git.meshlink.io Git - utcp/blob - utcp.c

   1 /*
   2     utcp.c -- Userspace TCP
   3     Copyright (C) 2014-2017 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)\
  49         do {\
  50                 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
  51                 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
  52                 if((r)->tv_usec < 0)\
  53                         (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
  54         } while (0)
  55 #endif
  56
  57 static inline size_t max(size_t a, size_t b) {
  58         return a > b ? a : b;
  59 }
  60
  61 #ifdef UTCP_DEBUG
  62 #include <stdarg.h>
  63
  64 static void debug(const char *format, ...) {
  65         va_list ap;
  66         va_start(ap, format);
  67         vfprintf(stderr, format, ap);
  68         va_end(ap);
  69 }
  70
  71 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
  72         struct hdr hdr;
  73
  74         if(len < sizeof(hdr)) {
  75                 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
  76                 return;
  77         }
  78
  79         memcpy(&hdr, pkt, sizeof(hdr));
  80         debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux);
  81
  82         if(hdr.ctl & SYN) {
  83                 debug("SYN");
  84         }
  85
  86         if(hdr.ctl & RST) {
  87                 debug("RST");
  88         }
  89
  90         if(hdr.ctl & FIN) {
  91                 debug("FIN");
  92         }
  93
  94         if(hdr.ctl & ACK) {
  95                 debug("ACK");
  96         }
  97
  98         if(len > sizeof(hdr)) {
  99                 uint32_t datalen = len - sizeof(hdr);
 100                 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 101                 char str[datalen * 2 + 1];
 102                 char *p = str;
 103
 104                 for(uint32_t i = 0; i < datalen; i++) {
 105                         *p++ = "0123456789ABCDEF"[data[i] >> 4];
 106                         *p++ = "0123456789ABCDEF"[data[i] & 15];
 107                 }
 108
 109                 *p = 0;
 110
 111                 debug(" data=%s", str);
 112         }
 113
 114         debug("\n");
 115 }
 116 #else
 117 #define debug(...) do {} while(0)
 118 #define print_packet(...) do {} while(0)
 119 #endif
 120
 121 static void set_state(struct utcp_connection *c, enum state state) {
 122         c->state = state;
 123
 124         if(state == ESTABLISHED) {
 125                 timerclear(&c->conn_timeout);
 126         }
 127
 128         debug("%p new state: %s\n", c->utcp, strstate[state]);
 129 }
 130
 131 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 132         if(seq != c->snd.last) {
 133                 return false;
 134         }
 135
 136         switch(c->state) {
 137         case FIN_WAIT_1:
 138         case CLOSING:
 139         case LAST_ACK:
 140                 return true;
 141
 142         default:
 143                 return false;
 144         }
 145 }
 146
 147 static bool is_reliable(struct utcp_connection *c) {
 148         return c->flags & UTCP_RELIABLE;
 149 }
 150
 151 static int32_t seqdiff(uint32_t a, uint32_t b) {
 152         return a - b;
 153 }
 154
 155 // Buffer functions
 156 // TODO: convert to ringbuffers to avoid memmove() operations.
 157
 158 // Store data into the buffer
 159 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 160         debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 161
 162         size_t required = offset + len;
 163
 164         if(required > buf->maxsize) {
 165                 if(offset >= buf->maxsize) {
 166                         return 0;
 167                 }
 168
 169                 len = buf->maxsize - offset;
 170                 required = buf->maxsize;
 171         }
 172
 173         if(required > buf->size) {
 174                 size_t newsize = buf->size;
 175
 176                 if(!newsize) {
 177                         newsize = required;
 178                 } else {
 179                         do {
 180                                 newsize *= 2;
 181                         } while(newsize < required);
 182                 }
 183
 184                 if(newsize > buf->maxsize) {
 185                         newsize = buf->maxsize;
 186                 }
 187
 188                 char *newdata = realloc(buf->data, newsize);
 189
 190                 if(!newdata) {
 191                         return -1;
 192                 }
 193
 194                 buf->data = newdata;
 195                 buf->size = newsize;
 196         }
 197
 198         memcpy(buf->data + offset, data, len);
 199
 200         if(required > buf->used) {
 201                 buf->used = required;
 202         }
 203
 204         return len;
 205 }
 206
 207 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 208         return buffer_put_at(buf, buf->used, data, len);
 209 }
 210
 211 // Get data from the buffer. data can be NULL.
 212 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 213         if(len > buf->used) {
 214                 len = buf->used;
 215         }
 216
 217         if(data) {
 218                 memcpy(data, buf->data, len);
 219         }
 220
 221         if(len < buf->used) {
 222                 memmove(buf->data, buf->data + len, buf->used - len);
 223         }
 224
 225         buf->used -= len;
 226         return len;
 227 }
 228
 229 // Copy data from the buffer without removing it.
 230 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 231         if(offset >= buf->used) {
 232                 return 0;
 233         }
 234
 235         if(offset + len > buf->used) {
 236                 len = buf->used - offset;
 237         }
 238
 239         memcpy(data, buf->data + offset, len);
 240         return len;
 241 }
 242
 243 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 244         memset(buf, 0, sizeof(*buf));
 245
 246         if(len) {
 247                 buf->data = malloc(len);
 248
 249                 if(!buf->data) {
 250                         return false;
 251                 }
 252         }
 253
 254         buf->size = len;
 255         buf->maxsize = maxlen;
 256         return true;
 257 }
 258
 259 static void buffer_exit(struct buffer *buf) {
 260         free(buf->data);
 261         memset(buf, 0, sizeof(*buf));
 262 }
 263
 264 static uint32_t buffer_free(const struct buffer *buf) {
 265         return buf->maxsize - buf->used;
 266 }
 267
 268 // Connections are stored in a sorted list.
 269 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 270
 271 static int compare(const void *va, const void *vb) {
 272         assert(va && vb);
 273
 274         const struct utcp_connection *a = *(struct utcp_connection **)va;
 275         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 276
 277         assert(a && b);
 278         assert(a->src && b->src);
 279
 280         int c = (int)a->src - (int)b->src;
 281
 282         if(c) {
 283                 return c;
 284         }
 285
 286         c = (int)a->dst - (int)b->dst;
 287         return c;
 288 }
 289
 290 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 291         if(!utcp->nconnections) {
 292                 return NULL;
 293         }
 294
 295         struct utcp_connection key = {
 296                 .src = src,
 297                 .dst = dst,
 298         }, *keyp = &key;
 299         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 300         return match ? *match : NULL;
 301 }
 302
 303 static void free_connection(struct utcp_connection *c) {
 304         struct utcp *utcp = c->utcp;
 305         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 306
 307         assert(cp);
 308
 309         int i = cp - utcp->connections;
 310         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 311         utcp->nconnections--;
 312
 313         buffer_exit(&c->rcvbuf);
 314         buffer_exit(&c->sndbuf);
 315         free(c);
 316 }
 317
 318 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 319         // Check whether this combination of src and dst is free
 320
 321         if(src) {
 322                 if(find_connection(utcp, src, dst)) {
 323                         errno = EADDRINUSE;
 324                         return NULL;
 325                 }
 326         } else { // If src == 0, generate a random port number with the high bit set
 327                 if(utcp->nconnections >= 32767) {
 328                         errno = ENOMEM;
 329                         return NULL;
 330                 }
 331
 332                 src = rand() | 0x8000;
 333
 334                 while(find_connection(utcp, src, dst)) {
 335                         src++;
 336                 }
 337         }
 338
 339         // Allocate memory for the new connection
 340
 341         if(utcp->nconnections >= utcp->nallocated) {
 342                 if(!utcp->nallocated) {
 343                         utcp->nallocated = 4;
 344                 } else {
 345                         utcp->nallocated *= 2;
 346                 }
 347
 348                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 349
 350                 if(!new_array) {
 351                         return NULL;
 352                 }
 353
 354                 utcp->connections = new_array;
 355         }
 356
 357         struct utcp_connection *c = calloc(1, sizeof(*c));
 358
 359         if(!c) {
 360                 return NULL;
 361         }
 362
 363         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 364                 free(c);
 365                 return NULL;
 366         }
 367
 368         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 369                 buffer_exit(&c->sndbuf);
 370                 free(c);
 371                 return NULL;
 372         }
 373
 374         // Fill in the details
 375
 376         c->src = src;
 377         c->dst = dst;
 378 #ifdef UTCP_DEBUG
 379         c->snd.iss = 0;
 380 #else
 381         c->snd.iss = rand();
 382 #endif
 383         c->snd.una = c->snd.iss;
 384         c->snd.nxt = c->snd.iss + 1;
 385         c->snd.last = c->snd.nxt;
 386         c->snd.cwnd = utcp->mtu;
 387         c->utcp = utcp;
 388
 389         // Add it to the sorted list of connections
 390
 391         utcp->connections[utcp->nconnections++] = c;
 392         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 393
 394         return c;
 395 }
 396
 397 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 398         if(a > b) {
 399                 return a - b;
 400         } else {
 401                 return b - a;
 402         }
 403 }
 404
 405 // Update RTT variables. See RFC 6298.
 406 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 407         if(!rtt) {
 408                 debug("invalid rtt\n");
 409                 return;
 410         }
 411
 412         struct utcp *utcp = c->utcp;
 413
 414         if(!utcp->srtt) {
 415                 utcp->srtt = rtt;
 416                 utcp->rttvar = rtt / 2;
 417         } else {
 418                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 419                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 420         }
 421
 422         utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
 423
 424         if(utcp->rto > MAX_RTO) {
 425                 utcp->rto = MAX_RTO;
 426         }
 427
 428         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 429 }
 430
 431 static void start_retransmit_timer(struct utcp_connection *c) {
 432         gettimeofday(&c->rtrx_timeout, NULL);
 433         c->rtrx_timeout.tv_usec += c->utcp->rto;
 434
 435         while(c->rtrx_timeout.tv_usec >= 1000000) {
 436                 c->rtrx_timeout.tv_usec -= 1000000;
 437                 c->rtrx_timeout.tv_sec++;
 438         }
 439
 440         debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
 441 }
 442
 443 static void stop_retransmit_timer(struct utcp_connection *c) {
 444         timerclear(&c->rtrx_timeout);
 445         debug("timeout cleared\n");
 446 }
 447
 448 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 449         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 450
 451         if(!c) {
 452                 return NULL;
 453         }
 454
 455         assert((flags & ~0x1f) == 0);
 456
 457         c->flags = flags;
 458         c->recv = recv;
 459         c->priv = priv;
 460
 461         struct {
 462                 struct hdr hdr;
 463                 uint8_t init[4];
 464         } pkt;
 465
 466         pkt.hdr.src = c->src;
 467         pkt.hdr.dst = c->dst;
 468         pkt.hdr.seq = c->snd.iss;
 469         pkt.hdr.ack = 0;
 470         pkt.hdr.wnd = c->rcvbuf.maxsize;
 471         pkt.hdr.ctl = SYN;
 472         pkt.hdr.aux = 0x0101;
 473         pkt.init[0] = 1;
 474         pkt.init[1] = 0;
 475         pkt.init[2] = 0;
 476         pkt.init[3] = flags & 0x7;
 477
 478         set_state(c, SYN_SENT);
 479
 480         print_packet(utcp, "send", &pkt, sizeof(pkt));
 481         utcp->send(utcp, &pkt, sizeof(pkt));
 482
 483         gettimeofday(&c->conn_timeout, NULL);
 484         c->conn_timeout.tv_sec += utcp->timeout;
 485
 486         start_retransmit_timer(c);
 487
 488         return c;
 489 }
 490
 491 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 492         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 493 }
 494
 495 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 496         if(c->reapable || c->state != SYN_RECEIVED) {
 497                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 498                 return;
 499         }
 500
 501         debug("%p accepted, %p %p\n", c, recv, priv);
 502         c->recv = recv;
 503         c->priv = priv;
 504         set_state(c, ESTABLISHED);
 505 }
 506
 507 static void ack(struct utcp_connection *c, bool sendatleastone) {
 508         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 509         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 510         debug("cwndleft = %d\n", cwndleft);
 511
 512         assert(left >= 0);
 513
 514         if(cwndleft <= 0) {
 515                 cwndleft = 0;
 516         }
 517
 518         if(cwndleft < left) {
 519                 left = cwndleft;
 520         }
 521
 522         if(!left && !sendatleastone) {
 523                 return;
 524         }
 525
 526         struct {
 527                 struct hdr hdr;
 528                 uint8_t data[];
 529         } *pkt;
 530
 531         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 532
 533         if(!pkt) {
 534                 return;
 535         }
 536
 537         pkt->hdr.src = c->src;
 538         pkt->hdr.dst = c->dst;
 539         pkt->hdr.ack = c->rcv.nxt;
 540         pkt->hdr.wnd = c->rcvbuf.maxsize;
 541         pkt->hdr.ctl = ACK;
 542         pkt->hdr.aux = 0;
 543
 544         do {
 545                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 546                 pkt->hdr.seq = c->snd.nxt;
 547
 548                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 549
 550                 c->snd.nxt += seglen;
 551                 left -= seglen;
 552
 553                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 554                         seglen--;
 555                         pkt->hdr.ctl |= FIN;
 556                 }
 557
 558                 if(!c->rtt_start.tv_sec) {
 559                         // Start RTT measurement
 560                         gettimeofday(&c->rtt_start, NULL);
 561                         c->rtt_seq = pkt->hdr.seq + seglen;
 562                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 563                 }
 564
 565                 print_packet(c->utcp, "send", pkt, sizeof(pkt->hdr) + seglen);
 566                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 567         } while(left);
 568
 569         free(pkt);
 570 }
 571
 572 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 573         if(c->reapable) {
 574                 debug("Error: send() called on closed connection %p\n", c);
 575                 errno = EBADF;
 576                 return -1;
 577         }
 578
 579         switch(c->state) {
 580         case CLOSED:
 581         case LISTEN:
 582                 debug("Error: send() called on unconnected connection %p\n", c);
 583                 errno = ENOTCONN;
 584                 return -1;
 585
 586         case SYN_SENT:
 587         case SYN_RECEIVED:
 588         case ESTABLISHED:
 589         case CLOSE_WAIT:
 590                 break;
 591
 592         case FIN_WAIT_1:
 593         case FIN_WAIT_2:
 594         case CLOSING:
 595         case LAST_ACK:
 596         case TIME_WAIT:
 597                 debug("Error: send() called on closing connection %p\n", c);
 598                 errno = EPIPE;
 599                 return -1;
 600         }
 601
 602         // Exit early if we have nothing to send.
 603
 604         if(!len) {
 605                 return 0;
 606         }
 607
 608         if(!data) {
 609                 errno = EFAULT;
 610                 return -1;
 611         }
 612
 613         // Check if we need to be able to buffer all data
 614
 615         if(c->flags & UTCP_NO_PARTIAL) {
 616                 if(len > buffer_free(&c->sndbuf)) {
 617                         if(len > c->sndbuf.maxsize) {
 618                                 errno = EMSGSIZE;
 619                                 return -1;
 620                         } else {
 621                                 errno = EWOULDBLOCK;
 622                                 return 0;
 623                         }
 624                 }
 625         }
 626
 627         // Add data to send buffer.
 628
 629         if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 630                 len = buffer_put(&c->sndbuf, data, len);
 631         } else {
 632                 return 0;
 633         }
 634
 635         if(len <= 0) {
 636                 if(is_reliable(c)) {
 637                         errno = EWOULDBLOCK;
 638                         return 0;
 639                 } else {
 640                         return len;
 641                 }
 642         }
 643
 644         c->snd.last += len;
 645
 646         // Don't send anything yet if the connection has not fully established yet
 647
 648         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 649                 return len;
 650         }
 651
 652         ack(c, false);
 653
 654         if(!is_reliable(c)) {
 655                 c->snd.una = c->snd.nxt = c->snd.last;
 656                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 657         }
 658
 659         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 660                 start_retransmit_timer(c);
 661         }
 662
 663         if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
 664                 gettimeofday(&c->conn_timeout, NULL);
 665                 c->conn_timeout.tv_sec += c->utcp->timeout;
 666         }
 667
 668         return len;
 669 }
 670
 671 static void swap_ports(struct hdr *hdr) {
 672         uint16_t tmp = hdr->src;
 673         hdr->src = hdr->dst;
 674         hdr->dst = tmp;
 675 }
 676
 677 static void retransmit(struct utcp_connection *c) {
 678         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 679                 debug("Retransmit() called but nothing to retransmit!\n");
 680                 stop_retransmit_timer(c);
 681                 return;
 682         }
 683
 684         struct utcp *utcp = c->utcp;
 685
 686         struct {
 687                 struct hdr hdr;
 688                 uint8_t data[];
 689         } *pkt;
 690
 691         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 692
 693         if(!pkt) {
 694                 return;
 695         }
 696
 697         pkt->hdr.src = c->src;
 698         pkt->hdr.dst = c->dst;
 699         pkt->hdr.wnd = c->rcvbuf.maxsize;
 700         pkt->hdr.aux = 0;
 701
 702         switch(c->state) {
 703         case SYN_SENT:
 704                 // Send our SYN again
 705                 pkt->hdr.seq = c->snd.iss;
 706                 pkt->hdr.ack = 0;
 707                 pkt->hdr.ctl = SYN;
 708                 pkt->hdr.aux = 0x0101;
 709                 pkt->data[0] = 1;
 710                 pkt->data[1] = 0;
 711                 pkt->data[2] = 0;
 712                 pkt->data[3] = c->flags & 0x7;
 713                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 714                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 715                 break;
 716
 717         case SYN_RECEIVED:
 718                 // Send SYNACK again
 719                 pkt->hdr.seq = c->snd.nxt;
 720                 pkt->hdr.ack = c->rcv.nxt;
 721                 pkt->hdr.ctl = SYN | ACK;
 722                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
 723                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 724                 break;
 725
 726         case ESTABLISHED:
 727         case FIN_WAIT_1:
 728         case CLOSE_WAIT:
 729         case CLOSING:
 730         case LAST_ACK:
 731                 // Send unacked data again.
 732                 pkt->hdr.seq = c->snd.una;
 733                 pkt->hdr.ack = c->rcv.nxt;
 734                 pkt->hdr.ctl = ACK;
 735                 uint32_t len = seqdiff(c->snd.last, c->snd.una);
 736
 737                 if(len > utcp->mtu) {
 738                         len = utcp->mtu;
 739                 }
 740
 741                 if(fin_wanted(c, c->snd.una + len)) {
 742                         len--;
 743                         pkt->hdr.ctl |= FIN;
 744                 }
 745
 746                 c->snd.nxt = c->snd.una + len;
 747                 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 748                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 749                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 750                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 751                 break;
 752
 753         case CLOSED:
 754         case LISTEN:
 755         case TIME_WAIT:
 756         case FIN_WAIT_2:
 757                 // We shouldn't need to retransmit anything in this state.
 758 #ifdef UTCP_DEBUG
 759                 abort();
 760 #endif
 761                 stop_retransmit_timer(c);
 762                 goto cleanup;
 763         }
 764
 765         start_retransmit_timer(c);
 766         utcp->rto *= 2;
 767
 768         if(utcp->rto > MAX_RTO) {
 769                 utcp->rto = MAX_RTO;
 770         }
 771
 772         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 773
 774 cleanup:
 775         free(pkt);
 776 }
 777
 778 /* Update receive buffer and SACK entries after consuming data.
 779  *
 780  * Situation:
 781  *
 782  * |.....0000..1111111111.....22222......3333|
 783  * |---------------^
 784  *
 785  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 786  * to remove data from the receive buffer. The idea is to substract "len"
 787  * from the offset of all the SACK entries, and then remove/cut down entries
 788  * that are shifted to before the start of the receive buffer.
 789  *
 790  * There are three cases:
 791  * - the SACK entry is after ^, in that case just change the offset.
 792  * - the SACK entry starts before and ends after ^, so we have to
 793  *   change both its offset and size.
 794  * - the SACK entry is completely before ^, in that case delete it.
 795  */
 796 static void sack_consume(struct utcp_connection *c, size_t len) {
 797         debug("sack_consume %lu\n", (unsigned long)len);
 798
 799         if(len > c->rcvbuf.used) {
 800                 debug("All SACK entries consumed");
 801                 c->sacks[0].len = 0;
 802                 return;
 803         }
 804
 805         buffer_get(&c->rcvbuf, NULL, len);
 806
 807         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 808                 if(len < c->sacks[i].offset) {
 809                         c->sacks[i].offset -= len;
 810                         i++;
 811                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 812                         c->sacks[i].len -= len - c->sacks[i].offset;
 813                         c->sacks[i].offset = 0;
 814                         i++;
 815                 } else {
 816                         if(i < NSACKS - 1) {
 817                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 818                                 c->sacks[NSACKS - 1].len = 0;
 819                         } else {
 820                                 c->sacks[i].len = 0;
 821                                 break;
 822                         }
 823                 }
 824         }
 825
 826         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 827                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 828         }
 829 }
 830
 831 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 832         debug("out of order packet, offset %u\n", offset);
 833         // Packet loss or reordering occured. Store the data in the buffer.
 834         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 835
 836         if(rxd < 0 || (size_t)rxd < len) {
 837                 abort();
 838         }
 839
 840         // Make note of where we put it.
 841         for(int i = 0; i < NSACKS; i++) {
 842                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 843                         debug("New SACK entry %d\n", i);
 844                         c->sacks[i].offset = offset;
 845                         c->sacks[i].len = rxd;
 846                         break;
 847                 } else if(offset < c->sacks[i].offset) {
 848                         if(offset + rxd < c->sacks[i].offset) { // insert before
 849                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 850                                         debug("Insert SACK entry at %d\n", i);
 851                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 852                                         c->sacks[i].offset = offset;
 853                                         c->sacks[i].len = rxd;
 854                                 } else {
 855                                         debug("SACK entries full, dropping packet\n");
 856                                 }
 857
 858                                 break;
 859                         } else { // merge
 860                                 debug("Merge with start of SACK entry at %d\n", i);
 861                                 c->sacks[i].offset = offset;
 862                                 break;
 863                         }
 864                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 865                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 866                                 debug("Merge with end of SACK entry at %d\n", i);
 867                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 868                                 // TODO: handle potential merge with next entry
 869                         }
 870
 871                         break;
 872                 }
 873         }
 874
 875         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 876                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 877         }
 878 }
 879
 880 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 881         // Check if we can process out-of-order data now.
 882         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 883                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 884                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 885                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 886                 data = c->rcvbuf.data;
 887         }
 888
 889         if(c->recv) {
 890                 ssize_t rxd = c->recv(c, data, len);
 891
 892                 if(rxd < 0 || (size_t)rxd != len) {
 893                         // TODO: handle the application not accepting all data.
 894                         abort();
 895                 }
 896         }
 897
 898         if(c->rcvbuf.used) {
 899                 sack_consume(c, len);
 900         }
 901
 902         c->rcv.nxt += len;
 903 }
 904
 905
 906 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 907         if(!is_reliable(c)) {
 908                 c->recv(c, data, len);
 909                 c->rcv.nxt = seq + len;
 910                 return;
 911         }
 912
 913         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 914
 915         if(offset + len > c->rcvbuf.maxsize) {
 916                 abort();
 917         }
 918
 919         if(offset) {
 920                 handle_out_of_order(c, offset, data, len);
 921         } else {
 922                 handle_in_order(c, data, len);
 923         }
 924 }
 925
 926
 927 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 928         const uint8_t *ptr = data;
 929
 930         if(!utcp) {
 931                 errno = EFAULT;
 932                 return -1;
 933         }
 934
 935         if(!len) {
 936                 return 0;
 937         }
 938
 939         if(!data) {
 940                 errno = EFAULT;
 941                 return -1;
 942         }
 943
 944         print_packet(utcp, "recv", data, len);
 945
 946         // Drop packets smaller than the header
 947
 948         struct hdr hdr;
 949
 950         if(len < sizeof(hdr)) {
 951                 errno = EBADMSG;
 952                 return -1;
 953         }
 954
 955         // Make a copy from the potentially unaligned data to a struct hdr
 956
 957         memcpy(&hdr, ptr, sizeof(hdr));
 958         ptr += sizeof(hdr);
 959         len -= sizeof(hdr);
 960
 961         // Drop packets with an unknown CTL flag
 962
 963         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 964                 errno = EBADMSG;
 965                 return -1;
 966         }
 967
 968         // Check for auxiliary headers
 969
 970         const uint8_t *init = NULL;
 971
 972         uint16_t aux = hdr.aux;
 973
 974         while(aux) {
 975                 size_t auxlen = 4 * (aux >> 8) & 0xf;
 976                 uint8_t auxtype = aux & 0xff;
 977
 978                 if(len < auxlen) {
 979                         errno = EBADMSG;
 980                         return -1;
 981                 }
 982
 983                 switch(auxtype) {
 984                 case AUX_INIT:
 985                         if(!(hdr.ctl & SYN) || auxlen != 4) {
 986                                 errno = EBADMSG;
 987                                 return -1;
 988                         }
 989
 990                         init = ptr;
 991                         break;
 992
 993                 default:
 994                         errno = EBADMSG;
 995                         return -1;
 996                 }
 997
 998                 len -= auxlen;
 999                 ptr += auxlen;
1000
1001                 if(!(aux & 0x800)) {
1002                         break;
1003                 }
1004
1005                 if(len < 2) {
1006                         errno = EBADMSG;
1007                         return -1;
1008                 }
1009
1010                 memcpy(&aux, ptr, 2);
1011                 len -= 2;
1012                 ptr += 2;
1013         }
1014
1015         bool has_data = len || (hdr.ctl & (SYN | FIN));
1016
1017         // Try to match the packet to an existing connection
1018
1019         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1020
1021         // Is it for a new connection?
1022
1023         if(!c) {
1024                 // Ignore RST packets
1025
1026                 if(hdr.ctl & RST) {
1027                         return 0;
1028                 }
1029
1030                 // Is it a SYN packet and are we LISTENing?
1031
1032                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1033                         // If we don't want to accept it, send a RST back
1034                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1035                                 len = 1;
1036                                 goto reset;
1037                         }
1038
1039                         // Try to allocate memory, otherwise send a RST back
1040                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1041
1042                         if(!c) {
1043                                 len = 1;
1044                                 goto reset;
1045                         }
1046
1047                         // Parse auxilliary information
1048                         if(init) {
1049                                 if(init[0] < 1) {
1050                                         len = 1;
1051                                         goto reset;
1052                                 }
1053
1054                                 c->flags = init[3] & 0x7;
1055                         } else {
1056                                 c->flags = UTCP_TCP;
1057                         }
1058
1059 synack:
1060                         // Return SYN+ACK, go to SYN_RECEIVED state
1061                         c->snd.wnd = hdr.wnd;
1062                         c->rcv.irs = hdr.seq;
1063                         c->rcv.nxt = c->rcv.irs + 1;
1064                         set_state(c, SYN_RECEIVED);
1065
1066                         struct {
1067                                 struct hdr hdr;
1068                                 uint8_t data[4];
1069                         } pkt;
1070
1071                         pkt.hdr.src = c->src;
1072                         pkt.hdr.dst = c->dst;
1073                         pkt.hdr.ack = c->rcv.irs + 1;
1074                         pkt.hdr.seq = c->snd.iss;
1075                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1076                         pkt.hdr.ctl = SYN | ACK;
1077
1078                         if(init) {
1079                                 pkt.hdr.aux = 0x0101;
1080                                 pkt.data[0] = 1;
1081                                 pkt.data[1] = 0;
1082                                 pkt.data[2] = 0;
1083                                 pkt.data[3] = c->flags & 0x7;
1084                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1085                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1086                         } else {
1087                                 pkt.hdr.aux = 0;
1088                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1089                                 utcp->send(utcp, &pkt, sizeof(hdr));
1090                         }
1091                 } else {
1092                         // No, we don't want your packets, send a RST back
1093                         len = 1;
1094                         goto reset;
1095                 }
1096
1097                 return 0;
1098         }
1099
1100         debug("%p state %s\n", c->utcp, strstate[c->state]);
1101
1102         // In case this is for a CLOSED connection, ignore the packet.
1103         // TODO: make it so incoming packets can never match a CLOSED connection.
1104
1105         if(c->state == CLOSED) {
1106                 debug("Got packet for closed connection\n");
1107                 return 0;
1108         }
1109
1110         // It is for an existing connection.
1111
1112         // 1. Drop invalid packets.
1113
1114         // 1a. Drop packets that should not happen in our current state.
1115
1116         switch(c->state) {
1117         case SYN_SENT:
1118         case SYN_RECEIVED:
1119         case ESTABLISHED:
1120         case FIN_WAIT_1:
1121         case FIN_WAIT_2:
1122         case CLOSE_WAIT:
1123         case CLOSING:
1124         case LAST_ACK:
1125         case TIME_WAIT:
1126                 break;
1127
1128         default:
1129 #ifdef UTCP_DEBUG
1130                 abort();
1131 #endif
1132                 break;
1133         }
1134
1135         // 1b. Discard data that is not in our receive window.
1136
1137         if(is_reliable(c)) {
1138                 bool acceptable;
1139
1140                 if(c->state == SYN_SENT) {
1141                         acceptable = true;
1142                 } else if(len == 0) {
1143                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1144                 } else {
1145                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1146
1147                         // cut already accepted front overlapping
1148                         if(rcv_offset < 0) {
1149                                 acceptable = len > (size_t) - rcv_offset;
1150
1151                                 if(acceptable) {
1152                                         ptr -= rcv_offset;
1153                                         len += rcv_offset;
1154                                         hdr.seq -= rcv_offset;
1155                                 }
1156                         } else {
1157                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1158                         }
1159                 }
1160
1161                 if(!acceptable) {
1162                         debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1163
1164                         // Ignore unacceptable RST packets.
1165                         if(hdr.ctl & RST) {
1166                                 return 0;
1167                         }
1168
1169                         // Otherwise, continue processing.
1170                         len = 0;
1171                 }
1172         }
1173
1174         c->snd.wnd = hdr.wnd; // TODO: move below
1175
1176         // 1c. Drop packets with an invalid ACK.
1177         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1178         // (= snd.una + c->sndbuf.used).
1179
1180         if(!is_reliable(c)) {
1181                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1182                         hdr.ack = c->snd.una;
1183                 }
1184         }
1185
1186         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1187                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1188
1189                 // Ignore unacceptable RST packets.
1190                 if(hdr.ctl & RST) {
1191                         return 0;
1192                 }
1193
1194                 goto reset;
1195         }
1196
1197         // 2. Handle RST packets
1198
1199         if(hdr.ctl & RST) {
1200                 switch(c->state) {
1201                 case SYN_SENT:
1202                         if(!(hdr.ctl & ACK)) {
1203                                 return 0;
1204                         }
1205
1206                         // The peer has refused our connection.
1207                         set_state(c, CLOSED);
1208                         errno = ECONNREFUSED;
1209
1210                         if(c->recv) {
1211                                 c->recv(c, NULL, 0);
1212                         }
1213
1214                         if(c->poll && !c->reapable) {
1215                                 c->poll(c, 0);
1216                         }
1217
1218                         return 0;
1219
1220                 case SYN_RECEIVED:
1221                         if(hdr.ctl & ACK) {
1222                                 return 0;
1223                         }
1224
1225                         // We haven't told the application about this connection yet. Silently delete.
1226                         free_connection(c);
1227                         return 0;
1228
1229                 case ESTABLISHED:
1230                 case FIN_WAIT_1:
1231                 case FIN_WAIT_2:
1232                 case CLOSE_WAIT:
1233                         if(hdr.ctl & ACK) {
1234                                 return 0;
1235                         }
1236
1237                         // The peer has aborted our connection.
1238                         set_state(c, CLOSED);
1239                         errno = ECONNRESET;
1240
1241                         if(c->recv) {
1242                                 c->recv(c, NULL, 0);
1243                         }
1244
1245                         if(c->poll && !c->reapable) {
1246                                 c->poll(c, 0);
1247                         }
1248
1249                         return 0;
1250
1251                 case CLOSING:
1252                 case LAST_ACK:
1253                 case TIME_WAIT:
1254                         if(hdr.ctl & ACK) {
1255                                 return 0;
1256                         }
1257
1258                         // As far as the application is concerned, the connection has already been closed.
1259                         // If it has called utcp_close() already, we can immediately free this connection.
1260                         if(c->reapable) {
1261                                 free_connection(c);
1262                                 return 0;
1263                         }
1264
1265                         // Otherwise, immediately move to the CLOSED state.
1266                         set_state(c, CLOSED);
1267                         return 0;
1268
1269                 default:
1270 #ifdef UTCP_DEBUG
1271                         abort();
1272 #endif
1273                         break;
1274                 }
1275         }
1276
1277         uint32_t advanced;
1278
1279         if(!(hdr.ctl & ACK)) {
1280                 advanced = 0;
1281                 goto skip_ack;
1282         }
1283
1284         // 3. Advance snd.una
1285
1286         advanced = seqdiff(hdr.ack, c->snd.una);
1287
1288         if(advanced) {
1289                 // RTT measurement
1290                 if(c->rtt_start.tv_sec) {
1291                         if(c->rtt_seq == hdr.ack) {
1292                                 struct timeval now, diff;
1293                                 gettimeofday(&now, NULL);
1294                                 timersub(&now, &c->rtt_start, &diff);
1295                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1296                                 c->rtt_start.tv_sec = 0;
1297                         } else if(c->rtt_seq < hdr.ack) {
1298                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1299                                 c->rtt_start.tv_sec = 0;
1300                         }
1301                 }
1302
1303                 int32_t data_acked = advanced;
1304
1305                 switch(c->state) {
1306                 case SYN_SENT:
1307                 case SYN_RECEIVED:
1308                         data_acked--;
1309                         break;
1310
1311                 // TODO: handle FIN as well.
1312                 default:
1313                         break;
1314                 }
1315
1316                 assert(data_acked >= 0);
1317
1318 #ifndef NDEBUG
1319                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1320                 assert(data_acked <= bufused);
1321 #endif
1322
1323                 if(data_acked) {
1324                         buffer_get(&c->sndbuf, NULL, data_acked);
1325                 }
1326
1327                 // Also advance snd.nxt if possible
1328                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1329                         c->snd.nxt = hdr.ack;
1330                 }
1331
1332                 c->snd.una = hdr.ack;
1333
1334                 c->dupack = 0;
1335                 c->snd.cwnd += utcp->mtu;
1336
1337                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1338                         c->snd.cwnd = c->sndbuf.maxsize;
1339                 }
1340
1341                 // Check if we have sent a FIN that is now ACKed.
1342                 switch(c->state) {
1343                 case FIN_WAIT_1:
1344                         if(c->snd.una == c->snd.last) {
1345                                 set_state(c, FIN_WAIT_2);
1346                         }
1347
1348                         break;
1349
1350                 case CLOSING:
1351                         if(c->snd.una == c->snd.last) {
1352                                 gettimeofday(&c->conn_timeout, NULL);
1353                                 c->conn_timeout.tv_sec += utcp->timeout;
1354                                 set_state(c, TIME_WAIT);
1355                         }
1356
1357                         break;
1358
1359                 default:
1360                         break;
1361                 }
1362         } else {
1363                 if(!len && is_reliable(c)) {
1364                         c->dupack++;
1365
1366                         if(c->dupack == 3) {
1367                                 debug("Triplicate ACK\n");
1368                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1369                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1370                                 //Reset the congestion window so we wait for ACKs.
1371                                 c->snd.nxt = c->snd.una;
1372                                 c->snd.cwnd = utcp->mtu;
1373                                 start_retransmit_timer(c);
1374                         }
1375                 }
1376         }
1377
1378         // 4. Update timers
1379
1380         if(advanced) {
1381                 if(c->snd.una == c->snd.last) {
1382                         stop_retransmit_timer(c);
1383                         timerclear(&c->conn_timeout);
1384                 } else if(is_reliable(c)) {
1385                         start_retransmit_timer(c);
1386                         gettimeofday(&c->conn_timeout, NULL);
1387                         c->conn_timeout.tv_sec += utcp->timeout;
1388                 }
1389         }
1390
1391 skip_ack:
1392         // 5. Process SYN stuff
1393
1394         if(hdr.ctl & SYN) {
1395                 switch(c->state) {
1396                 case SYN_SENT:
1397
1398                         // This is a SYNACK. It should always have ACKed the SYN.
1399                         if(!advanced) {
1400                                 goto reset;
1401                         }
1402
1403                         c->rcv.irs = hdr.seq;
1404                         c->rcv.nxt = hdr.seq;
1405
1406                         if(c->shut_wr) {
1407                                 c->snd.last++;
1408                                 set_state(c, FIN_WAIT_1);
1409                         } else {
1410                                 set_state(c, ESTABLISHED);
1411                         }
1412
1413                         // TODO: notify application of this somehow.
1414                         break;
1415
1416                 case SYN_RECEIVED:
1417                         // This is a retransmit of a SYN, send back the SYNACK.
1418                         goto synack;
1419
1420                 case ESTABLISHED:
1421                 case FIN_WAIT_1:
1422                 case FIN_WAIT_2:
1423                 case CLOSE_WAIT:
1424                 case CLOSING:
1425                 case LAST_ACK:
1426                 case TIME_WAIT:
1427                         // Ehm, no. We should never receive a second SYN.
1428                         return 0;
1429
1430                 default:
1431 #ifdef UTCP_DEBUG
1432                         abort();
1433 #endif
1434                         return 0;
1435                 }
1436
1437                 // SYN counts as one sequence number
1438                 c->rcv.nxt++;
1439         }
1440
1441         // 6. Process new data
1442
1443         if(c->state == SYN_RECEIVED) {
1444                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1445                 if(!advanced) {
1446                         goto reset;
1447                 }
1448
1449                 // Are we still LISTENing?
1450                 if(utcp->accept) {
1451                         utcp->accept(c, c->src);
1452                 }
1453
1454                 if(c->state != ESTABLISHED) {
1455                         set_state(c, CLOSED);
1456                         c->reapable = true;
1457                         goto reset;
1458                 }
1459         }
1460
1461         if(len) {
1462                 switch(c->state) {
1463                 case SYN_SENT:
1464                 case SYN_RECEIVED:
1465                         // This should never happen.
1466 #ifdef UTCP_DEBUG
1467                         abort();
1468 #endif
1469                         return 0;
1470
1471                 case ESTABLISHED:
1472                 case FIN_WAIT_1:
1473                 case FIN_WAIT_2:
1474                         break;
1475
1476                 case CLOSE_WAIT:
1477                 case CLOSING:
1478                 case LAST_ACK:
1479                 case TIME_WAIT:
1480                         // Ehm no, We should never receive more data after a FIN.
1481                         goto reset;
1482
1483                 default:
1484 #ifdef UTCP_DEBUG
1485                         abort();
1486 #endif
1487                         return 0;
1488                 }
1489
1490                 handle_incoming_data(c, hdr.seq, ptr, len);
1491         }
1492
1493         // 7. Process FIN stuff
1494
1495         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1496                 switch(c->state) {
1497                 case SYN_SENT:
1498                 case SYN_RECEIVED:
1499                         // This should never happen.
1500 #ifdef UTCP_DEBUG
1501                         abort();
1502 #endif
1503                         break;
1504
1505                 case ESTABLISHED:
1506                         set_state(c, CLOSE_WAIT);
1507                         break;
1508
1509                 case FIN_WAIT_1:
1510                         set_state(c, CLOSING);
1511                         break;
1512
1513                 case FIN_WAIT_2:
1514                         gettimeofday(&c->conn_timeout, NULL);
1515                         c->conn_timeout.tv_sec += utcp->timeout;
1516                         set_state(c, TIME_WAIT);
1517                         break;
1518
1519                 case CLOSE_WAIT:
1520                 case CLOSING:
1521                 case LAST_ACK:
1522                 case TIME_WAIT:
1523                         // Ehm, no. We should never receive a second FIN.
1524                         goto reset;
1525
1526                 default:
1527 #ifdef UTCP_DEBUG
1528                         abort();
1529 #endif
1530                         break;
1531                 }
1532
1533                 // FIN counts as one sequence number
1534                 c->rcv.nxt++;
1535                 len++;
1536
1537                 // Inform the application that the peer closed its end of the connection.
1538                 if(c->recv) {
1539                         errno = 0;
1540                         c->recv(c, NULL, 0);
1541                 }
1542         }
1543
1544         // Now we send something back if:
1545         // - we received data, so we have to send back an ACK
1546         //   -> sendatleastone = true
1547         // - or we got an ack, so we should maybe send a bit more data
1548         //   -> sendatleastone = false
1549
1550         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1551                 ack(c, has_data);
1552         }
1553
1554         return 0;
1555
1556 reset:
1557         swap_ports(&hdr);
1558         hdr.wnd = 0;
1559         hdr.aux = 0;
1560
1561         if(hdr.ctl & ACK) {
1562                 hdr.seq = hdr.ack;
1563                 hdr.ctl = RST;
1564         } else {
1565                 hdr.ack = hdr.seq + len;
1566                 hdr.seq = 0;
1567                 hdr.ctl = RST | ACK;
1568         }
1569
1570         print_packet(utcp, "send", &hdr, sizeof(hdr));
1571         utcp->send(utcp, &hdr, sizeof(hdr));
1572         return 0;
1573
1574 }
1575
1576 int utcp_shutdown(struct utcp_connection *c, int dir) {
1577         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1578
1579         if(!c) {
1580                 errno = EFAULT;
1581                 return -1;
1582         }
1583
1584         if(c->reapable) {
1585                 debug("Error: shutdown() called on closed connection %p\n", c);
1586                 errno = EBADF;
1587                 return -1;
1588         }
1589
1590         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1591                 errno = EINVAL;
1592                 return -1;
1593         }
1594
1595         // TCP does not have a provision for stopping incoming packets.
1596         // The best we can do is to just ignore them.
1597         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1598                 c->recv = NULL;
1599         }
1600
1601         // The rest of the code deals with shutting down writes.
1602         if(dir == UTCP_SHUT_RD) {
1603                 return 0;
1604         }
1605
1606         // Only process shutting down writes once.
1607         if(c->shut_wr) {
1608                 return 0;
1609         }
1610
1611         c->shut_wr = true;
1612
1613         switch(c->state) {
1614         case CLOSED:
1615         case LISTEN:
1616                 errno = ENOTCONN;
1617                 return -1;
1618
1619         case SYN_SENT:
1620                 return 0;
1621
1622         case SYN_RECEIVED:
1623         case ESTABLISHED:
1624                 set_state(c, FIN_WAIT_1);
1625                 break;
1626
1627         case FIN_WAIT_1:
1628         case FIN_WAIT_2:
1629                 return 0;
1630
1631         case CLOSE_WAIT:
1632                 set_state(c, CLOSING);
1633                 break;
1634
1635         case CLOSING:
1636         case LAST_ACK:
1637         case TIME_WAIT:
1638                 return 0;
1639         }
1640
1641         c->snd.last++;
1642
1643         ack(c, false);
1644
1645         if(!timerisset(&c->rtrx_timeout)) {
1646                 start_retransmit_timer(c);
1647         }
1648
1649         return 0;
1650 }
1651
1652 static bool reset_connection(struct utcp_connection *c) {
1653         if(!c) {
1654                 errno = EFAULT;
1655                 return false;
1656         }
1657
1658         if(c->reapable) {
1659                 debug("Error: abort() called on closed connection %p\n", c);
1660                 errno = EBADF;
1661                 return false;
1662         }
1663
1664         c->recv = NULL;
1665         c->poll = NULL;
1666
1667         switch(c->state) {
1668         case CLOSED:
1669                 return true;
1670
1671         case LISTEN:
1672         case SYN_SENT:
1673         case CLOSING:
1674         case LAST_ACK:
1675         case TIME_WAIT:
1676                 set_state(c, CLOSED);
1677                 return true;
1678
1679         case SYN_RECEIVED:
1680         case ESTABLISHED:
1681         case FIN_WAIT_1:
1682         case FIN_WAIT_2:
1683         case CLOSE_WAIT:
1684                 set_state(c, CLOSED);
1685                 break;
1686         }
1687
1688         // Send RST
1689
1690         struct hdr hdr;
1691
1692         hdr.src = c->src;
1693         hdr.dst = c->dst;
1694         hdr.seq = c->snd.nxt;
1695         hdr.ack = 0;
1696         hdr.wnd = 0;
1697         hdr.ctl = RST;
1698
1699         print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1700         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1701         return true;
1702 }
1703
1704 // Closes all the opened connections
1705 void utcp_abort_all_connections(struct utcp *utcp) {
1706         if(!utcp) {
1707                 errno = EINVAL;
1708                 return;
1709         }
1710
1711         for(int i = 0; i < utcp->nconnections; i++) {
1712                 struct utcp_connection *c = utcp->connections[i];
1713
1714                 if(c->reapable || c->state == CLOSED) {
1715                         continue;
1716                 }
1717
1718                 utcp_recv_t old_recv = c->recv;
1719                 utcp_poll_t old_poll = c->poll;
1720
1721                 reset_connection(c);
1722
1723                 if(old_recv) {
1724                         errno = 0;
1725                         old_recv(c, NULL, 0);
1726                 }
1727
1728                 if(old_poll && !c->reapable) {
1729                         errno = 0;
1730                         old_poll(c, 0);
1731                 }
1732         }
1733
1734         return;
1735 }
1736
1737 int utcp_close(struct utcp_connection *c) {
1738         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1739                 return -1;
1740         }
1741
1742         c->recv = NULL;
1743         c->poll = NULL;
1744         c->reapable = true;
1745         return 0;
1746 }
1747
1748 int utcp_abort(struct utcp_connection *c) {
1749         if(!reset_connection(c)) {
1750                 return -1;
1751         }
1752
1753         c->reapable = true;
1754         return 0;
1755 }
1756
1757 /* Handle timeouts.
1758  * One call to this function will loop through all connections,
1759  * checking if something needs to be resent or not.
1760  * The return value is the time to the next timeout in milliseconds,
1761  * or maybe a negative value if the timeout is infinite.
1762  */
1763 struct timeval utcp_timeout(struct utcp *utcp) {
1764         struct timeval now;
1765         gettimeofday(&now, NULL);
1766         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1767
1768         for(int i = 0; i < utcp->nconnections; i++) {
1769                 struct utcp_connection *c = utcp->connections[i];
1770
1771                 if(!c) {
1772                         continue;
1773                 }
1774
1775                 // delete connections that have been utcp_close()d.
1776                 if(c->state == CLOSED) {
1777                         if(c->reapable) {
1778                                 debug("Reaping %p\n", c);
1779                                 free_connection(c);
1780                                 i--;
1781                         }
1782
1783                         continue;
1784                 }
1785
1786                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1787                         errno = ETIMEDOUT;
1788                         c->state = CLOSED;
1789
1790                         if(c->recv) {
1791                                 c->recv(c, NULL, 0);
1792                         }
1793
1794                         if(c->poll && !c->reapable) {
1795                                 c->poll(c, 0);
1796                         }
1797
1798                         continue;
1799                 }
1800
1801                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1802                         debug("retransmit()\n");
1803                         retransmit(c);
1804                 }
1805
1806                 if(c->poll) {
1807                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1808                                 uint32_t len =  buffer_free(&c->sndbuf);
1809
1810                                 if(len) {
1811                                         c->poll(c, len);
1812                                 }
1813                         } else if(c->state == CLOSED) {
1814                                 c->poll(c, 0);
1815                         }
1816                 }
1817
1818                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1819                         next = c->conn_timeout;
1820                 }
1821
1822                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1823                         next = c->rtrx_timeout;
1824                 }
1825         }
1826
1827         struct timeval diff;
1828
1829         timersub(&next, &now, &diff);
1830
1831         return diff;
1832 }
1833
1834 bool utcp_is_active(struct utcp *utcp) {
1835         if(!utcp) {
1836                 return false;
1837         }
1838
1839         for(int i = 0; i < utcp->nconnections; i++)
1840                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1841                         return true;
1842                 }
1843
1844         return false;
1845 }
1846
1847 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1848         if(!send) {
1849                 errno = EFAULT;
1850                 return NULL;
1851         }
1852
1853         struct utcp *utcp = calloc(1, sizeof(*utcp));
1854
1855         if(!utcp) {
1856                 return NULL;
1857         }
1858
1859         utcp->accept = accept;
1860         utcp->pre_accept = pre_accept;
1861         utcp->send = send;
1862         utcp->priv = priv;
1863         utcp->mtu = DEFAULT_MTU;
1864         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1865         utcp->rto = START_RTO; // usec
1866
1867         return utcp;
1868 }
1869
1870 void utcp_exit(struct utcp *utcp) {
1871         if(!utcp) {
1872                 return;
1873         }
1874
1875         for(int i = 0; i < utcp->nconnections; i++) {
1876                 struct utcp_connection *c = utcp->connections[i];
1877
1878                 if(!c->reapable) {
1879                         if(c->recv) {
1880                                 c->recv(c, NULL, 0);
1881                         }
1882
1883                         if(c->poll && !c->reapable) {
1884                                 c->poll(c, 0);
1885                         }
1886                 }
1887
1888                 buffer_exit(&c->rcvbuf);
1889                 buffer_exit(&c->sndbuf);
1890                 free(c);
1891         }
1892
1893         free(utcp->connections);
1894         free(utcp);
1895 }
1896
1897 uint16_t utcp_get_mtu(struct utcp *utcp) {
1898         return utcp ? utcp->mtu : 0;
1899 }
1900
1901 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1902         // TODO: handle overhead of the header
1903         if(utcp) {
1904                 utcp->mtu = mtu;
1905         }
1906 }
1907
1908 void utcp_reset_timers(struct utcp *utcp) {
1909         if(!utcp) {
1910                 return;
1911         }
1912
1913         struct timeval now, then;
1914
1915         gettimeofday(&now, NULL);
1916
1917         then = now;
1918
1919         then.tv_sec += utcp->timeout;
1920
1921         for(int i = 0; i < utcp->nconnections; i++) {
1922                 struct utcp_connection *c = utcp->connections[i];
1923
1924                 if(c->reapable) {
1925                         continue;
1926                 }
1927
1928                 if(timerisset(&c->rtrx_timeout)) {
1929                         c->rtrx_timeout = now;
1930                 }
1931
1932                 if(timerisset(&c->conn_timeout)) {
1933                         c->conn_timeout = then;
1934                 }
1935
1936                 c->rtt_start.tv_sec = 0;
1937         }
1938
1939         if(utcp->rto > START_RTO) {
1940                 utcp->rto = START_RTO;
1941         }
1942 }
1943
1944 int utcp_get_user_timeout(struct utcp *u) {
1945         return u ? u->timeout : 0;
1946 }
1947
1948 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1949         if(u) {
1950                 u->timeout = timeout;
1951         }
1952 }
1953
1954 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1955         return c ? c->sndbuf.maxsize : 0;
1956 }
1957
1958 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1959         if(!c) {
1960                 return 0;
1961         }
1962
1963         switch(c->state) {
1964         case SYN_SENT:
1965         case SYN_RECEIVED:
1966         case ESTABLISHED:
1967         case CLOSE_WAIT:
1968                 return buffer_free(&c->sndbuf);
1969
1970         default:
1971                 return 0;
1972         }
1973 }
1974
1975 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1976         if(!c) {
1977                 return;
1978         }
1979
1980         c->sndbuf.maxsize = size;
1981
1982         if(c->sndbuf.maxsize != size) {
1983                 c->sndbuf.maxsize = -1;
1984         }
1985 }
1986
1987 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1988         return c ? c->rcvbuf.maxsize : 0;
1989 }
1990
1991 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1992         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1993                 return buffer_free(&c->rcvbuf);
1994         } else {
1995                 return 0;
1996         }
1997 }
1998
1999 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2000         if(!c) {
2001                 return;
2002         }
2003
2004         c->rcvbuf.maxsize = size;
2005
2006         if(c->rcvbuf.maxsize != size) {
2007                 c->rcvbuf.maxsize = -1;
2008         }
2009 }
2010
2011 size_t utcp_get_sendq(struct utcp_connection *c) {
2012         return c->sndbuf.used;
2013 }
2014
2015 size_t utcp_get_recvq(struct utcp_connection *c) {
2016         return c->rcvbuf.used;
2017 }
2018
2019 bool utcp_get_nodelay(struct utcp_connection *c) {
2020         return c ? c->nodelay : false;
2021 }
2022
2023 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2024         if(c) {
2025                 c->nodelay = nodelay;
2026         }
2027 }
2028
2029 bool utcp_get_keepalive(struct utcp_connection *c) {
2030         return c ? c->keepalive : false;
2031 }
2032
2033 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2034         if(c) {
2035                 c->keepalive = keepalive;
2036         }
2037 }
2038
2039 size_t utcp_get_outq(struct utcp_connection *c) {
2040         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2041 }
2042
2043 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2044         if(c) {
2045                 c->recv = recv;
2046         }
2047 }
2048
2049 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2050         if(c) {
2051                 c->poll = poll;
2052         }
2053 }
2054
2055 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2056         if(utcp) {
2057                 utcp->accept = accept;
2058                 utcp->pre_accept = pre_accept;
2059         }
2060 }
2061
2062 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2063         if(!c || c->reapable) {
2064                 return;
2065         }
2066
2067         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2068                 return;
2069         }
2070
2071         if(expect) {
2072                 // If we expect data, start the connection timer.
2073                 if(!timerisset(&c->conn_timeout)) {
2074                         gettimeofday(&c->conn_timeout, NULL);
2075                         c->conn_timeout.tv_sec += c->utcp->timeout;
2076                 }
2077         } else {
2078                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2079                 if(c->snd.una == c->snd.last) {
2080                         timerclear(&c->conn_timeout);
2081                 }
2082         }
2083 }
2084
2085 void utcp_offline(struct utcp *utcp, bool offline) {
2086         struct timeval now;
2087         gettimeofday(&now, NULL);
2088
2089         for(int i = 0; i < utcp->nconnections; i++) {
2090                 struct utcp_connection *c = utcp->connections[i];
2091
2092                 if(c->reapable) {
2093                         continue;
2094                 }
2095
2096                 utcp_expect_data(c, offline);
2097
2098                 if(!offline) {
2099                         if(timerisset(&c->rtrx_timeout)) {
2100                                 c->rtrx_timeout = now;
2101                         }
2102
2103                         utcp->connections[i]->rtt_start.tv_sec = 0;
2104                 }
2105         }
2106
2107         if(!offline && utcp->rto > START_RTO) {
2108                 utcp->rto = START_RTO;
2109         }
2110 }