git.meshlink.io Git - utcp/blob - utcp.c

   1 /*
   2     utcp.c -- Userspace TCP
   3     Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
   4
   5     This program is free software; you can redistribute it and/or modify
   6     it under the terms of the GNU General Public License as published by
   7     the Free Software Foundation; either version 2 of the License, or
   8     (at your option) any later version.
   9
  10     This program is distributed in the hope that it will be useful,
  11     but WITHOUT ANY WARRANTY; without even the implied warranty of
  12     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13     GNU General Public License for more details.
  14
  15     You should have received a copy of the GNU General Public License along
  16     with this program; if not, write to the Free Software Foundation, Inc.,
  17     51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18 */
  19
  20 #define _GNU_SOURCE
  21
  22 #include <assert.h>
  23 #include <errno.h>
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <stdint.h>
  27 #include <stdbool.h>
  28 #include <string.h>
  29 #include <unistd.h>
  30 #include <sys/time.h>
  31 #include <sys/socket.h>
  32
  33 #include "utcp_priv.h"
  34
  35 #ifndef EBADMSG
  36 #define EBADMSG         104
  37 #endif
  38
  39 #ifndef SHUT_RDWR
  40 #define SHUT_RDWR 2
  41 #endif
  42
  43 #ifdef poll
  44 #undef poll
  45 #endif
  46
  47 #ifndef timersub
  48 #define timersub(a, b, r)\
  49         do {\
  50                 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
  51                 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
  52                 if((r)->tv_usec < 0)\
  53                         (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
  54         } while (0)
  55 #endif
  56
  57 static inline size_t max(size_t a, size_t b) {
  58         return a > b ? a : b;
  59 }
  60
  61 #ifdef UTCP_DEBUG
  62 #include <stdarg.h>
  63
  64 static void debug(const char *format, ...) {
  65         va_list ap;
  66         va_start(ap, format);
  67         vfprintf(stderr, format, ap);
  68         va_end(ap);
  69 }
  70
  71 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
  72         struct hdr hdr;
  73
  74         if(len < sizeof(hdr)) {
  75                 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
  76                 return;
  77         }
  78
  79         memcpy(&hdr, pkt, sizeof(hdr));
  80         debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux);
  81
  82         if(hdr.ctl & SYN) {
  83                 debug("SYN");
  84         }
  85
  86         if(hdr.ctl & RST) {
  87                 debug("RST");
  88         }
  89
  90         if(hdr.ctl & FIN) {
  91                 debug("FIN");
  92         }
  93
  94         if(hdr.ctl & ACK) {
  95                 debug("ACK");
  96         }
  97
  98         if(len > sizeof(hdr)) {
  99                 uint32_t datalen = len - sizeof(hdr);
 100                 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 101                 char str[datalen * 2 + 1];
 102                 char *p = str;
 103
 104                 for(uint32_t i = 0; i < datalen; i++) {
 105                         *p++ = "0123456789ABCDEF"[data[i] >> 4];
 106                         *p++ = "0123456789ABCDEF"[data[i] & 15];
 107                 }
 108
 109                 *p = 0;
 110
 111                 debug(" data=%s", str);
 112         }
 113
 114         debug("\n");
 115 }
 116 #else
 117 #define debug(...) do {} while(0)
 118 #define print_packet(...) do {} while(0)
 119 #endif
 120
 121 static void set_state(struct utcp_connection *c, enum state state) {
 122         c->state = state;
 123
 124         if(state == ESTABLISHED) {
 125                 timerclear(&c->conn_timeout);
 126         }
 127
 128         debug("%p new state: %s\n", c->utcp, strstate[state]);
 129 }
 130
 131 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 132         if(seq != c->snd.last) {
 133                 return false;
 134         }
 135
 136         switch(c->state) {
 137         case FIN_WAIT_1:
 138         case CLOSING:
 139         case LAST_ACK:
 140                 return true;
 141
 142         default:
 143                 return false;
 144         }
 145 }
 146
 147 static bool is_reliable(struct utcp_connection *c) {
 148         return c->flags & UTCP_RELIABLE;
 149 }
 150
 151 static int32_t seqdiff(uint32_t a, uint32_t b) {
 152         return a - b;
 153 }
 154
 155 // Buffer functions
 156 // TODO: convert to ringbuffers to avoid memmove() operations.
 157
 158 // Store data into the buffer
 159 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 160         debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 161
 162         size_t required = offset + len;
 163
 164         if(required > buf->maxsize) {
 165                 if(offset >= buf->maxsize) {
 166                         return 0;
 167                 }
 168
 169                 len = buf->maxsize - offset;
 170                 required = buf->maxsize;
 171         }
 172
 173         if(required > buf->size) {
 174                 size_t newsize = buf->size;
 175
 176                 if(!newsize) {
 177                         newsize = required;
 178                 } else {
 179                         do {
 180                                 newsize *= 2;
 181                         } while(newsize < required);
 182                 }
 183
 184                 if(newsize > buf->maxsize) {
 185                         newsize = buf->maxsize;
 186                 }
 187
 188                 char *newdata = realloc(buf->data, newsize);
 189
 190                 if(!newdata) {
 191                         return -1;
 192                 }
 193
 194                 buf->data = newdata;
 195                 buf->size = newsize;
 196         }
 197
 198         memcpy(buf->data + offset, data, len);
 199
 200         if(required > buf->used) {
 201                 buf->used = required;
 202         }
 203
 204         return len;
 205 }
 206
 207 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 208         return buffer_put_at(buf, buf->used, data, len);
 209 }
 210
 211 // Get data from the buffer. data can be NULL.
 212 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 213         if(len > buf->used) {
 214                 len = buf->used;
 215         }
 216
 217         if(data) {
 218                 memcpy(data, buf->data, len);
 219         }
 220
 221         if(len < buf->used) {
 222                 memmove(buf->data, buf->data + len, buf->used - len);
 223         }
 224
 225         buf->used -= len;
 226         return len;
 227 }
 228
 229 // Copy data from the buffer without removing it.
 230 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 231         if(offset >= buf->used) {
 232                 return 0;
 233         }
 234
 235         if(offset + len > buf->used) {
 236                 len = buf->used - offset;
 237         }
 238
 239         memcpy(data, buf->data + offset, len);
 240         return len;
 241 }
 242
 243 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 244         memset(buf, 0, sizeof(*buf));
 245
 246         if(len) {
 247                 buf->data = malloc(len);
 248
 249                 if(!buf->data) {
 250                         return false;
 251                 }
 252         }
 253
 254         buf->size = len;
 255         buf->maxsize = maxlen;
 256         return true;
 257 }
 258
 259 static void buffer_exit(struct buffer *buf) {
 260         free(buf->data);
 261         memset(buf, 0, sizeof(*buf));
 262 }
 263
 264 static uint32_t buffer_free(const struct buffer *buf) {
 265         return buf->maxsize - buf->used;
 266 }
 267
 268 // Connections are stored in a sorted list.
 269 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 270
 271 static int compare(const void *va, const void *vb) {
 272         assert(va && vb);
 273
 274         const struct utcp_connection *a = *(struct utcp_connection **)va;
 275         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 276
 277         assert(a && b);
 278         assert(a->src && b->src);
 279
 280         int c = (int)a->src - (int)b->src;
 281
 282         if(c) {
 283                 return c;
 284         }
 285
 286         c = (int)a->dst - (int)b->dst;
 287         return c;
 288 }
 289
 290 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 291         if(!utcp->nconnections) {
 292                 return NULL;
 293         }
 294
 295         struct utcp_connection key = {
 296                 .src = src,
 297                 .dst = dst,
 298         }, *keyp = &key;
 299         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 300         return match ? *match : NULL;
 301 }
 302
 303 static void free_connection(struct utcp_connection *c) {
 304         struct utcp *utcp = c->utcp;
 305         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 306
 307         assert(cp);
 308
 309         int i = cp - utcp->connections;
 310         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 311         utcp->nconnections--;
 312
 313         buffer_exit(&c->rcvbuf);
 314         buffer_exit(&c->sndbuf);
 315         free(c);
 316 }
 317
 318 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 319         // Check whether this combination of src and dst is free
 320
 321         if(src) {
 322                 if(find_connection(utcp, src, dst)) {
 323                         errno = EADDRINUSE;
 324                         return NULL;
 325                 }
 326         } else { // If src == 0, generate a random port number with the high bit set
 327                 if(utcp->nconnections >= 32767) {
 328                         errno = ENOMEM;
 329                         return NULL;
 330                 }
 331
 332                 src = rand() | 0x8000;
 333
 334                 while(find_connection(utcp, src, dst)) {
 335                         src++;
 336                 }
 337         }
 338
 339         // Allocate memory for the new connection
 340
 341         if(utcp->nconnections >= utcp->nallocated) {
 342                 if(!utcp->nallocated) {
 343                         utcp->nallocated = 4;
 344                 } else {
 345                         utcp->nallocated *= 2;
 346                 }
 347
 348                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 349
 350                 if(!new_array) {
 351                         return NULL;
 352                 }
 353
 354                 utcp->connections = new_array;
 355         }
 356
 357         struct utcp_connection *c = calloc(1, sizeof(*c));
 358
 359         if(!c) {
 360                 return NULL;
 361         }
 362
 363         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 364                 free(c);
 365                 return NULL;
 366         }
 367
 368         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 369                 buffer_exit(&c->sndbuf);
 370                 free(c);
 371                 return NULL;
 372         }
 373
 374         // Fill in the details
 375
 376         c->src = src;
 377         c->dst = dst;
 378 #ifdef UTCP_DEBUG
 379         c->snd.iss = 0;
 380 #else
 381         c->snd.iss = rand();
 382 #endif
 383         c->snd.una = c->snd.iss;
 384         c->snd.nxt = c->snd.iss + 1;
 385         c->rcv.wnd = utcp->mtu;
 386         c->snd.last = c->snd.nxt;
 387         c->snd.cwnd = utcp->mtu;
 388         c->utcp = utcp;
 389
 390         // Add it to the sorted list of connections
 391
 392         utcp->connections[utcp->nconnections++] = c;
 393         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 394
 395         return c;
 396 }
 397
 398 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 399         if(a > b) {
 400                 return a - b;
 401         } else {
 402                 return b - a;
 403         }
 404 }
 405
 406 // Update RTT variables. See RFC 6298.
 407 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 408         if(!rtt) {
 409                 debug("invalid rtt\n");
 410                 return;
 411         }
 412
 413         struct utcp *utcp = c->utcp;
 414
 415         if(!utcp->srtt) {
 416                 utcp->srtt = rtt;
 417                 utcp->rttvar = rtt / 2;
 418         } else {
 419                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 420                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 421         }
 422
 423         utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
 424
 425         if(utcp->rto > MAX_RTO) {
 426                 utcp->rto = MAX_RTO;
 427         }
 428
 429         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 430 }
 431
 432 static void start_retransmit_timer(struct utcp_connection *c) {
 433         gettimeofday(&c->rtrx_timeout, NULL);
 434         c->rtrx_timeout.tv_usec += c->utcp->rto;
 435
 436         while(c->rtrx_timeout.tv_usec >= 1000000) {
 437                 c->rtrx_timeout.tv_usec -= 1000000;
 438                 c->rtrx_timeout.tv_sec++;
 439         }
 440
 441         debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
 442 }
 443
 444 static void stop_retransmit_timer(struct utcp_connection *c) {
 445         timerclear(&c->rtrx_timeout);
 446         debug("timeout cleared\n");
 447 }
 448
 449 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 450         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 451
 452         if(!c) {
 453                 return NULL;
 454         }
 455
 456         assert((flags & ~0x1f) == 0);
 457
 458         c->flags = flags;
 459         c->recv = recv;
 460         c->priv = priv;
 461
 462         struct {
 463                 struct hdr hdr;
 464                 uint8_t init[4];
 465         } pkt;
 466
 467         pkt.hdr.src = c->src;
 468         pkt.hdr.dst = c->dst;
 469         pkt.hdr.seq = c->snd.iss;
 470         pkt.hdr.ack = 0;
 471         pkt.hdr.wnd = c->rcv.wnd;
 472         pkt.hdr.ctl = SYN;
 473         pkt.hdr.aux = 0x0101;
 474         pkt.init[0] = 1;
 475         pkt.init[1] = 0;
 476         pkt.init[2] = 0;
 477         pkt.init[3] = flags & 0x7;
 478
 479         set_state(c, SYN_SENT);
 480
 481         print_packet(utcp, "send", &pkt, sizeof(pkt));
 482         utcp->send(utcp, &pkt, sizeof(pkt));
 483
 484         gettimeofday(&c->conn_timeout, NULL);
 485         c->conn_timeout.tv_sec += utcp->timeout;
 486
 487         start_retransmit_timer(c);
 488
 489         return c;
 490 }
 491
 492 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 493         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 494 }
 495
 496 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 497         if(c->reapable || c->state != SYN_RECEIVED) {
 498                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 499                 return;
 500         }
 501
 502         debug("%p accepted, %p %p\n", c, recv, priv);
 503         c->recv = recv;
 504         c->priv = priv;
 505         set_state(c, ESTABLISHED);
 506 }
 507
 508 static void ack(struct utcp_connection *c, bool sendatleastone) {
 509         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 510         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 511         debug("cwndleft = %d\n", cwndleft);
 512
 513         assert(left >= 0);
 514
 515         if(cwndleft <= 0) {
 516                 cwndleft = 0;
 517         }
 518
 519         if(cwndleft < left) {
 520                 left = cwndleft;
 521         }
 522
 523         if(!left && !sendatleastone) {
 524                 return;
 525         }
 526
 527         struct {
 528                 struct hdr hdr;
 529                 uint8_t data[];
 530         } *pkt = c->utcp->pkt;
 531
 532         pkt->hdr.src = c->src;
 533         pkt->hdr.dst = c->dst;
 534         pkt->hdr.ack = c->rcv.nxt;
 535         pkt->hdr.wnd = c->snd.wnd;
 536         pkt->hdr.ctl = ACK;
 537         pkt->hdr.aux = 0;
 538
 539         do {
 540                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 541                 pkt->hdr.seq = c->snd.nxt;
 542
 543                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 544
 545                 c->snd.nxt += seglen;
 546                 left -= seglen;
 547
 548                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 549                         seglen--;
 550                         pkt->hdr.ctl |= FIN;
 551                 }
 552
 553                 if(!c->rtt_start.tv_sec) {
 554                         // Start RTT measurement
 555                         gettimeofday(&c->rtt_start, NULL);
 556                         c->rtt_seq = pkt->hdr.seq + seglen;
 557                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 558                 }
 559
 560                 print_packet(c->utcp, "send", pkt, sizeof(pkt->hdr) + seglen);
 561                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 562         } while(left);
 563 }
 564
 565 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 566         if(c->reapable) {
 567                 debug("Error: send() called on closed connection %p\n", c);
 568                 errno = EBADF;
 569                 return -1;
 570         }
 571
 572         switch(c->state) {
 573         case CLOSED:
 574         case LISTEN:
 575                 debug("Error: send() called on unconnected connection %p\n", c);
 576                 errno = ENOTCONN;
 577                 return -1;
 578
 579         case SYN_SENT:
 580         case SYN_RECEIVED:
 581         case ESTABLISHED:
 582         case CLOSE_WAIT:
 583                 break;
 584
 585         case FIN_WAIT_1:
 586         case FIN_WAIT_2:
 587         case CLOSING:
 588         case LAST_ACK:
 589         case TIME_WAIT:
 590                 debug("Error: send() called on closing connection %p\n", c);
 591                 errno = EPIPE;
 592                 return -1;
 593         }
 594
 595         // Exit early if we have nothing to send.
 596
 597         if(!len) {
 598                 return 0;
 599         }
 600
 601         if(!data) {
 602                 errno = EFAULT;
 603                 return -1;
 604         }
 605
 606         // Check if we need to be able to buffer all data
 607
 608         if(c->flags & UTCP_NO_PARTIAL) {
 609                 if(len > buffer_free(&c->sndbuf)) {
 610                         if(len > c->sndbuf.maxsize) {
 611                                 errno = EMSGSIZE;
 612                                 return -1;
 613                         } else {
 614                                 errno = EWOULDBLOCK;
 615                                 return 0;
 616                         }
 617                 }
 618         }
 619
 620         // Add data to send buffer.
 621
 622         if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 623                 len = buffer_put(&c->sndbuf, data, len);
 624         } else {
 625                 return 0;
 626         }
 627
 628         if(len <= 0) {
 629                 if(is_reliable(c)) {
 630                         errno = EWOULDBLOCK;
 631                         return 0;
 632                 } else {
 633                         return len;
 634                 }
 635         }
 636
 637         c->snd.last += len;
 638
 639         // Don't send anything yet if the connection has not fully established yet
 640
 641         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 642                 return len;
 643         }
 644
 645         ack(c, false);
 646
 647         if(!is_reliable(c)) {
 648                 c->snd.una = c->snd.nxt = c->snd.last;
 649                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 650         }
 651
 652         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 653                 start_retransmit_timer(c);
 654         }
 655
 656         if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
 657                 gettimeofday(&c->conn_timeout, NULL);
 658                 c->conn_timeout.tv_sec += c->utcp->timeout;
 659         }
 660
 661         return len;
 662 }
 663
 664 static void swap_ports(struct hdr *hdr) {
 665         uint16_t tmp = hdr->src;
 666         hdr->src = hdr->dst;
 667         hdr->dst = tmp;
 668 }
 669
 670 static void retransmit(struct utcp_connection *c) {
 671         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 672                 debug("Retransmit() called but nothing to retransmit!\n");
 673                 stop_retransmit_timer(c);
 674                 return;
 675         }
 676
 677         struct utcp *utcp = c->utcp;
 678
 679         struct {
 680                 struct hdr hdr;
 681                 uint8_t data[];
 682         } *pkt = c->utcp->pkt;
 683
 684         pkt->hdr.src = c->src;
 685         pkt->hdr.dst = c->dst;
 686         pkt->hdr.wnd = c->rcv.wnd;
 687         pkt->hdr.aux = 0;
 688
 689         switch(c->state) {
 690         case SYN_SENT:
 691                 // Send our SYN again
 692                 pkt->hdr.seq = c->snd.iss;
 693                 pkt->hdr.ack = 0;
 694                 pkt->hdr.ctl = SYN;
 695                 pkt->hdr.aux = 0x0101;
 696                 pkt->data[0] = 1;
 697                 pkt->data[1] = 0;
 698                 pkt->data[2] = 0;
 699                 pkt->data[3] = c->flags & 0x7;
 700                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 701                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 702                 break;
 703
 704         case SYN_RECEIVED:
 705                 // Send SYNACK again
 706                 pkt->hdr.seq = c->snd.nxt;
 707                 pkt->hdr.ack = c->rcv.nxt;
 708                 pkt->hdr.ctl = SYN | ACK;
 709                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
 710                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 711                 break;
 712
 713         case ESTABLISHED:
 714         case FIN_WAIT_1:
 715         case CLOSE_WAIT:
 716         case CLOSING:
 717         case LAST_ACK:
 718                 // Send unacked data again.
 719                 pkt->hdr.seq = c->snd.una;
 720                 pkt->hdr.ack = c->rcv.nxt;
 721                 pkt->hdr.ctl = ACK;
 722                 uint32_t len = seqdiff(c->snd.last, c->snd.una);
 723
 724                 if(len > utcp->mtu) {
 725                         len = utcp->mtu;
 726                 }
 727
 728                 if(fin_wanted(c, c->snd.una + len)) {
 729                         len--;
 730                         pkt->hdr.ctl |= FIN;
 731                 }
 732
 733                 c->snd.nxt = c->snd.una + len;
 734                 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 735                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 736                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 737                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 738                 break;
 739
 740         case CLOSED:
 741         case LISTEN:
 742         case TIME_WAIT:
 743         case FIN_WAIT_2:
 744                 // We shouldn't need to retransmit anything in this state.
 745 #ifdef UTCP_DEBUG
 746                 abort();
 747 #endif
 748                 stop_retransmit_timer(c);
 749                 goto cleanup;
 750         }
 751
 752         start_retransmit_timer(c);
 753         utcp->rto *= 2;
 754
 755         if(utcp->rto > MAX_RTO) {
 756                 utcp->rto = MAX_RTO;
 757         }
 758
 759         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 760
 761 cleanup:
 762         return;
 763 }
 764
 765 /* Update receive buffer and SACK entries after consuming data.
 766  *
 767  * Situation:
 768  *
 769  * |.....0000..1111111111.....22222......3333|
 770  * |---------------^
 771  *
 772  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 773  * to remove data from the receive buffer. The idea is to substract "len"
 774  * from the offset of all the SACK entries, and then remove/cut down entries
 775  * that are shifted to before the start of the receive buffer.
 776  *
 777  * There are three cases:
 778  * - the SACK entry is after ^, in that case just change the offset.
 779  * - the SACK entry starts before and ends after ^, so we have to
 780  *   change both its offset and size.
 781  * - the SACK entry is completely before ^, in that case delete it.
 782  */
 783 static void sack_consume(struct utcp_connection *c, size_t len) {
 784         debug("sack_consume %lu\n", (unsigned long)len);
 785
 786         if(len > c->rcvbuf.used) {
 787                 debug("All SACK entries consumed");
 788                 c->sacks[0].len = 0;
 789                 return;
 790         }
 791
 792         buffer_get(&c->rcvbuf, NULL, len);
 793
 794         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 795                 if(len < c->sacks[i].offset) {
 796                         c->sacks[i].offset -= len;
 797                         i++;
 798                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 799                         c->sacks[i].len -= len - c->sacks[i].offset;
 800                         c->sacks[i].offset = 0;
 801                         i++;
 802                 } else {
 803                         if(i < NSACKS - 1) {
 804                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 805                                 c->sacks[NSACKS - 1].len = 0;
 806                         } else {
 807                                 c->sacks[i].len = 0;
 808                                 break;
 809                         }
 810                 }
 811         }
 812
 813         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 814                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 815         }
 816 }
 817
 818 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 819         debug("out of order packet, offset %u\n", offset);
 820         // Packet loss or reordering occured. Store the data in the buffer.
 821         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 822
 823         if(rxd < 0 || (size_t)rxd < len) {
 824                 abort();
 825         }
 826
 827         // Make note of where we put it.
 828         for(int i = 0; i < NSACKS; i++) {
 829                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 830                         debug("New SACK entry %d\n", i);
 831                         c->sacks[i].offset = offset;
 832                         c->sacks[i].len = rxd;
 833                         break;
 834                 } else if(offset < c->sacks[i].offset) {
 835                         if(offset + rxd < c->sacks[i].offset) { // insert before
 836                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 837                                         debug("Insert SACK entry at %d\n", i);
 838                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 839                                         c->sacks[i].offset = offset;
 840                                         c->sacks[i].len = rxd;
 841                                 } else {
 842                                         debug("SACK entries full, dropping packet\n");
 843                                 }
 844
 845                                 break;
 846                         } else { // merge
 847                                 debug("Merge with start of SACK entry at %d\n", i);
 848                                 c->sacks[i].offset = offset;
 849                                 break;
 850                         }
 851                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 852                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 853                                 debug("Merge with end of SACK entry at %d\n", i);
 854                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 855                                 // TODO: handle potential merge with next entry
 856                         }
 857
 858                         break;
 859                 }
 860         }
 861
 862         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 863                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 864         }
 865 }
 866
 867 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 868         // Check if we can process out-of-order data now.
 869         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 870                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 871                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 872                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 873                 data = c->rcvbuf.data;
 874         }
 875
 876         if(c->recv) {
 877                 ssize_t rxd = c->recv(c, data, len);
 878
 879                 if(rxd < 0 || (size_t)rxd != len) {
 880                         // TODO: handle the application not accepting all data.
 881                         abort();
 882                 }
 883         }
 884
 885         if(c->rcvbuf.used) {
 886                 sack_consume(c, len);
 887         }
 888
 889         c->rcv.nxt += len;
 890 }
 891
 892
 893 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 894         if(!is_reliable(c)) {
 895                 c->recv(c, data, len);
 896                 c->rcv.nxt = seq + len;
 897                 return;
 898         }
 899
 900         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 901
 902         if(offset + len > c->rcvbuf.maxsize) {
 903                 abort();
 904         }
 905
 906         if(offset) {
 907                 handle_out_of_order(c, offset, data, len);
 908         } else {
 909                 handle_in_order(c, data, len);
 910         }
 911 }
 912
 913
 914 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 915         const uint8_t *ptr = data;
 916
 917         if(!utcp) {
 918                 errno = EFAULT;
 919                 return -1;
 920         }
 921
 922         if(!len) {
 923                 return 0;
 924         }
 925
 926         if(!data) {
 927                 errno = EFAULT;
 928                 return -1;
 929         }
 930
 931         print_packet(utcp, "recv", data, len);
 932
 933         // Drop packets smaller than the header
 934
 935         struct hdr hdr;
 936
 937         if(len < sizeof(hdr)) {
 938                 errno = EBADMSG;
 939                 return -1;
 940         }
 941
 942         // Make a copy from the potentially unaligned data to a struct hdr
 943
 944         memcpy(&hdr, ptr, sizeof(hdr));
 945         ptr += sizeof(hdr);
 946         len -= sizeof(hdr);
 947
 948         // Drop packets with an unknown CTL flag
 949
 950         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 951                 errno = EBADMSG;
 952                 return -1;
 953         }
 954
 955         // Check for auxiliary headers
 956
 957         const uint8_t *init = NULL;
 958
 959         uint16_t aux = hdr.aux;
 960
 961         while(aux) {
 962                 size_t auxlen = 4 * (aux >> 8) & 0xf;
 963                 uint8_t auxtype = aux & 0xff;
 964
 965                 if(len < auxlen) {
 966                         errno = EBADMSG;
 967                         return -1;
 968                 }
 969
 970                 switch(auxtype) {
 971                 case AUX_INIT:
 972                         if(!(hdr.ctl & SYN) || auxlen != 4) {
 973                                 errno = EBADMSG;
 974                                 return -1;
 975                         }
 976
 977                         init = ptr;
 978                         break;
 979
 980                 default:
 981                         errno = EBADMSG;
 982                         return -1;
 983                 }
 984
 985                 len -= auxlen;
 986                 ptr += auxlen;
 987
 988                 if(!(aux & 0x800)) {
 989                         break;
 990                 }
 991
 992                 if(len < 2) {
 993                         errno = EBADMSG;
 994                         return -1;
 995                 }
 996
 997                 memcpy(&aux, ptr, 2);
 998                 len -= 2;
 999                 ptr += 2;
1000         }
1001
1002         bool has_data = len || (hdr.ctl & (SYN | FIN));
1003
1004         // Try to match the packet to an existing connection
1005
1006         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1007
1008         // Is it for a new connection?
1009
1010         if(!c) {
1011                 // Ignore RST packets
1012
1013                 if(hdr.ctl & RST) {
1014                         return 0;
1015                 }
1016
1017                 // Is it a SYN packet and are we LISTENing?
1018
1019                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1020                         // If we don't want to accept it, send a RST back
1021                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1022                                 len = 1;
1023                                 goto reset;
1024                         }
1025
1026                         // Try to allocate memory, otherwise send a RST back
1027                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1028
1029                         if(!c) {
1030                                 len = 1;
1031                                 goto reset;
1032                         }
1033
1034                         // Parse auxilliary information
1035                         if(init) {
1036                                 if(init[0] < 1) {
1037                                         len = 1;
1038                                         goto reset;
1039                                 }
1040
1041                                 c->flags = init[3] & 0x7;
1042                         } else {
1043                                 c->flags = UTCP_TCP;
1044                         }
1045
1046 synack:
1047                         // Return SYN+ACK, go to SYN_RECEIVED state
1048                         c->snd.wnd = hdr.wnd;
1049                         c->rcv.irs = hdr.seq;
1050                         c->rcv.nxt = c->rcv.irs + 1;
1051                         set_state(c, SYN_RECEIVED);
1052
1053                         struct {
1054                                 struct hdr hdr;
1055                                 uint8_t data[4];
1056                         } pkt;
1057
1058                         pkt.hdr.src = c->src;
1059                         pkt.hdr.dst = c->dst;
1060                         pkt.hdr.ack = c->rcv.irs + 1;
1061                         pkt.hdr.seq = c->snd.iss;
1062                         pkt.hdr.wnd = c->rcv.wnd;
1063                         pkt.hdr.ctl = SYN | ACK;
1064
1065                         if(init) {
1066                                 pkt.hdr.aux = 0x0101;
1067                                 pkt.data[0] = 1;
1068                                 pkt.data[1] = 0;
1069                                 pkt.data[2] = 0;
1070                                 pkt.data[3] = c->flags & 0x7;
1071                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1072                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1073                         } else {
1074                                 pkt.hdr.aux = 0;
1075                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1076                                 utcp->send(utcp, &pkt, sizeof(hdr));
1077                         }
1078                 } else {
1079                         // No, we don't want your packets, send a RST back
1080                         len = 1;
1081                         goto reset;
1082                 }
1083
1084                 return 0;
1085         }
1086
1087         debug("%p state %s\n", c->utcp, strstate[c->state]);
1088
1089         // In case this is for a CLOSED connection, ignore the packet.
1090         // TODO: make it so incoming packets can never match a CLOSED connection.
1091
1092         if(c->state == CLOSED) {
1093                 debug("Got packet for closed connection\n");
1094                 return 0;
1095         }
1096
1097         // It is for an existing connection.
1098
1099         // 1. Drop invalid packets.
1100
1101         // 1a. Drop packets that should not happen in our current state.
1102
1103         switch(c->state) {
1104         case SYN_SENT:
1105         case SYN_RECEIVED:
1106         case ESTABLISHED:
1107         case FIN_WAIT_1:
1108         case FIN_WAIT_2:
1109         case CLOSE_WAIT:
1110         case CLOSING:
1111         case LAST_ACK:
1112         case TIME_WAIT:
1113                 break;
1114
1115         default:
1116 #ifdef UTCP_DEBUG
1117                 abort();
1118 #endif
1119                 break;
1120         }
1121
1122         // 1b. Discard data that is not in our receive window.
1123
1124         if(is_reliable(c)) {
1125                 bool acceptable;
1126
1127                 if(c->state == SYN_SENT) {
1128                         acceptable = true;
1129                 } else if(len == 0) {
1130                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1131                 } else {
1132                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1133
1134                         // cut already accepted front overlapping
1135                         if(rcv_offset < 0) {
1136                                 acceptable = len > (size_t) - rcv_offset;
1137
1138                                 if(acceptable) {
1139                                         ptr -= rcv_offset;
1140                                         len += rcv_offset;
1141                                         hdr.seq -= rcv_offset;
1142                                 }
1143                         } else {
1144                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1145                         }
1146                 }
1147
1148                 if(!acceptable) {
1149                         debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1150
1151                         // Ignore unacceptable RST packets.
1152                         if(hdr.ctl & RST) {
1153                                 return 0;
1154                         }
1155
1156                         // Otherwise, continue processing.
1157                         len = 0;
1158                 }
1159         }
1160
1161         c->snd.wnd = hdr.wnd; // TODO: move below
1162
1163         // 1c. Drop packets with an invalid ACK.
1164         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1165         // (= snd.una + c->sndbuf.used).
1166
1167         if(!is_reliable(c)) {
1168                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1169                         hdr.ack = c->snd.una;
1170                 }
1171         }
1172
1173         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1174                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1175
1176                 // Ignore unacceptable RST packets.
1177                 if(hdr.ctl & RST) {
1178                         return 0;
1179                 }
1180
1181                 goto reset;
1182         }
1183
1184         // 2. Handle RST packets
1185
1186         if(hdr.ctl & RST) {
1187                 switch(c->state) {
1188                 case SYN_SENT:
1189                         if(!(hdr.ctl & ACK)) {
1190                                 return 0;
1191                         }
1192
1193                         // The peer has refused our connection.
1194                         set_state(c, CLOSED);
1195                         errno = ECONNREFUSED;
1196
1197                         if(c->recv) {
1198                                 c->recv(c, NULL, 0);
1199                         }
1200
1201                         if(c->poll && !c->reapable) {
1202                                 c->poll(c, 0);
1203                         }
1204
1205                         return 0;
1206
1207                 case SYN_RECEIVED:
1208                         if(hdr.ctl & ACK) {
1209                                 return 0;
1210                         }
1211
1212                         // We haven't told the application about this connection yet. Silently delete.
1213                         free_connection(c);
1214                         return 0;
1215
1216                 case ESTABLISHED:
1217                 case FIN_WAIT_1:
1218                 case FIN_WAIT_2:
1219                 case CLOSE_WAIT:
1220                         if(hdr.ctl & ACK) {
1221                                 return 0;
1222                         }
1223
1224                         // The peer has aborted our connection.
1225                         set_state(c, CLOSED);
1226                         errno = ECONNRESET;
1227
1228                         if(c->recv) {
1229                                 c->recv(c, NULL, 0);
1230                         }
1231
1232                         if(c->poll && !c->reapable) {
1233                                 c->poll(c, 0);
1234                         }
1235
1236                         return 0;
1237
1238                 case CLOSING:
1239                 case LAST_ACK:
1240                 case TIME_WAIT:
1241                         if(hdr.ctl & ACK) {
1242                                 return 0;
1243                         }
1244
1245                         // As far as the application is concerned, the connection has already been closed.
1246                         // If it has called utcp_close() already, we can immediately free this connection.
1247                         if(c->reapable) {
1248                                 free_connection(c);
1249                                 return 0;
1250                         }
1251
1252                         // Otherwise, immediately move to the CLOSED state.
1253                         set_state(c, CLOSED);
1254                         return 0;
1255
1256                 default:
1257 #ifdef UTCP_DEBUG
1258                         abort();
1259 #endif
1260                         break;
1261                 }
1262         }
1263
1264         uint32_t advanced;
1265
1266         if(!(hdr.ctl & ACK)) {
1267                 advanced = 0;
1268                 goto skip_ack;
1269         }
1270
1271         // 3. Advance snd.una
1272
1273         advanced = seqdiff(hdr.ack, c->snd.una);
1274
1275         if(advanced) {
1276                 // RTT measurement
1277                 if(c->rtt_start.tv_sec) {
1278                         if(c->rtt_seq == hdr.ack) {
1279                                 struct timeval now, diff;
1280                                 gettimeofday(&now, NULL);
1281                                 timersub(&now, &c->rtt_start, &diff);
1282                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1283                                 c->rtt_start.tv_sec = 0;
1284                         } else if(c->rtt_seq < hdr.ack) {
1285                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1286                                 c->rtt_start.tv_sec = 0;
1287                         }
1288                 }
1289
1290                 int32_t data_acked = advanced;
1291
1292                 switch(c->state) {
1293                 case SYN_SENT:
1294                 case SYN_RECEIVED:
1295                         data_acked--;
1296                         break;
1297
1298                 // TODO: handle FIN as well.
1299                 default:
1300                         break;
1301                 }
1302
1303                 assert(data_acked >= 0);
1304
1305 #ifndef NDEBUG
1306                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1307                 assert(data_acked <= bufused);
1308 #endif
1309
1310                 if(data_acked) {
1311                         buffer_get(&c->sndbuf, NULL, data_acked);
1312                 }
1313
1314                 // Also advance snd.nxt if possible
1315                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1316                         c->snd.nxt = hdr.ack;
1317                 }
1318
1319                 c->snd.una = hdr.ack;
1320
1321                 c->dupack = 0;
1322                 c->snd.cwnd += utcp->mtu;
1323
1324                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1325                         c->snd.cwnd = c->sndbuf.maxsize;
1326                 }
1327
1328                 // Check if we have sent a FIN that is now ACKed.
1329                 switch(c->state) {
1330                 case FIN_WAIT_1:
1331                         if(c->snd.una == c->snd.last) {
1332                                 set_state(c, FIN_WAIT_2);
1333                         }
1334
1335                         break;
1336
1337                 case CLOSING:
1338                         if(c->snd.una == c->snd.last) {
1339                                 gettimeofday(&c->conn_timeout, NULL);
1340                                 c->conn_timeout.tv_sec += utcp->timeout;
1341                                 set_state(c, TIME_WAIT);
1342                         }
1343
1344                         break;
1345
1346                 default:
1347                         break;
1348                 }
1349         } else {
1350                 if(!len && is_reliable(c)) {
1351                         c->dupack++;
1352
1353                         if(c->dupack == 3) {
1354                                 debug("Triplicate ACK\n");
1355                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1356                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1357                                 //Reset the congestion window so we wait for ACKs.
1358                                 c->snd.nxt = c->snd.una;
1359                                 c->snd.cwnd = utcp->mtu;
1360                                 start_retransmit_timer(c);
1361                         }
1362                 }
1363         }
1364
1365         // 4. Update timers
1366
1367         if(advanced) {
1368                 if(c->snd.una == c->snd.last) {
1369                         stop_retransmit_timer(c);
1370                         timerclear(&c->conn_timeout);
1371                 } else if(is_reliable(c)) {
1372                         start_retransmit_timer(c);
1373                         gettimeofday(&c->conn_timeout, NULL);
1374                         c->conn_timeout.tv_sec += utcp->timeout;
1375                 }
1376         }
1377
1378 skip_ack:
1379         // 5. Process SYN stuff
1380
1381         if(hdr.ctl & SYN) {
1382                 switch(c->state) {
1383                 case SYN_SENT:
1384
1385                         // This is a SYNACK. It should always have ACKed the SYN.
1386                         if(!advanced) {
1387                                 goto reset;
1388                         }
1389
1390                         c->rcv.irs = hdr.seq;
1391                         c->rcv.nxt = hdr.seq;
1392
1393                         if(c->shut_wr) {
1394                                 c->snd.last++;
1395                                 set_state(c, FIN_WAIT_1);
1396                         } else {
1397                                 set_state(c, ESTABLISHED);
1398                         }
1399
1400                         // TODO: notify application of this somehow.
1401                         break;
1402
1403                 case SYN_RECEIVED:
1404                         // This is a retransmit of a SYN, send back the SYNACK.
1405                         goto synack;
1406
1407                 case ESTABLISHED:
1408                 case FIN_WAIT_1:
1409                 case FIN_WAIT_2:
1410                 case CLOSE_WAIT:
1411                 case CLOSING:
1412                 case LAST_ACK:
1413                 case TIME_WAIT:
1414                         // Ehm, no. We should never receive a second SYN.
1415                         return 0;
1416
1417                 default:
1418 #ifdef UTCP_DEBUG
1419                         abort();
1420 #endif
1421                         return 0;
1422                 }
1423
1424                 // SYN counts as one sequence number
1425                 c->rcv.nxt++;
1426         }
1427
1428         // 6. Process new data
1429
1430         if(c->state == SYN_RECEIVED) {
1431                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1432                 if(!advanced) {
1433                         goto reset;
1434                 }
1435
1436                 // Are we still LISTENing?
1437                 if(utcp->accept) {
1438                         utcp->accept(c, c->src);
1439                 }
1440
1441                 if(c->state != ESTABLISHED) {
1442                         set_state(c, CLOSED);
1443                         c->reapable = true;
1444                         goto reset;
1445                 }
1446         }
1447
1448         if(len) {
1449                 switch(c->state) {
1450                 case SYN_SENT:
1451                 case SYN_RECEIVED:
1452                         // This should never happen.
1453 #ifdef UTCP_DEBUG
1454                         abort();
1455 #endif
1456                         return 0;
1457
1458                 case ESTABLISHED:
1459                 case FIN_WAIT_1:
1460                 case FIN_WAIT_2:
1461                         break;
1462
1463                 case CLOSE_WAIT:
1464                 case CLOSING:
1465                 case LAST_ACK:
1466                 case TIME_WAIT:
1467                         // Ehm no, We should never receive more data after a FIN.
1468                         goto reset;
1469
1470                 default:
1471 #ifdef UTCP_DEBUG
1472                         abort();
1473 #endif
1474                         return 0;
1475                 }
1476
1477                 handle_incoming_data(c, hdr.seq, ptr, len);
1478         }
1479
1480         // 7. Process FIN stuff
1481
1482         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1483                 switch(c->state) {
1484                 case SYN_SENT:
1485                 case SYN_RECEIVED:
1486                         // This should never happen.
1487 #ifdef UTCP_DEBUG
1488                         abort();
1489 #endif
1490                         break;
1491
1492                 case ESTABLISHED:
1493                         set_state(c, CLOSE_WAIT);
1494                         break;
1495
1496                 case FIN_WAIT_1:
1497                         set_state(c, CLOSING);
1498                         break;
1499
1500                 case FIN_WAIT_2:
1501                         gettimeofday(&c->conn_timeout, NULL);
1502                         c->conn_timeout.tv_sec += utcp->timeout;
1503                         set_state(c, TIME_WAIT);
1504                         break;
1505
1506                 case CLOSE_WAIT:
1507                 case CLOSING:
1508                 case LAST_ACK:
1509                 case TIME_WAIT:
1510                         // Ehm, no. We should never receive a second FIN.
1511                         goto reset;
1512
1513                 default:
1514 #ifdef UTCP_DEBUG
1515                         abort();
1516 #endif
1517                         break;
1518                 }
1519
1520                 // FIN counts as one sequence number
1521                 c->rcv.nxt++;
1522                 len++;
1523
1524                 // Inform the application that the peer closed its end of the connection.
1525                 if(c->recv) {
1526                         errno = 0;
1527                         c->recv(c, NULL, 0);
1528                 }
1529         }
1530
1531         // Now we send something back if:
1532         // - we received data, so we have to send back an ACK
1533         //   -> sendatleastone = true
1534         // - or we got an ack, so we should maybe send a bit more data
1535         //   -> sendatleastone = false
1536
1537         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1538                 ack(c, has_data);
1539         }
1540
1541         return 0;
1542
1543 reset:
1544         swap_ports(&hdr);
1545         hdr.wnd = 0;
1546         hdr.aux = 0;
1547
1548         if(hdr.ctl & ACK) {
1549                 hdr.seq = hdr.ack;
1550                 hdr.ctl = RST;
1551         } else {
1552                 hdr.ack = hdr.seq + len;
1553                 hdr.seq = 0;
1554                 hdr.ctl = RST | ACK;
1555         }
1556
1557         print_packet(utcp, "send", &hdr, sizeof(hdr));
1558         utcp->send(utcp, &hdr, sizeof(hdr));
1559         return 0;
1560
1561 }
1562
1563 int utcp_shutdown(struct utcp_connection *c, int dir) {
1564         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1565
1566         if(!c) {
1567                 errno = EFAULT;
1568                 return -1;
1569         }
1570
1571         if(c->reapable) {
1572                 debug("Error: shutdown() called on closed connection %p\n", c);
1573                 errno = EBADF;
1574                 return -1;
1575         }
1576
1577         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1578                 errno = EINVAL;
1579                 return -1;
1580         }
1581
1582         // TCP does not have a provision for stopping incoming packets.
1583         // The best we can do is to just ignore them.
1584         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1585                 c->recv = NULL;
1586         }
1587
1588         // The rest of the code deals with shutting down writes.
1589         if(dir == UTCP_SHUT_RD) {
1590                 return 0;
1591         }
1592
1593         // Only process shutting down writes once.
1594         if(c->shut_wr) {
1595                 return 0;
1596         }
1597
1598         c->shut_wr = true;
1599
1600         switch(c->state) {
1601         case CLOSED:
1602         case LISTEN:
1603                 errno = ENOTCONN;
1604                 return -1;
1605
1606         case SYN_SENT:
1607                 return 0;
1608
1609         case SYN_RECEIVED:
1610         case ESTABLISHED:
1611                 set_state(c, FIN_WAIT_1);
1612                 break;
1613
1614         case FIN_WAIT_1:
1615         case FIN_WAIT_2:
1616                 return 0;
1617
1618         case CLOSE_WAIT:
1619                 set_state(c, CLOSING);
1620                 break;
1621
1622         case CLOSING:
1623         case LAST_ACK:
1624         case TIME_WAIT:
1625                 return 0;
1626         }
1627
1628         c->snd.last++;
1629
1630         ack(c, false);
1631
1632         if(!timerisset(&c->rtrx_timeout)) {
1633                 start_retransmit_timer(c);
1634         }
1635
1636         return 0;
1637 }
1638
1639 static bool reset_connection(struct utcp_connection *c) {
1640         if(!c) {
1641                 errno = EFAULT;
1642                 return false;
1643         }
1644
1645         if(c->reapable) {
1646                 debug("Error: abort() called on closed connection %p\n", c);
1647                 errno = EBADF;
1648                 return false;
1649         }
1650
1651         c->recv = NULL;
1652         c->poll = NULL;
1653
1654         switch(c->state) {
1655         case CLOSED:
1656                 return true;
1657
1658         case LISTEN:
1659         case SYN_SENT:
1660         case CLOSING:
1661         case LAST_ACK:
1662         case TIME_WAIT:
1663                 set_state(c, CLOSED);
1664                 return true;
1665
1666         case SYN_RECEIVED:
1667         case ESTABLISHED:
1668         case FIN_WAIT_1:
1669         case FIN_WAIT_2:
1670         case CLOSE_WAIT:
1671                 set_state(c, CLOSED);
1672                 break;
1673         }
1674
1675         // Send RST
1676
1677         struct hdr hdr;
1678
1679         hdr.src = c->src;
1680         hdr.dst = c->dst;
1681         hdr.seq = c->snd.nxt;
1682         hdr.ack = 0;
1683         hdr.wnd = 0;
1684         hdr.ctl = RST;
1685
1686         print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1687         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1688         return true;
1689 }
1690
1691 // Closes all the opened connections
1692 void utcp_abort_all_connections(struct utcp *utcp) {
1693         if(!utcp) {
1694                 errno = EINVAL;
1695                 return;
1696         }
1697
1698         for(int i = 0; i < utcp->nconnections; i++) {
1699                 struct utcp_connection *c = utcp->connections[i];
1700
1701                 if(c->reapable || c->state == CLOSED) {
1702                         continue;
1703                 }
1704
1705                 utcp_recv_t old_recv = c->recv;
1706                 utcp_poll_t old_poll = c->poll;
1707
1708                 reset_connection(c);
1709
1710                 if(old_recv) {
1711                         errno = 0;
1712                         old_recv(c, NULL, 0);
1713                 }
1714
1715                 if(old_poll && !c->reapable) {
1716                         errno = 0;
1717                         old_poll(c, 0);
1718                 }
1719         }
1720
1721         return;
1722 }
1723
1724 int utcp_close(struct utcp_connection *c) {
1725         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1726                 return -1;
1727         }
1728
1729         c->recv = NULL;
1730         c->poll = NULL;
1731         c->reapable = true;
1732         return 0;
1733 }
1734
1735 int utcp_abort(struct utcp_connection *c) {
1736         if(!reset_connection(c)) {
1737                 return -1;
1738         }
1739
1740         c->reapable = true;
1741         return 0;
1742 }
1743
1744 /* Handle timeouts.
1745  * One call to this function will loop through all connections,
1746  * checking if something needs to be resent or not.
1747  * The return value is the time to the next timeout in milliseconds,
1748  * or maybe a negative value if the timeout is infinite.
1749  */
1750 struct timeval utcp_timeout(struct utcp *utcp) {
1751         struct timeval now;
1752         gettimeofday(&now, NULL);
1753         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1754
1755         for(int i = 0; i < utcp->nconnections; i++) {
1756                 struct utcp_connection *c = utcp->connections[i];
1757
1758                 if(!c) {
1759                         continue;
1760                 }
1761
1762                 // delete connections that have been utcp_close()d.
1763                 if(c->state == CLOSED) {
1764                         if(c->reapable) {
1765                                 debug("Reaping %p\n", c);
1766                                 free_connection(c);
1767                                 i--;
1768                         }
1769
1770                         continue;
1771                 }
1772
1773                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1774                         errno = ETIMEDOUT;
1775                         c->state = CLOSED;
1776
1777                         if(c->recv) {
1778                                 c->recv(c, NULL, 0);
1779                         }
1780
1781                         if(c->poll && !c->reapable) {
1782                                 c->poll(c, 0);
1783                         }
1784
1785                         continue;
1786                 }
1787
1788                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1789                         debug("retransmit()\n");
1790                         retransmit(c);
1791                 }
1792
1793                 if(c->poll) {
1794                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1795                                 uint32_t len =  buffer_free(&c->sndbuf);
1796
1797                                 if(len) {
1798                                         c->poll(c, len);
1799                                 }
1800                         } else if(c->state == CLOSED) {
1801                                 c->poll(c, 0);
1802                         }
1803                 }
1804
1805                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1806                         next = c->conn_timeout;
1807                 }
1808
1809                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1810                         next = c->rtrx_timeout;
1811                 }
1812         }
1813
1814         struct timeval diff;
1815
1816         timersub(&next, &now, &diff);
1817
1818         return diff;
1819 }
1820
1821 bool utcp_is_active(struct utcp *utcp) {
1822         if(!utcp) {
1823                 return false;
1824         }
1825
1826         for(int i = 0; i < utcp->nconnections; i++)
1827                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1828                         return true;
1829                 }
1830
1831         return false;
1832 }
1833
1834 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1835         if(!send) {
1836                 errno = EFAULT;
1837                 return NULL;
1838         }
1839
1840         struct utcp *utcp = calloc(1, sizeof(*utcp));
1841
1842         if(!utcp) {
1843                 return NULL;
1844         }
1845
1846         utcp->accept = accept;
1847         utcp->pre_accept = pre_accept;
1848         utcp->send = send;
1849         utcp->priv = priv;
1850         utcp_set_mtu(utcp, DEFAULT_MTU);
1851         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1852         utcp->rto = START_RTO; // usec
1853
1854         return utcp;
1855 }
1856
1857 void utcp_exit(struct utcp *utcp) {
1858         if(!utcp) {
1859                 return;
1860         }
1861
1862         for(int i = 0; i < utcp->nconnections; i++) {
1863                 struct utcp_connection *c = utcp->connections[i];
1864
1865                 if(!c->reapable) {
1866                         if(c->recv) {
1867                                 c->recv(c, NULL, 0);
1868                         }
1869
1870                         if(c->poll && !c->reapable) {
1871                                 c->poll(c, 0);
1872                         }
1873                 }
1874
1875                 buffer_exit(&c->rcvbuf);
1876                 buffer_exit(&c->sndbuf);
1877                 free(c);
1878         }
1879
1880         free(utcp->connections);
1881         free(utcp);
1882 }
1883
1884 uint16_t utcp_get_mtu(struct utcp *utcp) {
1885         return utcp ? utcp->mtu : 0;
1886 }
1887
1888 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1889         if (!utcp) {
1890                 return;
1891         }
1892
1893         if (mtu <= sizeof(struct hdr)) {
1894                 return;
1895         }
1896
1897         if (mtu > utcp->mtu) {
1898                 char *new = realloc(utcp->pkt, mtu);
1899                 if (!new)
1900                         return;
1901                 utcp->pkt = new;
1902         }
1903
1904         utcp->mtu = mtu;
1905 }
1906
1907 void utcp_reset_timers(struct utcp *utcp) {
1908         if(!utcp) {
1909                 return;
1910         }
1911
1912         struct timeval now, then;
1913
1914         gettimeofday(&now, NULL);
1915
1916         then = now;
1917
1918         then.tv_sec += utcp->timeout;
1919
1920         for(int i = 0; i < utcp->nconnections; i++) {
1921                 struct utcp_connection *c = utcp->connections[i];
1922
1923                 if(c->reapable) {
1924                         continue;
1925                 }
1926
1927                 if(timerisset(&c->rtrx_timeout)) {
1928                         c->rtrx_timeout = now;
1929                 }
1930
1931                 if(timerisset(&c->conn_timeout)) {
1932                         c->conn_timeout = then;
1933                 }
1934
1935                 c->rtt_start.tv_sec = 0;
1936         }
1937
1938         if(utcp->rto > START_RTO) {
1939                 utcp->rto = START_RTO;
1940         }
1941 }
1942
1943 int utcp_get_user_timeout(struct utcp *u) {
1944         return u ? u->timeout : 0;
1945 }
1946
1947 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1948         if(u) {
1949                 u->timeout = timeout;
1950         }
1951 }
1952
1953 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1954         return c ? c->sndbuf.maxsize : 0;
1955 }
1956
1957 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1958         if(!c) {
1959                 return 0;
1960         }
1961
1962         switch(c->state) {
1963         case SYN_SENT:
1964         case SYN_RECEIVED:
1965         case ESTABLISHED:
1966         case CLOSE_WAIT:
1967                 return buffer_free(&c->sndbuf);
1968
1969         default:
1970                 return 0;
1971         }
1972 }
1973
1974 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1975         if(!c) {
1976                 return;
1977         }
1978
1979         c->sndbuf.maxsize = size;
1980
1981         if(c->sndbuf.maxsize != size) {
1982                 c->sndbuf.maxsize = -1;
1983         }
1984 }
1985
1986 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1987         return c ? c->rcvbuf.maxsize : 0;
1988 }
1989
1990 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1991         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1992                 return buffer_free(&c->rcvbuf);
1993         } else {
1994                 return 0;
1995         }
1996 }
1997
1998 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1999         if(!c) {
2000                 return;
2001         }
2002
2003         c->rcvbuf.maxsize = size;
2004
2005         if(c->rcvbuf.maxsize != size) {
2006                 c->rcvbuf.maxsize = -1;
2007         }
2008 }
2009
2010 size_t utcp_get_sendq(struct utcp_connection *c) {
2011         return c->sndbuf.used;
2012 }
2013
2014 size_t utcp_get_recvq(struct utcp_connection *c) {
2015         return c->rcvbuf.used;
2016 }
2017
2018 bool utcp_get_nodelay(struct utcp_connection *c) {
2019         return c ? c->nodelay : false;
2020 }
2021
2022 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2023         if(c) {
2024                 c->nodelay = nodelay;
2025         }
2026 }
2027
2028 bool utcp_get_keepalive(struct utcp_connection *c) {
2029         return c ? c->keepalive : false;
2030 }
2031
2032 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2033         if(c) {
2034                 c->keepalive = keepalive;
2035         }
2036 }
2037
2038 size_t utcp_get_outq(struct utcp_connection *c) {
2039         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2040 }
2041
2042 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2043         if(c) {
2044                 c->recv = recv;
2045         }
2046 }
2047
2048 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2049         if(c) {
2050                 c->poll = poll;
2051         }
2052 }
2053
2054 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2055         if(utcp) {
2056                 utcp->accept = accept;
2057                 utcp->pre_accept = pre_accept;
2058         }
2059 }
2060
2061 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2062         if(!c || c->reapable) {
2063                 return;
2064         }
2065
2066         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2067                 return;
2068         }
2069
2070         if(expect) {
2071                 // If we expect data, start the connection timer.
2072                 if(!timerisset(&c->conn_timeout)) {
2073                         gettimeofday(&c->conn_timeout, NULL);
2074                         c->conn_timeout.tv_sec += c->utcp->timeout;
2075                 }
2076         } else {
2077                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2078                 if(c->snd.una == c->snd.last) {
2079                         timerclear(&c->conn_timeout);
2080                 }
2081         }
2082 }
2083
2084 void utcp_offline(struct utcp *utcp, bool offline) {
2085         struct timeval now;
2086         gettimeofday(&now, NULL);
2087
2088         for(int i = 0; i < utcp->nconnections; i++) {
2089                 struct utcp_connection *c = utcp->connections[i];
2090
2091                 if(c->reapable) {
2092                         continue;
2093                 }
2094
2095                 utcp_expect_data(c, offline);
2096
2097                 if(!offline) {
2098                         if(timerisset(&c->rtrx_timeout)) {
2099                                 c->rtrx_timeout = now;
2100                         }
2101
2102                         utcp->connections[i]->rtt_start.tv_sec = 0;
2103                 }
2104         }
2105
2106         if(!offline && utcp->rto > START_RTO) {
2107                 utcp->rto = START_RTO;
2108         }
2109 }