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 (%lu bytes)\n", utcp, dir, (unsigned long)len);
  74                 return;
  75         }
  76
  77         memcpy(&hdr, pkt, sizeof hdr);
  78         debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, (unsigned long)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                 const uint8_t *data = (uint8_t *)pkt + sizeof hdr;
  91                 char str[datalen * 2 + 1];
  92                 char *p = str;
  93
  94                 for(uint32_t i = 0; i < datalen; i++) {
  95                         *p++ = "0123456789ABCDEF"[data[i] >> 4];
  96                         *p++ = "0123456789ABCDEF"[data[i] & 15];
  97                 }
  98                 *p = 0;
  99
 100                 debug(" data=%s", str);
 101         }
 102
 103         debug("\n");
 104 }
 105 #else
 106 #define debug(...)
 107 #define print_packet(...)
 108 #endif
 109
 110 static void set_state(struct utcp_connection *c, enum state state) {
 111         c->state = state;
 112         if(state == ESTABLISHED)
 113                 timerclear(&c->conn_timeout);
 114         debug("%p new state: %s\n", c->utcp, strstate[state]);
 115 }
 116
 117 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 118         if(seq != c->snd.last)
 119                 return false;
 120         switch(c->state) {
 121         case FIN_WAIT_1:
 122         case CLOSING:
 123         case LAST_ACK:
 124                 return true;
 125         default:
 126                 return false;
 127         }
 128 }
 129
 130 static bool is_reliable(struct utcp_connection *c) {
 131         return c->flags & UTCP_RELIABLE;
 132 }
 133
 134 static inline void list_connections(struct utcp *utcp) {
 135         debug("%p has %d connections:\n", utcp, utcp->nconnections);
 136         for(int i = 0; i < utcp->nconnections; i++)
 137                 debug("  %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
 138 }
 139
 140 static int32_t seqdiff(uint32_t a, uint32_t b) {
 141         return a - b;
 142 }
 143
 144 // Buffer functions
 145 // TODO: convert to ringbuffers to avoid memmove() operations.
 146
 147 // Store data into the buffer
 148 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 149         debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 150
 151         size_t required = offset + len;
 152         if(required > buf->maxsize) {
 153                 if(offset >= buf->maxsize)
 154                         return 0;
 155                 len = buf->maxsize - offset;
 156                 required = buf->maxsize;
 157         }
 158
 159         if(required > buf->size) {
 160                 size_t newsize = buf->size;
 161                 if(!newsize) {
 162                         newsize = required;
 163                 } else {
 164                         do {
 165                                 newsize *= 2;
 166                         } while(newsize < required);
 167                 }
 168                 if(newsize > buf->maxsize)
 169                         newsize = buf->maxsize;
 170                 char *newdata = realloc(buf->data, newsize);
 171                 if(!newdata)
 172                         return -1;
 173                 buf->data = newdata;
 174                 buf->size = newsize;
 175         }
 176
 177         memcpy(buf->data + offset, data, len);
 178         if(required > buf->used)
 179                 buf->used = required;
 180         return len;
 181 }
 182
 183 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 184         return buffer_put_at(buf, buf->used, data, len);
 185 }
 186
 187 // Get data from the buffer. data can be NULL.
 188 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 189         if(len > buf->used)
 190                 len = buf->used;
 191         if(data)
 192                 memcpy(data, buf->data, len);
 193         if(len < buf->used)
 194                 memmove(buf->data, buf->data + len, buf->used - len);
 195         buf->used -= len;
 196         return len;
 197 }
 198
 199 // Copy data from the buffer without removing it.
 200 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 201         if(offset >= buf->used)
 202                 return 0;
 203         if(offset + len > buf->used)
 204                 len = buf->used - offset;
 205         memcpy(data, buf->data + offset, len);
 206         return len;
 207 }
 208
 209 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 210         memset(buf, 0, sizeof *buf);
 211         if(len) {
 212                 buf->data = malloc(len);
 213                 if(!buf->data)
 214                         return false;
 215         }
 216         buf->size = len;
 217         buf->maxsize = maxlen;
 218         return true;
 219 }
 220
 221 static void buffer_exit(struct buffer *buf) {
 222         free(buf->data);
 223         memset(buf, 0, sizeof *buf);
 224 }
 225
 226 static uint32_t buffer_free(const struct buffer *buf) {
 227         return buf->maxsize - buf->used;
 228 }
 229
 230 // Connections are stored in a sorted list.
 231 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 232
 233 static int compare(const void *va, const void *vb) {
 234         assert(va && vb);
 235
 236         const struct utcp_connection *a = *(struct utcp_connection **)va;
 237         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 238
 239         assert(a && b);
 240         assert(a->src && b->src);
 241
 242         int c = (int)a->src - (int)b->src;
 243         if(c)
 244                 return c;
 245         c = (int)a->dst - (int)b->dst;
 246         return c;
 247 }
 248
 249 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 250         if(!utcp->nconnections)
 251                 return NULL;
 252         struct utcp_connection key = {
 253                 .src = src,
 254                 .dst = dst,
 255         }, *keyp = &key;
 256         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 257         return match ? *match : NULL;
 258 }
 259
 260 static void free_connection(struct utcp_connection *c) {
 261         struct utcp *utcp = c->utcp;
 262         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 263
 264         assert(cp);
 265
 266         int i = cp - utcp->connections;
 267         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
 268         utcp->nconnections--;
 269
 270         buffer_exit(&c->rcvbuf);
 271         buffer_exit(&c->sndbuf);
 272         free(c);
 273 }
 274
 275 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 276         // Check whether this combination of src and dst is free
 277
 278         if(src) {
 279                 if(find_connection(utcp, src, dst)) {
 280                         errno = EADDRINUSE;
 281                         return NULL;
 282                 }
 283         } else { // If src == 0, generate a random port number with the high bit set
 284                 if(utcp->nconnections >= 32767) {
 285                         errno = ENOMEM;
 286                         return NULL;
 287                 }
 288                 src = rand() | 0x8000;
 289                 while(find_connection(utcp, src, dst))
 290                         src++;
 291         }
 292
 293         // Allocate memory for the new connection
 294
 295         if(utcp->nconnections >= utcp->nallocated) {
 296                 if(!utcp->nallocated)
 297                         utcp->nallocated = 4;
 298                 else
 299                         utcp->nallocated *= 2;
 300                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
 301                 if(!new_array)
 302                         return NULL;
 303                 utcp->connections = new_array;
 304         }
 305
 306         struct utcp_connection *c = calloc(1, sizeof *c);
 307         if(!c)
 308                 return NULL;
 309
 310         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 311                 free(c);
 312                 return NULL;
 313         }
 314
 315         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 316                 buffer_exit(&c->sndbuf);
 317                 free(c);
 318                 return NULL;
 319         }
 320
 321         // Fill in the details
 322
 323         c->src = src;
 324         c->dst = dst;
 325 #ifdef UTCP_DEBUG
 326         c->snd.iss = 0;
 327 #else
 328         c->snd.iss = rand();
 329 #endif
 330         c->snd.una = c->snd.iss;
 331         c->snd.nxt = c->snd.iss + 1;
 332         c->rcv.wnd = utcp->mtu;
 333         c->snd.last = c->snd.nxt;
 334         c->snd.cwnd = utcp->mtu;
 335         c->utcp = utcp;
 336
 337         // Add it to the sorted list of connections
 338
 339         utcp->connections[utcp->nconnections++] = c;
 340         qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 341
 342         return c;
 343 }
 344
 345 // Update RTT variables. See RFC 6298.
 346 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 347         if(!rtt) {
 348                 debug("invalid rtt\n");
 349                 return;
 350         }
 351
 352         struct utcp *utcp = c->utcp;
 353
 354         if(!utcp->srtt) {
 355                 utcp->srtt = rtt;
 356                 utcp->rttvar = rtt / 2;
 357                 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
 358         } else {
 359                 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
 360                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 361                 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
 362         }
 363
 364         if(utcp->rto > MAX_RTO)
 365                 utcp->rto = MAX_RTO;
 366
 367         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 368 }
 369
 370 static void start_retransmit_timer(struct utcp_connection *c) {
 371         gettimeofday(&c->rtrx_timeout, NULL);
 372         c->rtrx_timeout.tv_usec += c->utcp->rto;
 373         while(c->rtrx_timeout.tv_usec >= 1000000) {
 374                 c->rtrx_timeout.tv_usec -= 1000000;
 375                 c->rtrx_timeout.tv_sec++;
 376         }
 377         debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
 378 }
 379
 380 static void stop_retransmit_timer(struct utcp_connection *c) {
 381         timerclear(&c->rtrx_timeout);
 382         debug("timeout cleared\n");
 383 }
 384
 385 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 386         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 387         if(!c)
 388                 return NULL;
 389
 390         assert((flags & ~0xf) == 0);
 391
 392         c->flags = flags;
 393         c->recv = recv;
 394         c->priv = priv;
 395
 396         struct {
 397                 struct hdr hdr;
 398                 uint8_t init[4];
 399         } pkt;
 400
 401         pkt.hdr.src = c->src;
 402         pkt.hdr.dst = c->dst;
 403         pkt.hdr.seq = c->snd.iss;
 404         pkt.hdr.ack = 0;
 405         pkt.hdr.wnd = c->rcv.wnd;
 406         pkt.hdr.ctl = SYN;
 407         pkt.hdr.aux = 0x0101;
 408         pkt.init[0] = 1;
 409         pkt.init[1] = 0;
 410         pkt.init[2] = 0;
 411         pkt.init[3] = flags & 0x7;
 412
 413         set_state(c, SYN_SENT);
 414
 415         print_packet(utcp, "send", &pkt, sizeof pkt);
 416         utcp->send(utcp, &pkt, sizeof pkt);
 417
 418         gettimeofday(&c->conn_timeout, NULL);
 419         c->conn_timeout.tv_sec += utcp->timeout;
 420
 421         return c;
 422 }
 423
 424 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 425         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 426 }
 427
 428 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 429         if(c->reapable || c->state != SYN_RECEIVED) {
 430                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 431                 return;
 432         }
 433
 434         debug("%p accepted, %p %p\n", c, recv, priv);
 435         c->recv = recv;
 436         c->priv = priv;
 437         set_state(c, ESTABLISHED);
 438 }
 439
 440 static void ack(struct utcp_connection *c, bool sendatleastone) {
 441         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 442         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 443         debug("cwndleft = %d\n", cwndleft);
 444
 445         assert(left >= 0);
 446
 447         if(cwndleft <= 0)
 448                 cwndleft = 0;
 449
 450         if(cwndleft < left)
 451                 left = cwndleft;
 452
 453         if(!left && !sendatleastone)
 454                 return;
 455
 456         struct {
 457                 struct hdr hdr;
 458                 char data[];
 459         } *pkt;
 460
 461         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 462         if(!pkt)
 463                 return;
 464
 465         pkt->hdr.src = c->src;
 466         pkt->hdr.dst = c->dst;
 467         pkt->hdr.ack = c->rcv.nxt;
 468         pkt->hdr.wnd = c->snd.wnd;
 469         pkt->hdr.ctl = ACK;
 470         pkt->hdr.aux = 0;
 471
 472         do {
 473                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 474                 pkt->hdr.seq = c->snd.nxt;
 475
 476                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 477
 478                 c->snd.nxt += seglen;
 479                 left -= seglen;
 480
 481                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 482                         seglen--;
 483                         pkt->hdr.ctl |= FIN;
 484                 }
 485
 486                 if(!c->rtt_start.tv_sec) {
 487                         // Start RTT measurement
 488                         gettimeofday(&c->rtt_start, NULL);
 489                         c->rtt_seq = pkt->hdr.seq + seglen;
 490                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 491                 }
 492
 493                 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
 494                 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
 495         } while(left);
 496
 497         free(pkt);
 498 }
 499
 500 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 501         if(c->reapable) {
 502                 debug("Error: send() called on closed connection %p\n", c);
 503                 errno = EBADF;
 504                 return -1;
 505         }
 506
 507         switch(c->state) {
 508         case CLOSED:
 509         case LISTEN:
 510         case SYN_SENT:
 511         case SYN_RECEIVED:
 512                 debug("Error: send() called on unconnected connection %p\n", c);
 513                 errno = ENOTCONN;
 514                 return -1;
 515         case ESTABLISHED:
 516         case CLOSE_WAIT:
 517                 break;
 518         case FIN_WAIT_1:
 519         case FIN_WAIT_2:
 520         case CLOSING:
 521         case LAST_ACK:
 522         case TIME_WAIT:
 523                 debug("Error: send() called on closing connection %p\n", c);
 524                 errno = EPIPE;
 525                 return -1;
 526         }
 527
 528         // Exit early if we have nothing to send.
 529
 530         if(!len)
 531                 return 0;
 532
 533         if(!data) {
 534                 errno = EFAULT;
 535                 return -1;
 536         }
 537
 538         // Add data to send buffer.
 539
 540         len = buffer_put(&c->sndbuf, data, len);
 541         if(len <= 0) {
 542                 errno = EWOULDBLOCK;
 543                 return 0;
 544         }
 545
 546         c->snd.last += len;
 547         ack(c, false);
 548         if(!is_reliable(c)) {
 549                 c->snd.una = c->snd.nxt = c->snd.last;
 550                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 551         }
 552         if(is_reliable(c) && !timerisset(&c->rtrx_timeout))
 553                 start_retransmit_timer(c);
 554         return len;
 555 }
 556
 557 static void swap_ports(struct hdr *hdr) {
 558         uint16_t tmp = hdr->src;
 559         hdr->src = hdr->dst;
 560         hdr->dst = tmp;
 561 }
 562
 563 static void retransmit(struct utcp_connection *c) {
 564         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 565                 debug("Retransmit() called but nothing to retransmit!\n");
 566                 stop_retransmit_timer(c);
 567                 return;
 568         }
 569
 570         struct utcp *utcp = c->utcp;
 571
 572         struct {
 573                 struct hdr hdr;
 574                 char data[];
 575         } *pkt;
 576
 577         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 578         if(!pkt)
 579                 return;
 580
 581         pkt->hdr.src = c->src;
 582         pkt->hdr.dst = c->dst;
 583         pkt->hdr.wnd = c->rcv.wnd;
 584         pkt->hdr.aux = 0;
 585
 586         switch(c->state) {
 587                 case SYN_SENT:
 588                         // Send our SYN again
 589                         pkt->hdr.seq = c->snd.iss;
 590                         pkt->hdr.ack = 0;
 591                         pkt->hdr.ctl = SYN;
 592                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 593                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 594                         break;
 595
 596                 case SYN_RECEIVED:
 597                         // Send SYNACK again
 598                         pkt->hdr.seq = c->snd.nxt;
 599                         pkt->hdr.ack = c->rcv.nxt;
 600                         pkt->hdr.ctl = SYN | ACK;
 601                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 602                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 603                         break;
 604
 605                 case ESTABLISHED:
 606                 case FIN_WAIT_1:
 607                 case CLOSE_WAIT:
 608                 case CLOSING:
 609                 case LAST_ACK:
 610                         // Send unacked data again.
 611                         pkt->hdr.seq = c->snd.una;
 612                         pkt->hdr.ack = c->rcv.nxt;
 613                         pkt->hdr.ctl = ACK;
 614                         uint32_t len = seqdiff(c->snd.last, c->snd.una);
 615                         if(len > utcp->mtu)
 616                                 len = utcp->mtu;
 617                         if(fin_wanted(c, c->snd.una + len)) {
 618                                 len--;
 619                                 pkt->hdr.ctl |= FIN;
 620                         }
 621                         c->snd.nxt = c->snd.una + len;
 622                         c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 623                         buffer_copy(&c->sndbuf, pkt->data, 0, len);
 624                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
 625                         utcp->send(utcp, pkt, sizeof pkt->hdr + len);
 626                         break;
 627
 628                 case CLOSED:
 629                 case LISTEN:
 630                 case TIME_WAIT:
 631                 case FIN_WAIT_2:
 632                         // We shouldn't need to retransmit anything in this state.
 633 #ifdef UTCP_DEBUG
 634                         abort();
 635 #endif
 636                         stop_retransmit_timer(c);
 637                         goto cleanup;
 638         }
 639
 640         start_retransmit_timer(c);
 641         utcp->rto *= 2;
 642         if(utcp->rto > MAX_RTO)
 643                 utcp->rto = MAX_RTO;
 644         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 645
 646 cleanup:
 647         free(pkt);
 648 }
 649
 650 /* Update receive buffer and SACK entries after consuming data.
 651  *
 652  * Situation:
 653  *
 654  * |.....0000..1111111111.....22222......3333|
 655  * |---------------^
 656  *
 657  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 658  * to remove data from the receive buffer. The idea is to substract "len"
 659  * from the offset of all the SACK entries, and then remove/cut down entries
 660  * that are shifted to before the start of the receive buffer.
 661  *
 662  * There are three cases:
 663  * - the SACK entry is after ^, in that case just change the offset.
 664  * - the SACK entry starts before and ends after ^, so we have to
 665  *   change both its offset and size.
 666  * - the SACK entry is completely before ^, in that case delete it.
 667  */
 668 static void sack_consume(struct utcp_connection *c, size_t len) {
 669         debug("sack_consume %lu\n", (unsigned long)len);
 670         if(len > c->rcvbuf.used) {
 671                 debug("All SACK entries consumed");
 672                 c->sacks[0].len = 0;
 673                 return;
 674         }
 675
 676         buffer_get(&c->rcvbuf, NULL, len);
 677
 678         for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
 679                 if(len < c->sacks[i].offset) {
 680                         c->sacks[i].offset -= len;
 681                         i++;
 682                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 683                         c->sacks[i].len -= len - c->sacks[i].offset;
 684                         c->sacks[i].offset = 0;
 685                         i++;
 686                 } else {
 687                         if(i < NSACKS - 1) {
 688                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
 689                                 c->sacks[NSACKS - 1].len = 0;
 690                         } else {
 691                                 c->sacks[i].len = 0;
 692                                 break;
 693                         }
 694                 }
 695         }
 696
 697         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 698                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 699 }
 700
 701 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 702         debug("out of order packet, offset %u\n", offset);
 703         // Packet loss or reordering occured. Store the data in the buffer.
 704         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 705         if(rxd < len)
 706                 abort();
 707
 708         // Make note of where we put it.
 709         for(int i = 0; i < NSACKS; i++) {
 710                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 711                         debug("New SACK entry %d\n", i);
 712                         c->sacks[i].offset = offset;
 713                         c->sacks[i].len = rxd;
 714                         break;
 715                 } else if(offset < c->sacks[i].offset) {
 716                         if(offset + rxd < c->sacks[i].offset) { // insert before
 717                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 718                                         debug("Insert SACK entry at %d\n", i);
 719                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
 720                                         c->sacks[i].offset = offset;
 721                                         c->sacks[i].len = rxd;
 722                                 } else {
 723                                         debug("SACK entries full, dropping packet\n");
 724                                 }
 725                                 break;
 726                         } else { // merge
 727                                 debug("Merge with start of SACK entry at %d\n", i);
 728                                 c->sacks[i].offset = offset;
 729                                 break;
 730                         }
 731                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 732                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 733                                 debug("Merge with end of SACK entry at %d\n", i);
 734                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 735                                 // TODO: handle potential merge with next entry
 736                         }
 737                         break;
 738                 }
 739         }
 740
 741         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 742                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 743 }
 744
 745 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 746         // Check if we can process out-of-order data now.
 747         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 748                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 749                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 750                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 751                 data = c->rcvbuf.data;
 752         }
 753
 754         if(c->recv) {
 755                 ssize_t rxd = c->recv(c, data, len);
 756                 if(rxd != len) {
 757                         // TODO: handle the application not accepting all data.
 758                         abort();
 759                 }
 760         }
 761
 762         if(c->rcvbuf.used)
 763                 sack_consume(c, len);
 764
 765         c->rcv.nxt += len;
 766 }
 767
 768
 769 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 770         if(!is_reliable(c)) {
 771                 c->recv(c, data, len);
 772                 c->rcv.nxt = seq + len;
 773                 return;
 774         }
 775
 776         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 777         if(offset + len > c->rcvbuf.maxsize)
 778                 abort();
 779
 780         if(offset)
 781                 handle_out_of_order(c, offset, data, len);
 782         else
 783                 handle_in_order(c, data, len);
 784 }
 785
 786
 787 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 788         if(!utcp) {
 789                 errno = EFAULT;
 790                 return -1;
 791         }
 792
 793         if(!len)
 794                 return 0;
 795
 796         if(!data) {
 797                 errno = EFAULT;
 798                 return -1;
 799         }
 800
 801         print_packet(utcp, "recv", data, len);
 802
 803         // Drop packets smaller than the header
 804
 805         struct hdr hdr;
 806         if(len < sizeof hdr) {
 807                 errno = EBADMSG;
 808                 return -1;
 809         }
 810
 811         // Make a copy from the potentially unaligned data to a struct hdr
 812
 813         memcpy(&hdr, data, sizeof hdr);
 814         data += sizeof hdr;
 815         len -= sizeof hdr;
 816
 817         // Drop packets with an unknown CTL flag
 818
 819         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 820                 errno = EBADMSG;
 821                 return -1;
 822         }
 823
 824         // Try to match the packet to an existing connection
 825
 826         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
 827
 828         // Is it for a new connection?
 829
 830         if(!c) {
 831                 // Ignore RST packets
 832
 833                 if(hdr.ctl & RST)
 834                         return 0;
 835
 836                 // Is it a SYN packet and are we LISTENing?
 837
 838                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
 839                         // If we don't want to accept it, send a RST back
 840                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
 841                                 len = 1;
 842                                 goto reset;
 843                         }
 844
 845                         // Try to allocate memory, otherwise send a RST back
 846                         c = allocate_connection(utcp, hdr.dst, hdr.src);
 847                         if(!c) {
 848                                 len = 1;
 849                                 goto reset;
 850                         }
 851
 852                         // Parse auxilliary information
 853                         if(hdr.aux) {
 854                                 if(hdr.aux != 0x0101 || len < 4 || ((uint8_t *)data)[0] != 1) {
 855                                         len = 1;
 856                                         goto reset;
 857                                 }
 858                                 c->flags = ((uint8_t *)data)[3] & 0x7;
 859                                 data += 4;
 860                                 len -= 4;
 861                         } else {
 862                                 c->flags = UTCP_TCP;
 863                         }
 864
 865                         // Return SYN+ACK, go to SYN_RECEIVED state
 866                         c->snd.wnd = hdr.wnd;
 867                         c->rcv.irs = hdr.seq;
 868                         c->rcv.nxt = c->rcv.irs + 1;
 869                         set_state(c, SYN_RECEIVED);
 870
 871                         hdr.dst = c->dst;
 872                         hdr.src = c->src;
 873                         hdr.ack = c->rcv.irs + 1;
 874                         hdr.seq = c->snd.iss;
 875                         hdr.ctl = SYN | ACK;
 876                         print_packet(c->utcp, "send", &hdr, sizeof hdr);
 877                         utcp->send(utcp, &hdr, sizeof hdr);
 878                 } else {
 879                         // No, we don't want your packets, send a RST back
 880                         len = 1;
 881                         goto reset;
 882                 }
 883
 884                 return 0;
 885         }
 886
 887         debug("%p state %s\n", c->utcp, strstate[c->state]);
 888
 889         // In case this is for a CLOSED connection, ignore the packet.
 890         // TODO: make it so incoming packets can never match a CLOSED connection.
 891
 892         if(c->state == CLOSED) {
 893                 debug("Got packet for closed connection\n");
 894                 return 0;
 895         }
 896
 897         // It is for an existing connection.
 898
 899         uint32_t prevrcvnxt = c->rcv.nxt;
 900
 901         // 1. Drop invalid packets.
 902
 903         // 1a. Drop packets that should not happen in our current state.
 904
 905         switch(c->state) {
 906         case SYN_SENT:
 907         case SYN_RECEIVED:
 908         case ESTABLISHED:
 909         case FIN_WAIT_1:
 910         case FIN_WAIT_2:
 911         case CLOSE_WAIT:
 912         case CLOSING:
 913         case LAST_ACK:
 914         case TIME_WAIT:
 915                 break;
 916         default:
 917 #ifdef UTCP_DEBUG
 918                 abort();
 919 #endif
 920                 break;
 921         }
 922
 923         // 1b. Drop packets with a sequence number not in our receive window.
 924
 925         bool acceptable;
 926
 927         if(c->state == SYN_SENT)
 928                 acceptable = true;
 929         else if(len == 0)
 930                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
 931         else {
 932                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
 933
 934                 // cut already accepted front overlapping
 935                 if(rcv_offset < 0) {
 936                         acceptable = len > -rcv_offset;
 937                         if(acceptable) {
 938                                 data -= rcv_offset;
 939                                 len += rcv_offset;
 940                                 hdr.seq -= rcv_offset;
 941                         }
 942                 } else {
 943                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
 944                 }
 945         }
 946
 947         if(!acceptable) {
 948                 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
 949                 // Ignore unacceptable RST packets.
 950                 if(hdr.ctl & RST)
 951                         return 0;
 952                 // Otherwise, continue processing.
 953                 len = 0;
 954         }
 955
 956         c->snd.wnd = hdr.wnd; // TODO: move below
 957
 958         // 1c. Drop packets with an invalid ACK.
 959         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
 960         // (= snd.una + c->sndbuf.used).
 961
 962         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
 963                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 964                 // Ignore unacceptable RST packets.
 965                 if(hdr.ctl & RST)
 966                         return 0;
 967                 goto reset;
 968         }
 969
 970         // 2. Handle RST packets
 971
 972         if(hdr.ctl & RST) {
 973                 switch(c->state) {
 974                 case SYN_SENT:
 975                         if(!(hdr.ctl & ACK))
 976                                 return 0;
 977                         // The peer has refused our connection.
 978                         set_state(c, CLOSED);
 979                         errno = ECONNREFUSED;
 980                         if(c->recv)
 981                                 c->recv(c, NULL, 0);
 982                         return 0;
 983                 case SYN_RECEIVED:
 984                         if(hdr.ctl & ACK)
 985                                 return 0;
 986                         // We haven't told the application about this connection yet. Silently delete.
 987                         free_connection(c);
 988                         return 0;
 989                 case ESTABLISHED:
 990                 case FIN_WAIT_1:
 991                 case FIN_WAIT_2:
 992                 case CLOSE_WAIT:
 993                         if(hdr.ctl & ACK)
 994                                 return 0;
 995                         // The peer has aborted our connection.
 996                         set_state(c, CLOSED);
 997                         errno = ECONNRESET;
 998                         if(c->recv)
 999                                 c->recv(c, NULL, 0);
1000                         return 0;
1001                 case CLOSING:
1002                 case LAST_ACK:
1003                 case TIME_WAIT:
1004                         if(hdr.ctl & ACK)
1005                                 return 0;
1006                         // As far as the application is concerned, the connection has already been closed.
1007                         // If it has called utcp_close() already, we can immediately free this connection.
1008                         if(c->reapable) {
1009                                 free_connection(c);
1010                                 return 0;
1011                         }
1012                         // Otherwise, immediately move to the CLOSED state.
1013                         set_state(c, CLOSED);
1014                         return 0;
1015                 default:
1016 #ifdef UTCP_DEBUG
1017                         abort();
1018 #endif
1019                         break;
1020                 }
1021         }
1022
1023         // 3. Advance snd.una
1024
1025         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
1026         prevrcvnxt = c->rcv.nxt;
1027
1028         if(advanced) {
1029                 // RTT measurement
1030                 if(c->rtt_start.tv_sec) {
1031                         if(c->rtt_seq == hdr.ack) {
1032                                 struct timeval now, diff;
1033                                 gettimeofday(&now, NULL);
1034                                 timersub(&now, &c->rtt_start, &diff);
1035                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1036                                 c->rtt_start.tv_sec = 0;
1037                         } else if(c->rtt_seq < hdr.ack) {
1038                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1039                                 c->rtt_start.tv_sec = 0;
1040                         }
1041                 }
1042
1043                 int32_t data_acked = advanced;
1044
1045                 switch(c->state) {
1046                         case SYN_SENT:
1047                         case SYN_RECEIVED:
1048                                 data_acked--;
1049                                 break;
1050                         // TODO: handle FIN as well.
1051                         default:
1052                                 break;
1053                 }
1054
1055                 assert(data_acked >= 0);
1056
1057                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1058                 assert(data_acked <= bufused);
1059
1060                 if(data_acked)
1061                         buffer_get(&c->sndbuf, NULL, data_acked);
1062
1063                 // Also advance snd.nxt if possible
1064                 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1065                         c->snd.nxt = hdr.ack;
1066
1067                 c->snd.una = hdr.ack;
1068
1069                 c->dupack = 0;
1070                 c->snd.cwnd += utcp->mtu;
1071                 if(c->snd.cwnd > c->sndbuf.maxsize)
1072                         c->snd.cwnd = c->sndbuf.maxsize;
1073
1074                 // Check if we have sent a FIN that is now ACKed.
1075                 switch(c->state) {
1076                 case FIN_WAIT_1:
1077                         if(c->snd.una == c->snd.last)
1078                                 set_state(c, FIN_WAIT_2);
1079                         break;
1080                 case CLOSING:
1081                         if(c->snd.una == c->snd.last) {
1082                                 gettimeofday(&c->conn_timeout, NULL);
1083                                 c->conn_timeout.tv_sec += 60;
1084                                 set_state(c, TIME_WAIT);
1085                         }
1086                         break;
1087                 default:
1088                         break;
1089                 }
1090         } else {
1091                 if(!len && is_reliable(c)) {
1092                         c->dupack++;
1093                         if(c->dupack == 3) {
1094                                 debug("Triplicate ACK\n");
1095                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1096                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1097                                 //Reset the congestion window so we wait for ACKs.
1098                                 c->snd.nxt = c->snd.una;
1099                                 c->snd.cwnd = utcp->mtu;
1100                                 start_retransmit_timer(c);
1101                         }
1102                 }
1103         }
1104
1105         // 4. Update timers
1106
1107         if(advanced) {
1108                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1109                 if(c->snd.una == c->snd.last)
1110                         stop_retransmit_timer(c);
1111                 else if(is_reliable(c))
1112                         start_retransmit_timer(c);
1113         }
1114
1115         // 5. Process SYN stuff
1116
1117         if(hdr.ctl & SYN) {
1118                 switch(c->state) {
1119                 case SYN_SENT:
1120                         // This is a SYNACK. It should always have ACKed the SYN.
1121                         if(!advanced)
1122                                 goto reset;
1123                         c->rcv.irs = hdr.seq;
1124                         c->rcv.nxt = hdr.seq;
1125                         set_state(c, ESTABLISHED);
1126                         // TODO: notify application of this somehow.
1127                         break;
1128                 case SYN_RECEIVED:
1129                 case ESTABLISHED:
1130                 case FIN_WAIT_1:
1131                 case FIN_WAIT_2:
1132                 case CLOSE_WAIT:
1133                 case CLOSING:
1134                 case LAST_ACK:
1135                 case TIME_WAIT:
1136                         // Ehm, no. We should never receive a second SYN.
1137                         goto reset;
1138                 default:
1139 #ifdef UTCP_DEBUG
1140                         abort();
1141 #endif
1142                         return 0;
1143                 }
1144
1145                 // SYN counts as one sequence number
1146                 c->rcv.nxt++;
1147         }
1148
1149         // 6. Process new data
1150
1151         if(c->state == SYN_RECEIVED) {
1152                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1153                 if(!advanced)
1154                         goto reset;
1155
1156                 // Are we still LISTENing?
1157                 if(utcp->accept)
1158                         utcp->accept(c, c->src);
1159
1160                 if(c->state != ESTABLISHED) {
1161                         set_state(c, CLOSED);
1162                         c->reapable = true;
1163                         goto reset;
1164                 }
1165         }
1166
1167         if(len) {
1168                 switch(c->state) {
1169                 case SYN_SENT:
1170                 case SYN_RECEIVED:
1171                         // This should never happen.
1172 #ifdef UTCP_DEBUG
1173                         abort();
1174 #endif
1175                         return 0;
1176                 case ESTABLISHED:
1177                 case FIN_WAIT_1:
1178                 case FIN_WAIT_2:
1179                         break;
1180                 case CLOSE_WAIT:
1181                 case CLOSING:
1182                 case LAST_ACK:
1183                 case TIME_WAIT:
1184                         // Ehm no, We should never receive more data after a FIN.
1185                         goto reset;
1186                 default:
1187 #ifdef UTCP_DEBUG
1188                         abort();
1189 #endif
1190                         return 0;
1191                 }
1192
1193                 handle_incoming_data(c, hdr.seq, data, len);
1194         }
1195
1196         // 7. Process FIN stuff
1197
1198         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1199                 switch(c->state) {
1200                 case SYN_SENT:
1201                 case SYN_RECEIVED:
1202                         // This should never happen.
1203 #ifdef UTCP_DEBUG
1204                         abort();
1205 #endif
1206                         break;
1207                 case ESTABLISHED:
1208                         set_state(c, CLOSE_WAIT);
1209                         break;
1210                 case FIN_WAIT_1:
1211                         set_state(c, CLOSING);
1212                         break;
1213                 case FIN_WAIT_2:
1214                         gettimeofday(&c->conn_timeout, NULL);
1215                         c->conn_timeout.tv_sec += 60;
1216                         set_state(c, TIME_WAIT);
1217                         break;
1218                 case CLOSE_WAIT:
1219                 case CLOSING:
1220                 case LAST_ACK:
1221                 case TIME_WAIT:
1222                         // Ehm, no. We should never receive a second FIN.
1223                         goto reset;
1224                 default:
1225 #ifdef UTCP_DEBUG
1226                         abort();
1227 #endif
1228                         break;
1229                 }
1230
1231                 // FIN counts as one sequence number
1232                 c->rcv.nxt++;
1233                 len++;
1234
1235                 // Inform the application that the peer closed the connection.
1236                 if(c->recv) {
1237                         errno = 0;
1238                         c->recv(c, NULL, 0);
1239                 }
1240         }
1241
1242         // Now we send something back if:
1243         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1244         //   -> sendatleastone = true
1245         // - or we got an ack, so we should maybe send a bit more data
1246         //   -> sendatleastone = false
1247
1248         ack(c, len || prevrcvnxt != c->rcv.nxt);
1249         return 0;
1250
1251 reset:
1252         swap_ports(&hdr);
1253         hdr.wnd = 0;
1254         if(hdr.ctl & ACK) {
1255                 hdr.seq = hdr.ack;
1256                 hdr.ctl = RST;
1257         } else {
1258                 hdr.ack = hdr.seq + len;
1259                 hdr.seq = 0;
1260                 hdr.ctl = RST | ACK;
1261         }
1262         print_packet(utcp, "send", &hdr, sizeof hdr);
1263         utcp->send(utcp, &hdr, sizeof hdr);
1264         return 0;
1265
1266 }
1267
1268 int utcp_shutdown(struct utcp_connection *c, int dir) {
1269         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1270         if(!c) {
1271                 errno = EFAULT;
1272                 return -1;
1273         }
1274
1275         if(c->reapable) {
1276                 debug("Error: shutdown() called on closed connection %p\n", c);
1277                 errno = EBADF;
1278                 return -1;
1279         }
1280
1281         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1282                 errno = EINVAL;
1283                 return -1;
1284         }
1285
1286         // TCP does not have a provision for stopping incoming packets.
1287         // The best we can do is to just ignore them.
1288         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1289                 c->recv = NULL;
1290
1291         // The rest of the code deals with shutting down writes.
1292         if(dir == UTCP_SHUT_RD)
1293                 return 0;
1294
1295         switch(c->state) {
1296         case CLOSED:
1297         case LISTEN:
1298                 errno = ENOTCONN;
1299                 return -1;
1300
1301         case SYN_SENT:
1302                 set_state(c, CLOSED);
1303                 return 0;
1304
1305         case SYN_RECEIVED:
1306         case ESTABLISHED:
1307                 set_state(c, FIN_WAIT_1);
1308                 break;
1309         case FIN_WAIT_1:
1310         case FIN_WAIT_2:
1311                 return 0;
1312         case CLOSE_WAIT:
1313                 set_state(c, CLOSING);
1314                 break;
1315
1316         case CLOSING:
1317         case LAST_ACK:
1318         case TIME_WAIT:
1319                 return 0;
1320         }
1321
1322         c->snd.last++;
1323
1324         ack(c, false);
1325         if(!timerisset(&c->rtrx_timeout))
1326                 start_retransmit_timer(c);
1327         return 0;
1328 }
1329
1330 int utcp_close(struct utcp_connection *c) {
1331         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1332                 return -1;
1333         c->recv = NULL;
1334         c->poll = NULL;
1335         c->reapable = true;
1336         return 0;
1337 }
1338
1339 int utcp_abort(struct utcp_connection *c) {
1340         if(!c) {
1341                 errno = EFAULT;
1342                 return -1;
1343         }
1344
1345         if(c->reapable) {
1346                 debug("Error: abort() called on closed connection %p\n", c);
1347                 errno = EBADF;
1348                 return -1;
1349         }
1350
1351         c->recv = NULL;
1352         c->poll = NULL;
1353         c->reapable = true;
1354
1355         switch(c->state) {
1356         case CLOSED:
1357                 return 0;
1358         case LISTEN:
1359         case SYN_SENT:
1360         case CLOSING:
1361         case LAST_ACK:
1362         case TIME_WAIT:
1363                 set_state(c, CLOSED);
1364                 return 0;
1365
1366         case SYN_RECEIVED:
1367         case ESTABLISHED:
1368         case FIN_WAIT_1:
1369         case FIN_WAIT_2:
1370         case CLOSE_WAIT:
1371                 set_state(c, CLOSED);
1372                 break;
1373         }
1374
1375         // Send RST
1376
1377         struct hdr hdr;
1378
1379         hdr.src = c->src;
1380         hdr.dst = c->dst;
1381         hdr.seq = c->snd.nxt;
1382         hdr.ack = 0;
1383         hdr.wnd = 0;
1384         hdr.ctl = RST;
1385
1386         print_packet(c->utcp, "send", &hdr, sizeof hdr);
1387         c->utcp->send(c->utcp, &hdr, sizeof hdr);
1388         return 0;
1389 }
1390
1391 /* Handle timeouts.
1392  * One call to this function will loop through all connections,
1393  * checking if something needs to be resent or not.
1394  * The return value is the time to the next timeout in milliseconds,
1395  * or maybe a negative value if the timeout is infinite.
1396  */
1397 struct timeval utcp_timeout(struct utcp *utcp) {
1398         struct timeval now;
1399         gettimeofday(&now, NULL);
1400         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1401
1402         for(int i = 0; i < utcp->nconnections; i++) {
1403                 struct utcp_connection *c = utcp->connections[i];
1404                 if(!c)
1405                         continue;
1406
1407                 // delete connections that have been utcp_close()d.
1408                 if(c->state == CLOSED) {
1409                         if(c->reapable) {
1410                                 debug("Reaping %p\n", c);
1411                                 free_connection(c);
1412                                 i--;
1413                         }
1414                         continue;
1415                 }
1416
1417                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1418                         errno = ETIMEDOUT;
1419                         c->state = CLOSED;
1420                         if(c->recv)
1421                                 c->recv(c, NULL, 0);
1422                         continue;
1423                 }
1424
1425                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1426                         debug("retransmit()\n");
1427                         retransmit(c);
1428                 }
1429
1430                 if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1431                         c->poll(c, buffer_free(&c->sndbuf));
1432
1433                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1434                         next = c->conn_timeout;
1435
1436                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1437                         next = c->rtrx_timeout;
1438         }
1439
1440         struct timeval diff;
1441         timersub(&next, &now, &diff);
1442         return diff;
1443 }
1444
1445 bool utcp_is_active(struct utcp *utcp) {
1446         if(!utcp)
1447                 return false;
1448
1449         for(int i = 0; i < utcp->nconnections; i++)
1450                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1451                         return true;
1452
1453         return false;
1454 }
1455
1456 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1457         if(!send) {
1458                 errno = EFAULT;
1459                 return NULL;
1460         }
1461
1462         struct utcp *utcp = calloc(1, sizeof *utcp);
1463         if(!utcp)
1464                 return NULL;
1465
1466         utcp->accept = accept;
1467         utcp->pre_accept = pre_accept;
1468         utcp->send = send;
1469         utcp->priv = priv;
1470         utcp->mtu = DEFAULT_MTU;
1471         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1472         utcp->rto = START_RTO; // usec
1473
1474         return utcp;
1475 }
1476
1477 void utcp_exit(struct utcp *utcp) {
1478         if(!utcp)
1479                 return;
1480         for(int i = 0; i < utcp->nconnections; i++) {
1481                 if(!utcp->connections[i]->reapable)
1482                         debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1483                 buffer_exit(&utcp->connections[i]->rcvbuf);
1484                 buffer_exit(&utcp->connections[i]->sndbuf);
1485                 free(utcp->connections[i]);
1486         }
1487         free(utcp->connections);
1488         free(utcp);
1489 }
1490
1491 uint16_t utcp_get_mtu(struct utcp *utcp) {
1492         return utcp ? utcp->mtu : 0;
1493 }
1494
1495 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1496         // TODO: handle overhead of the header
1497         if(utcp)
1498                 utcp->mtu = mtu;
1499 }
1500
1501 int utcp_get_user_timeout(struct utcp *u) {
1502         return u ? u->timeout : 0;
1503 }
1504
1505 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1506         if(u)
1507                 u->timeout = timeout;
1508 }
1509
1510 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1511         return c ? c->sndbuf.maxsize : 0;
1512 }
1513
1514 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1515         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1516                 return buffer_free(&c->sndbuf);
1517         else
1518                 return 0;
1519 }
1520
1521 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1522         if(!c)
1523                 return;
1524         c->sndbuf.maxsize = size;
1525         if(c->sndbuf.maxsize != size)
1526                 c->sndbuf.maxsize = -1;
1527 }
1528
1529 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1530         return c ? c->rcvbuf.maxsize : 0;
1531 }
1532
1533 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1534         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1535                 return buffer_free(&c->rcvbuf);
1536         else
1537                 return 0;
1538 }
1539
1540 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1541         if(!c)
1542                 return;
1543         c->rcvbuf.maxsize = size;
1544         if(c->rcvbuf.maxsize != size)
1545                 c->rcvbuf.maxsize = -1;
1546 }
1547
1548 bool utcp_get_nodelay(struct utcp_connection *c) {
1549         return c ? c->nodelay : false;
1550 }
1551
1552 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1553         if(c)
1554                 c->nodelay = nodelay;
1555 }
1556
1557 bool utcp_get_keepalive(struct utcp_connection *c) {
1558         return c ? c->keepalive : false;
1559 }
1560
1561 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1562         if(c)
1563                 c->keepalive = keepalive;
1564 }
1565
1566 size_t utcp_get_outq(struct utcp_connection *c) {
1567         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1568 }
1569
1570 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1571         if(c)
1572                 c->recv = recv;
1573 }
1574
1575 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1576         if(c)
1577                 c->poll = poll;
1578 }
1579
1580 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1581         if(utcp) {
1582                 utcp->accept = accept;
1583                 utcp->pre_accept = pre_accept;
1584         }
1585 }