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
 565         switch(c->state) {
 566                 case SYN_SENT:
 567                         // Send our SYN again
 568                         pkt->hdr.seq = c->snd.iss;
 569                         pkt->hdr.ack = 0;
 570                         pkt->hdr.wnd = c->rcv.wnd;
 571                         pkt->hdr.ctl = SYN;
 572                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 573                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 574                         break;
 575
 576                 case SYN_RECEIVED:
 577                         // Send SYNACK again
 578                         pkt->hdr.seq = c->snd.nxt;
 579                         pkt->hdr.ack = c->rcv.nxt;
 580                         pkt->hdr.ctl = SYN | ACK;
 581                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 582                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 583                         break;
 584
 585                 case ESTABLISHED:
 586                 case FIN_WAIT_1:
 587                 case CLOSE_WAIT:
 588                 case CLOSING:
 589                 case LAST_ACK:
 590                         // Send unacked data again.
 591                         pkt->hdr.seq = c->snd.una;
 592                         pkt->hdr.ack = c->rcv.nxt;
 593                         pkt->hdr.ctl = ACK;
 594                         uint32_t len = seqdiff(c->snd.last, c->snd.una);
 595                         if(len > utcp->mtu)
 596                                 len = utcp->mtu;
 597                         if(fin_wanted(c, c->snd.una + len)) {
 598                                 len--;
 599                                 pkt->hdr.ctl |= FIN;
 600                         }
 601                         c->snd.nxt = c->snd.una + len;
 602                         c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 603                         buffer_copy(&c->sndbuf, pkt->data, 0, len);
 604                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
 605                         utcp->send(utcp, pkt, sizeof pkt->hdr + len);
 606                         break;
 607
 608                 case CLOSED:
 609                 case LISTEN:
 610                 case TIME_WAIT:
 611                 case FIN_WAIT_2:
 612                         // We shouldn't need to retransmit anything in this state.
 613 #ifdef UTCP_DEBUG
 614                         abort();
 615 #endif
 616                         stop_retransmit_timer(c);
 617                         goto cleanup;
 618         }
 619
 620         start_retransmit_timer(c);
 621         utcp->rto *= 2;
 622         if(utcp->rto > MAX_RTO)
 623                 utcp->rto = MAX_RTO;
 624         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 625
 626 cleanup:
 627         free(pkt);
 628 }
 629
 630 // Update receive buffer and SACK entries after consuming data.
 631 static void sack_consume(struct utcp_connection *c, size_t len) {
 632         debug("sack_consume %zu\n", len);
 633         if(len > c->rcvbuf.used)
 634                 abort();
 635
 636         buffer_get(&c->rcvbuf, NULL, len);
 637
 638         for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
 639                 if(len < c->sacks[i].offset) {
 640                         c->sacks[i].offset -= len;
 641                         i++;
 642                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 643                         c->sacks[i].offset = 0;
 644                         c->sacks[i].len -= len - c->sacks[i].offset;
 645                         i++;
 646                 } else {
 647                         if(i < NSACKS - 1) {
 648                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
 649                                 c->sacks[i + 1].len = 0;
 650                         } else {
 651                                 c->sacks[i].len = 0;
 652                                 break;
 653                         }
 654                 }
 655         }
 656
 657         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 658                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 659 }
 660
 661 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 662         debug("out of order packet, offset %u\n", offset);
 663         // Packet loss or reordering occured. Store the data in the buffer.
 664         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 665         if(rxd < len)
 666                 abort();
 667
 668         // Make note of where we put it.
 669         for(int i = 0; i < NSACKS; i++) {
 670                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 671                         debug("New SACK entry %d\n", i);
 672                         c->sacks[i].offset = offset;
 673                         c->sacks[i].len = rxd;
 674                         break;
 675                 } else if(offset < c->sacks[i].offset) {
 676                         if(offset + rxd < c->sacks[i].offset) { // insert before
 677                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 678                                         debug("Insert SACK entry at %d\n", i);
 679                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
 680                                         c->sacks[i].offset = offset;
 681                                         c->sacks[i].len = rxd;
 682                                 }
 683                                 break;
 684                         } else { // merge
 685                                 debug("Merge with start of SACK entry at %d\n", i);
 686                                 c->sacks[i].offset = offset;
 687                                 break;
 688                         }
 689                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 690                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 691                                 debug("Merge with end of SACK entry at %d\n", i);
 692                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 693                                 // TODO: handle potential merge with next entry
 694                         }
 695                         break;
 696                 }
 697         }
 698
 699         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 700                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 701 }
 702
 703 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 704         // Check if we can process out-of-order data now.
 705         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 706                 debug("incoming packet len %zu connected with SACK at %u\n", len, c->sacks[0].offset);
 707                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 708                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 709                 data = c->rcvbuf.data;
 710         }
 711
 712         if(c->recv) {
 713                 ssize_t rxd = c->recv(c, data, len);
 714                 if(rxd != len) {
 715                         // TODO: handle the application not accepting all data.
 716                         abort();
 717                 }
 718         }
 719
 720         if(c->rcvbuf.used)
 721                 sack_consume(c, len);
 722
 723         c->rcv.nxt += len;
 724 }
 725
 726
 727 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 728         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 729         if(offset + len > c->rcvbuf.maxsize)
 730                 abort();
 731
 732         if(offset)
 733                 handle_out_of_order(c, offset, data, len);
 734         else
 735                 handle_in_order(c, data, len);
 736 }
 737
 738
 739 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 740         if(!utcp) {
 741                 errno = EFAULT;
 742                 return -1;
 743         }
 744
 745         if(!len)
 746                 return 0;
 747
 748         if(!data) {
 749                 errno = EFAULT;
 750                 return -1;
 751         }
 752
 753         print_packet(utcp, "recv", data, len);
 754
 755         // Drop packets smaller than the header
 756
 757         struct hdr hdr;
 758         if(len < sizeof hdr) {
 759                 errno = EBADMSG;
 760                 return -1;
 761         }
 762
 763         // Make a copy from the potentially unaligned data to a struct hdr
 764
 765         memcpy(&hdr, data, sizeof hdr);
 766         data += sizeof hdr;
 767         len -= sizeof hdr;
 768
 769         // Drop packets with an unknown CTL flag
 770
 771         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 772                 errno = EBADMSG;
 773                 return -1;
 774         }
 775
 776         // Try to match the packet to an existing connection
 777
 778         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
 779
 780         // Is it for a new connection?
 781
 782         if(!c) {
 783                 // Ignore RST packets
 784
 785                 if(hdr.ctl & RST)
 786                         return 0;
 787
 788                 // Is it a SYN packet and are we LISTENing?
 789
 790                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
 791                         // If we don't want to accept it, send a RST back
 792                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
 793                                 len = 1;
 794                                 goto reset;
 795                         }
 796
 797                         // Try to allocate memory, otherwise send a RST back
 798                         c = allocate_connection(utcp, hdr.dst, hdr.src);
 799                         if(!c) {
 800                                 len = 1;
 801                                 goto reset;
 802                         }
 803
 804                         // Return SYN+ACK, go to SYN_RECEIVED state
 805                         c->snd.wnd = hdr.wnd;
 806                         c->rcv.irs = hdr.seq;
 807                         c->rcv.nxt = c->rcv.irs + 1;
 808                         set_state(c, SYN_RECEIVED);
 809
 810                         hdr.dst = c->dst;
 811                         hdr.src = c->src;
 812                         hdr.ack = c->rcv.irs + 1;
 813                         hdr.seq = c->snd.iss;
 814                         hdr.ctl = SYN | ACK;
 815                         print_packet(c->utcp, "send", &hdr, sizeof hdr);
 816                         utcp->send(utcp, &hdr, sizeof hdr);
 817                 } else {
 818                         // No, we don't want your packets, send a RST back
 819                         len = 1;
 820                         goto reset;
 821                 }
 822
 823                 return 0;
 824         }
 825
 826         debug("%p state %s\n", c->utcp, strstate[c->state]);
 827
 828         // In case this is for a CLOSED connection, ignore the packet.
 829         // TODO: make it so incoming packets can never match a CLOSED connection.
 830
 831         if(c->state == CLOSED)
 832                 return 0;
 833
 834         // It is for an existing connection.
 835
 836         uint32_t prevrcvnxt = c->rcv.nxt;
 837
 838         // 1. Drop invalid packets.
 839
 840         // 1a. Drop packets that should not happen in our current state.
 841
 842         switch(c->state) {
 843         case SYN_SENT:
 844         case SYN_RECEIVED:
 845         case ESTABLISHED:
 846         case FIN_WAIT_1:
 847         case FIN_WAIT_2:
 848         case CLOSE_WAIT:
 849         case CLOSING:
 850         case LAST_ACK:
 851         case TIME_WAIT:
 852                 break;
 853         default:
 854 #ifdef UTCP_DEBUG
 855                 abort();
 856 #endif
 857                 break;
 858         }
 859
 860         // 1b. Drop packets with a sequence number not in our receive window.
 861
 862         bool acceptable;
 863
 864         if(c->state == SYN_SENT)
 865                 acceptable = true;
 866         else if(len == 0)
 867                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
 868         else {
 869                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
 870
 871                 // cut already accepted front overlapping
 872                 if(rcv_offset < 0) {
 873                         acceptable = rcv_offset + len >= 0;
 874                         if(acceptable) {
 875                                 data -= rcv_offset;
 876                                 len += rcv_offset;
 877                         }
 878                 }
 879
 880                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
 881         }
 882
 883         if(!acceptable) {
 884                 debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
 885                 // Ignore unacceptable RST packets.
 886                 if(hdr.ctl & RST)
 887                         return 0;
 888                 // Otherwise, send an ACK back in the hope things improve.
 889                 ack(c, true);
 890                 return 0;
 891         }
 892
 893         c->snd.wnd = hdr.wnd; // TODO: move below
 894
 895         // 1c. Drop packets with an invalid ACK.
 896         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
 897         // (= snd.una + c->sndbuf.used).
 898
 899         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
 900                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 901                 // Ignore unacceptable RST packets.
 902                 if(hdr.ctl & RST)
 903                         return 0;
 904                 goto reset;
 905         }
 906
 907         // 2. Handle RST packets
 908
 909         if(hdr.ctl & RST) {
 910                 switch(c->state) {
 911                 case SYN_SENT:
 912                         if(!(hdr.ctl & ACK))
 913                                 return 0;
 914                         // The peer has refused our connection.
 915                         set_state(c, CLOSED);
 916                         errno = ECONNREFUSED;
 917                         if(c->recv)
 918                                 c->recv(c, NULL, 0);
 919                         return 0;
 920                 case SYN_RECEIVED:
 921                         if(hdr.ctl & ACK)
 922                                 return 0;
 923                         // We haven't told the application about this connection yet. Silently delete.
 924                         free_connection(c);
 925                         return 0;
 926                 case ESTABLISHED:
 927                 case FIN_WAIT_1:
 928                 case FIN_WAIT_2:
 929                 case CLOSE_WAIT:
 930                         if(hdr.ctl & ACK)
 931                                 return 0;
 932                         // The peer has aborted our connection.
 933                         set_state(c, CLOSED);
 934                         errno = ECONNRESET;
 935                         if(c->recv)
 936                                 c->recv(c, NULL, 0);
 937                         return 0;
 938                 case CLOSING:
 939                 case LAST_ACK:
 940                 case TIME_WAIT:
 941                         if(hdr.ctl & ACK)
 942                                 return 0;
 943                         // As far as the application is concerned, the connection has already been closed.
 944                         // If it has called utcp_close() already, we can immediately free this connection.
 945                         if(c->reapable) {
 946                                 free_connection(c);
 947                                 return 0;
 948                         }
 949                         // Otherwise, immediately move to the CLOSED state.
 950                         set_state(c, CLOSED);
 951                         return 0;
 952                 default:
 953 #ifdef UTCP_DEBUG
 954                         abort();
 955 #endif
 956                         break;
 957                 }
 958         }
 959
 960         // 3. Advance snd.una
 961
 962         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
 963         prevrcvnxt = c->rcv.nxt;
 964
 965         if(advanced) {
 966                 // RTT measurement
 967                 if(c->rtt_start.tv_sec) {
 968                         if(c->rtt_seq == hdr.ack) {
 969                                 struct timeval now, diff;
 970                                 gettimeofday(&now, NULL);
 971                                 timersub(&now, &c->rtt_start, &diff);
 972                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
 973                                 c->rtt_start.tv_sec = 0;
 974                         } else if(c->rtt_seq < hdr.ack) {
 975                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
 976                                 c->rtt_start.tv_sec = 0;
 977                         }
 978                 }
 979
 980                 int32_t data_acked = advanced;
 981
 982                 switch(c->state) {
 983                         case SYN_SENT:
 984                         case SYN_RECEIVED:
 985                                 data_acked--;
 986                                 break;
 987                         // TODO: handle FIN as well.
 988                         default:
 989                                 break;
 990                 }
 991
 992                 assert(data_acked >= 0);
 993
 994                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
 995                 assert(data_acked <= bufused);
 996
 997                 if(data_acked)
 998                         buffer_get(&c->sndbuf, NULL, data_acked);
 999
1000                 // Also advance snd.nxt if possible
1001                 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1002                         c->snd.nxt = hdr.ack;
1003
1004                 c->snd.una = hdr.ack;
1005
1006                 c->dupack = 0;
1007                 c->snd.cwnd += utcp->mtu;
1008                 if(c->snd.cwnd > c->sndbuf.maxsize)
1009                         c->snd.cwnd = c->sndbuf.maxsize;
1010
1011                 // Check if we have sent a FIN that is now ACKed.
1012                 switch(c->state) {
1013                 case FIN_WAIT_1:
1014                         if(c->snd.una == c->snd.last)
1015                                 set_state(c, FIN_WAIT_2);
1016                         break;
1017                 case CLOSING:
1018                         if(c->snd.una == c->snd.last) {
1019                                 gettimeofday(&c->conn_timeout, NULL);
1020                                 c->conn_timeout.tv_sec += 60;
1021                                 set_state(c, TIME_WAIT);
1022                         }
1023                         break;
1024                 default:
1025                         break;
1026                 }
1027         } else {
1028                 if(!len) {
1029                         c->dupack++;
1030                         if(c->dupack == 3) {
1031                                 debug("Triplicate ACK\n");
1032                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1033                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1034                                 //Reset the congestion window so we wait for ACKs.
1035                                 c->snd.nxt = c->snd.una;
1036                                 c->snd.cwnd = utcp->mtu;
1037                                 start_retransmit_timer(c);
1038                         }
1039                 }
1040         }
1041
1042         // 4. Update timers
1043
1044         if(advanced) {
1045                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1046                 if(c->snd.una == c->snd.last)
1047                         stop_retransmit_timer(c);
1048                 else
1049                         start_retransmit_timer(c);
1050         }
1051
1052         // 5. Process SYN stuff
1053
1054         if(hdr.ctl & SYN) {
1055                 switch(c->state) {
1056                 case SYN_SENT:
1057                         // This is a SYNACK. It should always have ACKed the SYN.
1058                         if(!advanced)
1059                                 goto reset;
1060                         c->rcv.irs = hdr.seq;
1061                         c->rcv.nxt = hdr.seq;
1062                         set_state(c, ESTABLISHED);
1063                         // TODO: notify application of this somehow.
1064                         break;
1065                 case SYN_RECEIVED:
1066                 case ESTABLISHED:
1067                 case FIN_WAIT_1:
1068                 case FIN_WAIT_2:
1069                 case CLOSE_WAIT:
1070                 case CLOSING:
1071                 case LAST_ACK:
1072                 case TIME_WAIT:
1073                         // Ehm, no. We should never receive a second SYN.
1074                         goto reset;
1075                 default:
1076 #ifdef UTCP_DEBUG
1077                         abort();
1078 #endif
1079                         return 0;
1080                 }
1081
1082                 // SYN counts as one sequence number
1083                 c->rcv.nxt++;
1084         }
1085
1086         // 6. Process new data
1087
1088         if(c->state == SYN_RECEIVED) {
1089                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1090                 if(!advanced)
1091                         goto reset;
1092
1093                 // Are we still LISTENing?
1094                 if(utcp->accept)
1095                         utcp->accept(c, c->src);
1096
1097                 if(c->state != ESTABLISHED) {
1098                         set_state(c, CLOSED);
1099                         c->reapable = true;
1100                         goto reset;
1101                 }
1102         }
1103
1104         if(len) {
1105                 switch(c->state) {
1106                 case SYN_SENT:
1107                 case SYN_RECEIVED:
1108                         // This should never happen.
1109 #ifdef UTCP_DEBUG
1110                         abort();
1111 #endif
1112                         return 0;
1113                 case ESTABLISHED:
1114                 case FIN_WAIT_1:
1115                 case FIN_WAIT_2:
1116                         break;
1117                 case CLOSE_WAIT:
1118                 case CLOSING:
1119                 case LAST_ACK:
1120                 case TIME_WAIT:
1121                         // Ehm no, We should never receive more data after a FIN.
1122                         goto reset;
1123                 default:
1124 #ifdef UTCP_DEBUG
1125                         abort();
1126 #endif
1127                         return 0;
1128                 }
1129
1130                 handle_incoming_data(c, hdr.seq, data, len);
1131         }
1132
1133         // 7. Process FIN stuff
1134
1135         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
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                         break;
1144                 case ESTABLISHED:
1145                         set_state(c, CLOSE_WAIT);
1146                         break;
1147                 case FIN_WAIT_1:
1148                         set_state(c, CLOSING);
1149                         break;
1150                 case FIN_WAIT_2:
1151                         gettimeofday(&c->conn_timeout, NULL);
1152                         c->conn_timeout.tv_sec += 60;
1153                         set_state(c, TIME_WAIT);
1154                         break;
1155                 case CLOSE_WAIT:
1156                 case CLOSING:
1157                 case LAST_ACK:
1158                 case TIME_WAIT:
1159                         // Ehm, no. We should never receive a second FIN.
1160                         goto reset;
1161                 default:
1162 #ifdef UTCP_DEBUG
1163                         abort();
1164 #endif
1165                         break;
1166                 }
1167
1168                 // FIN counts as one sequence number
1169                 c->rcv.nxt++;
1170                 len++;
1171
1172                 // Inform the application that the peer closed the connection.
1173                 if(c->recv) {
1174                         errno = 0;
1175                         c->recv(c, NULL, 0);
1176                 }
1177         }
1178
1179         // Now we send something back if:
1180         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1181         //   -> sendatleastone = true
1182         // - or we got an ack, so we should maybe send a bit more data
1183         //   -> sendatleastone = false
1184
1185         ack(c, len || prevrcvnxt != c->rcv.nxt);
1186         return 0;
1187
1188 reset:
1189         swap_ports(&hdr);
1190         hdr.wnd = 0;
1191         if(hdr.ctl & ACK) {
1192                 hdr.seq = hdr.ack;
1193                 hdr.ctl = RST;
1194         } else {
1195                 hdr.ack = hdr.seq + len;
1196                 hdr.seq = 0;
1197                 hdr.ctl = RST | ACK;
1198         }
1199         print_packet(utcp, "send", &hdr, sizeof hdr);
1200         utcp->send(utcp, &hdr, sizeof hdr);
1201         return 0;
1202
1203 }
1204
1205 int utcp_shutdown(struct utcp_connection *c, int dir) {
1206         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1207         if(!c) {
1208                 errno = EFAULT;
1209                 return -1;
1210         }
1211
1212         if(c->reapable) {
1213                 debug("Error: shutdown() called on closed connection %p\n", c);
1214                 errno = EBADF;
1215                 return -1;
1216         }
1217
1218         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1219                 errno = EINVAL;
1220                 return -1;
1221         }
1222
1223         // TCP does not have a provision for stopping incoming packets.
1224         // The best we can do is to just ignore them.
1225         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1226                 c->recv = NULL;
1227
1228         // The rest of the code deals with shutting down writes.
1229         if(dir == UTCP_SHUT_RD)
1230                 return 0;
1231
1232         switch(c->state) {
1233         case CLOSED:
1234         case LISTEN:
1235                 errno = ENOTCONN;
1236                 return -1;
1237
1238         case SYN_SENT:
1239                 set_state(c, CLOSED);
1240                 return 0;
1241
1242         case SYN_RECEIVED:
1243         case ESTABLISHED:
1244                 set_state(c, FIN_WAIT_1);
1245                 break;
1246         case FIN_WAIT_1:
1247         case FIN_WAIT_2:
1248                 return 0;
1249         case CLOSE_WAIT:
1250                 set_state(c, CLOSING);
1251                 break;
1252
1253         case CLOSING:
1254         case LAST_ACK:
1255         case TIME_WAIT:
1256                 return 0;
1257         }
1258
1259         c->snd.last++;
1260
1261         ack(c, false);
1262         return 0;
1263 }
1264
1265 int utcp_close(struct utcp_connection *c) {
1266         if(utcp_shutdown(c, SHUT_RDWR))
1267                 return -1;
1268         c->recv = NULL;
1269         c->poll = NULL;
1270         c->reapable = true;
1271         return 0;
1272 }
1273
1274 int utcp_abort(struct utcp_connection *c) {
1275         if(!c) {
1276                 errno = EFAULT;
1277                 return -1;
1278         }
1279
1280         if(c->reapable) {
1281                 debug("Error: abort() called on closed connection %p\n", c);
1282                 errno = EBADF;
1283                 return -1;
1284         }
1285
1286         c->recv = NULL;
1287         c->poll = NULL;
1288         c->reapable = true;
1289
1290         switch(c->state) {
1291         case CLOSED:
1292                 return 0;
1293         case LISTEN:
1294         case SYN_SENT:
1295         case CLOSING:
1296         case LAST_ACK:
1297         case TIME_WAIT:
1298                 set_state(c, CLOSED);
1299                 return 0;
1300
1301         case SYN_RECEIVED:
1302         case ESTABLISHED:
1303         case FIN_WAIT_1:
1304         case FIN_WAIT_2:
1305         case CLOSE_WAIT:
1306                 set_state(c, CLOSED);
1307                 break;
1308         }
1309
1310         // Send RST
1311
1312         struct hdr hdr;
1313
1314         hdr.src = c->src;
1315         hdr.dst = c->dst;
1316         hdr.seq = c->snd.nxt;
1317         hdr.ack = 0;
1318         hdr.wnd = 0;
1319         hdr.ctl = RST;
1320
1321         print_packet(c->utcp, "send", &hdr, sizeof hdr);
1322         c->utcp->send(c->utcp, &hdr, sizeof hdr);
1323         return 0;
1324 }
1325
1326 /* Handle timeouts.
1327  * One call to this function will loop through all connections,
1328  * checking if something needs to be resent or not.
1329  * The return value is the time to the next timeout in milliseconds,
1330  * or maybe a negative value if the timeout is infinite.
1331  */
1332 struct timeval utcp_timeout(struct utcp *utcp) {
1333         struct timeval now;
1334         gettimeofday(&now, NULL);
1335         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1336
1337         for(int i = 0; i < utcp->nconnections; i++) {
1338                 struct utcp_connection *c = utcp->connections[i];
1339                 if(!c)
1340                         continue;
1341
1342                 // delete connections that have been utcp_close()d.
1343                 if(c->state == CLOSED) {
1344                         if(c->reapable) {
1345                                 debug("Reaping %p\n", c);
1346                                 free_connection(c);
1347                                 i--;
1348                         }
1349                         continue;
1350                 }
1351
1352                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1353                         errno = ETIMEDOUT;
1354                         c->state = CLOSED;
1355                         if(c->recv)
1356                                 c->recv(c, NULL, 0);
1357                         continue;
1358                 }
1359
1360                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1361                         debug("retransmit()\n");
1362                         retransmit(c);
1363                 }
1364
1365                 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1366                         c->poll(c, buffer_free(&c->sndbuf));
1367
1368                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1369                         next = c->conn_timeout;
1370
1371                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1372                         next = c->rtrx_timeout;
1373         }
1374
1375         struct timeval diff;
1376         timersub(&next, &now, &diff);
1377         return diff;
1378 }
1379
1380 bool utcp_is_active(struct utcp *utcp) {
1381         if(!utcp)
1382                 return false;
1383
1384         for(int i = 0; i < utcp->nconnections; i++)
1385                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1386                         return true;
1387
1388         return false;
1389 }
1390
1391 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1392         if(!send) {
1393                 errno = EFAULT;
1394                 return NULL;
1395         }
1396
1397         struct utcp *utcp = calloc(1, sizeof *utcp);
1398         if(!utcp)
1399                 return NULL;
1400
1401         utcp->accept = accept;
1402         utcp->pre_accept = pre_accept;
1403         utcp->send = send;
1404         utcp->priv = priv;
1405         utcp->mtu = DEFAULT_MTU;
1406         utcp->timeout = DEFAULT_USER_TIMEOUT; // s
1407         utcp->rto = START_RTO; // us
1408
1409         return utcp;
1410 }
1411
1412 void utcp_exit(struct utcp *utcp) {
1413         if(!utcp)
1414                 return;
1415         for(int i = 0; i < utcp->nconnections; i++) {
1416                 if(!utcp->connections[i]->reapable)
1417                         debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1418                 buffer_exit(&utcp->connections[i]->sndbuf);
1419                 free(utcp->connections[i]);
1420         }
1421         free(utcp->connections);
1422         free(utcp);
1423 }
1424
1425 uint16_t utcp_get_mtu(struct utcp *utcp) {
1426         return utcp ? utcp->mtu : 0;
1427 }
1428
1429 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1430         // TODO: handle overhead of the header
1431         if(utcp)
1432                 utcp->mtu = mtu;
1433 }
1434
1435 int utcp_get_user_timeout(struct utcp *u) {
1436         return u ? u->timeout : 0;
1437 }
1438
1439 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1440         if(u)
1441                 u->timeout = timeout;
1442 }
1443
1444 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1445         return c ? c->sndbuf.maxsize : 0;
1446 }
1447
1448 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1449         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1450                 return buffer_free(&c->sndbuf);
1451         else
1452                 return 0;
1453 }
1454
1455 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1456         if(!c)
1457                 return;
1458         c->sndbuf.maxsize = size;
1459         if(c->sndbuf.maxsize != size)
1460                 c->sndbuf.maxsize = -1;
1461 }
1462
1463 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1464         return c ? c->rcvbuf.maxsize : 0;
1465 }
1466
1467 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1468         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1469                 return buffer_free(&c->rcvbuf);
1470         else
1471                 return 0;
1472 }
1473
1474 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1475         if(!c)
1476                 return;
1477         c->rcvbuf.maxsize = size;
1478         if(c->rcvbuf.maxsize != size)
1479                 c->rcvbuf.maxsize = -1;
1480 }
1481
1482 bool utcp_get_nodelay(struct utcp_connection *c) {
1483         return c ? c->nodelay : false;
1484 }
1485
1486 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1487         if(c)
1488                 c->nodelay = nodelay;
1489 }
1490
1491 bool utcp_get_keepalive(struct utcp_connection *c) {
1492         return c ? c->keepalive : false;
1493 }
1494
1495 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1496         if(c)
1497                 c->keepalive = keepalive;
1498 }
1499
1500 size_t utcp_get_outq(struct utcp_connection *c) {
1501         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1502 }
1503
1504 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1505         if(c)
1506                 c->recv = recv;
1507 }
1508
1509 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1510         if(c)
1511                 c->poll = poll;
1512 }
1513
1514 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1515         if(utcp) {
1516                 utcp->accept = accept;
1517                 utcp->pre_accept = pre_accept;
1518         }
1519 }