( iobuf );
 303                 }
 304
 305                 /* Remove from list and drop reference */
 306                 stop_timer ( &tcp->timer );
 307                 list_del ( &tcp->list );
 308                 ref_put ( &tcp->refcnt );
 309                 DBGC ( tcp, "TCP %p connection deleted\n", tcp );
 310                 return;
 311         }
 312
 313         /* If we have not had our SYN acknowledged (i.e. we are in
 314          * SYN_RCVD), pretend that it has been acknowledged so that we
 315          * can send a FIN without breaking things.
 316          */
 317         if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 318                 tcp_rx_ack ( tcp, ( tcp->snd_seq + 1 ), 0 );
 319
 320         /* If we have no data remaining to send, start sending FIN */
 321         if ( list_empty ( &tcp->queue ) ) {
 322                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 323                 tcp_dump_state ( tcp );
 324         }
 325 }
 326
 327 /***************************************************************************
 328  *
 329  * Transmit data path
 330  *
 331  ***************************************************************************
 332  */
 333
 334 /**
 335  * Calculate transmission window
 336  *
 337  * @v tcp               TCP connection
 338  * @ret len             Maximum length that can be sent in a single packet
 339  */
 340 static size_t tcp_xmit_win ( struct tcp_connection *tcp ) {
 341         size_t len;
 342
 343         /* Not ready if we're not in a suitable connection state */
 344         if ( ! TCP_CAN_SEND_DATA ( tcp->tcp_state ) )
 345                 return 0;
 346
 347         /* Length is the minimum of the receiver's window and the path MTU */
 348         len = tcp->snd_win;
 349         if ( len > TCP_PATH_MTU )
 350                 len = TCP_PATH_MTU;
 351
 352         return len;
 353 }
 354
 355 /**
 356  * Process TCP transmit queue
 357  *
 358  * @v tcp               TCP connection
 359  * @v max_len           Maximum length to process
 360  * @v dest              I/O buffer to fill with data, or NULL
 361  * @v remove            Remove data from queue
 362  * @ret len             Length of data processed
 363  *
 364  * This processes at most @c max_len bytes from the TCP connection's
 365  * transmit queue.  Data will be copied into the @c dest I/O buffer
 366  * (if provided) and, if @c remove is true, removed from the transmit
 367  * queue.
 368  */
 369 static size_t tcp_process_queue ( struct tcp_connection *tcp, size_t max_len,
 370                                   struct io_buffer *dest, int remove ) {
 371         struct io_buffer *iobuf;
 372         struct io_buffer *tmp;
 373         size_t frag_len;
 374         size_t len = 0;
 375
 376         list_for_each_entry_safe ( iobuf, tmp, &tcp->queue, list ) {
 377                 frag_len = iob_len ( iobuf );
 378                 if ( frag_len > max_len )
 379                         frag_len = max_len;
 380                 if ( dest ) {
 381                         memcpy ( iob_put ( dest, frag_len ), iobuf->data,
 382                                  frag_len );
 383                 }
 384                 if ( remove ) {
 385                         iob_pull ( iobuf, frag_len );
 386                         if ( ! iob_len ( iobuf ) ) {
 387                                 list_del ( &iobuf->list );
 388                                 free_iob ( iobuf );
 389                         }
 390                 }
 391                 len += frag_len;
 392                 max_len -= frag_len;
 393         }
 394         return len;
 395 }
 396
 397 /**
 398  * Transmit any outstanding data
 399  *
 400  * @v tcp               TCP connection
 401  *
 402  * Transmits any outstanding data on the connection.
 403  *
 404  * Note that even if an error is returned, the retransmission timer
 405  * will have been started if necessary, and so the stack will
 406  * eventually attempt to retransmit the failed packet.
 407  */
 408 static int tcp_xmit ( struct tcp_connection *tcp ) {
 409         struct io_buffer *iobuf;
 410         struct tcp_header *tcphdr;
 411         struct tcp_mss_option *mssopt;
 412         struct tcp_timestamp_padded_option *tsopt;
 413         void *payload;
 414         unsigned int flags;
 415         size_t len = 0;
 416         uint32_t seq_len;
 417         uint32_t app_win;
 418         uint32_t max_rcv_win;
 419         int rc;
 420
 421         /* If retransmission timer is already running, do nothing */
 422         if ( timer_running ( &tcp->timer ) )
 423                 return 0;
 424
 425         /* Calculate both the actual (payload) and sequence space
 426          * lengths that we wish to transmit.
 427          */
 428         if ( TCP_CAN_SEND_DATA ( tcp->tcp_state ) ) {
 429                 len = tcp_process_queue ( tcp, tcp_xmit_win ( tcp ),
 430                                           NULL, 0 );
 431         }
 432         seq_len = len;
 433         flags = TCP_FLAGS_SENDING ( tcp->tcp_state );
 434         if ( flags & ( TCP_SYN | TCP_FIN ) ) {
 435                 /* SYN or FIN consume one byte, and we can never send both */
 436                 assert ( ! ( ( flags & TCP_SYN ) && ( flags & TCP_FIN ) ) );
 437                 seq_len++;
 438         }
 439         tcp->snd_sent = seq_len;
 440
 441         /* If we have nothing to transmit, stop now */
 442         if ( ( seq_len == 0 ) && ! ( tcp->flags & TCP_ACK_PENDING ) )
 443                 return 0;
 444
 445         /* If we are transmitting anything that requires
 446          * acknowledgement (i.e. consumes sequence space), start the
 447          * retransmission timer.  Do this before attempting to
 448          * allocate the I/O buffer, in case allocation itself fails.
 449          */
 450         if ( seq_len )
 451                 start_timer ( &tcp->timer );
 452
 453         /* Allocate I/O buffer */
 454         iobuf = alloc_iob ( len + MAX_HDR_LEN );
 455         if ( ! iobuf ) {
 456                 DBGC ( tcp, "TCP %p could not allocate iobuf for %08x..%08x "
 457                        "%08x\n", tcp, tcp->snd_seq, ( tcp->snd_seq + seq_len ),
 458                        tcp->rcv_ack );
 459                 return -ENOMEM;
 460         }
 461         iob_reserve ( iobuf, MAX_HDR_LEN );
 462
 463         /* Fill data payload from transmit queue */
 464         tcp_process_queue ( tcp, len, iobuf, 0 );
 465
 466         /* Expand receive window if possible */
 467         max_rcv_win = ( ( freemem * 3 ) / 4 );
 468         if ( max_rcv_win > TCP_MAX_WINDOW_SIZE )
 469                 max_rcv_win = TCP_MAX_WINDOW_SIZE;
 470         app_win = xfer_window ( &tcp->xfer );
 471         if ( max_rcv_win > app_win )
 472                 max_rcv_win = app_win;
 473         max_rcv_win &= ~0x03; /* Keep everything dword-aligned */
 474         if ( tcp->rcv_win < max_rcv_win )
 475                 tcp->rcv_win = max_rcv_win;
 476
 477         /* Fill up the TCP header */
 478         payload = iobuf->data;
 479         if ( flags & TCP_SYN ) {
 480                 mssopt = iob_push ( iobuf, sizeof ( *mssopt ) );
 481                 mssopt->kind = TCP_OPTION_MSS;
 482                 mssopt->length = sizeof ( *mssopt );
 483                 mssopt->mss = htons ( TCP_MSS );
 484         }
 485         if ( ( flags & TCP_SYN ) || ( tcp->flags & TCP_TS_ENABLED ) ) {
 486                 tsopt = iob_push ( iobuf, sizeof ( *tsopt ) );
 487                 memset ( tsopt->nop, TCP_OPTION_NOP, sizeof ( tsopt->nop ) );
 488                 tsopt->tsopt.kind = TCP_OPTION_TS;
 489                 tsopt->tsopt.length = sizeof ( tsopt->tsopt );
 490                 tsopt->tsopt.tsval = htonl ( currticks() );
 491                 tsopt->tsopt.tsecr = htonl ( tcp->ts_recent );
 492         }
 493         if ( ! ( flags & TCP_SYN ) )
 494                 flags |= TCP_PSH;
 495         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 496         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 497         tcphdr->src = htons ( tcp->local_port );
 498         tcphdr->dest = tcp->peer.st_port;
 499         tcphdr->seq = htonl ( tcp->snd_seq );
 500         tcphdr->ack = htonl ( tcp->rcv_ack );
 501         tcphdr->hlen = ( ( payload - iobuf->data ) << 2 );
 502         tcphdr->flags = flags;
 503         tcphdr->win = htons ( tcp->rcv_win );
 504         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 505
 506         /* Dump header */
 507         DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4zd",
 508                 tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 509                 ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) + seq_len ),
 510                 ntohl ( tcphdr->ack ), len );
 511         tcp_dump_flags ( tcp, tcphdr->flags );
 512         DBGC2 ( tcp, "\n" );
 513
 514         /* Transmit packet */
 515         if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, &tcp->peer, NULL,
 516                                &tcphdr->csum ) ) != 0 ) {
 517                 DBGC ( tcp, "TCP %p could not transmit %08x..%08x %08x: %s\n",
 518                        tcp, tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ),
 519                        tcp->rcv_ack, strerror ( rc ) );
 520                 return rc;
 521         }
 522
 523         /* Clear ACK-pending flag */
 524         tcp->flags &= ~TCP_ACK_PENDING;
 525
 526         return 0;
 527 }
 528
 529 /**
 530  * Retransmission timer expired
 531  *
 532  * @v timer             Retransmission timer
 533  * @v over              Failure indicator
 534  */
 535 static void tcp_expired ( struct retry_timer *timer, int over ) {
 536         struct tcp_connection *tcp =
 537                 container_of ( timer, struct tcp_connection, timer );
 538
 539         DBGC ( tcp, "TCP %p timer %s in %s for %08x..%08x %08x\n", tcp,
 540                ( over ? "expired" : "fired" ), tcp_state ( tcp->tcp_state ),
 541                tcp->snd_seq, ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
 542
 543         assert ( ( tcp->tcp_state == TCP_SYN_SENT ) ||
 544                  ( tcp->tcp_state == TCP_SYN_RCVD ) ||
 545                  ( tcp->tcp_state == TCP_ESTABLISHED ) ||
 546                  ( tcp->tcp_state == TCP_FIN_WAIT_1 ) ||
 547                  ( tcp->tcp_state == TCP_CLOSE_WAIT ) ||
 548                  ( tcp->tcp_state == TCP_CLOSING_OR_LAST_ACK ) );
 549
 550         if ( over ) {
 551                 /* If we have finally timed out and given up,
 552                  * terminate the connection
 553                  */
 554                 tcp->tcp_state = TCP_CLOSED;
 555                 tcp_dump_state ( tcp );
 556                 tcp_close ( tcp, -ETIMEDOUT );
 557         } else {
 558                 /* Otherwise, retransmit the packet */
 559                 tcp_xmit ( tcp );
 560         }
 561 }
 562
 563 /**
 564  * Shutdown timer expired
 565  *
 566  * @v timer             Shutdown timer
 567  * @v over              Failure indicator
 568  */
 569 static void tcp_wait_expired ( struct retry_timer *timer, int over __unused ) {
 570         struct tcp_connection *tcp =
 571                 container_of ( timer, struct tcp_connection, wait );
 572
 573         assert ( tcp->tcp_state == TCP_TIME_WAIT );
 574
 575         DBGC ( tcp, "TCP %p wait complete in %s for %08x..%08x %08x\n", tcp,
 576                tcp_state ( tcp->tcp_state ), tcp->snd_seq,
 577                ( tcp->snd_seq + tcp->snd_sent ), tcp->rcv_ack );
 578
 579         tcp->tcp_state = TCP_CLOSED;
 580         tcp_dump_state ( tcp );
 581         tcp_close ( tcp, 0 );
 582 }
 583
 584 /**
 585  * Send RST response to incoming packet
 586  *
 587  * @v in_tcphdr         TCP header of incoming packet
 588  * @ret rc              Return status code
 589  */
 590 static int tcp_xmit_reset ( struct tcp_connection *tcp,
 591                             struct sockaddr_tcpip *st_dest,
 592                             struct tcp_header *in_tcphdr ) {
 593         struct io_buffer *iobuf;
 594         struct tcp_header *tcphdr;
 595         int rc;
 596
 597         /* Allocate space for dataless TX buffer */
 598         iobuf = alloc_iob ( MAX_HDR_LEN );
 599         if ( ! iobuf ) {
 600                 DBGC ( tcp, "TCP %p could not allocate iobuf for RST "
 601                        "%08x..%08x %08x\n", tcp, ntohl ( in_tcphdr->ack ),
 602                        ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ) );
 603                 return -ENOMEM;
 604         }
 605         iob_reserve ( iobuf, MAX_HDR_LEN );
 606
 607         /* Construct RST response */
 608         tcphdr = iob_push ( iobuf, sizeof ( *tcphdr ) );
 609         memset ( tcphdr, 0, sizeof ( *tcphdr ) );
 610         tcphdr->src = in_tcphdr->dest;
 611         tcphdr->dest = in_tcphdr->src;
 612         tcphdr->seq = in_tcphdr->ack;
 613         tcphdr->ack = in_tcphdr->seq;
 614         tcphdr->hlen = ( ( sizeof ( *tcphdr ) / 4 ) << 4 );
 615         tcphdr->flags = ( TCP_RST | TCP_ACK );
 616         tcphdr->win = htons ( TCP_MAX_WINDOW_SIZE );
 617         tcphdr->csum = tcpip_chksum ( iobuf->data, iob_len ( iobuf ) );
 618
 619         /* Dump header */
 620         DBGC2 ( tcp, "TCP %p TX %d->%d %08x..%08x           %08x %4d",
 621                 tcp, ntohs ( tcphdr->src ), ntohs ( tcphdr->dest ),
 622                 ntohl ( tcphdr->seq ), ( ntohl ( tcphdr->seq ) ),
 623                 ntohl ( tcphdr->ack ), 0 );
 624         tcp_dump_flags ( tcp, tcphdr->flags );
 625         DBGC2 ( tcp, "\n" );
 626
 627         /* Transmit packet */
 628         if ( ( rc = tcpip_tx ( iobuf, &tcp_protocol, NULL, st_dest,
 629                                NULL, &tcphdr->csum ) ) != 0 ) {
 630                 DBGC ( tcp, "TCP %p could not transmit RST %08x..%08x %08x: "
 631                        "%s\n", tcp, ntohl ( in_tcphdr->ack ),
 632                        ntohl ( in_tcphdr->ack ), ntohl ( in_tcphdr->seq ),
 633                        strerror ( rc ) );
 634                 return rc;
 635         }
 636
 637         return 0;
 638 }
 639
 640 /***************************************************************************
 641  *
 642  * Receive data path
 643  *
 644  ***************************************************************************
 645  */
 646
 647 /**
 648  * Identify TCP connection by local port number
 649  *
 650  * @v local_port        Local port
 651  * @ret tcp             TCP connection, or NULL
 652  */
 653 static struct tcp_connection * tcp_demux ( unsigned int local_port ) {
 654         struct tcp_connection *tcp;
 655
 656         list_for_each_entry ( tcp, &tcp_conns, list ) {
 657                 if ( tcp->local_port == local_port )
 658                         return tcp;
 659         }
 660         return NULL;
 661 }
 662
 663 /**
 664  * Parse TCP received options
 665  *
 666  * @v tcp               TCP connection
 667  * @v data              Raw options data
 668  * @v len               Raw options length
 669  * @v options           Options structure to fill in
 670  */
 671 static void tcp_rx_opts ( struct tcp_connection *tcp, const void *data,
 672                           size_t len, struct tcp_options *options ) {
 673         const void *end = ( data + len );
 674         const struct tcp_option *option;
 675         unsigned int kind;
 676
 677         memset ( options, 0, sizeof ( *options ) );
 678         while ( data < end ) {
 679                 option = data;
 680                 kind = option->kind;
 681                 if ( kind == TCP_OPTION_END )
 682                         return;
 683                 if ( kind == TCP_OPTION_NOP ) {
 684                         data++;
 685                         continue;
 686                 }
 687                 switch ( kind ) {
 688                 case TCP_OPTION_MSS:
 689                         options->mssopt = data;
 690                         break;
 691                 case TCP_OPTION_TS:
 692                         options->tsopt = data;
 693                         break;
 694                 default:
 695                         DBGC ( tcp, "TCP %p received unknown option %d\n",
 696                                tcp, kind );
 697                         break;
 698                 }
 699                 data += option->length;
 700         }
 701 }
 702
 703 /**
 704  * Consume received sequence space
 705  *
 706  * @v tcp               TCP connection
 707  * @v seq_len           Sequence space length to consume
 708  */
 709 static void tcp_rx_seq ( struct tcp_connection *tcp, uint32_t seq_len ) {
 710         tcp->rcv_ack += seq_len;
 711         if ( tcp->rcv_win > seq_len ) {
 712                 tcp->rcv_win -= seq_len;
 713         } else {
 714                 tcp->rcv_win = 0;
 715         }
 716         tcp->flags |= TCP_ACK_PENDING;
 717 }
 718
 719 /**
 720  * Handle TCP received SYN
 721  *
 722  * @v tcp               TCP connection
 723  * @v seq               SEQ value (in host-endian order)
 724  * @v options           TCP options
 725  * @ret rc              Return status code
 726  */
 727 static int tcp_rx_syn ( struct tcp_connection *tcp, uint32_t seq,
 728                         struct tcp_options *options ) {
 729
 730         /* Synchronise sequence numbers on first SYN */
 731         if ( ! ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) ) {
 732                 tcp->rcv_ack = seq;
 733                 if ( options->tsopt )
 734                         tcp->flags |= TCP_TS_ENABLED;
 735         }
 736
 737         /* Ignore duplicate SYN */
 738         if ( ( tcp->rcv_ack - seq ) > 0 )
 739                 return 0;
 740
 741         /* Acknowledge SYN */
 742         tcp_rx_seq ( tcp, 1 );
 743
 744         /* Mark SYN as received and start sending ACKs with each packet */
 745         tcp->tcp_state |= ( TCP_STATE_SENT ( TCP_ACK ) |
 746                             TCP_STATE_RCVD ( TCP_SYN ) );
 747
 748         return 0;
 749 }
 750
 751 /**
 752  * Handle TCP received ACK
 753  *
 754  * @v tcp               TCP connection
 755  * @v ack               ACK value (in host-endian order)
 756  * @v win               WIN value (in host-endian order)
 757  * @ret rc              Return status code
 758  */
 759 static int tcp_rx_ack ( struct tcp_connection *tcp, uint32_t ack,
 760                         uint32_t win ) {
 761         uint32_t ack_len = ( ack - tcp->snd_seq );
 762         size_t len;
 763         unsigned int acked_flags;
 764
 765         /* Check for out-of-range or old duplicate ACKs */
 766         if ( ack_len > tcp->snd_sent ) {
 767                 DBGC ( tcp, "TCP %p received ACK for %08x..%08x, "
 768                        "sent only %08x..%08x\n", tcp, tcp->snd_seq,
 769                        ( tcp->snd_seq + ack_len ), tcp->snd_seq,
 770                        ( tcp->snd_seq + tcp->snd_sent ) );
 771
 772                 if ( TCP_HAS_BEEN_ESTABLISHED ( tcp->tcp_state ) ) {
 773                         /* Just ignore what might be old duplicate ACKs */
 774                         return 0;
 775                 } else {
 776                         /* Send RST if an out-of-range ACK is received
 777                          * on a not-yet-established connection, as per
 778                          * RFC 793.
 779                          */
 780                         return -EINVAL;
 781                 }
 782         }
 783
 784         /* Ignore ACKs that don't actually acknowledge any new data.
 785          * (In particular, do not stop the retransmission timer; this
 786          * avoids creating a sorceror's apprentice syndrome when a
 787          * duplicate ACK is received and we still have data in our
 788          * transmit queue.)
 789          */
 790         if ( ack_len == 0 )
 791                 return 0;
 792
 793         /* Stop the retransmission timer */
 794         stop_timer ( &tcp->timer );
 795
 796         /* Determine acknowledged flags and data length */
 797         len = ack_len;
 798         acked_flags = ( TCP_FLAGS_SENDING ( tcp->tcp_state ) &
 799                         ( TCP_SYN | TCP_FIN ) );
 800         if ( acked_flags )
 801                 len--;
 802
 803         /* Update SEQ and sent counters, and window size */
 804         tcp->snd_seq = ack;
 805         tcp->snd_sent = 0;
 806         tcp->snd_win = win;
 807
 808         /* Remove any acknowledged data from transmit queue */
 809         tcp_process_queue ( tcp, len, NULL, 1 );
 810
 811         /* Mark SYN/FIN as acknowledged if applicable. */
 812         if ( acked_flags )
 813                 tcp->tcp_state |= TCP_STATE_ACKED ( acked_flags );
 814
 815         /* Start sending FIN if we've had all possible data ACKed */
 816         if ( list_empty ( &tcp->queue ) && ( tcp->flags & TCP_XFER_CLOSED ) )
 817                 tcp->tcp_state |= TCP_STATE_SENT ( TCP_FIN );
 818
 819         return 0;
 820 }
 821
 822 /**
 823  * Handle TCP received data
 824  *
 825  * @v tcp               TCP connection
 826  * @v seq               SEQ value (in host-endian order)
 827  * @v iobuf             I/O buffer
 828  * @ret rc              Return status code
 829  *
 830  * This function takes ownership of the I/O buffer.
 831  */
 832 static int tcp_rx_data ( struct tcp_connection *tcp, uint32_t seq,
 833                          struct io_buffer *iobuf ) {
 834         uint32_t already_rcvd;
 835         uint32_t len;
 836         int rc;
 837
 838         /* Ignore duplicate or out-of-order data */
 839         already_rcvd = ( tcp->rcv_ack - seq );
 840         len = iob_len ( iobuf );
 841         if ( already_rcvd >= len ) {
 842                 free_iob ( iobuf );
 843                 return 0;
 844         }
 845         iob_pull ( iobuf, already_rcvd );
 846         len -= already_rcvd;
 847
 848         /* Acknowledge new data */
 849         tcp_rx_seq ( tcp, len );
 850
 851         /* Deliver data to application */
 852         if ( ( rc = xfer_deliver_iob ( &tcp->xfer, iobuf ) ) != 0 ) {
 853                 DBGC ( tcp, "TCP %p could not deliver %08x..%08x: %s\n",
 854                        tcp, seq, ( seq + len ), strerror ( rc ) );
 855                 return rc;
 856         }
 857
 858         return 0;
 859 }
 860
 861 /**
 862  * Handle TCP received FIN
 863  *
 864  * @v tcp               TCP connection
 865  * @v seq               SEQ value (in host-endian order)
 866  * @ret rc              Return status code
 867  */
 868 static int tcp_rx_fin ( struct tcp_connection *tcp, uint32_t seq ) {
 869
 870         /* Ignore duplicate or out-of-order FIN */
 871         if ( ( tcp->rcv_ack - seq ) > 0 )
 872                 return 0;
 873
 874         /* Acknowledge FIN */
 875         tcp_rx_seq ( tcp, 1 );
 876
 877         /* Mark FIN as received */
 878         tcp->tcp_state |= TCP_STATE_RCVD ( TCP_FIN );
 879
 880         /* Close connection */
 881         tcp_close ( tcp, 0 );
 882
 883         return 0;
 884 }
 885
 886 /**
 887  * Handle TCP received RST
 888  *
 889  * @v tcp               TCP connection
 890  * @v seq               SEQ value (in host-endian order)
 891  * @ret rc              Return status code
 892  */
 893 static int tcp_rx_rst ( struct tcp_connection *tcp, uint32_t seq ) {
 894
 895         /* Accept RST only if it falls within the window.  If we have
 896          * not yet received a SYN, then we have no window to test
 897          * against, so fall back to checking that our SYN has been
 898          * ACKed.
 899          */
 900         if ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) {
 901                 if ( ( seq - tcp->rcv_ack ) >= tcp->rcv_win )
 902                         return 0;
 903         } else {
 904                 if ( ! ( tcp->tcp_state & TCP_STATE_ACKED ( TCP_SYN ) ) )
 905                         return 0;
 906         }
 907
 908         /* Abort connection */
 909         tcp->tcp_state = TCP_CLOSED;
 910         tcp_dump_state ( tcp );
 911         tcp_close ( tcp, -ECONNRESET );
 912
 913         DBGC ( tcp, "TCP %p connection reset by peer\n", tcp );
 914         return -ECONNRESET;
 915 }
 916
 917 /**
 918  * Process received packet
 919  *
 920  * @v iobuf             I/O buffer
 921  * @v st_src            Partially-filled source address
 922  * @v st_dest           Partially-filled destination address
 923  * @v pshdr_csum        Pseudo-header checksum
 924  * @ret rc              Return status code
 925   */
 926 static int tcp_rx ( struct io_buffer *iobuf,
 927                     struct sockaddr_tcpip *st_src,
 928                     struct sockaddr_tcpip *st_dest __unused,
 929                     uint16_t pshdr_csum ) {
 930         struct tcp_header *tcphdr = iobuf->data;
 931         struct tcp_connection *tcp;
 932         struct tcp_options options;
 933         size_t hlen;
 934         uint16_t csum;
 935         uint32_t seq;
 936         uint32_t ack;
 937         uint32_t win;
 938         uint32_t ts_recent;
 939         unsigned int flags;
 940         size_t len;
 941         int rc;
 942
 943         /* Sanity check packet */
 944         if ( iob_len ( iobuf ) < sizeof ( *tcphdr ) ) {
 945                 DBG ( "TCP packet too short at %zd bytes (min %zd bytes)\n",
 946                       iob_len ( iobuf ), sizeof ( *tcphdr ) );
 947                 rc = -EINVAL;
 948                 goto discard;
 949         }
 950         hlen = ( ( tcphdr->hlen & TCP_MASK_HLEN ) / 16 ) * 4;
 951         if ( hlen < sizeof ( *tcphdr ) ) {
 952                 DBG ( "TCP header too short at %zd bytes (min %zd bytes)\n",
 953                       hlen, sizeof ( *tcphdr ) );
 954                 rc = -EINVAL;
 955                 goto discard;
 956         }
 957         if ( hlen > iob_len ( iobuf ) ) {
 958                 DBG ( "TCP header too long at %zd bytes (max %zd bytes)\n",
 959                       hlen, iob_len ( iobuf ) );
 960                 rc = -EINVAL;
 961                 goto discard;
 962         }
 963         csum = tcpip_continue_chksum ( pshdr_csum, iobuf->data,
 964                                        iob_len ( iobuf ) );
 965         if ( csum != 0 ) {
 966                 DBG ( "TCP checksum incorrect (is %04x including checksum "
 967                       "field, should be 0000)\n", csum );
 968                 rc = -EINVAL;
 969                 goto discard;
 970         }
 971
 972         /* Parse parameters from header and strip header */
 973         tcp = tcp_demux ( ntohs ( tcphdr->dest ) );
 974         seq = ntohl ( tcphdr->seq );
 975         ack = ntohl ( tcphdr->ack );
 976         win = ntohs ( tcphdr->win );
 977         flags = tcphdr->flags;
 978         tcp_rx_opts ( tcp, ( ( ( void * ) tcphdr ) + sizeof ( *tcphdr ) ),
 979                       ( hlen - sizeof ( *tcphdr ) ), &options );
 980         ts_recent = ( options.tsopt ?
 981                       ntohl ( options.tsopt->tsval ) : tcp->ts_recent );
 982         iob_pull ( iobuf, hlen );
 983         len = iob_len ( iobuf );
 984
 985         /* Dump header */
 986         DBGC2 ( tcp, "TCP %p RX %d<-%d           %08x %08x..%08zx %4zd",
 987                 tcp, ntohs ( tcphdr->dest ), ntohs ( tcphdr->src ),
 988                 ntohl ( tcphdr->ack ), ntohl ( tcphdr->seq ),
 989                 ( ntohl ( tcphdr->seq ) + len +
 990                   ( ( tcphdr->flags & ( TCP_SYN | TCP_FIN ) ) ? 1 : 0 )), len);
 991         tcp_dump_flags ( tcp, tcphdr->flags );
 992         DBGC2 ( tcp, "\n" );
 993
 994         /* If no connection was found, send RST */
 995         if ( ! tcp ) {
 996                 tcp_xmit_reset ( tcp, st_src, tcphdr );
 997                 rc = -ENOTCONN;
 998                 goto discard;
 999         }
1000
1001         /* Handle ACK, if present */
1002         if ( flags & TCP_ACK ) {
1003                 if ( ( rc = tcp_rx_ack ( tcp, ack, win ) ) != 0 ) {
1004                         tcp_xmit_reset ( tcp, st_src, tcphdr );
1005                         goto discard;
1006                 }
1007         }
1008
1009         /* Force an ACK if this packet is out of order */
1010         if ( ( tcp->tcp_state & TCP_STATE_RCVD ( TCP_SYN ) ) &&
1011              ( seq != tcp->rcv_ack ) ) {
1012                 tcp->flags |= TCP_ACK_PENDING;
1013         }
1014
1015         /* Handle SYN, if present */
1016         if ( flags & TCP_SYN ) {
1017                 tcp_rx_syn ( tcp, seq, &options );
1018                 seq++;
1019         }
1020
1021         /* Handle RST, if present */
1022         if ( flags & TCP_RST ) {
1023                 if ( ( rc = tcp_rx_rst ( tcp, seq ) ) != 0 )
1024                         goto discard;
1025         }
1026
1027         /* Handle new data, if any */
1028         tcp_rx_data ( tcp, seq, iob_disown ( iobuf ) );
1029         seq += len;
1030
1031         /* Handle FIN, if present */
1032         if ( flags & TCP_FIN ) {
1033                 tcp_rx_fin ( tcp, seq );
1034                 seq++;
1035         }
1036
1037         /* Update timestamp, if applicable */
1038         if ( seq == tcp->rcv_ack )
1039                 tcp->ts_recent = ts_recent;
1040
1041         /* Dump out any state change as a result of the received packet */
1042         tcp_dump_state ( tcp );
1043
1044         /* Send out any pending data */
1045         tcp_xmit ( tcp );
1046
1047         /* If this packet was the last we expect to receive, set up
1048          * timer to expire and cause the connection to be freed.
1049          */
1050         if ( TCP_CLOSED_GRACEFULLY ( tcp->tcp_state ) ) {
1051                 stop_timer ( &tcp->wait );
1052                 start_timer_fixed ( &tcp->wait, ( 2 * TCP_MSL ) );
1053         }
1054
1055         return 0;
1056
1057  discard:
1058         /* Free received packet */
1059         free_iob ( iobuf );
1060         return rc;
1061 }
1062
1063 /** TCP protocol */
1064 struct tcpip_protocol tcp_protocol __tcpip_protocol = {
1065         .name = "TCP",
1066         .rx = tcp_rx,
1067         .tcpip_proto = IP_TCP,
1068 };
1069
1070 /***************************************************************************
1071  *
1072  * Data transfer interface
1073  *
1074  ***************************************************************************
1075  */
1076
1077 /**
1078  * Close interface
1079  *
1080  * @v xfer              Data transfer interface
1081  * @v rc                Reason for close
1082  */
1083 static void tcp_xfer_close ( struct xfer_interface *xfer, int rc ) {
1084         struct tcp_connection *tcp =
1085                 container_of ( xfer, struct tcp_connection, xfer );
1086
1087         /* Close data transfer interface */
1088         tcp_close ( tcp, rc );
1089
1090         /* Transmit FIN, if possible */
1091         tcp_xmit ( tcp );
1092 }
1093
1094 /**
1095  * Check flow control window
1096  *
1097  * @v xfer              Data transfer interface
1098  * @ret len             Length of window
1099  */
1100 static size_t tcp_xfer_window ( struct xfer_interface *xfer ) {
1101         struct tcp_connection *tcp =
1102                 container_of ( xfer, struct tcp_connection, xfer );
1103
1104         /* Not ready if data queue is non-empty.  This imposes a limit
1105          * of only one unACKed packet in the TX queue at any time; we
1106          * do this to conserve memory usage.
1107          */
1108         if ( ! list_empty ( &tcp->queue ) )
1109                 return 0;
1110
1111         /* Return TCP window length */
1112         return tcp_xmit_win ( tcp );
1113 }
1114
1115 /**
1116  * Deliver datagram as I/O buffer
1117  *
1118  * @v xfer              Data transfer interface
1119  * @v iobuf             Datagram I/O buffer
1120  * @v meta              Data transfer metadata
1121  * @ret rc              Return status code
1122  */
1123 static int tcp_xfer_deliver_iob ( struct xfer_interface *xfer,
1124                                   struct io_buffer *iobuf,
1125                                   struct xfer_metadata *meta __unused ) {
1126         struct tcp_connection *tcp =
1127                 container_of ( xfer, struct tcp_connection, xfer );
1128
1129         /* Enqueue packet */
1130         list_add_tail ( &iobuf->list, &tcp->queue );
1131
1132         /* Transmit data, if possible */
1133         tcp_xmit ( tcp );
1134
1135         return 0;
1136 }
1137
1138 /** TCP data transfer interface operations */
1139 static struct xfer_interface_operations tcp_xfer_operations = {
1140         .close          = tcp_xfer_close,
1141         .vredirect      = ignore_xfer_vredirect,
1142         .window         = tcp_xfer_window,
1143         .alloc_iob      = default_xfer_alloc_iob,
1144         .deliver_iob    = tcp_xfer_deliver_iob,
1145         .deliver_raw    = xfer_deliver_as_iob,
1146 };
1147
1148 /***************************************************************************
1149  *
1150  * Openers
1151  *
1152  ***************************************************************************
1153  */
1154
1155 /** TCP socket opener */
1156 struct socket_opener tcp_socket_opener __socket_opener = {
1157         .semantics      = TCP_SOCK_STREAM,
1158         .family         = AF_INET,
1159         .open           = tcp_open,
1160 };
1161
1162 /** Linkage hack */
1163 int tcp_sock_stream = TCP_SOCK_STREAM;
1164
1165 /**
1166  * Open TCP URI
1167  *
1168  * @v xfer              Data transfer interface
1169  * @v uri               URI
1170  * @ret rc              Return status code
1171  */
1172 static int tcp_open_uri ( struct xfer_interface *xfer, struct uri *uri ) {
1173         struct sockaddr_tcpip peer;
1174
1175         /* Sanity check */
1176         if ( ! uri->host )
1177                 return -EINVAL;
1178
1179         memset ( &peer, 0, sizeof ( peer ) );
1180         peer.st_port = htons ( uri_port ( uri, 0 ) );
1181         return xfer_open_named_socket ( xfer, SOCK_STREAM,
1182                                         ( struct sockaddr * ) &peer,
1183                                         uri->host, NULL );
1184 }
1185
1186 /** TCP URI opener */
1187 struct uri_opener tcp_uri_opener __uri_opener = {
1188         .scheme         = "tcp",
1189         .open           = tcp_open_uri,
1190 };
1191