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 += USEC_PER_SEC;\
  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                 debug("Got packet for closed connection\n");
 862                 return 0;
 863         }
 864
 865         // It is for an existing connection.
 866
 867         uint32_t prevrcvnxt = c->rcv.nxt;
 868
 869         // 1. Drop invalid packets.
 870
 871         // 1a. Drop packets that should not happen in our current state.
 872
 873         switch(c->state) {
 874         case SYN_SENT:
 875         case SYN_RECEIVED:
 876         case ESTABLISHED:
 877         case FIN_WAIT_1:
 878         case FIN_WAIT_2:
 879         case CLOSE_WAIT:
 880         case CLOSING:
 881         case LAST_ACK:
 882         case TIME_WAIT:
 883                 break;
 884         default:
 885 #ifdef UTCP_DEBUG
 886                 abort();
 887 #endif
 888                 break;
 889         }
 890
 891         // 1b. Drop packets with a sequence number not in our receive window.
 892
 893         bool acceptable;
 894
 895         if(c->state == SYN_SENT)
 896                 acceptable = true;
 897         else if(len == 0)
 898                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
 899         else {
 900                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
 901
 902                 // cut already accepted front overlapping
 903                 if(rcv_offset < 0) {
 904                         acceptable = rcv_offset + len >= 0;
 905                         if(acceptable) {
 906                                 data -= rcv_offset;
 907                                 len += rcv_offset;
 908                         }
 909                 }
 910
 911                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
 912         }
 913
 914         if(!acceptable) {
 915                 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
 916                 // Ignore unacceptable RST packets.
 917                 if(hdr.ctl & RST)
 918                         return 0;
 919                 // Otherwise, send an ACK back in the hope things improve.
 920                 ack(c, true);
 921                 return 0;
 922         }
 923
 924         c->snd.wnd = hdr.wnd; // TODO: move below
 925
 926         // 1c. Drop packets with an invalid ACK.
 927         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
 928         // (= snd.una + c->sndbuf.used).
 929
 930         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
 931                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 932                 // Ignore unacceptable RST packets.
 933                 if(hdr.ctl & RST)
 934                         return 0;
 935                 goto reset;
 936         }
 937
 938         // 2. Handle RST packets
 939
 940         if(hdr.ctl & RST) {
 941                 switch(c->state) {
 942                 case SYN_SENT:
 943                         if(!(hdr.ctl & ACK))
 944                                 return 0;
 945                         // The peer has refused our connection.
 946                         set_state(c, CLOSED);
 947                         errno = ECONNREFUSED;
 948                         if(c->recv)
 949                                 c->recv(c, NULL, 0);
 950                         return 0;
 951                 case SYN_RECEIVED:
 952                         if(hdr.ctl & ACK)
 953                                 return 0;
 954                         // We haven't told the application about this connection yet. Silently delete.
 955                         free_connection(c);
 956                         return 0;
 957                 case ESTABLISHED:
 958                 case FIN_WAIT_1:
 959                 case FIN_WAIT_2:
 960                 case CLOSE_WAIT:
 961                         if(hdr.ctl & ACK)
 962                                 return 0;
 963                         // The peer has aborted our connection.
 964                         set_state(c, CLOSED);
 965                         errno = ECONNRESET;
 966                         if(c->recv)
 967                                 c->recv(c, NULL, 0);
 968                         return 0;
 969                 case CLOSING:
 970                 case LAST_ACK:
 971                 case TIME_WAIT:
 972                         if(hdr.ctl & ACK)
 973                                 return 0;
 974                         // As far as the application is concerned, the connection has already been closed.
 975                         // If it has called utcp_close() already, we can immediately free this connection.
 976                         if(c->reapable) {
 977                                 free_connection(c);
 978                                 return 0;
 979                         }
 980                         // Otherwise, immediately move to the CLOSED state.
 981                         set_state(c, CLOSED);
 982                         return 0;
 983                 default:
 984 #ifdef UTCP_DEBUG
 985                         abort();
 986 #endif
 987                         break;
 988                 }
 989         }
 990
 991         // 3. Advance snd.una
 992
 993         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
 994         prevrcvnxt = c->rcv.nxt;
 995
 996         if(advanced) {
 997                 // RTT measurement
 998                 if(c->rtt_start.tv_sec) {
 999                         if(c->rtt_seq == hdr.ack) {
1000                                 struct timeval now, diff;
1001                                 gettimeofday(&now, NULL);
1002                                 timersub(&now, &c->rtt_start, &diff);
1003                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1004                                 c->rtt_start.tv_sec = 0;
1005                         } else if(c->rtt_seq < hdr.ack) {
1006                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1007                                 c->rtt_start.tv_sec = 0;
1008                         }
1009                 }
1010
1011                 int32_t data_acked = advanced;
1012
1013                 switch(c->state) {
1014                         case SYN_SENT:
1015                         case SYN_RECEIVED:
1016                                 data_acked--;
1017                                 break;
1018                         // TODO: handle FIN as well.
1019                         default:
1020                                 break;
1021                 }
1022
1023                 assert(data_acked >= 0);
1024
1025                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1026                 assert(data_acked <= bufused);
1027
1028                 if(data_acked)
1029                         buffer_get(&c->sndbuf, NULL, data_acked);
1030
1031                 // Also advance snd.nxt if possible
1032                 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1033                         c->snd.nxt = hdr.ack;
1034
1035                 c->snd.una = hdr.ack;
1036
1037                 c->dupack = 0;
1038                 c->snd.cwnd += utcp->mtu;
1039                 if(c->snd.cwnd > c->sndbuf.maxsize)
1040                         c->snd.cwnd = c->sndbuf.maxsize;
1041
1042                 // Check if we have sent a FIN that is now ACKed.
1043                 switch(c->state) {
1044                 case FIN_WAIT_1:
1045                         if(c->snd.una == c->snd.last)
1046                                 set_state(c, FIN_WAIT_2);
1047                         break;
1048                 case CLOSING:
1049                         if(c->snd.una == c->snd.last) {
1050                                 gettimeofday(&c->conn_timeout, NULL);
1051                                 c->conn_timeout.tv_sec += 60;
1052                                 set_state(c, TIME_WAIT);
1053                         }
1054                         break;
1055                 default:
1056                         break;
1057                 }
1058         } else {
1059                 if(!len) {
1060                         c->dupack++;
1061                         if(c->dupack == 3) {
1062                                 debug("Triplicate ACK\n");
1063                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1064                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1065                                 //Reset the congestion window so we wait for ACKs.
1066                                 c->snd.nxt = c->snd.una;
1067                                 c->snd.cwnd = utcp->mtu;
1068                                 start_retransmit_timer(c);
1069                         }
1070                 }
1071         }
1072
1073         // 4. Update timers
1074
1075         if(advanced) {
1076                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1077                 if(c->snd.una == c->snd.last)
1078                         stop_retransmit_timer(c);
1079                 else
1080                         start_retransmit_timer(c);
1081         }
1082
1083         // 5. Process SYN stuff
1084
1085         if(hdr.ctl & SYN) {
1086                 switch(c->state) {
1087                 case SYN_SENT:
1088                         // This is a SYNACK. It should always have ACKed the SYN.
1089                         if(!advanced)
1090                                 goto reset;
1091                         c->rcv.irs = hdr.seq;
1092                         c->rcv.nxt = hdr.seq;
1093                         set_state(c, ESTABLISHED);
1094                         // TODO: notify application of this somehow.
1095                         break;
1096                 case SYN_RECEIVED:
1097                 case ESTABLISHED:
1098                 case FIN_WAIT_1:
1099                 case FIN_WAIT_2:
1100                 case CLOSE_WAIT:
1101                 case CLOSING:
1102                 case LAST_ACK:
1103                 case TIME_WAIT:
1104                         // Ehm, no. We should never receive a second SYN.
1105                         goto reset;
1106                 default:
1107 #ifdef UTCP_DEBUG
1108                         abort();
1109 #endif
1110                         return 0;
1111                 }
1112
1113                 // SYN counts as one sequence number
1114                 c->rcv.nxt++;
1115         }
1116
1117         // 6. Process new data
1118
1119         if(c->state == SYN_RECEIVED) {
1120                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1121                 if(!advanced)
1122                         goto reset;
1123
1124                 // Are we still LISTENing?
1125                 if(utcp->accept)
1126                         utcp->accept(c, c->src);
1127
1128                 if(c->state != ESTABLISHED) {
1129                         set_state(c, CLOSED);
1130                         c->reapable = true;
1131                         goto reset;
1132                 }
1133         }
1134
1135         if(len) {
1136                 switch(c->state) {
1137                 case SYN_SENT:
1138                 case SYN_RECEIVED:
1139                         // This should never happen.
1140 #ifdef UTCP_DEBUG
1141                         abort();
1142 #endif
1143                         return 0;
1144                 case ESTABLISHED:
1145                 case FIN_WAIT_1:
1146                 case FIN_WAIT_2:
1147                         break;
1148                 case CLOSE_WAIT:
1149                 case CLOSING:
1150                 case LAST_ACK:
1151                 case TIME_WAIT:
1152                         // Ehm no, We should never receive more data after a FIN.
1153                         goto reset;
1154                 default:
1155 #ifdef UTCP_DEBUG
1156                         abort();
1157 #endif
1158                         return 0;
1159                 }
1160
1161                 handle_incoming_data(c, hdr.seq, data, len);
1162         }
1163
1164         // 7. Process FIN stuff
1165
1166         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1167                 switch(c->state) {
1168                 case SYN_SENT:
1169                 case SYN_RECEIVED:
1170                         // This should never happen.
1171 #ifdef UTCP_DEBUG
1172                         abort();
1173 #endif
1174                         break;
1175                 case ESTABLISHED:
1176                         set_state(c, CLOSE_WAIT);
1177                         break;
1178                 case FIN_WAIT_1:
1179                         set_state(c, CLOSING);
1180                         break;
1181                 case FIN_WAIT_2:
1182                         gettimeofday(&c->conn_timeout, NULL);
1183                         c->conn_timeout.tv_sec += 60;
1184                         set_state(c, TIME_WAIT);
1185                         break;
1186                 case CLOSE_WAIT:
1187                 case CLOSING:
1188                 case LAST_ACK:
1189                 case TIME_WAIT:
1190                         // Ehm, no. We should never receive a second FIN.
1191                         goto reset;
1192                 default:
1193 #ifdef UTCP_DEBUG
1194                         abort();
1195 #endif
1196                         break;
1197                 }
1198
1199                 // FIN counts as one sequence number
1200                 c->rcv.nxt++;
1201                 len++;
1202
1203                 // Inform the application that the peer closed the connection.
1204                 if(c->recv) {
1205                         errno = 0;
1206                         c->recv(c, NULL, 0);
1207                 }
1208         }
1209
1210         // Now we send something back if:
1211         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1212         //   -> sendatleastone = true
1213         // - or we got an ack, so we should maybe send a bit more data
1214         //   -> sendatleastone = false
1215
1216         ack(c, len || prevrcvnxt != c->rcv.nxt);
1217         return 0;
1218
1219 reset:
1220         swap_ports(&hdr);
1221         hdr.wnd = 0;
1222         if(hdr.ctl & ACK) {
1223                 hdr.seq = hdr.ack;
1224                 hdr.ctl = RST;
1225         } else {
1226                 hdr.ack = hdr.seq + len;
1227                 hdr.seq = 0;
1228                 hdr.ctl = RST | ACK;
1229         }
1230         print_packet(utcp, "send", &hdr, sizeof hdr);
1231         utcp->send(utcp, &hdr, sizeof hdr);
1232         return 0;
1233
1234 }
1235
1236 int utcp_shutdown(struct utcp_connection *c, int dir) {
1237         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1238         if(!c) {
1239                 errno = EFAULT;
1240                 return -1;
1241         }
1242
1243         if(c->reapable) {
1244                 debug("Error: shutdown() called on closed connection %p\n", c);
1245                 errno = EBADF;
1246                 return -1;
1247         }
1248
1249         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1250                 errno = EINVAL;
1251                 return -1;
1252         }
1253
1254         // TCP does not have a provision for stopping incoming packets.
1255         // The best we can do is to just ignore them.
1256         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1257                 c->recv = NULL;
1258
1259         // The rest of the code deals with shutting down writes.
1260         if(dir == UTCP_SHUT_RD)
1261                 return 0;
1262
1263         switch(c->state) {
1264         case CLOSED:
1265         case LISTEN:
1266                 errno = ENOTCONN;
1267                 return -1;
1268
1269         case SYN_SENT:
1270                 set_state(c, CLOSED);
1271                 return 0;
1272
1273         case SYN_RECEIVED:
1274         case ESTABLISHED:
1275                 set_state(c, FIN_WAIT_1);
1276                 break;
1277         case FIN_WAIT_1:
1278         case FIN_WAIT_2:
1279                 return 0;
1280         case CLOSE_WAIT:
1281                 set_state(c, CLOSING);
1282                 break;
1283
1284         case CLOSING:
1285         case LAST_ACK:
1286         case TIME_WAIT:
1287                 return 0;
1288         }
1289
1290         c->snd.last++;
1291
1292         ack(c, false);
1293         if(!timerisset(&c->rtrx_timeout))
1294                 start_retransmit_timer(c);
1295         return 0;
1296 }
1297
1298 int utcp_close(struct utcp_connection *c) {
1299         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1300                 return -1;
1301         c->recv = NULL;
1302         c->poll = NULL;
1303         c->reapable = true;
1304         return 0;
1305 }
1306
1307 int utcp_abort(struct utcp_connection *c) {
1308         if(!c) {
1309                 errno = EFAULT;
1310                 return -1;
1311         }
1312
1313         if(c->reapable) {
1314                 debug("Error: abort() called on closed connection %p\n", c);
1315                 errno = EBADF;
1316                 return -1;
1317         }
1318
1319         c->recv = NULL;
1320         c->poll = NULL;
1321         c->reapable = true;
1322
1323         switch(c->state) {
1324         case CLOSED:
1325                 return 0;
1326         case LISTEN:
1327         case SYN_SENT:
1328         case CLOSING:
1329         case LAST_ACK:
1330         case TIME_WAIT:
1331                 set_state(c, CLOSED);
1332                 return 0;
1333
1334         case SYN_RECEIVED:
1335         case ESTABLISHED:
1336         case FIN_WAIT_1:
1337         case FIN_WAIT_2:
1338         case CLOSE_WAIT:
1339                 set_state(c, CLOSED);
1340                 break;
1341         }
1342
1343         // Send RST
1344
1345         struct hdr hdr;
1346
1347         hdr.src = c->src;
1348         hdr.dst = c->dst;
1349         hdr.seq = c->snd.nxt;
1350         hdr.ack = 0;
1351         hdr.wnd = 0;
1352         hdr.ctl = RST;
1353
1354         print_packet(c->utcp, "send", &hdr, sizeof hdr);
1355         c->utcp->send(c->utcp, &hdr, sizeof hdr);
1356         return 0;
1357 }
1358
1359 /* Handle timeouts.
1360  * One call to this function will loop through all connections,
1361  * checking if something needs to be resent or not.
1362  * The return value is the time to the next timeout in milliseconds,
1363  * or maybe a negative value if the timeout is infinite.
1364  */
1365 struct timeval utcp_timeout(struct utcp *utcp) {
1366         struct timeval now;
1367         gettimeofday(&now, NULL);
1368         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1369
1370         for(int i = 0; i < utcp->nconnections; i++) {
1371                 struct utcp_connection *c = utcp->connections[i];
1372                 if(!c)
1373                         continue;
1374
1375                 // delete connections that have been utcp_close()d.
1376                 if(c->state == CLOSED) {
1377                         if(c->reapable) {
1378                                 debug("Reaping %p\n", c);
1379                                 free_connection(c);
1380                                 i--;
1381                         }
1382                         continue;
1383                 }
1384
1385                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1386                         errno = ETIMEDOUT;
1387                         c->state = CLOSED;
1388                         if(c->recv)
1389                                 c->recv(c, NULL, 0);
1390                         continue;
1391                 }
1392
1393                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1394                         debug("retransmit()\n");
1395                         retransmit(c);
1396                 }
1397
1398                 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1399                         c->poll(c, buffer_free(&c->sndbuf));
1400
1401                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1402                         next = c->conn_timeout;
1403
1404                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1405                         next = c->rtrx_timeout;
1406         }
1407
1408         struct timeval diff;
1409         timersub(&next, &now, &diff);
1410         return diff;
1411 }
1412
1413 bool utcp_is_active(struct utcp *utcp) {
1414         if(!utcp)
1415                 return false;
1416
1417         for(int i = 0; i < utcp->nconnections; i++)
1418                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1419                         return true;
1420
1421         return false;
1422 }
1423
1424 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1425         if(!send) {
1426                 errno = EFAULT;
1427                 return NULL;
1428         }
1429
1430         struct utcp *utcp = calloc(1, sizeof *utcp);
1431         if(!utcp)
1432                 return NULL;
1433
1434         utcp->accept = accept;
1435         utcp->pre_accept = pre_accept;
1436         utcp->send = send;
1437         utcp->priv = priv;
1438         utcp->mtu = DEFAULT_MTU;
1439         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1440         utcp->rto = START_RTO; // usec
1441
1442         return utcp;
1443 }
1444
1445 void utcp_exit(struct utcp *utcp) {
1446         if(!utcp)
1447                 return;
1448         for(int i = 0; i < utcp->nconnections; i++) {
1449                 if(!utcp->connections[i]->reapable)
1450                         debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1451                 buffer_exit(&utcp->connections[i]->rcvbuf);
1452                 buffer_exit(&utcp->connections[i]->sndbuf);
1453                 free(utcp->connections[i]);
1454         }
1455         free(utcp->connections);
1456         free(utcp);
1457 }
1458
1459 uint16_t utcp_get_mtu(struct utcp *utcp) {
1460         return utcp ? utcp->mtu : 0;
1461 }
1462
1463 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1464         // TODO: handle overhead of the header
1465         if(utcp)
1466                 utcp->mtu = mtu;
1467 }
1468
1469 int utcp_get_user_timeout(struct utcp *u) {
1470         return u ? u->timeout : 0;
1471 }
1472
1473 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1474         if(u)
1475                 u->timeout = timeout;
1476 }
1477
1478 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1479         return c ? c->sndbuf.maxsize : 0;
1480 }
1481
1482 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1483         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1484                 return buffer_free(&c->sndbuf);
1485         else
1486                 return 0;
1487 }
1488
1489 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1490         if(!c)
1491                 return;
1492         c->sndbuf.maxsize = size;
1493         if(c->sndbuf.maxsize != size)
1494                 c->sndbuf.maxsize = -1;
1495 }
1496
1497 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1498         return c ? c->rcvbuf.maxsize : 0;
1499 }
1500
1501 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1502         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1503                 return buffer_free(&c->rcvbuf);
1504         else
1505                 return 0;
1506 }
1507
1508 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1509         if(!c)
1510                 return;
1511         c->rcvbuf.maxsize = size;
1512         if(c->rcvbuf.maxsize != size)
1513                 c->rcvbuf.maxsize = -1;
1514 }
1515
1516 bool utcp_get_nodelay(struct utcp_connection *c) {
1517         return c ? c->nodelay : false;
1518 }
1519
1520 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1521         if(c)
1522                 c->nodelay = nodelay;
1523 }
1524
1525 bool utcp_get_keepalive(struct utcp_connection *c) {
1526         return c ? c->keepalive : false;
1527 }
1528
1529 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1530         if(c)
1531                 c->keepalive = keepalive;
1532 }
1533
1534 size_t utcp_get_outq(struct utcp_connection *c) {
1535         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1536 }
1537
1538 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1539         if(c)
1540                 c->recv = recv;
1541 }
1542
1543 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1544         if(c)
1545                 c->poll = poll;
1546 }
1547
1548 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1549         if(utcp) {
1550                 utcp->accept = accept;
1551                 utcp->pre_accept = pre_accept;
1552         }
1553 }