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