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