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

   1 /*
   2     utcp.c -- Userspace TCP
   3     Copyright (C) 2014 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) do {\
  49         (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
  50         (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
  51         if((r)->tv_usec < 0)\
  52                 (r)->tv_sec--, (r)->tv_usec += 1000000;\
  53 } while (0)
  54 #endif
  55
  56 #ifndef max
  57 #define max(a, b) ((a) > (b) ? (a) : (b))
  58 #endif
  59
  60 #ifdef UTCP_DEBUG
  61 #include <stdarg.h>
  62
  63 static void debug(const char *format, ...) {
  64         va_list ap;
  65         va_start(ap, format);
  66         vfprintf(stderr, format, ap);
  67         va_end(ap);
  68 }
  69
  70 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
  71         struct hdr hdr;
  72         if(len < sizeof hdr) {
  73                 debug("%p %s: short packet (%zu bytes)\n", utcp, dir, len);
  74                 return;
  75         }
  76
  77         memcpy(&hdr, pkt, sizeof hdr);
  78         fprintf (stderr, "%p %s: len=%zu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
  79         if(hdr.ctl & SYN)
  80                 debug("SYN");
  81         if(hdr.ctl & RST)
  82                 debug("RST");
  83         if(hdr.ctl & FIN)
  84                 debug("FIN");
  85         if(hdr.ctl & ACK)
  86                 debug("ACK");
  87
  88         if(len > sizeof hdr) {
  89                 debug(" data=");
  90                 for(int i = sizeof hdr; i < len; i++) {
  91                         const char *data = pkt;
  92                         debug("%c", data[i] >= 32 ? data[i] : '.');
  93                 }
  94         }
  95
  96         debug("\n");
  97 }
  98 #else
  99 #define debug(...)
 100 #define print_packet(...)
 101 #endif
 102
 103 static void set_state(struct utcp_connection *c, enum state state) {
 104         c->state = state;
 105         if(state == ESTABLISHED)
 106                 timerclear(&c->conn_timeout);
 107         debug("%p new state: %s\n", c->utcp, strstate[state]);
 108 }
 109
 110 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 111         if(seq != c->snd.last)
 112                 return false;
 113         switch(c->state) {
 114         case FIN_WAIT_1:
 115         case CLOSING:
 116         case LAST_ACK:
 117                 return true;
 118         default:
 119                 return false;
 120         }
 121 }
 122
 123 static inline void list_connections(struct utcp *utcp) {
 124         debug("%p has %d connections:\n", utcp, utcp->nconnections);
 125         for(int i = 0; i < utcp->nconnections; i++)
 126                 debug("  %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
 127 }
 128
 129 static int32_t seqdiff(uint32_t a, uint32_t b) {
 130         return a - b;
 131 }
 132
 133 // Buffer functions
 134 // TODO: convert to ringbuffers to avoid memmove() operations.
 135
 136 // Store data into the buffer
 137 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 138         if(buf->maxsize <= buf->used)
 139                 return 0;
 140
 141         debug("buffer_put_at %zu %zu %zu\n", buf->used, offset, len);
 142
 143         size_t required = offset + len;
 144         if(required > buf->maxsize) {
 145                 if(offset >= buf->maxsize)
 146                         return 0;
 147                 abort();
 148                 len = buf->maxsize - offset;
 149                 required = buf->maxsize;
 150         }
 151
 152         if(required > buf->size) {
 153                 size_t newsize = buf->size;
 154                 if(!newsize) {
 155                         newsize = required;
 156                 } else {
 157                         do {
 158                                 newsize *= 2;
 159                         } while(newsize < buf->used + len);
 160                 }
 161                 if(newsize > buf->maxsize)
 162                         newsize = buf->maxsize;
 163                 char *newdata = realloc(buf->data, newsize);
 164                 if(!newdata)
 165                         return -1;
 166                 buf->data = newdata;
 167                 buf->size = newsize;
 168         }
 169
 170         memcpy(buf->data + offset, data, len);
 171         if(required > buf->used)
 172                 buf->used = required;
 173         return len;
 174 }
 175
 176 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 177         return buffer_put_at(buf, buf->used, data, len);
 178 }
 179
 180 // Get data from the buffer. data can be NULL.
 181 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 182         if(len > buf->used)
 183                 len = buf->used;
 184         if(data)
 185                 memcpy(data, buf->data, len);
 186         if(len < buf->used)
 187                 memmove(buf->data, buf->data + len, buf->used - len);
 188         buf->used -= len;
 189         return len;
 190 }
 191
 192 // Copy data from the buffer without removing it.
 193 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 194         if(offset >= buf->used)
 195                 return 0;
 196         if(offset + len > buf->used)
 197                 len = buf->used - offset;
 198         memcpy(data, buf->data + offset, len);
 199         return len;
 200 }
 201
 202 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 203         memset(buf, 0, sizeof *buf);
 204         if(len) {
 205                 buf->data = malloc(len);
 206                 if(!buf->data)
 207                         return false;
 208         }
 209         buf->size = len;
 210         buf->maxsize = maxlen;
 211         return true;
 212 }
 213
 214 static void buffer_exit(struct buffer *buf) {
 215         free(buf->data);
 216         memset(buf, 0, sizeof *buf);
 217 }
 218
 219 static uint32_t buffer_free(const struct buffer *buf) {
 220         return buf->maxsize - buf->used;
 221 }
 222
 223 // Connections are stored in a sorted list.
 224 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 225
 226 static int compare(const void *va, const void *vb) {
 227         assert(va && vb);
 228
 229         const struct utcp_connection *a = *(struct utcp_connection **)va;
 230         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 231
 232         assert(a && b);
 233         assert(a->src && b->src);
 234
 235         int c = (int)a->src - (int)b->src;
 236         if(c)
 237                 return c;
 238         c = (int)a->dst - (int)b->dst;
 239         return c;
 240 }
 241
 242 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 243         if(!utcp->nconnections)
 244                 return NULL;
 245         struct utcp_connection key = {
 246                 .src = src,
 247                 .dst = dst,
 248         }, *keyp = &key;
 249         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 250         return match ? *match : NULL;
 251 }
 252
 253 static void free_connection(struct utcp_connection *c) {
 254         struct utcp *utcp = c->utcp;
 255         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 256
 257         assert(cp);
 258
 259         int i = cp - utcp->connections;
 260         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
 261         utcp->nconnections--;
 262
 263         buffer_exit(&c->sndbuf);
 264         free(c);
 265 }
 266
 267 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 268         // Check whether this combination of src and dst is free
 269
 270         if(src) {
 271                 if(find_connection(utcp, src, dst)) {
 272                         errno = EADDRINUSE;
 273                         return NULL;
 274                 }
 275         } else { // If src == 0, generate a random port number with the high bit set
 276                 if(utcp->nconnections >= 32767) {
 277                         errno = ENOMEM;
 278                         return NULL;
 279                 }
 280                 src = rand() | 0x8000;
 281                 while(find_connection(utcp, src, dst))
 282                         src++;
 283         }
 284
 285         // Allocate memory for the new connection
 286
 287         if(utcp->nconnections >= utcp->nallocated) {
 288                 if(!utcp->nallocated)
 289                         utcp->nallocated = 4;
 290                 else
 291                         utcp->nallocated *= 2;
 292                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
 293                 if(!new_array)
 294                         return NULL;
 295                 utcp->connections = new_array;
 296         }
 297
 298         struct utcp_connection *c = calloc(1, sizeof *c);
 299         if(!c)
 300                 return NULL;
 301
 302         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 303                 free(c);
 304                 return NULL;
 305         }
 306
 307         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 308                 free(c);
 309                 return NULL;
 310         }
 311
 312         // Fill in the details
 313
 314         c->src = src;
 315         c->dst = dst;
 316 #ifdef UTCP_DEBUG
 317         c->snd.iss = 0;
 318 #else
 319         c->snd.iss = rand();
 320 #endif
 321         c->snd.una = c->snd.iss;
 322         c->snd.nxt = c->snd.iss + 1;
 323         c->rcv.wnd = utcp->mtu;
 324         c->snd.last = c->snd.nxt;
 325         c->snd.cwnd = utcp->mtu;
 326         c->utcp = utcp;
 327
 328         // Add it to the sorted list of connections
 329
 330         utcp->connections[utcp->nconnections++] = c;
 331         qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 332
 333         return c;
 334 }
 335
 336 // Update RTT variables. See RFC 6298.
 337 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 338         if(!rtt) {
 339                 debug("invalid rtt\n");
 340                 return;
 341         }
 342
 343         struct utcp *utcp = c->utcp;
 344
 345         if(!utcp->srtt) {
 346                 utcp->srtt = rtt;
 347                 utcp->rttvar = rtt / 2;
 348                 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
 349         } else {
 350                 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
 351                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 352                 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
 353         }
 354
 355         if(utcp->rto > MAX_RTO)
 356                 utcp->rto = MAX_RTO;
 357
 358         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 359 }
 360
 361 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 362         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 363         if(!c)
 364                 return NULL;
 365
 366         c->recv = recv;
 367         c->priv = priv;
 368
 369         struct hdr hdr;
 370
 371         hdr.src = c->src;
 372         hdr.dst = c->dst;
 373         hdr.seq = c->snd.iss;
 374         hdr.ack = 0;
 375         hdr.wnd = c->rcv.wnd;
 376         hdr.ctl = SYN;
 377         hdr.aux = 0;
 378
 379         set_state(c, SYN_SENT);
 380
 381         print_packet(utcp, "send", &hdr, sizeof hdr);
 382         utcp->send(utcp, &hdr, sizeof hdr);
 383
 384         gettimeofday(&c->conn_timeout, NULL);
 385         c->conn_timeout.tv_sec += utcp->timeout;
 386
 387         return c;
 388 }
 389
 390 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 391         if(c->reapable || c->state != SYN_RECEIVED) {
 392                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 393                 return;
 394         }
 395
 396         debug("%p accepted, %p %p\n", c, recv, priv);
 397         c->recv = recv;
 398         c->priv = priv;
 399         set_state(c, ESTABLISHED);
 400 }
 401
 402 static void ack(struct utcp_connection *c, bool sendatleastone) {
 403         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 404         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 405         debug("cwndleft = %d\n", cwndleft);
 406
 407         assert(left >= 0);
 408
 409         if(cwndleft <= 0)
 410                 cwndleft = 0;
 411
 412         if(cwndleft < left)
 413                 left = cwndleft;
 414
 415         if(!left && !sendatleastone)
 416                 return;
 417
 418         struct {
 419                 struct hdr hdr;
 420                 char data[];
 421         } *pkt;
 422
 423         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 424         if(!pkt)
 425                 return;
 426
 427         pkt->hdr.src = c->src;
 428         pkt->hdr.dst = c->dst;
 429         pkt->hdr.ack = c->rcv.nxt;
 430         pkt->hdr.wnd = c->snd.wnd;
 431         pkt->hdr.ctl = ACK;
 432         pkt->hdr.aux = 0;
 433
 434         do {
 435                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 436                 pkt->hdr.seq = c->snd.nxt;
 437
 438                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 439
 440                 c->snd.nxt += seglen;
 441                 left -= seglen;
 442
 443                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 444                         seglen--;
 445                         pkt->hdr.ctl |= FIN;
 446                 }
 447
 448                 if(!c->rtt_start.tv_sec) {
 449                         // Start RTT measurement
 450                         gettimeofday(&c->rtt_start, NULL);
 451                         c->rtt_seq = pkt->hdr.seq + seglen;
 452                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 453                 }
 454
 455                 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
 456                 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
 457         } while(left);
 458
 459         free(pkt);
 460 }
 461
 462 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 463         if(c->reapable) {
 464                 debug("Error: send() called on closed connection %p\n", c);
 465                 errno = EBADF;
 466                 return -1;
 467         }
 468
 469         switch(c->state) {
 470         case CLOSED:
 471         case LISTEN:
 472         case SYN_SENT:
 473         case SYN_RECEIVED:
 474                 debug("Error: send() called on unconnected connection %p\n", c);
 475                 errno = ENOTCONN;
 476                 return -1;
 477         case ESTABLISHED:
 478         case CLOSE_WAIT:
 479                 break;
 480         case FIN_WAIT_1:
 481         case FIN_WAIT_2:
 482         case CLOSING:
 483         case LAST_ACK:
 484         case TIME_WAIT:
 485                 debug("Error: send() called on closing connection %p\n", c);
 486                 errno = EPIPE;
 487                 return -1;
 488         }
 489
 490         // Add data to send buffer
 491
 492         if(!len)
 493                 return 0;
 494
 495         if(!data) {
 496                 errno = EFAULT;
 497                 return -1;
 498         }
 499
 500         len = buffer_put(&c->sndbuf, data, len);
 501         if(len <= 0) {
 502                 errno = EWOULDBLOCK;
 503                 return 0;
 504         }
 505
 506         c->snd.last += len;
 507         ack(c, false);
 508         return len;
 509 }
 510
 511 static void swap_ports(struct hdr *hdr) {
 512         uint16_t tmp = hdr->src;
 513         hdr->src = hdr->dst;
 514         hdr->dst = tmp;
 515 }
 516
 517 static void retransmit(struct utcp_connection *c) {
 518         if(c->state == CLOSED || c->snd.nxt == c->snd.una)
 519                 return;
 520
 521         struct utcp *utcp = c->utcp;
 522
 523         struct {
 524                 struct hdr hdr;
 525                 char data[];
 526         } *pkt;
 527
 528         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 529         if(!pkt)
 530                 return;
 531
 532         pkt->hdr.src = c->src;
 533         pkt->hdr.dst = c->dst;
 534
 535         switch(c->state) {
 536                 case SYN_SENT:
 537                         // Send our SYN again
 538                         pkt->hdr.seq = c->snd.iss;
 539                         pkt->hdr.ack = 0;
 540                         pkt->hdr.wnd = c->rcv.wnd;
 541                         pkt->hdr.ctl = SYN;
 542                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 543                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 544                         break;
 545
 546                 case SYN_RECEIVED:
 547                         // Send SYNACK again
 548                         pkt->hdr.seq = c->snd.nxt;
 549                         pkt->hdr.ack = c->rcv.nxt;
 550                         pkt->hdr.ctl = SYN | ACK;
 551                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 552                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 553                         break;
 554
 555                 case ESTABLISHED:
 556                 case FIN_WAIT_1:
 557                 case CLOSE_WAIT:
 558                 case CLOSING:
 559                 case LAST_ACK:
 560                         // Send unacked data again.
 561                         pkt->hdr.seq = c->snd.una;
 562                         pkt->hdr.ack = c->rcv.nxt;
 563                         pkt->hdr.ctl = ACK;
 564                         uint32_t len = seqdiff(c->snd.last, c->snd.una);
 565                         if(len > utcp->mtu)
 566                                 len = utcp->mtu;
 567                         if(fin_wanted(c, c->snd.una + len)) {
 568                                 len--;
 569                                 pkt->hdr.ctl |= FIN;
 570                         }
 571                         c->snd.nxt = c->snd.una + len;
 572                         c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 573                         buffer_copy(&c->sndbuf, pkt->data, 0, len);
 574                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
 575                         utcp->send(utcp, pkt, sizeof pkt->hdr + len);
 576                         break;
 577
 578                 case CLOSED:
 579                 case LISTEN:
 580                 case TIME_WAIT:
 581                 case FIN_WAIT_2:
 582                         // We shouldn't need to retransmit anything in this state.
 583 #ifdef UTCP_DEBUG
 584                         abort();
 585 #endif
 586                         timerclear(&c->rtrx_timeout);
 587                         break;
 588         }
 589
 590         utcp->rto *= 2;
 591         if(utcp->rto > MAX_RTO)
 592                 utcp->rto = MAX_RTO;
 593         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 594         free(pkt);
 595 }
 596
 597 // Update receive buffer and SACK entries after consuming data.
 598 static void sack_consume(struct utcp_connection *c, size_t len) {
 599         debug("sack_consume %zu\n", len);
 600         if(len > c->rcvbuf.used)
 601                 abort();
 602
 603         buffer_get(&c->rcvbuf, NULL, len);
 604
 605         for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
 606                 if(len < c->sacks[i].offset) {
 607                         c->sacks[i].offset -= len;
 608                         i++;
 609                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 610                         c->sacks[i].offset = 0;
 611                         c->sacks[i].len -= len - c->sacks[i].offset;
 612                         i++;
 613                 } else {
 614                         if(i < NSACKS - 1) {
 615                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
 616                                 c->sacks[i + 1].len = 0;
 617                         } else {
 618                                 c->sacks[i].len = 0;
 619                                 break;
 620                         }
 621                 }
 622         }
 623
 624         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 625                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 626 }
 627
 628 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 629         debug("out of order packet, offset %u\n", offset);
 630         // Packet loss or reordering occured. Store the data in the buffer.
 631         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 632         if(rxd < len)
 633                 abort();
 634
 635         // Make note of where we put it.
 636         for(int i = 0; i < NSACKS; i++) {
 637                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 638                         debug("New SACK entry %d\n", i);
 639                         c->sacks[i].offset = offset;
 640                         c->sacks[i].len = rxd;
 641                         break;
 642                 } else if(offset < c->sacks[i].offset) {
 643                         if(offset + rxd < c->sacks[i].offset) { // insert before
 644                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 645                                         debug("Insert SACK entry at %d\n", i);
 646                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
 647                                         c->sacks[i].offset = offset;
 648                                         c->sacks[i].len = rxd;
 649                                 }
 650                                 break;
 651                         } else { // merge
 652                                 debug("Merge with start of SACK entry at %d\n", i);
 653                                 c->sacks[i].offset = offset;
 654                                 break;
 655                         }
 656                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 657                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 658                                 debug("Merge with end of SACK entry at %d\n", i);
 659                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 660                                 // TODO: handle potential merge with next entry
 661                         }
 662                         break;
 663                 }
 664         }
 665
 666         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 667                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 668 }
 669
 670 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 671         // Check if we can process out-of-order data now.
 672         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 673                 debug("incoming packet len %zu connected with SACK at %u\n", len, c->sacks[0].offset);
 674                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 675                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 676                 data = c->rcvbuf.data;
 677         }
 678
 679         if(c->recv) {
 680                 ssize_t rxd = c->recv(c, data, len);
 681                 if(rxd != len) {
 682                         // TODO: handle the application not accepting all data.
 683                         abort();
 684                 }
 685         }
 686
 687         if(c->rcvbuf.used)
 688                 sack_consume(c, len);
 689
 690         c->rcv.nxt += len;
 691 }
 692
 693
 694 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 695         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 696         if(offset + len > c->rcvbuf.maxsize)
 697                 abort();
 698
 699         if(offset)
 700                 handle_out_of_order(c, offset, data, len);
 701         else
 702                 handle_in_order(c, data, len);
 703 }
 704
 705
 706 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 707         if(!utcp) {
 708                 errno = EFAULT;
 709                 return -1;
 710         }
 711
 712         if(!len)
 713                 return 0;
 714
 715         if(!data) {
 716                 errno = EFAULT;
 717                 return -1;
 718         }
 719
 720         print_packet(utcp, "recv", data, len);
 721
 722         // Drop packets smaller than the header
 723
 724         struct hdr hdr;
 725         if(len < sizeof hdr) {
 726                 errno = EBADMSG;
 727                 return -1;
 728         }
 729
 730         // Make a copy from the potentially unaligned data to a struct hdr
 731
 732         memcpy(&hdr, data, sizeof hdr);
 733         data += sizeof hdr;
 734         len -= sizeof hdr;
 735
 736         // Drop packets with an unknown CTL flag
 737
 738         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 739                 errno = EBADMSG;
 740                 return -1;
 741         }
 742
 743         // Try to match the packet to an existing connection
 744
 745         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
 746
 747         // Is it for a new connection?
 748
 749         if(!c) {
 750                 // Ignore RST packets
 751
 752                 if(hdr.ctl & RST)
 753                         return 0;
 754
 755                 // Is it a SYN packet and are we LISTENing?
 756
 757                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
 758                         // If we don't want to accept it, send a RST back
 759                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
 760                                 len = 1;
 761                                 goto reset;
 762                         }
 763
 764                         // Try to allocate memory, otherwise send a RST back
 765                         c = allocate_connection(utcp, hdr.dst, hdr.src);
 766                         if(!c) {
 767                                 len = 1;
 768                                 goto reset;
 769                         }
 770
 771                         // Return SYN+ACK, go to SYN_RECEIVED state
 772                         c->snd.wnd = hdr.wnd;
 773                         c->rcv.irs = hdr.seq;
 774                         c->rcv.nxt = c->rcv.irs + 1;
 775                         set_state(c, SYN_RECEIVED);
 776
 777                         hdr.dst = c->dst;
 778                         hdr.src = c->src;
 779                         hdr.ack = c->rcv.irs + 1;
 780                         hdr.seq = c->snd.iss;
 781                         hdr.ctl = SYN | ACK;
 782                         print_packet(c->utcp, "send", &hdr, sizeof hdr);
 783                         utcp->send(utcp, &hdr, sizeof hdr);
 784                 } else {
 785                         // No, we don't want your packets, send a RST back
 786                         len = 1;
 787                         goto reset;
 788                 }
 789
 790                 return 0;
 791         }
 792
 793         debug("%p state %s\n", c->utcp, strstate[c->state]);
 794
 795         // In case this is for a CLOSED connection, ignore the packet.
 796         // TODO: make it so incoming packets can never match a CLOSED connection.
 797
 798         if(c->state == CLOSED)
 799                 return 0;
 800
 801         // It is for an existing connection.
 802
 803         uint32_t prevrcvnxt = c->rcv.nxt;
 804
 805         // 1. Drop invalid packets.
 806
 807         // 1a. Drop packets that should not happen in our current state.
 808
 809         switch(c->state) {
 810         case SYN_SENT:
 811         case SYN_RECEIVED:
 812         case ESTABLISHED:
 813         case FIN_WAIT_1:
 814         case FIN_WAIT_2:
 815         case CLOSE_WAIT:
 816         case CLOSING:
 817         case LAST_ACK:
 818         case TIME_WAIT:
 819                 break;
 820         default:
 821 #ifdef UTCP_DEBUG
 822                 abort();
 823 #endif
 824                 break;
 825         }
 826
 827         // 1b. Drop packets with a sequence number not in our receive window.
 828
 829         bool acceptable;
 830
 831         if(c->state == SYN_SENT)
 832                 acceptable = true;
 833
 834         // TODO: handle packets overlapping c->rcv.nxt.
 835 #if 1
 836         // Only use this when accepting out-of-order packets.
 837         else if(len == 0)
 838                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
 839         else
 840                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
 841 #else
 842         if(c->state != SYN_SENT)
 843                 acceptable = hdr.seq == c->rcv.nxt;
 844 #endif
 845
 846         if(!acceptable) {
 847                 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
 848                 // Ignore unacceptable RST packets.
 849                 if(hdr.ctl & RST)
 850                         return 0;
 851                 // Otherwise, send an ACK back in the hope things improve.
 852                 ack(c, true);
 853                 return 0;
 854         }
 855
 856         c->snd.wnd = hdr.wnd; // TODO: move below
 857
 858         // 1c. Drop packets with an invalid ACK.
 859         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
 860         // (= snd.una + c->sndbuf.used).
 861
 862         if(hdr.ctl & ACK &&
 863                         ((seqdiff(hdr.ack, c->snd.una + c->sndbuf.used) > 0 &&
 864                           seqdiff(hdr.ack, c->snd.nxt) > 0) // TODO: simplify this if
 865                          || seqdiff(hdr.ack, c->snd.una) < 0)) {
 866                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 867                 // Ignore unacceptable RST packets.
 868                 if(hdr.ctl & RST)
 869                         return 0;
 870                 goto reset;
 871         }
 872
 873         // 2. Handle RST packets
 874
 875         if(hdr.ctl & RST) {
 876                 switch(c->state) {
 877                 case SYN_SENT:
 878                         if(!(hdr.ctl & ACK))
 879                                 return 0;
 880                         // The peer has refused our connection.
 881                         set_state(c, CLOSED);
 882                         errno = ECONNREFUSED;
 883                         if(c->recv)
 884                                 c->recv(c, NULL, 0);
 885                         return 0;
 886                 case SYN_RECEIVED:
 887                         if(hdr.ctl & ACK)
 888                                 return 0;
 889                         // We haven't told the application about this connection yet. Silently delete.
 890                         free_connection(c);
 891                         return 0;
 892                 case ESTABLISHED:
 893                 case FIN_WAIT_1:
 894                 case FIN_WAIT_2:
 895                 case CLOSE_WAIT:
 896                         if(hdr.ctl & ACK)
 897                                 return 0;
 898                         // The peer has aborted our connection.
 899                         set_state(c, CLOSED);
 900                         errno = ECONNRESET;
 901                         if(c->recv)
 902                                 c->recv(c, NULL, 0);
 903                         return 0;
 904                 case CLOSING:
 905                 case LAST_ACK:
 906                 case TIME_WAIT:
 907                         if(hdr.ctl & ACK)
 908                                 return 0;
 909                         // As far as the application is concerned, the connection has already been closed.
 910                         // If it has called utcp_close() already, we can immediately free this connection.
 911                         if(c->reapable) {
 912                                 free_connection(c);
 913                                 return 0;
 914                         }
 915                         // Otherwise, immediately move to the CLOSED state.
 916                         set_state(c, CLOSED);
 917                         return 0;
 918                 default:
 919 #ifdef UTCP_DEBUG
 920                         abort();
 921 #endif
 922                         break;
 923                 }
 924         }
 925
 926         // 3. Advance snd.una
 927
 928         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
 929         prevrcvnxt = c->rcv.nxt;
 930
 931         if(advanced) {
 932                 // RTT measurement
 933                 if(c->rtt_start.tv_sec) {
 934                         if(c->rtt_seq == hdr.ack) {
 935                                 struct timeval now, diff;
 936                                 gettimeofday(&now, NULL);
 937                                 timersub(&now, &c->rtt_start, &diff);
 938                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
 939                                 c->rtt_start.tv_sec = 0;
 940                         } else if(c->rtt_seq < hdr.ack) {
 941                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
 942                                 c->rtt_start.tv_sec = 0;
 943                         }
 944                 }
 945
 946                 int32_t data_acked = advanced;
 947
 948                 switch(c->state) {
 949                         case SYN_SENT:
 950                         case SYN_RECEIVED:
 951                                 data_acked--;
 952                                 break;
 953                         // TODO: handle FIN as well.
 954                         default:
 955                                 break;
 956                 }
 957
 958                 assert(data_acked >= 0);
 959
 960                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
 961                 assert(data_acked <= bufused);
 962
 963                 if(data_acked)
 964                         buffer_get(&c->sndbuf, NULL, data_acked);
 965
 966                 // Also advance snd.nxt if possible
 967                 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
 968                         c->snd.nxt = hdr.ack;
 969
 970                 c->snd.una = hdr.ack;
 971
 972                 c->dupack = 0;
 973                 c->snd.cwnd += utcp->mtu;
 974                 if(c->snd.cwnd > c->sndbuf.maxsize)
 975                         c->snd.cwnd = c->sndbuf.maxsize;
 976
 977                 // Check if we have sent a FIN that is now ACKed.
 978                 switch(c->state) {
 979                 case FIN_WAIT_1:
 980                         if(c->snd.una == c->snd.last)
 981                                 set_state(c, FIN_WAIT_2);
 982                         break;
 983                 case CLOSING:
 984                         if(c->snd.una == c->snd.last) {
 985                                 gettimeofday(&c->conn_timeout, NULL);
 986                                 c->conn_timeout.tv_sec += 60;
 987                                 set_state(c, TIME_WAIT);
 988                         }
 989                         break;
 990                 default:
 991                         break;
 992                 }
 993         } else {
 994                 if(!len) {
 995                         c->dupack++;
 996                         if(c->dupack == 3) {
 997                                 debug("Triplicate ACK\n");
 998                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
 999                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1000                                 //Reset the congestion window so we wait for ACKs.
1001                                 c->snd.nxt = c->snd.una;
1002                                 c->snd.cwnd = utcp->mtu;
1003                         }
1004                 }
1005         }
1006
1007         // 4. Update timers
1008
1009         if(advanced) {
1010                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1011                 if(c->snd.una == c->snd.nxt)
1012                         timerclear(&c->rtrx_timeout);
1013         }
1014
1015         // 5. Process SYN stuff
1016
1017         if(hdr.ctl & SYN) {
1018                 switch(c->state) {
1019                 case SYN_SENT:
1020                         // This is a SYNACK. It should always have ACKed the SYN.
1021                         if(!advanced)
1022                                 goto reset;
1023                         c->rcv.irs = hdr.seq;
1024                         c->rcv.nxt = hdr.seq;
1025                         set_state(c, ESTABLISHED);
1026                         // TODO: notify application of this somehow.
1027                         break;
1028                 case SYN_RECEIVED:
1029                 case ESTABLISHED:
1030                 case FIN_WAIT_1:
1031                 case FIN_WAIT_2:
1032                 case CLOSE_WAIT:
1033                 case CLOSING:
1034                 case LAST_ACK:
1035                 case TIME_WAIT:
1036                         // Ehm, no. We should never receive a second SYN.
1037                         goto reset;
1038                 default:
1039 #ifdef UTCP_DEBUG
1040                         abort();
1041 #endif
1042                         return 0;
1043                 }
1044
1045                 // SYN counts as one sequence number
1046                 c->rcv.nxt++;
1047         }
1048
1049         // 6. Process new data
1050
1051         if(c->state == SYN_RECEIVED) {
1052                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1053                 if(!advanced)
1054                         goto reset;
1055
1056                 // Are we still LISTENing?
1057                 if(utcp->accept)
1058                         utcp->accept(c, c->src);
1059
1060                 if(c->state != ESTABLISHED) {
1061                         set_state(c, CLOSED);
1062                         c->reapable = true;
1063                         goto reset;
1064                 }
1065         }
1066
1067         if(len) {
1068                 switch(c->state) {
1069                 case SYN_SENT:
1070                 case SYN_RECEIVED:
1071                         // This should never happen.
1072 #ifdef UTCP_DEBUG
1073                         abort();
1074 #endif
1075                         return 0;
1076                 case ESTABLISHED:
1077                 case FIN_WAIT_1:
1078                 case FIN_WAIT_2:
1079                         break;
1080                 case CLOSE_WAIT:
1081                 case CLOSING:
1082                 case LAST_ACK:
1083                 case TIME_WAIT:
1084                         // Ehm no, We should never receive more data after a FIN.
1085                         goto reset;
1086                 default:
1087 #ifdef UTCP_DEBUG
1088                         abort();
1089 #endif
1090                         return 0;
1091                 }
1092
1093                 handle_incoming_data(c, hdr.seq, data, len);
1094         }
1095
1096         // 7. Process FIN stuff
1097
1098         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1099                 switch(c->state) {
1100                 case SYN_SENT:
1101                 case SYN_RECEIVED:
1102                         // This should never happen.
1103 #ifdef UTCP_DEBUG
1104                         abort();
1105 #endif
1106                         break;
1107                 case ESTABLISHED:
1108                         set_state(c, CLOSE_WAIT);
1109                         break;
1110                 case FIN_WAIT_1:
1111                         set_state(c, CLOSING);
1112                         break;
1113                 case FIN_WAIT_2:
1114                         gettimeofday(&c->conn_timeout, NULL);
1115                         c->conn_timeout.tv_sec += 60;
1116                         set_state(c, TIME_WAIT);
1117                         break;
1118                 case CLOSE_WAIT:
1119                 case CLOSING:
1120                 case LAST_ACK:
1121                 case TIME_WAIT:
1122                         // Ehm, no. We should never receive a second FIN.
1123                         goto reset;
1124                 default:
1125 #ifdef UTCP_DEBUG
1126                         abort();
1127 #endif
1128                         break;
1129                 }
1130
1131                 // FIN counts as one sequence number
1132                 c->rcv.nxt++;
1133                 len++;
1134
1135                 // Inform the application that the peer closed the connection.
1136                 if(c->recv) {
1137                         errno = 0;
1138                         c->recv(c, NULL, 0);
1139                 }
1140         }
1141
1142         // Now we send something back if:
1143         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1144         //   -> sendatleastone = true
1145         // - or we got an ack, so we should maybe send a bit more data
1146         //   -> sendatleastone = false
1147
1148 ack:
1149         ack(c, prevrcvnxt != c->rcv.nxt);
1150         return 0;
1151
1152 reset:
1153         swap_ports(&hdr);
1154         hdr.wnd = 0;
1155         if(hdr.ctl & ACK) {
1156                 hdr.seq = hdr.ack;
1157                 hdr.ctl = RST;
1158         } else {
1159                 hdr.ack = hdr.seq + len;
1160                 hdr.seq = 0;
1161                 hdr.ctl = RST | ACK;
1162         }
1163         print_packet(utcp, "send", &hdr, sizeof hdr);
1164         utcp->send(utcp, &hdr, sizeof hdr);
1165         return 0;
1166
1167 }
1168
1169 int utcp_shutdown(struct utcp_connection *c, int dir) {
1170         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1171         if(!c) {
1172                 errno = EFAULT;
1173                 return -1;
1174         }
1175
1176         if(c->reapable) {
1177                 debug("Error: shutdown() called on closed connection %p\n", c);
1178                 errno = EBADF;
1179                 return -1;
1180         }
1181
1182         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1183                 errno = EINVAL;
1184                 return -1;
1185         }
1186
1187         // TCP does not have a provision for stopping incoming packets.
1188         // The best we can do is to just ignore them.
1189         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1190                 c->recv = NULL;
1191
1192         // The rest of the code deals with shutting down writes.
1193         if(dir == UTCP_SHUT_RD)
1194                 return 0;
1195
1196         switch(c->state) {
1197         case CLOSED:
1198         case LISTEN:
1199                 errno = ENOTCONN;
1200                 return -1;
1201
1202         case SYN_SENT:
1203                 set_state(c, CLOSED);
1204                 return 0;
1205
1206         case SYN_RECEIVED:
1207         case ESTABLISHED:
1208                 set_state(c, FIN_WAIT_1);
1209                 break;
1210         case FIN_WAIT_1:
1211         case FIN_WAIT_2:
1212                 return 0;
1213         case CLOSE_WAIT:
1214                 set_state(c, CLOSING);
1215                 break;
1216
1217         case CLOSING:
1218         case LAST_ACK:
1219         case TIME_WAIT:
1220                 return 0;
1221         }
1222
1223         c->snd.last++;
1224
1225         ack(c, false);
1226         return 0;
1227 }
1228
1229 int utcp_close(struct utcp_connection *c) {
1230         if(utcp_shutdown(c, SHUT_RDWR))
1231                 return -1;
1232         c->recv = NULL;
1233         c->poll = NULL;
1234         c->reapable = true;
1235         return 0;
1236 }
1237
1238 int utcp_abort(struct utcp_connection *c) {
1239         if(!c) {
1240                 errno = EFAULT;
1241                 return -1;
1242         }
1243
1244         if(c->reapable) {
1245                 debug("Error: abort() called on closed connection %p\n", c);
1246                 errno = EBADF;
1247                 return -1;
1248         }
1249
1250         c->recv = NULL;
1251         c->poll = NULL;
1252         c->reapable = true;
1253
1254         switch(c->state) {
1255         case CLOSED:
1256                 return 0;
1257         case LISTEN:
1258         case SYN_SENT:
1259         case CLOSING:
1260         case LAST_ACK:
1261         case TIME_WAIT:
1262                 set_state(c, CLOSED);
1263                 return 0;
1264
1265         case SYN_RECEIVED:
1266         case ESTABLISHED:
1267         case FIN_WAIT_1:
1268         case FIN_WAIT_2:
1269         case CLOSE_WAIT:
1270                 set_state(c, CLOSED);
1271                 break;
1272         }
1273
1274         // Send RST
1275
1276         struct hdr hdr;
1277
1278         hdr.src = c->src;
1279         hdr.dst = c->dst;
1280         hdr.seq = c->snd.nxt;
1281         hdr.ack = 0;
1282         hdr.wnd = 0;
1283         hdr.ctl = RST;
1284
1285         print_packet(c->utcp, "send", &hdr, sizeof hdr);
1286         c->utcp->send(c->utcp, &hdr, sizeof hdr);
1287         return 0;
1288 }
1289
1290 /* Handle timeouts.
1291  * One call to this function will loop through all connections,
1292  * checking if something needs to be resent or not.
1293  * The return value is the time to the next timeout in milliseconds,
1294  * or maybe a negative value if the timeout is infinite.
1295  */
1296 struct timeval utcp_timeout(struct utcp *utcp) {
1297         struct timeval now;
1298         gettimeofday(&now, NULL);
1299         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1300
1301         for(int i = 0; i < utcp->nconnections; i++) {
1302                 struct utcp_connection *c = utcp->connections[i];
1303                 if(!c)
1304                         continue;
1305
1306                 if(c->state == CLOSED) {
1307                         if(c->reapable) {
1308                                 debug("Reaping %p\n", c);
1309                                 free_connection(c);
1310                                 i--;
1311                         }
1312                         continue;
1313                 }
1314
1315                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1316                         errno = ETIMEDOUT;
1317                         c->state = CLOSED;
1318                         if(c->recv)
1319                                 c->recv(c, NULL, 0);
1320                         continue;
1321                 }
1322
1323                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1324                         retransmit(c);
1325                 }
1326
1327                 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1328                         c->poll(c, buffer_free(&c->sndbuf));
1329
1330                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1331                         next = c->conn_timeout;
1332
1333                 if(c->snd.nxt != c->snd.una) {
1334                         c->rtrx_timeout = now;
1335                         c->rtrx_timeout.tv_sec++;
1336                 } else {
1337                         timerclear(&c->rtrx_timeout);
1338                 }
1339
1340                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1341                         next = c->rtrx_timeout;
1342         }
1343
1344         struct timeval diff;
1345         timersub(&next, &now, &diff);
1346         return diff;
1347 }
1348
1349 bool utcp_is_active(struct utcp *utcp) {
1350         if(!utcp)
1351                 return false;
1352
1353         for(int i = 0; i < utcp->nconnections; i++)
1354                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1355                         return true;
1356
1357         return false;
1358 }
1359
1360 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1361         if(!send) {
1362                 errno = EFAULT;
1363                 return NULL;
1364         }
1365
1366         struct utcp *utcp = calloc(1, sizeof *utcp);
1367         if(!utcp)
1368                 return NULL;
1369
1370         utcp->accept = accept;
1371         utcp->pre_accept = pre_accept;
1372         utcp->send = send;
1373         utcp->priv = priv;
1374         utcp->mtu = DEFAULT_MTU;
1375         utcp->timeout = DEFAULT_USER_TIMEOUT; // s
1376         utcp->rto = START_RTO; // us
1377
1378         return utcp;
1379 }
1380
1381 void utcp_exit(struct utcp *utcp) {
1382         if(!utcp)
1383                 return;
1384         for(int i = 0; i < utcp->nconnections; i++) {
1385                 if(!utcp->connections[i]->reapable)
1386                         debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1387                 buffer_exit(&utcp->connections[i]->sndbuf);
1388                 free(utcp->connections[i]);
1389         }
1390         free(utcp->connections);
1391         free(utcp);
1392 }
1393
1394 uint16_t utcp_get_mtu(struct utcp *utcp) {
1395         return utcp ? utcp->mtu : 0;
1396 }
1397
1398 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1399         // TODO: handle overhead of the header
1400         if(utcp)
1401                 utcp->mtu = mtu;
1402 }
1403
1404 int utcp_get_user_timeout(struct utcp *u) {
1405         return u ? u->timeout : 0;
1406 }
1407
1408 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1409         if(u)
1410                 u->timeout = timeout;
1411 }
1412
1413 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1414         return c ? c->sndbuf.maxsize : 0;
1415 }
1416
1417 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1418         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1419                 return buffer_free(&c->sndbuf);
1420         else
1421                 return 0;
1422 }
1423
1424 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1425         if(!c)
1426                 return;
1427         c->sndbuf.maxsize = size;
1428         if(c->sndbuf.maxsize != size)
1429                 c->sndbuf.maxsize = -1;
1430 }
1431
1432 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1433         return c ? c->rcvbuf.maxsize : 0;
1434 }
1435
1436 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1437         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1438                 return buffer_free(&c->rcvbuf);
1439         else
1440                 return 0;
1441 }
1442
1443 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1444         if(!c)
1445                 return;
1446         c->rcvbuf.maxsize = size;
1447         if(c->rcvbuf.maxsize != size)
1448                 c->rcvbuf.maxsize = -1;
1449 }
1450
1451 bool utcp_get_nodelay(struct utcp_connection *c) {
1452         return c ? c->nodelay : false;
1453 }
1454
1455 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1456         if(c)
1457                 c->nodelay = nodelay;
1458 }
1459
1460 bool utcp_get_keepalive(struct utcp_connection *c) {
1461         return c ? c->keepalive : false;
1462 }
1463
1464 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1465         if(c)
1466                 c->keepalive = keepalive;
1467 }
1468
1469 size_t utcp_get_outq(struct utcp_connection *c) {
1470         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1471 }
1472
1473 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1474         if(c)
1475                 c->recv = recv;
1476 }
1477
1478 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1479         if(c)
1480                 c->poll = poll;
1481 }
1482
1483 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1484         if(utcp) {
1485                 utcp->accept = accept;
1486                 utcp->pre_accept = pre_accept;
1487         }
1488 }