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 static inline size_t max(size_t a, size_t b) {
  57         return a > b ? a : b;
  58 }
  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                 uint32_t datalen = len - sizeof hdr;
  90                 uint8_t *str = malloc((datalen << 1) + 7);
  91                 if(!str) {
  92                         debug("out of memory");
  93                         return;
  94                 }
  95                 memcpy(str, " data=", 6);
  96                 uint8_t *strptr = str + 6;
  97                 const uint8_t *data = pkt;
  98                 const uint8_t *dataend = data + datalen;
  99
 100                 while(data != dataend) {
 101                         *strptr = (*data >> 4) > 9? (*data >> 4) + 55 : (*data >> 4) + 48;
 102                         ++strptr;
 103                         *strptr = (*data & 0xf) > 9? (*data & 0xf) + 55 : (*data & 0xf) + 48;
 104                         ++strptr;
 105                         ++data;
 106                 }
 107                 *strptr = 0;
 108
 109                 debug(str);
 110                 free(str);
 111         }
 112
 113         debug("\n");
 114 }
 115 #else
 116 #define debug(...)
 117 #define print_packet(...)
 118 #endif
 119
 120 static void set_state(struct utcp_connection *c, enum state state) {
 121         c->state = state;
 122         if(state == ESTABLISHED)
 123                 timerclear(&c->conn_timeout);
 124         debug("%p new state: %s\n", c->utcp, strstate[state]);
 125 }
 126
 127 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 128         if(seq != c->snd.last)
 129                 return false;
 130         switch(c->state) {
 131         case FIN_WAIT_1:
 132         case CLOSING:
 133         case LAST_ACK:
 134                 return true;
 135         default:
 136                 return false;
 137         }
 138 }
 139
 140 static inline void list_connections(struct utcp *utcp) {
 141         debug("%p has %d connections:\n", utcp, utcp->nconnections);
 142         for(int i = 0; i < utcp->nconnections; i++)
 143                 debug("  %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
 144 }
 145
 146 static int32_t seqdiff(uint32_t a, uint32_t b) {
 147         return a - b;
 148 }
 149
 150 // Buffer functions
 151 // TODO: convert to ringbuffers to avoid memmove() operations.
 152
 153 // Store data into the buffer
 154 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 155         if(buf->maxsize <= buf->used)
 156                 return 0;
 157
 158         debug("buffer_put_at %zu %zu %zu\n", buf->used, offset, len);
 159
 160         size_t required = offset + len;
 161         if(required > buf->maxsize) {
 162                 if(offset >= buf->maxsize)
 163                         return 0;
 164                 abort();
 165                 len = buf->maxsize - offset;
 166                 required = buf->maxsize;
 167         }
 168
 169         if(required > buf->size) {
 170                 size_t newsize = buf->size;
 171                 if(!newsize) {
 172                         newsize = required;
 173                 } else {
 174                         do {
 175                                 newsize *= 2;
 176                         } while(newsize < buf->used + len);
 177                 }
 178                 if(newsize > buf->maxsize)
 179                         newsize = buf->maxsize;
 180                 char *newdata = realloc(buf->data, newsize);
 181                 if(!newdata)
 182                         return -1;
 183                 buf->data = newdata;
 184                 buf->size = newsize;
 185         }
 186
 187         memcpy(buf->data + offset, data, len);
 188         if(required > buf->used)
 189                 buf->used = required;
 190         return len;
 191 }
 192
 193 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 194         return buffer_put_at(buf, buf->used, data, len);
 195 }
 196
 197 // Get data from the buffer. data can be NULL.
 198 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 199         if(len > buf->used)
 200                 len = buf->used;
 201         if(data)
 202                 memcpy(data, buf->data, len);
 203         if(len < buf->used)
 204                 memmove(buf->data, buf->data + len, buf->used - len);
 205         buf->used -= len;
 206         return len;
 207 }
 208
 209 // Copy data from the buffer without removing it.
 210 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 211         if(offset >= buf->used)
 212                 return 0;
 213         if(offset + len > buf->used)
 214                 len = buf->used - offset;
 215         memcpy(data, buf->data + offset, len);
 216         return len;
 217 }
 218
 219 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 220         memset(buf, 0, sizeof *buf);
 221         if(len) {
 222                 buf->data = malloc(len);
 223                 if(!buf->data)
 224                         return false;
 225         }
 226         buf->size = len;
 227         buf->maxsize = maxlen;
 228         return true;
 229 }
 230
 231 static void buffer_exit(struct buffer *buf) {
 232         free(buf->data);
 233         memset(buf, 0, sizeof *buf);
 234 }
 235
 236 static uint32_t buffer_free(const struct buffer *buf) {
 237         return buf->maxsize - buf->used;
 238 }
 239
 240 // Connections are stored in a sorted list.
 241 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 242
 243 static int compare(const void *va, const void *vb) {
 244         assert(va && vb);
 245
 246         const struct utcp_connection *a = *(struct utcp_connection **)va;
 247         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 248
 249         assert(a && b);
 250         assert(a->src && b->src);
 251
 252         int c = (int)a->src - (int)b->src;
 253         if(c)
 254                 return c;
 255         c = (int)a->dst - (int)b->dst;
 256         return c;
 257 }
 258
 259 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 260         if(!utcp->nconnections)
 261                 return NULL;
 262         struct utcp_connection key = {
 263                 .src = src,
 264                 .dst = dst,
 265         }, *keyp = &key;
 266         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 267         return match ? *match : NULL;
 268 }
 269
 270 static void free_connection(struct utcp_connection *c) {
 271         struct utcp *utcp = c->utcp;
 272         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 273
 274         assert(cp);
 275
 276         int i = cp - utcp->connections;
 277         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
 278         utcp->nconnections--;
 279
 280         buffer_exit(&c->rcvbuf);
 281         buffer_exit(&c->sndbuf);
 282         free(c);
 283 }
 284
 285 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 286         // Check whether this combination of src and dst is free
 287
 288         if(src) {
 289                 if(find_connection(utcp, src, dst)) {
 290                         errno = EADDRINUSE;
 291                         return NULL;
 292                 }
 293         } else { // If src == 0, generate a random port number with the high bit set
 294                 if(utcp->nconnections >= 32767) {
 295                         errno = ENOMEM;
 296                         return NULL;
 297                 }
 298                 src = rand() | 0x8000;
 299                 while(find_connection(utcp, src, dst))
 300                         src++;
 301         }
 302
 303         // Allocate memory for the new connection
 304
 305         if(utcp->nconnections >= utcp->nallocated) {
 306                 if(!utcp->nallocated)
 307                         utcp->nallocated = 4;
 308                 else
 309                         utcp->nallocated *= 2;
 310                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
 311                 if(!new_array)
 312                         return NULL;
 313                 utcp->connections = new_array;
 314         }
 315
 316         struct utcp_connection *c = calloc(1, sizeof *c);
 317         if(!c)
 318                 return NULL;
 319
 320         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 321                 free(c);
 322                 return NULL;
 323         }
 324
 325         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 326                 buffer_exit(&c->sndbuf);
 327                 free(c);
 328                 return NULL;
 329         }
 330
 331         // Fill in the details
 332
 333         c->src = src;
 334         c->dst = dst;
 335 #ifdef UTCP_DEBUG
 336         c->snd.iss = 0;
 337 #else
 338         c->snd.iss = rand();
 339 #endif
 340         c->snd.una = c->snd.iss;
 341         c->snd.nxt = c->snd.iss + 1;
 342         c->rcv.wnd = utcp->mtu;
 343         c->snd.last = c->snd.nxt;
 344         c->snd.cwnd = utcp->mtu;
 345         c->utcp = utcp;
 346
 347         // Add it to the sorted list of connections
 348
 349         utcp->connections[utcp->nconnections++] = c;
 350         qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 351
 352         return c;
 353 }
 354
 355 // Update RTT variables. See RFC 6298.
 356 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 357         if(!rtt) {
 358                 debug("invalid rtt\n");
 359                 return;
 360         }
 361
 362         struct utcp *utcp = c->utcp;
 363
 364         if(!utcp->srtt) {
 365                 utcp->srtt = rtt;
 366                 utcp->rttvar = rtt / 2;
 367                 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
 368         } else {
 369                 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
 370                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 371                 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
 372         }
 373
 374         if(utcp->rto > MAX_RTO)
 375                 utcp->rto = MAX_RTO;
 376
 377         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 378 }
 379
 380 static void start_retransmit_timer(struct utcp_connection *c) {
 381         gettimeofday(&c->rtrx_timeout, NULL);
 382         c->rtrx_timeout.tv_usec += c->utcp->rto;
 383         while(c->rtrx_timeout.tv_usec >= 1000000) {
 384                 c->rtrx_timeout.tv_usec -= 1000000;
 385                 c->rtrx_timeout.tv_sec++;
 386         }
 387         debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
 388 }
 389
 390 static void stop_retransmit_timer(struct utcp_connection *c) {
 391         timerclear(&c->rtrx_timeout);
 392         debug("timeout cleared\n");
 393 }
 394
 395 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 396         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 397         if(!c)
 398                 return NULL;
 399
 400         c->recv = recv;
 401         c->priv = priv;
 402
 403         struct hdr hdr;
 404
 405         hdr.src = c->src;
 406         hdr.dst = c->dst;
 407         hdr.seq = c->snd.iss;
 408         hdr.ack = 0;
 409         hdr.wnd = c->rcv.wnd;
 410         hdr.ctl = SYN;
 411         hdr.aux = 0;
 412
 413         set_state(c, SYN_SENT);
 414
 415         print_packet(utcp, "send", &hdr, sizeof hdr);
 416         utcp->send(utcp, &hdr, sizeof hdr);
 417
 418         gettimeofday(&c->conn_timeout, NULL);
 419         c->conn_timeout.tv_sec += utcp->timeout;
 420
 421         return c;
 422 }
 423
 424 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 425         if(c->reapable || c->state != SYN_RECEIVED) {
 426                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 427                 return;
 428         }
 429
 430         debug("%p accepted, %p %p\n", c, recv, priv);
 431         c->recv = recv;
 432         c->priv = priv;
 433         set_state(c, ESTABLISHED);
 434 }
 435
 436 static void ack(struct utcp_connection *c, bool sendatleastone) {
 437         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 438         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 439         debug("cwndleft = %d\n", cwndleft);
 440
 441         assert(left >= 0);
 442
 443         if(cwndleft <= 0)
 444                 cwndleft = 0;
 445
 446         if(cwndleft < left)
 447                 left = cwndleft;
 448
 449         if(!left && !sendatleastone)
 450                 return;
 451
 452         struct {
 453                 struct hdr hdr;
 454                 char data[];
 455         } *pkt;
 456
 457         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 458         if(!pkt)
 459                 return;
 460
 461         pkt->hdr.src = c->src;
 462         pkt->hdr.dst = c->dst;
 463         pkt->hdr.ack = c->rcv.nxt;
 464         pkt->hdr.wnd = c->snd.wnd;
 465         pkt->hdr.ctl = ACK;
 466         pkt->hdr.aux = 0;
 467
 468         do {
 469                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 470                 pkt->hdr.seq = c->snd.nxt;
 471
 472                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 473
 474                 c->snd.nxt += seglen;
 475                 left -= seglen;
 476
 477                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 478                         seglen--;
 479                         pkt->hdr.ctl |= FIN;
 480                 }
 481
 482                 if(!c->rtt_start.tv_sec) {
 483                         // Start RTT measurement
 484                         gettimeofday(&c->rtt_start, NULL);
 485                         c->rtt_seq = pkt->hdr.seq + seglen;
 486                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 487                 }
 488
 489                 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
 490                 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
 491         } while(left);
 492
 493         free(pkt);
 494 }
 495
 496 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 497         if(c->reapable) {
 498                 debug("Error: send() called on closed connection %p\n", c);
 499                 errno = EBADF;
 500                 return -1;
 501         }
 502
 503         switch(c->state) {
 504         case CLOSED:
 505         case LISTEN:
 506         case SYN_SENT:
 507         case SYN_RECEIVED:
 508                 debug("Error: send() called on unconnected connection %p\n", c);
 509                 errno = ENOTCONN;
 510                 return -1;
 511         case ESTABLISHED:
 512         case CLOSE_WAIT:
 513                 break;
 514         case FIN_WAIT_1:
 515         case FIN_WAIT_2:
 516         case CLOSING:
 517         case LAST_ACK:
 518         case TIME_WAIT:
 519                 debug("Error: send() called on closing connection %p\n", c);
 520                 errno = EPIPE;
 521                 return -1;
 522         }
 523
 524         // Add data to send buffer
 525
 526         if(!len)
 527                 return 0;
 528
 529         if(!data) {
 530                 errno = EFAULT;
 531                 return -1;
 532         }
 533
 534         len = buffer_put(&c->sndbuf, data, len);
 535         if(len <= 0) {
 536                 errno = EWOULDBLOCK;
 537                 return 0;
 538         }
 539
 540         c->snd.last += len;
 541         ack(c, false);
 542         if(!timerisset(&c->rtrx_timeout))
 543                 start_retransmit_timer(c);
 544         return len;
 545 }
 546
 547 static void swap_ports(struct hdr *hdr) {
 548         uint16_t tmp = hdr->src;
 549         hdr->src = hdr->dst;
 550         hdr->dst = tmp;
 551 }
 552
 553 static void retransmit(struct utcp_connection *c) {
 554         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 555                 debug("Retransmit() called but nothing to retransmit!\n");
 556                 stop_retransmit_timer(c);
 557                 return;
 558         }
 559
 560         struct utcp *utcp = c->utcp;
 561
 562         struct {
 563                 struct hdr hdr;
 564                 char data[];
 565         } *pkt;
 566
 567         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 568         if(!pkt)
 569                 return;
 570
 571         pkt->hdr.src = c->src;
 572         pkt->hdr.dst = c->dst;
 573         pkt->hdr.wnd = c->rcv.wnd;
 574         pkt->hdr.aux = 0;
 575
 576         switch(c->state) {
 577                 case SYN_SENT:
 578                         // Send our SYN again
 579                         pkt->hdr.seq = c->snd.iss;
 580                         pkt->hdr.ack = 0;
 581                         pkt->hdr.ctl = SYN;
 582                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 583                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 584                         break;
 585
 586                 case SYN_RECEIVED:
 587                         // Send SYNACK again
 588                         pkt->hdr.seq = c->snd.nxt;
 589                         pkt->hdr.ack = c->rcv.nxt;
 590                         pkt->hdr.ctl = SYN | ACK;
 591                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 592                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 593                         break;
 594
 595                 case ESTABLISHED:
 596                 case FIN_WAIT_1:
 597                 case CLOSE_WAIT:
 598                 case CLOSING:
 599                 case LAST_ACK:
 600                         // Send unacked data again.
 601                         pkt->hdr.seq = c->snd.una;
 602                         pkt->hdr.ack = c->rcv.nxt;
 603                         pkt->hdr.ctl = ACK;
 604                         uint32_t len = seqdiff(c->snd.last, c->snd.una);
 605                         if(len > utcp->mtu)
 606                                 len = utcp->mtu;
 607                         if(fin_wanted(c, c->snd.una + len)) {
 608                                 len--;
 609                                 pkt->hdr.ctl |= FIN;
 610                         }
 611                         c->snd.nxt = c->snd.una + len;
 612                         c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 613                         buffer_copy(&c->sndbuf, pkt->data, 0, len);
 614                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
 615                         utcp->send(utcp, pkt, sizeof pkt->hdr + len);
 616                         break;
 617
 618                 case CLOSED:
 619                 case LISTEN:
 620                 case TIME_WAIT:
 621                 case FIN_WAIT_2:
 622                         // We shouldn't need to retransmit anything in this state.
 623 #ifdef UTCP_DEBUG
 624                         abort();
 625 #endif
 626                         stop_retransmit_timer(c);
 627                         goto cleanup;
 628         }
 629
 630         start_retransmit_timer(c);
 631         utcp->rto *= 2;
 632         if(utcp->rto > MAX_RTO)
 633                 utcp->rto = MAX_RTO;
 634         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 635
 636 cleanup:
 637         free(pkt);
 638 }
 639
 640 /* Update receive buffer and SACK entries after consuming data.
 641  *
 642  * Situation:
 643  *
 644  * |.....0000..1111111111.....22222......3333|
 645  * |---------------^
 646  *
 647  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 648  * to remove data from the receive buffer. The idea is to substract "len"
 649  * from the offset of all the SACK entries, and then remove/cut down entries
 650  * that are shifted to before the start of the receive buffer.
 651  *
 652  * There are three cases:
 653  * - the SACK entry is after ^, in that case just change the offset.
 654  * - the SACK entry starts before and ends after ^, so we have to
 655  *   change both its offset and size.
 656  * - the SACK entry is completely before ^, in that case delete it.
 657  */
 658 static void sack_consume(struct utcp_connection *c, size_t len) {
 659         debug("sack_consume %zu\n", len);
 660         if(len > c->rcvbuf.used)
 661                 abort();
 662
 663         buffer_get(&c->rcvbuf, NULL, len);
 664
 665         for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
 666                 if(len < c->sacks[i].offset) {
 667                         c->sacks[i].offset -= len;
 668                         i++;
 669                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 670                         c->sacks[i].len -= len - c->sacks[i].offset;
 671                         c->sacks[i].offset = 0;
 672                         i++;
 673                 } else {
 674                         if(i < NSACKS - 1) {
 675                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
 676                                 c->sacks[NSACKS - 1].len = 0;
 677                         } else {
 678                                 c->sacks[i].len = 0;
 679                                 break;
 680                         }
 681                 }
 682         }
 683
 684         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 685                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 686 }
 687
 688 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 689         debug("out of order packet, offset %u\n", offset);
 690         // Packet loss or reordering occured. Store the data in the buffer.
 691         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 692         if(rxd < len)
 693                 abort();
 694
 695         // Make note of where we put it.
 696         for(int i = 0; i < NSACKS; i++) {
 697                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 698                         debug("New SACK entry %d\n", i);
 699                         c->sacks[i].offset = offset;
 700                         c->sacks[i].len = rxd;
 701                         break;
 702                 } else if(offset < c->sacks[i].offset) {
 703                         if(offset + rxd < c->sacks[i].offset) { // insert before
 704                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 705                                         debug("Insert SACK entry at %d\n", i);
 706                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
 707                                         c->sacks[i].offset = offset;
 708                                         c->sacks[i].len = rxd;
 709                                 } else {
 710                                         debug("SACK entries full, dropping packet\n");
 711                                 }
 712                                 break;
 713                         } else { // merge
 714                                 debug("Merge with start of SACK entry at %d\n", i);
 715                                 c->sacks[i].offset = offset;
 716                                 break;
 717                         }
 718                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 719                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 720                                 debug("Merge with end of SACK entry at %d\n", i);
 721                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 722                                 // TODO: handle potential merge with next entry
 723                         }
 724                         break;
 725                 }
 726         }
 727
 728         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 729                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 730 }
 731
 732 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 733         // Check if we can process out-of-order data now.
 734         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 735                 debug("incoming packet len %zu connected with SACK at %u\n", len, c->sacks[0].offset);
 736                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 737                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 738                 data = c->rcvbuf.data;
 739         }
 740
 741         if(c->recv) {
 742                 ssize_t rxd = c->recv(c, data, len);
 743                 if(rxd != len) {
 744                         // TODO: handle the application not accepting all data.
 745                         abort();
 746                 }
 747         }
 748
 749         if(c->rcvbuf.used)
 750                 sack_consume(c, len);
 751
 752         c->rcv.nxt += len;
 753 }
 754
 755
 756 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 757         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 758         if(offset + len > c->rcvbuf.maxsize)
 759                 abort();
 760
 761         if(offset)
 762                 handle_out_of_order(c, offset, data, len);
 763         else
 764                 handle_in_order(c, data, len);
 765 }
 766
 767
 768 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 769         if(!utcp) {
 770                 errno = EFAULT;
 771                 return -1;
 772         }
 773
 774         if(!len)
 775                 return 0;
 776
 777         if(!data) {
 778                 errno = EFAULT;
 779                 return -1;
 780         }
 781
 782         print_packet(utcp, "recv", data, len);
 783
 784         // Drop packets smaller than the header
 785
 786         struct hdr hdr;
 787         if(len < sizeof hdr) {
 788                 errno = EBADMSG;
 789                 return -1;
 790         }
 791
 792         // Make a copy from the potentially unaligned data to a struct hdr
 793
 794         memcpy(&hdr, data, sizeof hdr);
 795         data += sizeof hdr;
 796         len -= sizeof hdr;
 797
 798         // Drop packets with an unknown CTL flag
 799
 800         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 801                 errno = EBADMSG;
 802                 return -1;
 803         }
 804
 805         // Try to match the packet to an existing connection
 806
 807         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
 808
 809         // Is it for a new connection?
 810
 811         if(!c) {
 812                 // Ignore RST packets
 813
 814                 if(hdr.ctl & RST)
 815                         return 0;
 816
 817                 // Is it a SYN packet and are we LISTENing?
 818
 819                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
 820                         // If we don't want to accept it, send a RST back
 821                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
 822                                 len = 1;
 823                                 goto reset;
 824                         }
 825
 826                         // Try to allocate memory, otherwise send a RST back
 827                         c = allocate_connection(utcp, hdr.dst, hdr.src);
 828                         if(!c) {
 829                                 len = 1;
 830                                 goto reset;
 831                         }
 832
 833                         // Return SYN+ACK, go to SYN_RECEIVED state
 834                         c->snd.wnd = hdr.wnd;
 835                         c->rcv.irs = hdr.seq;
 836                         c->rcv.nxt = c->rcv.irs + 1;
 837                         set_state(c, SYN_RECEIVED);
 838
 839                         hdr.dst = c->dst;
 840                         hdr.src = c->src;
 841                         hdr.ack = c->rcv.irs + 1;
 842                         hdr.seq = c->snd.iss;
 843                         hdr.ctl = SYN | ACK;
 844                         print_packet(c->utcp, "send", &hdr, sizeof hdr);
 845                         utcp->send(utcp, &hdr, sizeof hdr);
 846                 } else {
 847                         // No, we don't want your packets, send a RST back
 848                         len = 1;
 849                         goto reset;
 850                 }
 851
 852                 return 0;
 853         }
 854
 855         debug("%p state %s\n", c->utcp, strstate[c->state]);
 856
 857         // In case this is for a CLOSED connection, ignore the packet.
 858         // TODO: make it so incoming packets can never match a CLOSED connection.
 859
 860         if(c->state == CLOSED)
 861                 return 0;
 862
 863         // It is for an existing connection.
 864
 865         uint32_t prevrcvnxt = c->rcv.nxt;
 866
 867         // 1. Drop invalid packets.
 868
 869         // 1a. Drop packets that should not happen in our current state.
 870
 871         switch(c->state) {
 872         case SYN_SENT:
 873         case SYN_RECEIVED:
 874         case ESTABLISHED:
 875         case FIN_WAIT_1:
 876         case FIN_WAIT_2:
 877         case CLOSE_WAIT:
 878         case CLOSING:
 879         case LAST_ACK:
 880         case TIME_WAIT:
 881                 break;
 882         default:
 883 #ifdef UTCP_DEBUG
 884                 abort();
 885 #endif
 886                 break;
 887         }
 888
 889         // 1b. Drop packets with a sequence number not in our receive window.
 890
 891         bool acceptable;
 892
 893         if(c->state == SYN_SENT)
 894                 acceptable = true;
 895         else if(len == 0)
 896                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
 897         else {
 898                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
 899
 900                 // cut already accepted front overlapping
 901                 if(rcv_offset < 0) {
 902                         acceptable = rcv_offset + len >= 0;
 903                         if(acceptable) {
 904                                 data -= rcv_offset;
 905                                 len += rcv_offset;
 906                         }
 907                 }
 908
 909                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
 910         }
 911
 912         if(!acceptable) {
 913                 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
 914                 // Ignore unacceptable RST packets.
 915                 if(hdr.ctl & RST)
 916                         return 0;
 917                 // Otherwise, send an ACK back in the hope things improve.
 918                 ack(c, true);
 919                 return 0;
 920         }
 921
 922         c->snd.wnd = hdr.wnd; // TODO: move below
 923
 924         // 1c. Drop packets with an invalid ACK.
 925         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
 926         // (= snd.una + c->sndbuf.used).
 927
 928         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
 929                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 930                 // Ignore unacceptable RST packets.
 931                 if(hdr.ctl & RST)
 932                         return 0;
 933                 goto reset;
 934         }
 935
 936         // 2. Handle RST packets
 937
 938         if(hdr.ctl & RST) {
 939                 switch(c->state) {
 940                 case SYN_SENT:
 941                         if(!(hdr.ctl & ACK))
 942                                 return 0;
 943                         // The peer has refused our connection.
 944                         set_state(c, CLOSED);
 945                         errno = ECONNREFUSED;
 946                         if(c->recv)
 947                                 c->recv(c, NULL, 0);
 948                         return 0;
 949                 case SYN_RECEIVED:
 950                         if(hdr.ctl & ACK)
 951                                 return 0;
 952                         // We haven't told the application about this connection yet. Silently delete.
 953                         free_connection(c);
 954                         return 0;
 955                 case ESTABLISHED:
 956                 case FIN_WAIT_1:
 957                 case FIN_WAIT_2:
 958                 case CLOSE_WAIT:
 959                         if(hdr.ctl & ACK)
 960                                 return 0;
 961                         // The peer has aborted our connection.
 962                         set_state(c, CLOSED);
 963                         errno = ECONNRESET;
 964                         if(c->recv)
 965                                 c->recv(c, NULL, 0);
 966                         return 0;
 967                 case CLOSING:
 968                 case LAST_ACK:
 969                 case TIME_WAIT:
 970                         if(hdr.ctl & ACK)
 971                                 return 0;
 972                         // As far as the application is concerned, the connection has already been closed.
 973                         // If it has called utcp_close() already, we can immediately free this connection.
 974                         if(c->reapable) {
 975                                 free_connection(c);
 976                                 return 0;
 977                         }
 978                         // Otherwise, immediately move to the CLOSED state.
 979                         set_state(c, CLOSED);
 980                         return 0;
 981                 default:
 982 #ifdef UTCP_DEBUG
 983                         abort();
 984 #endif
 985                         break;
 986                 }
 987         }
 988
 989         // 3. Advance snd.una
 990
 991         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
 992         prevrcvnxt = c->rcv.nxt;
 993
 994         if(advanced) {
 995                 // RTT measurement
 996                 if(c->rtt_start.tv_sec) {
 997                         if(c->rtt_seq == hdr.ack) {
 998                                 struct timeval now, diff;
 999                                 gettimeofday(&now, NULL);
1000                                 timersub(&now, &c->rtt_start, &diff);
1001                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1002                                 c->rtt_start.tv_sec = 0;
1003                         } else if(c->rtt_seq < hdr.ack) {
1004                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1005                                 c->rtt_start.tv_sec = 0;
1006                         }
1007                 }
1008
1009                 int32_t data_acked = advanced;
1010
1011                 switch(c->state) {
1012                         case SYN_SENT:
1013                         case SYN_RECEIVED:
1014                                 data_acked--;
1015                                 break;
1016                         // TODO: handle FIN as well.
1017                         default:
1018                                 break;
1019                 }
1020
1021                 assert(data_acked >= 0);
1022
1023                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1024                 assert(data_acked <= bufused);
1025
1026                 if(data_acked)
1027                         buffer_get(&c->sndbuf, NULL, data_acked);
1028
1029                 // Also advance snd.nxt if possible
1030                 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1031                         c->snd.nxt = hdr.ack;
1032
1033                 c->snd.una = hdr.ack;
1034
1035                 c->dupack = 0;
1036                 c->snd.cwnd += utcp->mtu;
1037                 if(c->snd.cwnd > c->sndbuf.maxsize)
1038                         c->snd.cwnd = c->sndbuf.maxsize;
1039
1040                 // Check if we have sent a FIN that is now ACKed.
1041                 switch(c->state) {
1042                 case FIN_WAIT_1:
1043                         if(c->snd.una == c->snd.last)
1044                                 set_state(c, FIN_WAIT_2);
1045                         break;
1046                 case CLOSING:
1047                         if(c->snd.una == c->snd.last) {
1048                                 gettimeofday(&c->conn_timeout, NULL);
1049                                 c->conn_timeout.tv_sec += 60;
1050                                 set_state(c, TIME_WAIT);
1051                         }
1052                         break;
1053                 default:
1054                         break;
1055                 }
1056         } else {
1057                 if(!len) {
1058                         c->dupack++;
1059                         if(c->dupack == 3) {
1060                                 debug("Triplicate ACK\n");
1061                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1062                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1063                                 //Reset the congestion window so we wait for ACKs.
1064                                 c->snd.nxt = c->snd.una;
1065                                 c->snd.cwnd = utcp->mtu;
1066                                 start_retransmit_timer(c);
1067                         }
1068                 }
1069         }
1070
1071         // 4. Update timers
1072
1073         if(advanced) {
1074                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1075                 if(c->snd.una == c->snd.last)
1076                         stop_retransmit_timer(c);
1077                 else
1078                         start_retransmit_timer(c);
1079         }
1080
1081         // 5. Process SYN stuff
1082
1083         if(hdr.ctl & SYN) {
1084                 switch(c->state) {
1085                 case SYN_SENT:
1086                         // This is a SYNACK. It should always have ACKed the SYN.
1087                         if(!advanced)
1088                                 goto reset;
1089                         c->rcv.irs = hdr.seq;
1090                         c->rcv.nxt = hdr.seq;
1091                         set_state(c, ESTABLISHED);
1092                         // TODO: notify application of this somehow.
1093                         break;
1094                 case SYN_RECEIVED:
1095                 case ESTABLISHED:
1096                 case FIN_WAIT_1:
1097                 case FIN_WAIT_2:
1098                 case CLOSE_WAIT:
1099                 case CLOSING:
1100                 case LAST_ACK:
1101                 case TIME_WAIT:
1102                         // Ehm, no. We should never receive a second SYN.
1103                         goto reset;
1104                 default:
1105 #ifdef UTCP_DEBUG
1106                         abort();
1107 #endif
1108                         return 0;
1109                 }
1110
1111                 // SYN counts as one sequence number
1112                 c->rcv.nxt++;
1113         }
1114
1115         // 6. Process new data
1116
1117         if(c->state == SYN_RECEIVED) {
1118                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1119                 if(!advanced)
1120                         goto reset;
1121
1122                 // Are we still LISTENing?
1123                 if(utcp->accept)
1124                         utcp->accept(c, c->src);
1125
1126                 if(c->state != ESTABLISHED) {
1127                         set_state(c, CLOSED);
1128                         c->reapable = true;
1129                         goto reset;
1130                 }
1131         }
1132
1133         if(len) {
1134                 switch(c->state) {
1135                 case SYN_SENT:
1136                 case SYN_RECEIVED:
1137                         // This should never happen.
1138 #ifdef UTCP_DEBUG
1139                         abort();
1140 #endif
1141                         return 0;
1142                 case ESTABLISHED:
1143                 case FIN_WAIT_1:
1144                 case FIN_WAIT_2:
1145                         break;
1146                 case CLOSE_WAIT:
1147                 case CLOSING:
1148                 case LAST_ACK:
1149                 case TIME_WAIT:
1150                         // Ehm no, We should never receive more data after a FIN.
1151                         goto reset;
1152                 default:
1153 #ifdef UTCP_DEBUG
1154                         abort();
1155 #endif
1156                         return 0;
1157                 }
1158
1159                 handle_incoming_data(c, hdr.seq, data, len);
1160         }
1161
1162         // 7. Process FIN stuff
1163
1164         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1165                 switch(c->state) {
1166                 case SYN_SENT:
1167                 case SYN_RECEIVED:
1168                         // This should never happen.
1169 #ifdef UTCP_DEBUG
1170                         abort();
1171 #endif
1172                         break;
1173                 case ESTABLISHED:
1174                         set_state(c, CLOSE_WAIT);
1175                         break;
1176                 case FIN_WAIT_1:
1177                         set_state(c, CLOSING);
1178                         break;
1179                 case FIN_WAIT_2:
1180                         gettimeofday(&c->conn_timeout, NULL);
1181                         c->conn_timeout.tv_sec += 60;
1182                         set_state(c, TIME_WAIT);
1183                         break;
1184                 case CLOSE_WAIT:
1185                 case CLOSING:
1186                 case LAST_ACK:
1187                 case TIME_WAIT:
1188                         // Ehm, no. We should never receive a second FIN.
1189                         goto reset;
1190                 default:
1191 #ifdef UTCP_DEBUG
1192                         abort();
1193 #endif
1194                         break;
1195                 }
1196
1197                 // FIN counts as one sequence number
1198                 c->rcv.nxt++;
1199                 len++;
1200
1201                 // Inform the application that the peer closed the connection.
1202                 if(c->recv) {
1203                         errno = 0;
1204                         c->recv(c, NULL, 0);
1205                 }
1206         }
1207
1208         // Now we send something back if:
1209         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1210         //   -> sendatleastone = true
1211         // - or we got an ack, so we should maybe send a bit more data
1212         //   -> sendatleastone = false
1213
1214         ack(c, len || prevrcvnxt != c->rcv.nxt);
1215         return 0;
1216
1217 reset:
1218         swap_ports(&hdr);
1219         hdr.wnd = 0;
1220         if(hdr.ctl & ACK) {
1221                 hdr.seq = hdr.ack;
1222                 hdr.ctl = RST;
1223         } else {
1224                 hdr.ack = hdr.seq + len;
1225                 hdr.seq = 0;
1226                 hdr.ctl = RST | ACK;
1227         }
1228         print_packet(utcp, "send", &hdr, sizeof hdr);
1229         utcp->send(utcp, &hdr, sizeof hdr);
1230         return 0;
1231
1232 }
1233
1234 int utcp_shutdown(struct utcp_connection *c, int dir) {
1235         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1236         if(!c) {
1237                 errno = EFAULT;
1238                 return -1;
1239         }
1240
1241         if(c->reapable) {
1242                 debug("Error: shutdown() called on closed connection %p\n", c);
1243                 errno = EBADF;
1244                 return -1;
1245         }
1246
1247         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1248                 errno = EINVAL;
1249                 return -1;
1250         }
1251
1252         // TCP does not have a provision for stopping incoming packets.
1253         // The best we can do is to just ignore them.
1254         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1255                 c->recv = NULL;
1256
1257         // The rest of the code deals with shutting down writes.
1258         if(dir == UTCP_SHUT_RD)
1259                 return 0;
1260
1261         switch(c->state) {
1262         case CLOSED:
1263         case LISTEN:
1264                 errno = ENOTCONN;
1265                 return -1;
1266
1267         case SYN_SENT:
1268                 set_state(c, CLOSED);
1269                 return 0;
1270
1271         case SYN_RECEIVED:
1272         case ESTABLISHED:
1273                 set_state(c, FIN_WAIT_1);
1274                 break;
1275         case FIN_WAIT_1:
1276         case FIN_WAIT_2:
1277                 return 0;
1278         case CLOSE_WAIT:
1279                 set_state(c, CLOSING);
1280                 break;
1281
1282         case CLOSING:
1283         case LAST_ACK:
1284         case TIME_WAIT:
1285                 return 0;
1286         }
1287
1288         c->snd.last++;
1289
1290         ack(c, false);
1291         if(!timerisset(&c->rtrx_timeout))
1292                 start_retransmit_timer(c);
1293         return 0;
1294 }
1295
1296 int utcp_close(struct utcp_connection *c) {
1297         if(utcp_shutdown(c, SHUT_RDWR))
1298                 return -1;
1299         c->recv = NULL;
1300         c->poll = NULL;
1301         c->reapable = true;
1302         return 0;
1303 }
1304
1305 int utcp_abort(struct utcp_connection *c) {
1306         if(!c) {
1307                 errno = EFAULT;
1308                 return -1;
1309         }
1310
1311         if(c->reapable) {
1312                 debug("Error: abort() called on closed connection %p\n", c);
1313                 errno = EBADF;
1314                 return -1;
1315         }
1316
1317         c->recv = NULL;
1318         c->poll = NULL;
1319         c->reapable = true;
1320
1321         switch(c->state) {
1322         case CLOSED:
1323                 return 0;
1324         case LISTEN:
1325         case SYN_SENT:
1326         case CLOSING:
1327         case LAST_ACK:
1328         case TIME_WAIT:
1329                 set_state(c, CLOSED);
1330                 return 0;
1331
1332         case SYN_RECEIVED:
1333         case ESTABLISHED:
1334         case FIN_WAIT_1:
1335         case FIN_WAIT_2:
1336         case CLOSE_WAIT:
1337                 set_state(c, CLOSED);
1338                 break;
1339         }
1340
1341         // Send RST
1342
1343         struct hdr hdr;
1344
1345         hdr.src = c->src;
1346         hdr.dst = c->dst;
1347         hdr.seq = c->snd.nxt;
1348         hdr.ack = 0;
1349         hdr.wnd = 0;
1350         hdr.ctl = RST;
1351
1352         print_packet(c->utcp, "send", &hdr, sizeof hdr);
1353         c->utcp->send(c->utcp, &hdr, sizeof hdr);
1354         return 0;
1355 }
1356
1357 /* Handle timeouts.
1358  * One call to this function will loop through all connections,
1359  * checking if something needs to be resent or not.
1360  * The return value is the time to the next timeout in milliseconds,
1361  * or maybe a negative value if the timeout is infinite.
1362  */
1363 struct timeval utcp_timeout(struct utcp *utcp) {
1364         struct timeval now;
1365         gettimeofday(&now, NULL);
1366         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1367
1368         for(int i = 0; i < utcp->nconnections; i++) {
1369                 struct utcp_connection *c = utcp->connections[i];
1370                 if(!c)
1371                         continue;
1372
1373                 // delete connections that have been utcp_close()d.
1374                 if(c->state == CLOSED) {
1375                         if(c->reapable) {
1376                                 debug("Reaping %p\n", c);
1377                                 free_connection(c);
1378                                 i--;
1379                         }
1380                         continue;
1381                 }
1382
1383                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1384                         errno = ETIMEDOUT;
1385                         c->state = CLOSED;
1386                         if(c->recv)
1387                                 c->recv(c, NULL, 0);
1388                         continue;
1389                 }
1390
1391                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1392                         debug("retransmit()\n");
1393                         retransmit(c);
1394                 }
1395
1396                 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1397                         c->poll(c, buffer_free(&c->sndbuf));
1398
1399                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1400                         next = c->conn_timeout;
1401
1402                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1403                         next = c->rtrx_timeout;
1404         }
1405
1406         struct timeval diff;
1407         timersub(&next, &now, &diff);
1408         return diff;
1409 }
1410
1411 bool utcp_is_active(struct utcp *utcp) {
1412         if(!utcp)
1413                 return false;
1414
1415         for(int i = 0; i < utcp->nconnections; i++)
1416                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1417                         return true;
1418
1419         return false;
1420 }
1421
1422 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1423         if(!send) {
1424                 errno = EFAULT;
1425                 return NULL;
1426         }
1427
1428         struct utcp *utcp = calloc(1, sizeof *utcp);
1429         if(!utcp)
1430                 return NULL;
1431
1432         utcp->accept = accept;
1433         utcp->pre_accept = pre_accept;
1434         utcp->send = send;
1435         utcp->priv = priv;
1436         utcp->mtu = DEFAULT_MTU;
1437         utcp->timeout = DEFAULT_USER_TIMEOUT; // s
1438         utcp->rto = START_RTO; // us
1439
1440         return utcp;
1441 }
1442
1443 void utcp_exit(struct utcp *utcp) {
1444         if(!utcp)
1445                 return;
1446         for(int i = 0; i < utcp->nconnections; i++) {
1447                 if(!utcp->connections[i]->reapable)
1448                         debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1449                 buffer_exit(&utcp->connections[i]->rcvbuf);
1450                 buffer_exit(&utcp->connections[i]->sndbuf);
1451                 free(utcp->connections[i]);
1452         }
1453         free(utcp->connections);
1454         free(utcp);
1455 }
1456
1457 uint16_t utcp_get_mtu(struct utcp *utcp) {
1458         return utcp ? utcp->mtu : 0;
1459 }
1460
1461 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1462         // TODO: handle overhead of the header
1463         if(utcp)
1464                 utcp->mtu = mtu;
1465 }
1466
1467 int utcp_get_user_timeout(struct utcp *u) {
1468         return u ? u->timeout : 0;
1469 }
1470
1471 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1472         if(u)
1473                 u->timeout = timeout;
1474 }
1475
1476 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1477         return c ? c->sndbuf.maxsize : 0;
1478 }
1479
1480 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1481         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1482                 return buffer_free(&c->sndbuf);
1483         else
1484                 return 0;
1485 }
1486
1487 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1488         if(!c)
1489                 return;
1490         c->sndbuf.maxsize = size;
1491         if(c->sndbuf.maxsize != size)
1492                 c->sndbuf.maxsize = -1;
1493 }
1494
1495 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1496         return c ? c->rcvbuf.maxsize : 0;
1497 }
1498
1499 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1500         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1501                 return buffer_free(&c->rcvbuf);
1502         else
1503                 return 0;
1504 }
1505
1506 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1507         if(!c)
1508                 return;
1509         c->rcvbuf.maxsize = size;
1510         if(c->rcvbuf.maxsize != size)
1511                 c->rcvbuf.maxsize = -1;
1512 }
1513
1514 bool utcp_get_nodelay(struct utcp_connection *c) {
1515         return c ? c->nodelay : false;
1516 }
1517
1518 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1519         if(c)
1520                 c->nodelay = nodelay;
1521 }
1522
1523 bool utcp_get_keepalive(struct utcp_connection *c) {
1524         return c ? c->keepalive : false;
1525 }
1526
1527 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1528         if(c)
1529                 c->keepalive = keepalive;
1530 }
1531
1532 size_t utcp_get_outq(struct utcp_connection *c) {
1533         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1534 }
1535
1536 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1537         if(c)
1538                 c->recv = recv;
1539 }
1540
1541 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1542         if(c)
1543                 c->poll = poll;
1544 }
1545
1546 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1547         if(utcp) {
1548                 utcp->accept = accept;
1549                 utcp->pre_accept = pre_accept;
1550         }
1551 }