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                 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
 419         } else {
 420                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 421                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 422                 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
 423         }
 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;
 531
 532         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 533
 534         if(!pkt) {
 535                 return;
 536         }
 537
 538         pkt->hdr.src = c->src;
 539         pkt->hdr.dst = c->dst;
 540         pkt->hdr.ack = c->rcv.nxt;
 541         pkt->hdr.wnd = c->snd.wnd;
 542         pkt->hdr.ctl = ACK;
 543         pkt->hdr.aux = 0;
 544
 545         do {
 546                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 547                 pkt->hdr.seq = c->snd.nxt;
 548
 549                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 550
 551                 c->snd.nxt += seglen;
 552                 left -= seglen;
 553
 554                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 555                         seglen--;
 556                         pkt->hdr.ctl |= FIN;
 557                 }
 558
 559                 if(!c->rtt_start.tv_sec) {
 560                         // Start RTT measurement
 561                         gettimeofday(&c->rtt_start, NULL);
 562                         c->rtt_seq = pkt->hdr.seq + seglen;
 563                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 564                 }
 565
 566                 print_packet(c->utcp, "send", pkt, sizeof(pkt->hdr) + seglen);
 567                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 568         } while(left);
 569
 570         free(pkt);
 571 }
 572
 573 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 574         if(c->reapable) {
 575                 debug("Error: send() called on closed connection %p\n", c);
 576                 errno = EBADF;
 577                 return -1;
 578         }
 579
 580         switch(c->state) {
 581         case CLOSED:
 582         case LISTEN:
 583                 debug("Error: send() called on unconnected connection %p\n", c);
 584                 errno = ENOTCONN;
 585                 return -1;
 586
 587         case SYN_SENT:
 588         case SYN_RECEIVED:
 589         case ESTABLISHED:
 590         case CLOSE_WAIT:
 591                 break;
 592
 593         case FIN_WAIT_1:
 594         case FIN_WAIT_2:
 595         case CLOSING:
 596         case LAST_ACK:
 597         case TIME_WAIT:
 598                 debug("Error: send() called on closing connection %p\n", c);
 599                 errno = EPIPE;
 600                 return -1;
 601         }
 602
 603         // Exit early if we have nothing to send.
 604
 605         if(!len) {
 606                 return 0;
 607         }
 608
 609         if(!data) {
 610                 errno = EFAULT;
 611                 return -1;
 612         }
 613
 614         // Check if we need to be able to buffer all data
 615
 616         if(c->flags & UTCP_NO_PARTIAL) {
 617                 if(len > buffer_free(&c->sndbuf)) {
 618                         if(len > c->sndbuf.maxsize) {
 619                                 errno = EMSGSIZE;
 620                                 return -1;
 621                         } else {
 622                                 errno = EWOULDBLOCK;
 623                                 return 0;
 624                         }
 625                 }
 626         }
 627
 628         // Add data to send buffer.
 629
 630         if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 631                 len = buffer_put(&c->sndbuf, data, len);
 632         }
 633
 634         if(len <= 0) {
 635                 if(is_reliable(c)) {
 636                         errno = EWOULDBLOCK;
 637                         return 0;
 638                 } else {
 639                         return len;
 640                 }
 641         }
 642
 643         c->snd.last += len;
 644
 645         // Don't send anything yet if the connection has not fully established yet
 646
 647         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 648                 return len;
 649         }
 650
 651         ack(c, false);
 652
 653         if(!is_reliable(c)) {
 654                 c->snd.una = c->snd.nxt = c->snd.last;
 655                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 656         }
 657
 658         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 659                 start_retransmit_timer(c);
 660         }
 661
 662         if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
 663                 gettimeofday(&c->conn_timeout, NULL);
 664                 c->conn_timeout.tv_sec += c->utcp->timeout;
 665         }
 666
 667         return len;
 668 }
 669
 670 static void swap_ports(struct hdr *hdr) {
 671         uint16_t tmp = hdr->src;
 672         hdr->src = hdr->dst;
 673         hdr->dst = tmp;
 674 }
 675
 676 static void retransmit(struct utcp_connection *c) {
 677         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 678                 debug("Retransmit() called but nothing to retransmit!\n");
 679                 stop_retransmit_timer(c);
 680                 return;
 681         }
 682
 683         struct utcp *utcp = c->utcp;
 684
 685         struct {
 686                 struct hdr hdr;
 687                 uint8_t data[];
 688         } *pkt;
 689
 690         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 691
 692         if(!pkt) {
 693                 return;
 694         }
 695
 696         pkt->hdr.src = c->src;
 697         pkt->hdr.dst = c->dst;
 698         pkt->hdr.wnd = c->rcv.wnd;
 699         pkt->hdr.aux = 0;
 700
 701         switch(c->state) {
 702         case SYN_SENT:
 703                 // Send our SYN again
 704                 pkt->hdr.seq = c->snd.iss;
 705                 pkt->hdr.ack = 0;
 706                 pkt->hdr.ctl = SYN;
 707                 pkt->hdr.aux = 0x0101;
 708                 pkt->data[0] = 1;
 709                 pkt->data[1] = 0;
 710                 pkt->data[2] = 0;
 711                 pkt->data[3] = c->flags & 0x7;
 712                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 713                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 714                 break;
 715
 716         case SYN_RECEIVED:
 717                 // Send SYNACK again
 718                 pkt->hdr.seq = c->snd.nxt;
 719                 pkt->hdr.ack = c->rcv.nxt;
 720                 pkt->hdr.ctl = SYN | ACK;
 721                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
 722                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 723                 break;
 724
 725         case ESTABLISHED:
 726         case FIN_WAIT_1:
 727         case CLOSE_WAIT:
 728         case CLOSING:
 729         case LAST_ACK:
 730                 // Send unacked data again.
 731                 pkt->hdr.seq = c->snd.una;
 732                 pkt->hdr.ack = c->rcv.nxt;
 733                 pkt->hdr.ctl = ACK;
 734                 uint32_t len = seqdiff(c->snd.last, c->snd.una);
 735
 736                 if(len > utcp->mtu) {
 737                         len = utcp->mtu;
 738                 }
 739
 740                 if(fin_wanted(c, c->snd.una + len)) {
 741                         len--;
 742                         pkt->hdr.ctl |= FIN;
 743                 }
 744
 745                 c->snd.nxt = c->snd.una + len;
 746                 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 747                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 748                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 749                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 750                 break;
 751
 752         case CLOSED:
 753         case LISTEN:
 754         case TIME_WAIT:
 755         case FIN_WAIT_2:
 756                 // We shouldn't need to retransmit anything in this state.
 757 #ifdef UTCP_DEBUG
 758                 abort();
 759 #endif
 760                 stop_retransmit_timer(c);
 761                 goto cleanup;
 762         }
 763
 764         start_retransmit_timer(c);
 765         utcp->rto *= 2;
 766
 767         if(utcp->rto > MAX_RTO) {
 768                 utcp->rto = MAX_RTO;
 769         }
 770
 771         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 772
 773 cleanup:
 774         free(pkt);
 775 }
 776
 777 /* Update receive buffer and SACK entries after consuming data.
 778  *
 779  * Situation:
 780  *
 781  * |.....0000..1111111111.....22222......3333|
 782  * |---------------^
 783  *
 784  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 785  * to remove data from the receive buffer. The idea is to substract "len"
 786  * from the offset of all the SACK entries, and then remove/cut down entries
 787  * that are shifted to before the start of the receive buffer.
 788  *
 789  * There are three cases:
 790  * - the SACK entry is after ^, in that case just change the offset.
 791  * - the SACK entry starts before and ends after ^, so we have to
 792  *   change both its offset and size.
 793  * - the SACK entry is completely before ^, in that case delete it.
 794  */
 795 static void sack_consume(struct utcp_connection *c, size_t len) {
 796         debug("sack_consume %lu\n", (unsigned long)len);
 797
 798         if(len > c->rcvbuf.used) {
 799                 debug("All SACK entries consumed");
 800                 c->sacks[0].len = 0;
 801                 return;
 802         }
 803
 804         buffer_get(&c->rcvbuf, NULL, len);
 805
 806         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 807                 if(len < c->sacks[i].offset) {
 808                         c->sacks[i].offset -= len;
 809                         i++;
 810                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 811                         c->sacks[i].len -= len - c->sacks[i].offset;
 812                         c->sacks[i].offset = 0;
 813                         i++;
 814                 } else {
 815                         if(i < NSACKS - 1) {
 816                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 817                                 c->sacks[NSACKS - 1].len = 0;
 818                         } else {
 819                                 c->sacks[i].len = 0;
 820                                 break;
 821                         }
 822                 }
 823         }
 824
 825         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 826                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 827         }
 828 }
 829
 830 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 831         debug("out of order packet, offset %u\n", offset);
 832         // Packet loss or reordering occured. Store the data in the buffer.
 833         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 834
 835         if(rxd < 0 || (size_t)rxd < len) {
 836                 abort();
 837         }
 838
 839         // Make note of where we put it.
 840         for(int i = 0; i < NSACKS; i++) {
 841                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 842                         debug("New SACK entry %d\n", i);
 843                         c->sacks[i].offset = offset;
 844                         c->sacks[i].len = rxd;
 845                         break;
 846                 } else if(offset < c->sacks[i].offset) {
 847                         if(offset + rxd < c->sacks[i].offset) { // insert before
 848                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 849                                         debug("Insert SACK entry at %d\n", i);
 850                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 851                                         c->sacks[i].offset = offset;
 852                                         c->sacks[i].len = rxd;
 853                                 } else {
 854                                         debug("SACK entries full, dropping packet\n");
 855                                 }
 856
 857                                 break;
 858                         } else { // merge
 859                                 debug("Merge with start of SACK entry at %d\n", i);
 860                                 c->sacks[i].offset = offset;
 861                                 break;
 862                         }
 863                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 864                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 865                                 debug("Merge with end of SACK entry at %d\n", i);
 866                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 867                                 // TODO: handle potential merge with next entry
 868                         }
 869
 870                         break;
 871                 }
 872         }
 873
 874         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 875                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 876         }
 877 }
 878
 879 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 880         // Check if we can process out-of-order data now.
 881         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 882                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 883                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 884                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 885                 data = c->rcvbuf.data;
 886         }
 887
 888         if(c->recv) {
 889                 ssize_t rxd = c->recv(c, data, len);
 890
 891                 if(rxd < 0 || (size_t)rxd != len) {
 892                         // TODO: handle the application not accepting all data.
 893                         abort();
 894                 }
 895         }
 896
 897         if(c->rcvbuf.used) {
 898                 sack_consume(c, len);
 899         }
 900
 901         c->rcv.nxt += len;
 902 }
 903
 904
 905 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 906         if(!is_reliable(c)) {
 907                 c->recv(c, data, len);
 908                 c->rcv.nxt = seq + len;
 909                 return;
 910         }
 911
 912         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 913
 914         if(offset + len > c->rcvbuf.maxsize) {
 915                 abort();
 916         }
 917
 918         if(offset) {
 919                 handle_out_of_order(c, offset, data, len);
 920         } else {
 921                 handle_in_order(c, data, len);
 922         }
 923 }
 924
 925
 926 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 927         const uint8_t *ptr = data;
 928
 929         if(!utcp) {
 930                 errno = EFAULT;
 931                 return -1;
 932         }
 933
 934         if(!len) {
 935                 return 0;
 936         }
 937
 938         if(!data) {
 939                 errno = EFAULT;
 940                 return -1;
 941         }
 942
 943         print_packet(utcp, "recv", data, len);
 944
 945         // Drop packets smaller than the header
 946
 947         struct hdr hdr;
 948
 949         if(len < sizeof(hdr)) {
 950                 errno = EBADMSG;
 951                 return -1;
 952         }
 953
 954         // Make a copy from the potentially unaligned data to a struct hdr
 955
 956         memcpy(&hdr, ptr, sizeof(hdr));
 957         ptr += sizeof(hdr);
 958         len -= sizeof(hdr);
 959
 960         // Drop packets with an unknown CTL flag
 961
 962         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 963                 errno = EBADMSG;
 964                 return -1;
 965         }
 966
 967         // Check for auxiliary headers
 968
 969         const uint8_t *init = NULL;
 970
 971         uint16_t aux = hdr.aux;
 972
 973         while(aux) {
 974                 size_t auxlen = 4 * (aux >> 8) & 0xf;
 975                 uint8_t auxtype = aux & 0xff;
 976
 977                 if(len < auxlen) {
 978                         errno = EBADMSG;
 979                         return -1;
 980                 }
 981
 982                 switch(auxtype) {
 983                 case AUX_INIT:
 984                         if(!(hdr.ctl & SYN) || auxlen != 4) {
 985                                 errno = EBADMSG;
 986                                 return -1;
 987                         }
 988
 989                         init = ptr;
 990                         break;
 991
 992                 default:
 993                         errno = EBADMSG;
 994                         return -1;
 995                 }
 996
 997                 len -= auxlen;
 998                 ptr += auxlen;
 999
1000                 if(!(aux & 0x800)) {
1001                         break;
1002                 }
1003
1004                 if(len < 2) {
1005                         errno = EBADMSG;
1006                         return -1;
1007                 }
1008
1009                 memcpy(&aux, ptr, 2);
1010                 len -= 2;
1011                 ptr += 2;
1012         }
1013
1014         bool has_data = len || (hdr.ctl & (SYN | FIN));
1015
1016         // Try to match the packet to an existing connection
1017
1018         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1019
1020         // Is it for a new connection?
1021
1022         if(!c) {
1023                 // Ignore RST packets
1024
1025                 if(hdr.ctl & RST) {
1026                         return 0;
1027                 }
1028
1029                 // Is it a SYN packet and are we LISTENing?
1030
1031                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1032                         // If we don't want to accept it, send a RST back
1033                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1034                                 len = 1;
1035                                 goto reset;
1036                         }
1037
1038                         // Try to allocate memory, otherwise send a RST back
1039                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1040
1041                         if(!c) {
1042                                 len = 1;
1043                                 goto reset;
1044                         }
1045
1046                         // Parse auxilliary information
1047                         if(init) {
1048                                 if(init[0] < 1) {
1049                                         len = 1;
1050                                         goto reset;
1051                                 }
1052
1053                                 c->flags = init[3] & 0x7;
1054                         } else {
1055                                 c->flags = UTCP_TCP;
1056                         }
1057
1058                         // Return SYN+ACK, go to SYN_RECEIVED state
1059                         c->snd.wnd = hdr.wnd;
1060                         c->rcv.irs = hdr.seq;
1061                         c->rcv.nxt = c->rcv.irs + 1;
1062                         set_state(c, SYN_RECEIVED);
1063
1064                         struct {
1065                                 struct hdr hdr;
1066                                 uint8_t data[4];
1067                         } pkt;
1068
1069                         pkt.hdr.src = c->src;
1070                         pkt.hdr.dst = c->dst;
1071                         pkt.hdr.ack = c->rcv.irs + 1;
1072                         pkt.hdr.seq = c->snd.iss;
1073                         pkt.hdr.wnd = c->rcv.wnd;
1074                         pkt.hdr.ctl = SYN | ACK;
1075
1076                         if(init) {
1077                                 pkt.hdr.aux = 0x0101;
1078                                 pkt.data[0] = 1;
1079                                 pkt.data[1] = 0;
1080                                 pkt.data[2] = 0;
1081                                 pkt.data[3] = c->flags & 0x7;
1082                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1083                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1084                         } else {
1085                                 pkt.hdr.aux = 0;
1086                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1087                                 utcp->send(utcp, &pkt, sizeof(hdr));
1088                         }
1089                 } else {
1090                         // No, we don't want your packets, send a RST back
1091                         len = 1;
1092                         goto reset;
1093                 }
1094
1095                 return 0;
1096         }
1097
1098         debug("%p state %s\n", c->utcp, strstate[c->state]);
1099
1100         // In case this is for a CLOSED connection, ignore the packet.
1101         // TODO: make it so incoming packets can never match a CLOSED connection.
1102
1103         if(c->state == CLOSED) {
1104                 debug("Got packet for closed connection\n");
1105                 return 0;
1106         }
1107
1108         // It is for an existing connection.
1109
1110         // 1. Drop invalid packets.
1111
1112         // 1a. Drop packets that should not happen in our current state.
1113
1114         switch(c->state) {
1115         case SYN_SENT:
1116         case SYN_RECEIVED:
1117         case ESTABLISHED:
1118         case FIN_WAIT_1:
1119         case FIN_WAIT_2:
1120         case CLOSE_WAIT:
1121         case CLOSING:
1122         case LAST_ACK:
1123         case TIME_WAIT:
1124                 break;
1125
1126         default:
1127 #ifdef UTCP_DEBUG
1128                 abort();
1129 #endif
1130                 break;
1131         }
1132
1133         // 1b. Discard data that is not in our receive window.
1134
1135         if(is_reliable(c)) {
1136                 bool acceptable;
1137
1138                 if(c->state == SYN_SENT) {
1139                         acceptable = true;
1140                 } else if(len == 0) {
1141                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1142                 } else {
1143                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1144
1145                         // cut already accepted front overlapping
1146                         if(rcv_offset < 0) {
1147                                 acceptable = len > (size_t) - rcv_offset;
1148
1149                                 if(acceptable) {
1150                                         ptr -= rcv_offset;
1151                                         len += rcv_offset;
1152                                         hdr.seq -= rcv_offset;
1153                                 }
1154                         } else {
1155                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1156                         }
1157                 }
1158
1159                 if(!acceptable) {
1160                         debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1161
1162                         // Ignore unacceptable RST packets.
1163                         if(hdr.ctl & RST) {
1164                                 return 0;
1165                         }
1166
1167                         // Otherwise, continue processing.
1168                         len = 0;
1169                 }
1170         }
1171
1172         c->snd.wnd = hdr.wnd; // TODO: move below
1173
1174         // 1c. Drop packets with an invalid ACK.
1175         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1176         // (= snd.una + c->sndbuf.used).
1177
1178         if(!is_reliable(c)) {
1179                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1180                         hdr.ack = c->snd.una;
1181                 }
1182         }
1183
1184         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1185                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1186
1187                 // Ignore unacceptable RST packets.
1188                 if(hdr.ctl & RST) {
1189                         return 0;
1190                 }
1191
1192                 goto reset;
1193         }
1194
1195         // 2. Handle RST packets
1196
1197         if(hdr.ctl & RST) {
1198                 switch(c->state) {
1199                 case SYN_SENT:
1200                         if(!(hdr.ctl & ACK)) {
1201                                 return 0;
1202                         }
1203
1204                         // The peer has refused our connection.
1205                         set_state(c, CLOSED);
1206                         errno = ECONNREFUSED;
1207
1208                         if(c->recv) {
1209                                 c->recv(c, NULL, 0);
1210                         }
1211
1212                         if(c->poll && !c->reapable) {
1213                                 c->poll(c, 0);
1214                         }
1215
1216                         return 0;
1217
1218                 case SYN_RECEIVED:
1219                         if(hdr.ctl & ACK) {
1220                                 return 0;
1221                         }
1222
1223                         // We haven't told the application about this connection yet. Silently delete.
1224                         free_connection(c);
1225                         return 0;
1226
1227                 case ESTABLISHED:
1228                 case FIN_WAIT_1:
1229                 case FIN_WAIT_2:
1230                 case CLOSE_WAIT:
1231                         if(hdr.ctl & ACK) {
1232                                 return 0;
1233                         }
1234
1235                         // The peer has aborted our connection.
1236                         set_state(c, CLOSED);
1237                         errno = ECONNRESET;
1238
1239                         if(c->recv) {
1240                                 c->recv(c, NULL, 0);
1241                         }
1242
1243                         if(c->poll && !c->reapable) {
1244                                 c->poll(c, 0);
1245                         }
1246
1247                         return 0;
1248
1249                 case CLOSING:
1250                 case LAST_ACK:
1251                 case TIME_WAIT:
1252                         if(hdr.ctl & ACK) {
1253                                 return 0;
1254                         }
1255
1256                         // As far as the application is concerned, the connection has already been closed.
1257                         // If it has called utcp_close() already, we can immediately free this connection.
1258                         if(c->reapable) {
1259                                 free_connection(c);
1260                                 return 0;
1261                         }
1262
1263                         // Otherwise, immediately move to the CLOSED state.
1264                         set_state(c, CLOSED);
1265                         return 0;
1266
1267                 default:
1268 #ifdef UTCP_DEBUG
1269                         abort();
1270 #endif
1271                         break;
1272                 }
1273         }
1274
1275         uint32_t advanced;
1276
1277         if(!(hdr.ctl & ACK)) {
1278                 advanced = 0;
1279                 goto skip_ack;
1280         }
1281
1282         // 3. Advance snd.una
1283
1284         advanced = seqdiff(hdr.ack, c->snd.una);
1285
1286         if(advanced) {
1287                 // RTT measurement
1288                 if(c->rtt_start.tv_sec) {
1289                         if(c->rtt_seq == hdr.ack) {
1290                                 struct timeval now, diff;
1291                                 gettimeofday(&now, NULL);
1292                                 timersub(&now, &c->rtt_start, &diff);
1293                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1294                                 c->rtt_start.tv_sec = 0;
1295                         } else if(c->rtt_seq < hdr.ack) {
1296                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1297                                 c->rtt_start.tv_sec = 0;
1298                         }
1299                 }
1300
1301                 int32_t data_acked = advanced;
1302
1303                 switch(c->state) {
1304                 case SYN_SENT:
1305                 case SYN_RECEIVED:
1306                         data_acked--;
1307                         break;
1308
1309                 // TODO: handle FIN as well.
1310                 default:
1311                         break;
1312                 }
1313
1314                 assert(data_acked >= 0);
1315
1316                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1317                 assert(data_acked <= bufused);
1318
1319                 if(data_acked) {
1320                         buffer_get(&c->sndbuf, NULL, data_acked);
1321                 }
1322
1323                 // Also advance snd.nxt if possible
1324                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1325                         c->snd.nxt = hdr.ack;
1326                 }
1327
1328                 c->snd.una = hdr.ack;
1329
1330                 c->dupack = 0;
1331                 c->snd.cwnd += utcp->mtu;
1332
1333                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1334                         c->snd.cwnd = c->sndbuf.maxsize;
1335                 }
1336
1337                 // Check if we have sent a FIN that is now ACKed.
1338                 switch(c->state) {
1339                 case FIN_WAIT_1:
1340                         if(c->snd.una == c->snd.last) {
1341                                 set_state(c, FIN_WAIT_2);
1342                         }
1343
1344                         break;
1345
1346                 case CLOSING:
1347                         if(c->snd.una == c->snd.last) {
1348                                 gettimeofday(&c->conn_timeout, NULL);
1349                                 c->conn_timeout.tv_sec += 60;
1350                                 set_state(c, TIME_WAIT);
1351                         }
1352
1353                         break;
1354
1355                 default:
1356                         break;
1357                 }
1358         } else {
1359                 if(!len && is_reliable(c)) {
1360                         c->dupack++;
1361
1362                         if(c->dupack == 3) {
1363                                 debug("Triplicate ACK\n");
1364                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1365                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1366                                 //Reset the congestion window so we wait for ACKs.
1367                                 c->snd.nxt = c->snd.una;
1368                                 c->snd.cwnd = utcp->mtu;
1369                                 start_retransmit_timer(c);
1370                         }
1371                 }
1372         }
1373
1374         // 4. Update timers
1375
1376         if(advanced) {
1377                 if(c->snd.una == c->snd.last) {
1378                         stop_retransmit_timer(c);
1379                         timerclear(&c->conn_timeout);
1380                 } else if(is_reliable(c)) {
1381                         start_retransmit_timer(c);
1382                         gettimeofday(&c->conn_timeout, NULL);
1383                         c->conn_timeout.tv_sec += utcp->timeout;
1384                 }
1385         }
1386
1387 skip_ack:
1388         // 5. Process SYN stuff
1389
1390         if(hdr.ctl & SYN) {
1391                 switch(c->state) {
1392                 case SYN_SENT:
1393
1394                         // This is a SYNACK. It should always have ACKed the SYN.
1395                         if(!advanced) {
1396                                 goto reset;
1397                         }
1398
1399                         c->rcv.irs = hdr.seq;
1400                         c->rcv.nxt = hdr.seq;
1401
1402                         if(c->shut_wr) {
1403                                 c->snd.last++;
1404                                 set_state(c, FIN_WAIT_1);
1405                         } else {
1406                                 set_state(c, ESTABLISHED);
1407                         }
1408
1409                         // TODO: notify application of this somehow.
1410                         break;
1411
1412                 case SYN_RECEIVED:
1413                 case ESTABLISHED:
1414                 case FIN_WAIT_1:
1415                 case FIN_WAIT_2:
1416                 case CLOSE_WAIT:
1417                 case CLOSING:
1418                 case LAST_ACK:
1419                 case TIME_WAIT:
1420                         // Ehm, no. We should never receive a second SYN.
1421                         return 0;
1422
1423                 default:
1424 #ifdef UTCP_DEBUG
1425                         abort();
1426 #endif
1427                         return 0;
1428                 }
1429
1430                 // SYN counts as one sequence number
1431                 c->rcv.nxt++;
1432         }
1433
1434         // 6. Process new data
1435
1436         if(c->state == SYN_RECEIVED) {
1437                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1438                 if(!advanced) {
1439                         goto reset;
1440                 }
1441
1442                 // Are we still LISTENing?
1443                 if(utcp->accept) {
1444                         utcp->accept(c, c->src);
1445                 }
1446
1447                 if(c->state != ESTABLISHED) {
1448                         set_state(c, CLOSED);
1449                         c->reapable = true;
1450                         goto reset;
1451                 }
1452         }
1453
1454         if(len) {
1455                 switch(c->state) {
1456                 case SYN_SENT:
1457                 case SYN_RECEIVED:
1458                         // This should never happen.
1459 #ifdef UTCP_DEBUG
1460                         abort();
1461 #endif
1462                         return 0;
1463
1464                 case ESTABLISHED:
1465                 case FIN_WAIT_1:
1466                 case FIN_WAIT_2:
1467                         break;
1468
1469                 case CLOSE_WAIT:
1470                 case CLOSING:
1471                 case LAST_ACK:
1472                 case TIME_WAIT:
1473                         // Ehm no, We should never receive more data after a FIN.
1474                         goto reset;
1475
1476                 default:
1477 #ifdef UTCP_DEBUG
1478                         abort();
1479 #endif
1480                         return 0;
1481                 }
1482
1483                 handle_incoming_data(c, hdr.seq, ptr, len);
1484         }
1485
1486         // 7. Process FIN stuff
1487
1488         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1489                 switch(c->state) {
1490                 case SYN_SENT:
1491                 case SYN_RECEIVED:
1492                         // This should never happen.
1493 #ifdef UTCP_DEBUG
1494                         abort();
1495 #endif
1496                         break;
1497
1498                 case ESTABLISHED:
1499                         set_state(c, CLOSE_WAIT);
1500                         break;
1501
1502                 case FIN_WAIT_1:
1503                         set_state(c, CLOSING);
1504                         break;
1505
1506                 case FIN_WAIT_2:
1507                         gettimeofday(&c->conn_timeout, NULL);
1508                         c->conn_timeout.tv_sec += 60;
1509                         set_state(c, TIME_WAIT);
1510                         break;
1511
1512                 case CLOSE_WAIT:
1513                 case CLOSING:
1514                 case LAST_ACK:
1515                 case TIME_WAIT:
1516                         // Ehm, no. We should never receive a second FIN.
1517                         goto reset;
1518
1519                 default:
1520 #ifdef UTCP_DEBUG
1521                         abort();
1522 #endif
1523                         break;
1524                 }
1525
1526                 // FIN counts as one sequence number
1527                 c->rcv.nxt++;
1528                 len++;
1529
1530                 // Inform the application that the peer closed its end of the connection.
1531                 if(c->recv) {
1532                         errno = 0;
1533                         c->recv(c, NULL, 0);
1534                 }
1535         }
1536
1537         // Now we send something back if:
1538         // - we received data, so we have to send back an ACK
1539         //   -> sendatleastone = true
1540         // - or we got an ack, so we should maybe send a bit more data
1541         //   -> sendatleastone = false
1542
1543         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1544                 ack(c, has_data);
1545         }
1546
1547         return 0;
1548
1549 reset:
1550         swap_ports(&hdr);
1551         hdr.wnd = 0;
1552         hdr.aux = 0;
1553
1554         if(hdr.ctl & ACK) {
1555                 hdr.seq = hdr.ack;
1556                 hdr.ctl = RST;
1557         } else {
1558                 hdr.ack = hdr.seq + len;
1559                 hdr.seq = 0;
1560                 hdr.ctl = RST | ACK;
1561         }
1562
1563         print_packet(utcp, "send", &hdr, sizeof(hdr));
1564         utcp->send(utcp, &hdr, sizeof(hdr));
1565         return 0;
1566
1567 }
1568
1569 int utcp_shutdown(struct utcp_connection *c, int dir) {
1570         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1571
1572         if(!c) {
1573                 errno = EFAULT;
1574                 return -1;
1575         }
1576
1577         if(c->reapable) {
1578                 debug("Error: shutdown() called on closed connection %p\n", c);
1579                 errno = EBADF;
1580                 return -1;
1581         }
1582
1583         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1584                 errno = EINVAL;
1585                 return -1;
1586         }
1587
1588         // TCP does not have a provision for stopping incoming packets.
1589         // The best we can do is to just ignore them.
1590         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1591                 c->recv = NULL;
1592         }
1593
1594         // The rest of the code deals with shutting down writes.
1595         if(dir == UTCP_SHUT_RD) {
1596                 return 0;
1597         }
1598
1599         // Only process shutting down writes once.
1600         if(c->shut_wr) {
1601                 return 0;
1602         }
1603
1604         c->shut_wr = true;
1605
1606         switch(c->state) {
1607         case CLOSED:
1608         case LISTEN:
1609                 errno = ENOTCONN;
1610                 return -1;
1611
1612         case SYN_SENT:
1613                 return 0;
1614
1615         case SYN_RECEIVED:
1616         case ESTABLISHED:
1617                 set_state(c, FIN_WAIT_1);
1618                 break;
1619
1620         case FIN_WAIT_1:
1621         case FIN_WAIT_2:
1622                 return 0;
1623
1624         case CLOSE_WAIT:
1625                 set_state(c, CLOSING);
1626                 break;
1627
1628         case CLOSING:
1629         case LAST_ACK:
1630         case TIME_WAIT:
1631                 return 0;
1632         }
1633
1634         c->snd.last++;
1635
1636         ack(c, false);
1637
1638         if(!timerisset(&c->rtrx_timeout)) {
1639                 start_retransmit_timer(c);
1640         }
1641
1642         return 0;
1643 }
1644
1645 static bool reset_connection(struct utcp_connection *c) {
1646         if(!c) {
1647                 errno = EFAULT;
1648                 return false;
1649         }
1650
1651         if(c->reapable) {
1652                 debug("Error: abort() called on closed connection %p\n", c);
1653                 errno = EBADF;
1654                 return false;
1655         }
1656
1657         c->recv = NULL;
1658         c->poll = NULL;
1659
1660         switch(c->state) {
1661         case CLOSED:
1662                 return true;
1663
1664         case LISTEN:
1665         case SYN_SENT:
1666         case CLOSING:
1667         case LAST_ACK:
1668         case TIME_WAIT:
1669                 set_state(c, CLOSED);
1670                 return true;
1671
1672         case SYN_RECEIVED:
1673         case ESTABLISHED:
1674         case FIN_WAIT_1:
1675         case FIN_WAIT_2:
1676         case CLOSE_WAIT:
1677                 set_state(c, CLOSED);
1678                 break;
1679         }
1680
1681         // Send RST
1682
1683         struct hdr hdr;
1684
1685         hdr.src = c->src;
1686         hdr.dst = c->dst;
1687         hdr.seq = c->snd.nxt;
1688         hdr.ack = 0;
1689         hdr.wnd = 0;
1690         hdr.ctl = RST;
1691
1692         print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1693         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1694         return true;
1695 }
1696
1697 // Closes all the opened connections
1698 void utcp_abort_all_connections(struct utcp *utcp) {
1699         if(!utcp) {
1700                 errno = EINVAL;
1701                 return;
1702         }
1703
1704         for(int i = 0; i < utcp->nconnections; i++) {
1705                 struct utcp_connection *c = utcp->connections[i];
1706
1707                 if(c->reapable || c->state == CLOSED) {
1708                         continue;
1709                 }
1710
1711                 utcp_recv_t old_recv = c->recv;
1712                 utcp_poll_t old_poll = c->poll;
1713
1714                 reset_connection(c);
1715
1716                 if(old_recv) {
1717                         errno = 0;
1718                         old_recv(c, NULL, 0);
1719                 }
1720
1721                 if(old_poll && !c->reapable) {
1722                         errno = 0;
1723                         old_poll(c, 0);
1724                 }
1725         }
1726
1727         return;
1728 }
1729
1730 int utcp_close(struct utcp_connection *c) {
1731         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1732                 return -1;
1733         }
1734
1735         c->recv = NULL;
1736         c->poll = NULL;
1737         c->reapable = true;
1738         return 0;
1739 }
1740
1741 int utcp_abort(struct utcp_connection *c) {
1742         if(!reset_connection(c)) {
1743                 return -1;
1744         }
1745
1746         c->reapable = true;
1747         return 0;
1748 }
1749
1750 /* Handle timeouts.
1751  * One call to this function will loop through all connections,
1752  * checking if something needs to be resent or not.
1753  * The return value is the time to the next timeout in milliseconds,
1754  * or maybe a negative value if the timeout is infinite.
1755  */
1756 struct timeval utcp_timeout(struct utcp *utcp) {
1757         struct timeval now;
1758         gettimeofday(&now, NULL);
1759         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1760
1761         for(int i = 0; i < utcp->nconnections; i++) {
1762                 struct utcp_connection *c = utcp->connections[i];
1763
1764                 if(!c) {
1765                         continue;
1766                 }
1767
1768                 // delete connections that have been utcp_close()d.
1769                 if(c->state == CLOSED) {
1770                         if(c->reapable) {
1771                                 debug("Reaping %p\n", c);
1772                                 free_connection(c);
1773                                 i--;
1774                         }
1775
1776                         continue;
1777                 }
1778
1779                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1780                         errno = ETIMEDOUT;
1781                         c->state = CLOSED;
1782
1783                         if(c->recv) {
1784                                 c->recv(c, NULL, 0);
1785                         }
1786
1787                         if(c->poll && !c->reapable) {
1788                                 c->poll(c, 0);
1789                         }
1790
1791                         continue;
1792                 }
1793
1794                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1795                         debug("retransmit()\n");
1796                         retransmit(c);
1797                 }
1798
1799                 if(c->poll) {
1800                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1801                                 uint32_t len =  buffer_free(&c->sndbuf);
1802
1803                                 if(len) {
1804                                         c->poll(c, len);
1805                                 }
1806                         } else if(c->state == CLOSED) {
1807                                 c->poll(c, 0);
1808                         }
1809                 }
1810
1811                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1812                         next = c->conn_timeout;
1813                 }
1814
1815                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1816                         next = c->rtrx_timeout;
1817                 }
1818         }
1819
1820         struct timeval diff;
1821
1822         timersub(&next, &now, &diff);
1823
1824         return diff;
1825 }
1826
1827 bool utcp_is_active(struct utcp *utcp) {
1828         if(!utcp) {
1829                 return false;
1830         }
1831
1832         for(int i = 0; i < utcp->nconnections; i++)
1833                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1834                         return true;
1835                 }
1836
1837         return false;
1838 }
1839
1840 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1841         if(!send) {
1842                 errno = EFAULT;
1843                 return NULL;
1844         }
1845
1846         struct utcp *utcp = calloc(1, sizeof(*utcp));
1847
1848         if(!utcp) {
1849                 return NULL;
1850         }
1851
1852         utcp->accept = accept;
1853         utcp->pre_accept = pre_accept;
1854         utcp->send = send;
1855         utcp->priv = priv;
1856         utcp->mtu = DEFAULT_MTU;
1857         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1858         utcp->rto = START_RTO; // usec
1859
1860         return utcp;
1861 }
1862
1863 void utcp_exit(struct utcp *utcp) {
1864         if(!utcp) {
1865                 return;
1866         }
1867
1868         for(int i = 0; i < utcp->nconnections; i++) {
1869                 struct utcp_connection *c = utcp->connections[i];
1870
1871                 if(!c->reapable) {
1872                         if(c->recv) {
1873                                 c->recv(c, NULL, 0);
1874                         }
1875
1876                         if(c->poll && !c->reapable) {
1877                                 c->poll(c, 0);
1878                         }
1879                 }
1880
1881                 buffer_exit(&c->rcvbuf);
1882                 buffer_exit(&c->sndbuf);
1883                 free(c);
1884         }
1885
1886         free(utcp->connections);
1887         free(utcp);
1888 }
1889
1890 uint16_t utcp_get_mtu(struct utcp *utcp) {
1891         return utcp ? utcp->mtu : 0;
1892 }
1893
1894 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1895         // TODO: handle overhead of the header
1896         if(utcp) {
1897                 utcp->mtu = mtu;
1898         }
1899 }
1900
1901 void utcp_reset_timers(struct utcp *utcp) {
1902         if(!utcp) {
1903                 return;
1904         }
1905
1906         struct timeval now, then;
1907
1908         gettimeofday(&now, NULL);
1909
1910         then = now;
1911
1912         then.tv_sec += utcp->timeout;
1913
1914         for(int i = 0; i < utcp->nconnections; i++) {
1915                 struct utcp_connection *c = utcp->connections[i];
1916
1917                 if(c->reapable) {
1918                         continue;
1919                 }
1920
1921                 c->rtrx_timeout = now;
1922                 c->conn_timeout = then;
1923                 c->rtt_start.tv_sec = 0;
1924         }
1925
1926         if(utcp->rto > START_RTO) {
1927                 utcp->rto = START_RTO;
1928         }
1929 }
1930
1931 int utcp_get_user_timeout(struct utcp *u) {
1932         return u ? u->timeout : 0;
1933 }
1934
1935 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1936         if(u) {
1937                 u->timeout = timeout;
1938         }
1939 }
1940
1941 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1942         return c ? c->sndbuf.maxsize : 0;
1943 }
1944
1945 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1946         if(!c) {
1947                 return 0;
1948         }
1949
1950         switch(c->state) {
1951         case SYN_SENT:
1952         case SYN_RECEIVED:
1953         case ESTABLISHED:
1954         case CLOSE_WAIT:
1955                 return buffer_free(&c->sndbuf);
1956
1957         default:
1958                 return 0;
1959         }
1960 }
1961
1962 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1963         if(!c) {
1964                 return;
1965         }
1966
1967         c->sndbuf.maxsize = size;
1968
1969         if(c->sndbuf.maxsize != size) {
1970                 c->sndbuf.maxsize = -1;
1971         }
1972 }
1973
1974 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1975         return c ? c->rcvbuf.maxsize : 0;
1976 }
1977
1978 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1979         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1980                 return buffer_free(&c->rcvbuf);
1981         } else {
1982                 return 0;
1983         }
1984 }
1985
1986 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1987         if(!c) {
1988                 return;
1989         }
1990
1991         c->rcvbuf.maxsize = size;
1992
1993         if(c->rcvbuf.maxsize != size) {
1994                 c->rcvbuf.maxsize = -1;
1995         }
1996 }
1997
1998 size_t utcp_get_sendq(struct utcp_connection *c) {
1999         return c->sndbuf.used;
2000 }
2001
2002 size_t utcp_get_recvq(struct utcp_connection *c) {
2003         return c->rcvbuf.used;
2004 }
2005
2006 bool utcp_get_nodelay(struct utcp_connection *c) {
2007         return c ? c->nodelay : false;
2008 }
2009
2010 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2011         if(c) {
2012                 c->nodelay = nodelay;
2013         }
2014 }
2015
2016 bool utcp_get_keepalive(struct utcp_connection *c) {
2017         return c ? c->keepalive : false;
2018 }
2019
2020 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2021         if(c) {
2022                 c->keepalive = keepalive;
2023         }
2024 }
2025
2026 size_t utcp_get_outq(struct utcp_connection *c) {
2027         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2028 }
2029
2030 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2031         if(c) {
2032                 c->recv = recv;
2033         }
2034 }
2035
2036 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2037         if(c) {
2038                 c->poll = poll;
2039         }
2040 }
2041
2042 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2043         if(utcp) {
2044                 utcp->accept = accept;
2045                 utcp->pre_accept = pre_accept;
2046         }
2047 }
2048
2049 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2050         if(!c || c->reapable) {
2051                 return;
2052         }
2053
2054         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2055                 return;
2056         }
2057
2058         if(expect) {
2059                 // If we expect data, start the connection timer.
2060                 if(!timerisset(&c->conn_timeout)) {
2061                         gettimeofday(&c->conn_timeout, NULL);
2062                         c->conn_timeout.tv_sec += c->utcp->timeout;
2063                 }
2064         } else {
2065                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2066                 if(c->snd.una == c->snd.last) {
2067                         timerclear(&c->conn_timeout);
2068                 }
2069         }
2070 }
2071
2072 void utcp_offline(struct utcp *utcp, bool offline) {
2073         for(int i = 0; i < utcp->nconnections; i++) {
2074                 struct utcp_connection *c = utcp->connections[i];
2075
2076                 if(!c->reapable) {
2077                         utcp_expect_data(c, offline);
2078
2079                         // If we are online again, reset the retransmission timers, but keep the connection timeout as it is,
2080                         // to prevent peers toggling online/offline state frequently from keeping connections alive
2081                         // if there is no progress in sending actual data.
2082                         if(!offline) {
2083                                 gettimeofday(&utcp->connections[i]->rtrx_timeout, NULL);
2084                                 utcp->connections[i]->rtt_start.tv_sec = 0;
2085                         }
2086                 }
2087         }
2088
2089         if(!offline && utcp->rto > START_RTO) {
2090                 utcp->rto = START_RTO;
2091         }
2092 }