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