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 & ~0xf) == 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         case SYN_SENT:
 584         case SYN_RECEIVED:
 585                 debug("Error: send() called on unconnected connection %p\n", c);
 586                 errno = ENOTCONN;
 587                 return -1;
 588
 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         // Add data to send buffer.
 615
 616         len = buffer_put(&c->sndbuf, data, len);
 617
 618         if(len <= 0) {
 619                 errno = EWOULDBLOCK;
 620                 return 0;
 621         }
 622
 623         c->snd.last += len;
 624         ack(c, false);
 625
 626         if(!is_reliable(c)) {
 627                 c->snd.una = c->snd.nxt = c->snd.last;
 628                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 629         }
 630
 631         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 632                 start_retransmit_timer(c);
 633         }
 634
 635         return len;
 636 }
 637
 638 static void swap_ports(struct hdr *hdr) {
 639         uint16_t tmp = hdr->src;
 640         hdr->src = hdr->dst;
 641         hdr->dst = tmp;
 642 }
 643
 644 static void retransmit(struct utcp_connection *c) {
 645         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 646                 debug("Retransmit() called but nothing to retransmit!\n");
 647                 stop_retransmit_timer(c);
 648                 return;
 649         }
 650
 651         struct utcp *utcp = c->utcp;
 652
 653         struct {
 654                 struct hdr hdr;
 655                 uint8_t data[];
 656         } *pkt;
 657
 658         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 659
 660         if(!pkt) {
 661                 return;
 662         }
 663
 664         pkt->hdr.src = c->src;
 665         pkt->hdr.dst = c->dst;
 666         pkt->hdr.wnd = c->rcv.wnd;
 667         pkt->hdr.aux = 0;
 668
 669         switch(c->state) {
 670         case SYN_SENT:
 671                 // Send our SYN again
 672                 pkt->hdr.seq = c->snd.iss;
 673                 pkt->hdr.ack = 0;
 674                 pkt->hdr.ctl = SYN;
 675                 pkt->hdr.aux = 0x0101;
 676                 pkt->data[0] = 1;
 677                 pkt->data[1] = 0;
 678                 pkt->data[2] = 0;
 679                 pkt->data[3] = c->flags & 0x7;
 680                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 681                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 682                 break;
 683
 684         case SYN_RECEIVED:
 685                 // Send SYNACK again
 686                 pkt->hdr.seq = c->snd.nxt;
 687                 pkt->hdr.ack = c->rcv.nxt;
 688                 pkt->hdr.ctl = SYN | ACK;
 689                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
 690                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 691                 break;
 692
 693         case ESTABLISHED:
 694         case FIN_WAIT_1:
 695         case CLOSE_WAIT:
 696         case CLOSING:
 697         case LAST_ACK:
 698                 // Send unacked data again.
 699                 pkt->hdr.seq = c->snd.una;
 700                 pkt->hdr.ack = c->rcv.nxt;
 701                 pkt->hdr.ctl = ACK;
 702                 uint32_t len = seqdiff(c->snd.last, c->snd.una);
 703
 704                 if(len > utcp->mtu) {
 705                         len = utcp->mtu;
 706                 }
 707
 708                 if(fin_wanted(c, c->snd.una + len)) {
 709                         len--;
 710                         pkt->hdr.ctl |= FIN;
 711                 }
 712
 713                 c->snd.nxt = c->snd.una + len;
 714                 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 715                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 716                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 717                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 718                 break;
 719
 720         case CLOSED:
 721         case LISTEN:
 722         case TIME_WAIT:
 723         case FIN_WAIT_2:
 724                 // We shouldn't need to retransmit anything in this state.
 725 #ifdef UTCP_DEBUG
 726                 abort();
 727 #endif
 728                 stop_retransmit_timer(c);
 729                 goto cleanup;
 730         }
 731
 732         start_retransmit_timer(c);
 733         utcp->rto *= 2;
 734
 735         if(utcp->rto > MAX_RTO) {
 736                 utcp->rto = MAX_RTO;
 737         }
 738
 739         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 740
 741 cleanup:
 742         free(pkt);
 743 }
 744
 745 /* Update receive buffer and SACK entries after consuming data.
 746  *
 747  * Situation:
 748  *
 749  * |.....0000..1111111111.....22222......3333|
 750  * |---------------^
 751  *
 752  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 753  * to remove data from the receive buffer. The idea is to substract "len"
 754  * from the offset of all the SACK entries, and then remove/cut down entries
 755  * that are shifted to before the start of the receive buffer.
 756  *
 757  * There are three cases:
 758  * - the SACK entry is after ^, in that case just change the offset.
 759  * - the SACK entry starts before and ends after ^, so we have to
 760  *   change both its offset and size.
 761  * - the SACK entry is completely before ^, in that case delete it.
 762  */
 763 static void sack_consume(struct utcp_connection *c, size_t len) {
 764         debug("sack_consume %lu\n", (unsigned long)len);
 765
 766         if(len > c->rcvbuf.used) {
 767                 debug("All SACK entries consumed");
 768                 c->sacks[0].len = 0;
 769                 return;
 770         }
 771
 772         buffer_get(&c->rcvbuf, NULL, len);
 773
 774         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 775                 if(len < c->sacks[i].offset) {
 776                         c->sacks[i].offset -= len;
 777                         i++;
 778                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 779                         c->sacks[i].len -= len - c->sacks[i].offset;
 780                         c->sacks[i].offset = 0;
 781                         i++;
 782                 } else {
 783                         if(i < NSACKS - 1) {
 784                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 785                                 c->sacks[NSACKS - 1].len = 0;
 786                         } else {
 787                                 c->sacks[i].len = 0;
 788                                 break;
 789                         }
 790                 }
 791         }
 792
 793         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 794                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 795         }
 796 }
 797
 798 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 799         debug("out of order packet, offset %u\n", offset);
 800         // Packet loss or reordering occured. Store the data in the buffer.
 801         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 802
 803         if(rxd < 0 || (size_t)rxd < len) {
 804                 abort();
 805         }
 806
 807         // Make note of where we put it.
 808         for(int i = 0; i < NSACKS; i++) {
 809                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 810                         debug("New SACK entry %d\n", i);
 811                         c->sacks[i].offset = offset;
 812                         c->sacks[i].len = rxd;
 813                         break;
 814                 } else if(offset < c->sacks[i].offset) {
 815                         if(offset + rxd < c->sacks[i].offset) { // insert before
 816                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 817                                         debug("Insert SACK entry at %d\n", i);
 818                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 819                                         c->sacks[i].offset = offset;
 820                                         c->sacks[i].len = rxd;
 821                                 } else {
 822                                         debug("SACK entries full, dropping packet\n");
 823                                 }
 824
 825                                 break;
 826                         } else { // merge
 827                                 debug("Merge with start of SACK entry at %d\n", i);
 828                                 c->sacks[i].offset = offset;
 829                                 break;
 830                         }
 831                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 832                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 833                                 debug("Merge with end of SACK entry at %d\n", i);
 834                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 835                                 // TODO: handle potential merge with next entry
 836                         }
 837
 838                         break;
 839                 }
 840         }
 841
 842         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 843                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 844         }
 845 }
 846
 847 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 848         // Check if we can process out-of-order data now.
 849         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 850                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 851                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 852                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 853                 data = c->rcvbuf.data;
 854         }
 855
 856         if(c->recv) {
 857                 ssize_t rxd = c->recv(c, data, len);
 858
 859                 if(rxd < 0 || (size_t)rxd != len) {
 860                         // TODO: handle the application not accepting all data.
 861                         abort();
 862                 }
 863         }
 864
 865         if(c->rcvbuf.used) {
 866                 sack_consume(c, len);
 867         }
 868
 869         c->rcv.nxt += len;
 870 }
 871
 872
 873 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 874         if(!is_reliable(c)) {
 875                 c->recv(c, data, len);
 876                 c->rcv.nxt = seq + len;
 877                 return;
 878         }
 879
 880         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 881
 882         if(offset + len > c->rcvbuf.maxsize) {
 883                 abort();
 884         }
 885
 886         if(offset) {
 887                 handle_out_of_order(c, offset, data, len);
 888         } else {
 889                 handle_in_order(c, data, len);
 890         }
 891 }
 892
 893
 894 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 895         if(!utcp) {
 896                 errno = EFAULT;
 897                 return -1;
 898         }
 899
 900         if(!len) {
 901                 return 0;
 902         }
 903
 904         if(!data) {
 905                 errno = EFAULT;
 906                 return -1;
 907         }
 908
 909         print_packet(utcp, "recv", data, len);
 910
 911         // Drop packets smaller than the header
 912
 913         struct hdr hdr;
 914
 915         if(len < sizeof(hdr)) {
 916                 errno = EBADMSG;
 917                 return -1;
 918         }
 919
 920         // Make a copy from the potentially unaligned data to a struct hdr
 921
 922         memcpy(&hdr, data, sizeof(hdr));
 923         data += sizeof(hdr);
 924         len -= sizeof(hdr);
 925
 926         // Drop packets with an unknown CTL flag
 927
 928         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 929                 errno = EBADMSG;
 930                 return -1;
 931         }
 932
 933         // Check for auxiliary headers
 934
 935         const uint8_t *init = NULL;
 936
 937         uint16_t aux = hdr.aux;
 938
 939         while(aux) {
 940                 size_t auxlen = 4 * (aux >> 8) & 0xf;
 941                 uint8_t auxtype = aux & 0xff;
 942
 943                 if(len < auxlen) {
 944                         errno = EBADMSG;
 945                         return -1;
 946                 }
 947
 948                 switch(auxtype) {
 949                 case AUX_INIT:
 950                         if(!(hdr.ctl & SYN) || auxlen != 4) {
 951                                 errno = EBADMSG;
 952                                 return -1;
 953                         }
 954
 955                         init = data;
 956                         break;
 957
 958                 default:
 959                         errno = EBADMSG;
 960                         return -1;
 961                 }
 962
 963                 len -= auxlen;
 964                 data += auxlen;
 965
 966                 if(!(aux & 0x800)) {
 967                         break;
 968                 }
 969
 970                 if(len < 2) {
 971                         errno = EBADMSG;
 972                         return -1;
 973                 }
 974
 975                 memcpy(&aux, data, 2);
 976                 len -= 2;
 977                 data += 2;
 978         }
 979
 980         // Try to match the packet to an existing connection
 981
 982         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
 983
 984         // Is it for a new connection?
 985
 986         if(!c) {
 987                 // Ignore RST packets
 988
 989                 if(hdr.ctl & RST) {
 990                         return 0;
 991                 }
 992
 993                 // Is it a SYN packet and are we LISTENing?
 994
 995                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
 996                         // If we don't want to accept it, send a RST back
 997                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
 998                                 len = 1;
 999                                 goto reset;
1000                         }
1001
1002                         // Try to allocate memory, otherwise send a RST back
1003                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1004
1005                         if(!c) {
1006                                 len = 1;
1007                                 goto reset;
1008                         }
1009
1010                         // Parse auxilliary information
1011                         if(init) {
1012                                 if(init[0] < 1) {
1013                                         len = 1;
1014                                         goto reset;
1015                                 }
1016
1017                                 c->flags = init[3] & 0x7;
1018                         } else {
1019                                 c->flags = UTCP_TCP;
1020                         }
1021
1022                         // Return SYN+ACK, go to SYN_RECEIVED state
1023                         c->snd.wnd = hdr.wnd;
1024                         c->rcv.irs = hdr.seq;
1025                         c->rcv.nxt = c->rcv.irs + 1;
1026                         set_state(c, SYN_RECEIVED);
1027
1028                         struct {
1029                                 struct hdr hdr;
1030                                 uint8_t data[4];
1031                         } pkt;
1032
1033                         pkt.hdr.src = c->src;
1034                         pkt.hdr.dst = c->dst;
1035                         pkt.hdr.ack = c->rcv.irs + 1;
1036                         pkt.hdr.seq = c->snd.iss;
1037                         pkt.hdr.wnd = c->rcv.wnd;
1038                         pkt.hdr.ctl = SYN | ACK;
1039
1040                         if(init) {
1041                                 pkt.hdr.aux = 0x0101;
1042                                 pkt.data[0] = 1;
1043                                 pkt.data[1] = 0;
1044                                 pkt.data[2] = 0;
1045                                 pkt.data[3] = c->flags & 0x7;
1046                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1047                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1048                         } else {
1049                                 pkt.hdr.aux = 0;
1050                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1051                                 utcp->send(utcp, &pkt, sizeof(hdr));
1052                         }
1053                 } else {
1054                         // No, we don't want your packets, send a RST back
1055                         len = 1;
1056                         goto reset;
1057                 }
1058
1059                 return 0;
1060         }
1061
1062         debug("%p state %s\n", c->utcp, strstate[c->state]);
1063
1064         // In case this is for a CLOSED connection, ignore the packet.
1065         // TODO: make it so incoming packets can never match a CLOSED connection.
1066
1067         if(c->state == CLOSED) {
1068                 debug("Got packet for closed connection\n");
1069                 return 0;
1070         }
1071
1072         // It is for an existing connection.
1073
1074         uint32_t prevrcvnxt = c->rcv.nxt;
1075
1076         // 1. Drop invalid packets.
1077
1078         // 1a. Drop packets that should not happen in our current state.
1079
1080         switch(c->state) {
1081         case SYN_SENT:
1082         case SYN_RECEIVED:
1083         case ESTABLISHED:
1084         case FIN_WAIT_1:
1085         case FIN_WAIT_2:
1086         case CLOSE_WAIT:
1087         case CLOSING:
1088         case LAST_ACK:
1089         case TIME_WAIT:
1090                 break;
1091
1092         default:
1093 #ifdef UTCP_DEBUG
1094                 abort();
1095 #endif
1096                 break;
1097         }
1098
1099         // 1b. Drop packets with a sequence number not in our receive window.
1100
1101         bool acceptable;
1102
1103         if(c->state == SYN_SENT) {
1104                 acceptable = true;
1105         } else if(len == 0) {
1106                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1107         } else {
1108                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1109
1110                 // cut already accepted front overlapping
1111                 if(rcv_offset < 0) {
1112                         acceptable = len > (size_t) - rcv_offset;
1113
1114                         if(acceptable) {
1115                                 data -= rcv_offset;
1116                                 len += rcv_offset;
1117                                 hdr.seq -= rcv_offset;
1118                         }
1119                 } else {
1120                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1121                 }
1122         }
1123
1124         if(!acceptable) {
1125                 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1126
1127                 // Ignore unacceptable RST packets.
1128                 if(hdr.ctl & RST) {
1129                         return 0;
1130                 }
1131
1132                 // Otherwise, continue processing.
1133                 len = 0;
1134         }
1135
1136         c->snd.wnd = hdr.wnd; // TODO: move below
1137
1138         // 1c. Drop packets with an invalid ACK.
1139         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1140         // (= snd.una + c->sndbuf.used).
1141
1142         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1143                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1144
1145                 // Ignore unacceptable RST packets.
1146                 if(hdr.ctl & RST) {
1147                         return 0;
1148                 }
1149
1150                 goto reset;
1151         }
1152
1153         // 2. Handle RST packets
1154
1155         if(hdr.ctl & RST) {
1156                 switch(c->state) {
1157                 case SYN_SENT:
1158                         if(!(hdr.ctl & ACK)) {
1159                                 return 0;
1160                         }
1161
1162                         // The peer has refused our connection.
1163                         set_state(c, CLOSED);
1164                         errno = ECONNREFUSED;
1165
1166                         if(c->recv) {
1167                                 c->recv(c, NULL, 0);
1168                         }
1169
1170                         return 0;
1171
1172                 case SYN_RECEIVED:
1173                         if(hdr.ctl & ACK) {
1174                                 return 0;
1175                         }
1176
1177                         // We haven't told the application about this connection yet. Silently delete.
1178                         free_connection(c);
1179                         return 0;
1180
1181                 case ESTABLISHED:
1182                 case FIN_WAIT_1:
1183                 case FIN_WAIT_2:
1184                 case CLOSE_WAIT:
1185                         if(hdr.ctl & ACK) {
1186                                 return 0;
1187                         }
1188
1189                         // The peer has aborted our connection.
1190                         set_state(c, CLOSED);
1191                         errno = ECONNRESET;
1192
1193                         if(c->recv) {
1194                                 c->recv(c, NULL, 0);
1195                         }
1196
1197                         return 0;
1198
1199                 case CLOSING:
1200                 case LAST_ACK:
1201                 case TIME_WAIT:
1202                         if(hdr.ctl & ACK) {
1203                                 return 0;
1204                         }
1205
1206                         // As far as the application is concerned, the connection has already been closed.
1207                         // If it has called utcp_close() already, we can immediately free this connection.
1208                         if(c->reapable) {
1209                                 free_connection(c);
1210                                 return 0;
1211                         }
1212
1213                         // Otherwise, immediately move to the CLOSED state.
1214                         set_state(c, CLOSED);
1215                         return 0;
1216
1217                 default:
1218 #ifdef UTCP_DEBUG
1219                         abort();
1220 #endif
1221                         break;
1222                 }
1223         }
1224
1225         if(!(hdr.ctl & ACK)) {
1226                 goto skip_ack;
1227         }
1228
1229         // 3. Advance snd.una
1230
1231         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
1232         prevrcvnxt = c->rcv.nxt;
1233
1234         if(advanced) {
1235                 // RTT measurement
1236                 if(c->rtt_start.tv_sec) {
1237                         if(c->rtt_seq == hdr.ack) {
1238                                 struct timeval now, diff;
1239                                 gettimeofday(&now, NULL);
1240                                 timersub(&now, &c->rtt_start, &diff);
1241                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1242                                 c->rtt_start.tv_sec = 0;
1243                         } else if(c->rtt_seq < hdr.ack) {
1244                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1245                                 c->rtt_start.tv_sec = 0;
1246                         }
1247                 }
1248
1249                 int32_t data_acked = advanced;
1250
1251                 switch(c->state) {
1252                 case SYN_SENT:
1253                 case SYN_RECEIVED:
1254                         data_acked--;
1255                         break;
1256
1257                 // TODO: handle FIN as well.
1258                 default:
1259                         break;
1260                 }
1261
1262                 assert(data_acked >= 0);
1263
1264                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1265                 assert(data_acked <= bufused);
1266
1267                 if(data_acked) {
1268                         buffer_get(&c->sndbuf, NULL, data_acked);
1269                 }
1270
1271                 // Also advance snd.nxt if possible
1272                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1273                         c->snd.nxt = hdr.ack;
1274                 }
1275
1276                 c->snd.una = hdr.ack;
1277
1278                 c->dupack = 0;
1279                 c->snd.cwnd += utcp->mtu;
1280
1281                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1282                         c->snd.cwnd = c->sndbuf.maxsize;
1283                 }
1284
1285                 // Check if we have sent a FIN that is now ACKed.
1286                 switch(c->state) {
1287                 case FIN_WAIT_1:
1288                         if(c->snd.una == c->snd.last) {
1289                                 set_state(c, FIN_WAIT_2);
1290                         }
1291
1292                         break;
1293
1294                 case CLOSING:
1295                         if(c->snd.una == c->snd.last) {
1296                                 gettimeofday(&c->conn_timeout, NULL);
1297                                 c->conn_timeout.tv_sec += 60;
1298                                 set_state(c, TIME_WAIT);
1299                         }
1300
1301                         break;
1302
1303                 default:
1304                         break;
1305                 }
1306         } else {
1307                 if(!len && is_reliable(c)) {
1308                         c->dupack++;
1309
1310                         if(c->dupack == 3) {
1311                                 debug("Triplicate ACK\n");
1312                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1313                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1314                                 //Reset the congestion window so we wait for ACKs.
1315                                 c->snd.nxt = c->snd.una;
1316                                 c->snd.cwnd = utcp->mtu;
1317                                 start_retransmit_timer(c);
1318                         }
1319                 }
1320         }
1321
1322         // 4. Update timers
1323
1324         if(advanced) {
1325                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1326
1327                 if(c->snd.una == c->snd.last) {
1328                         stop_retransmit_timer(c);
1329                 } else if(is_reliable(c)) {
1330                         start_retransmit_timer(c);
1331                 }
1332         }
1333
1334 skip_ack:
1335         // 5. Process SYN stuff
1336
1337         if(hdr.ctl & SYN) {
1338                 switch(c->state) {
1339                 case SYN_SENT:
1340
1341                         // This is a SYNACK. It should always have ACKed the SYN.
1342                         if(!advanced) {
1343                                 goto reset;
1344                         }
1345
1346                         c->rcv.irs = hdr.seq;
1347                         c->rcv.nxt = hdr.seq;
1348                         set_state(c, ESTABLISHED);
1349                         // TODO: notify application of this somehow.
1350                         break;
1351
1352                 case SYN_RECEIVED:
1353                 case ESTABLISHED:
1354                 case FIN_WAIT_1:
1355                 case FIN_WAIT_2:
1356                 case CLOSE_WAIT:
1357                 case CLOSING:
1358                 case LAST_ACK:
1359                 case TIME_WAIT:
1360                         // Ehm, no. We should never receive a second SYN.
1361                         return 0;
1362
1363                 default:
1364 #ifdef UTCP_DEBUG
1365                         abort();
1366 #endif
1367                         return 0;
1368                 }
1369
1370                 // SYN counts as one sequence number
1371                 c->rcv.nxt++;
1372         }
1373
1374         // 6. Process new data
1375
1376         if(c->state == SYN_RECEIVED) {
1377                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1378                 if(!advanced) {
1379                         goto reset;
1380                 }
1381
1382                 // Are we still LISTENing?
1383                 if(utcp->accept) {
1384                         utcp->accept(c, c->src);
1385                 }
1386
1387                 if(c->state != ESTABLISHED) {
1388                         set_state(c, CLOSED);
1389                         c->reapable = true;
1390                         goto reset;
1391                 }
1392         }
1393
1394         if(len) {
1395                 switch(c->state) {
1396                 case SYN_SENT:
1397                 case SYN_RECEIVED:
1398                         // This should never happen.
1399 #ifdef UTCP_DEBUG
1400                         abort();
1401 #endif
1402                         return 0;
1403
1404                 case ESTABLISHED:
1405                 case FIN_WAIT_1:
1406                 case FIN_WAIT_2:
1407                         break;
1408
1409                 case CLOSE_WAIT:
1410                 case CLOSING:
1411                 case LAST_ACK:
1412                 case TIME_WAIT:
1413                         // Ehm no, We should never receive more data after a FIN.
1414                         goto reset;
1415
1416                 default:
1417 #ifdef UTCP_DEBUG
1418                         abort();
1419 #endif
1420                         return 0;
1421                 }
1422
1423                 handle_incoming_data(c, hdr.seq, data, len);
1424         }
1425
1426         // 7. Process FIN stuff
1427
1428         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1429                 switch(c->state) {
1430                 case SYN_SENT:
1431                 case SYN_RECEIVED:
1432                         // This should never happen.
1433 #ifdef UTCP_DEBUG
1434                         abort();
1435 #endif
1436                         break;
1437
1438                 case ESTABLISHED:
1439                         set_state(c, CLOSE_WAIT);
1440                         break;
1441
1442                 case FIN_WAIT_1:
1443                         set_state(c, CLOSING);
1444                         break;
1445
1446                 case FIN_WAIT_2:
1447                         gettimeofday(&c->conn_timeout, NULL);
1448                         c->conn_timeout.tv_sec += 60;
1449                         set_state(c, TIME_WAIT);
1450                         break;
1451
1452                 case CLOSE_WAIT:
1453                 case CLOSING:
1454                 case LAST_ACK:
1455                 case TIME_WAIT:
1456                         // Ehm, no. We should never receive a second FIN.
1457                         goto reset;
1458
1459                 default:
1460 #ifdef UTCP_DEBUG
1461                         abort();
1462 #endif
1463                         break;
1464                 }
1465
1466                 // FIN counts as one sequence number
1467                 c->rcv.nxt++;
1468                 len++;
1469
1470                 // Inform the application that the peer closed the connection.
1471                 if(c->recv) {
1472                         errno = 0;
1473                         c->recv(c, NULL, 0);
1474                 }
1475         }
1476
1477         // Now we send something back if:
1478         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1479         //   -> sendatleastone = true
1480         // - or we got an ack, so we should maybe send a bit more data
1481         //   -> sendatleastone = false
1482
1483         ack(c, len || prevrcvnxt != c->rcv.nxt);
1484         return 0;
1485
1486 reset:
1487         swap_ports(&hdr);
1488         hdr.wnd = 0;
1489         hdr.aux = 0;
1490
1491         if(hdr.ctl & ACK) {
1492                 hdr.seq = hdr.ack;
1493                 hdr.ctl = RST;
1494         } else {
1495                 hdr.ack = hdr.seq + len;
1496                 hdr.seq = 0;
1497                 hdr.ctl = RST | ACK;
1498         }
1499
1500         print_packet(utcp, "send", &hdr, sizeof(hdr));
1501         utcp->send(utcp, &hdr, sizeof(hdr));
1502         return 0;
1503
1504 }
1505
1506 int utcp_shutdown(struct utcp_connection *c, int dir) {
1507         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1508
1509         if(!c) {
1510                 errno = EFAULT;
1511                 return -1;
1512         }
1513
1514         if(c->reapable) {
1515                 debug("Error: shutdown() called on closed connection %p\n", c);
1516                 errno = EBADF;
1517                 return -1;
1518         }
1519
1520         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1521                 errno = EINVAL;
1522                 return -1;
1523         }
1524
1525         // TCP does not have a provision for stopping incoming packets.
1526         // The best we can do is to just ignore them.
1527         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1528                 c->recv = NULL;
1529         }
1530
1531         // The rest of the code deals with shutting down writes.
1532         if(dir == UTCP_SHUT_RD) {
1533                 return 0;
1534         }
1535
1536         switch(c->state) {
1537         case CLOSED:
1538         case LISTEN:
1539                 errno = ENOTCONN;
1540                 return -1;
1541
1542         case SYN_SENT:
1543                 set_state(c, CLOSED);
1544                 return 0;
1545
1546         case SYN_RECEIVED:
1547         case ESTABLISHED:
1548                 set_state(c, FIN_WAIT_1);
1549                 break;
1550
1551         case FIN_WAIT_1:
1552         case FIN_WAIT_2:
1553                 return 0;
1554
1555         case CLOSE_WAIT:
1556                 set_state(c, CLOSING);
1557                 break;
1558
1559         case CLOSING:
1560         case LAST_ACK:
1561         case TIME_WAIT:
1562                 return 0;
1563         }
1564
1565         c->snd.last++;
1566
1567         ack(c, false);
1568
1569         if(!timerisset(&c->rtrx_timeout)) {
1570                 start_retransmit_timer(c);
1571         }
1572
1573         return 0;
1574 }
1575
1576 int utcp_close(struct utcp_connection *c) {
1577         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1578                 return -1;
1579         }
1580
1581         c->recv = NULL;
1582         c->poll = NULL;
1583         c->reapable = true;
1584         return 0;
1585 }
1586
1587 int utcp_abort(struct utcp_connection *c) {
1588         if(!c) {
1589                 errno = EFAULT;
1590                 return -1;
1591         }
1592
1593         if(c->reapable) {
1594                 debug("Error: abort() called on closed connection %p\n", c);
1595                 errno = EBADF;
1596                 return -1;
1597         }
1598
1599         c->recv = NULL;
1600         c->poll = NULL;
1601         c->reapable = true;
1602
1603         switch(c->state) {
1604         case CLOSED:
1605                 return 0;
1606
1607         case LISTEN:
1608         case SYN_SENT:
1609         case CLOSING:
1610         case LAST_ACK:
1611         case TIME_WAIT:
1612                 set_state(c, CLOSED);
1613                 return 0;
1614
1615         case SYN_RECEIVED:
1616         case ESTABLISHED:
1617         case FIN_WAIT_1:
1618         case FIN_WAIT_2:
1619         case CLOSE_WAIT:
1620                 set_state(c, CLOSED);
1621                 break;
1622         }
1623
1624         // Send RST
1625
1626         struct hdr hdr;
1627
1628         hdr.src = c->src;
1629         hdr.dst = c->dst;
1630         hdr.seq = c->snd.nxt;
1631         hdr.ack = 0;
1632         hdr.wnd = 0;
1633         hdr.ctl = RST;
1634
1635         print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1636         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1637         return 0;
1638 }
1639
1640 /* Handle timeouts.
1641  * One call to this function will loop through all connections,
1642  * checking if something needs to be resent or not.
1643  * The return value is the time to the next timeout in milliseconds,
1644  * or maybe a negative value if the timeout is infinite.
1645  */
1646 struct timeval utcp_timeout(struct utcp *utcp) {
1647         struct timeval now;
1648         gettimeofday(&now, NULL);
1649         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1650
1651         for(int i = 0; i < utcp->nconnections; i++) {
1652                 struct utcp_connection *c = utcp->connections[i];
1653
1654                 if(!c) {
1655                         continue;
1656                 }
1657
1658                 // delete connections that have been utcp_close()d.
1659                 if(c->state == CLOSED) {
1660                         if(c->reapable) {
1661                                 debug("Reaping %p\n", c);
1662                                 free_connection(c);
1663                                 i--;
1664                         }
1665
1666                         continue;
1667                 }
1668
1669                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1670                         errno = ETIMEDOUT;
1671                         c->state = CLOSED;
1672
1673                         if(c->recv) {
1674                                 c->recv(c, NULL, 0);
1675                         }
1676
1677                         continue;
1678                 }
1679
1680                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1681                         debug("retransmit()\n");
1682                         retransmit(c);
1683                 }
1684
1685                 if(c->poll) {
1686                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1687                                 uint32_t len =  buffer_free(&c->sndbuf);
1688
1689                                 if(len) {
1690                                         c->poll(c, len);
1691                                 }
1692                         } else if(c->state == CLOSED) {
1693                                 c->poll(c, 0);
1694                         }
1695                 }
1696
1697                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1698                         next = c->conn_timeout;
1699                 }
1700
1701                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1702                         next = c->rtrx_timeout;
1703                 }
1704         }
1705
1706         struct timeval diff;
1707
1708         timersub(&next, &now, &diff);
1709
1710         return diff;
1711 }
1712
1713 bool utcp_is_active(struct utcp *utcp) {
1714         if(!utcp) {
1715                 return false;
1716         }
1717
1718         for(int i = 0; i < utcp->nconnections; i++)
1719                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1720                         return true;
1721                 }
1722
1723         return false;
1724 }
1725
1726 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1727         if(!send) {
1728                 errno = EFAULT;
1729                 return NULL;
1730         }
1731
1732         struct utcp *utcp = calloc(1, sizeof(*utcp));
1733
1734         if(!utcp) {
1735                 return NULL;
1736         }
1737
1738         utcp->accept = accept;
1739         utcp->pre_accept = pre_accept;
1740         utcp->send = send;
1741         utcp->priv = priv;
1742         utcp->mtu = DEFAULT_MTU;
1743         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1744         utcp->rto = START_RTO; // usec
1745
1746         return utcp;
1747 }
1748
1749 void utcp_exit(struct utcp *utcp) {
1750         if(!utcp) {
1751                 return;
1752         }
1753
1754         for(int i = 0; i < utcp->nconnections; i++) {
1755                 struct utcp_connection *c = utcp->connections[i];
1756
1757                 if(!c->reapable)
1758                         if(c->recv) {
1759                                 c->recv(c, NULL, 0);
1760                         }
1761
1762                 buffer_exit(&c->rcvbuf);
1763                 buffer_exit(&c->sndbuf);
1764                 free(c);
1765         }
1766
1767         free(utcp->connections);
1768         free(utcp);
1769 }
1770
1771 uint16_t utcp_get_mtu(struct utcp *utcp) {
1772         return utcp ? utcp->mtu : 0;
1773 }
1774
1775 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1776         // TODO: handle overhead of the header
1777         if(utcp) {
1778                 utcp->mtu = mtu;
1779         }
1780 }
1781
1782 void utcp_reset_timers(struct utcp *utcp) {
1783         if(!utcp) {
1784                 return;
1785         }
1786
1787         struct timeval now, then;
1788
1789         gettimeofday(&now, NULL);
1790
1791         then = now;
1792
1793         then.tv_sec += utcp->timeout;
1794
1795         for(int i = 0; i < utcp->nconnections; i++) {
1796                 utcp->connections[i]->rtrx_timeout = now;
1797                 utcp->connections[i]->conn_timeout = then;
1798                 utcp->connections[i]->rtt_start.tv_sec = 0;
1799         }
1800
1801         if(utcp->rto > START_RTO) {
1802                 utcp->rto = START_RTO;
1803         }
1804 }
1805
1806 int utcp_get_user_timeout(struct utcp *u) {
1807         return u ? u->timeout : 0;
1808 }
1809
1810 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1811         if(u) {
1812                 u->timeout = timeout;
1813         }
1814 }
1815
1816 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1817         return c ? c->sndbuf.maxsize : 0;
1818 }
1819
1820 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1821         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1822                 return buffer_free(&c->sndbuf);
1823         } else {
1824                 return 0;
1825         }
1826 }
1827
1828 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1829         if(!c) {
1830                 return;
1831         }
1832
1833         c->sndbuf.maxsize = size;
1834
1835         if(c->sndbuf.maxsize != size) {
1836                 c->sndbuf.maxsize = -1;
1837         }
1838 }
1839
1840 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1841         return c ? c->rcvbuf.maxsize : 0;
1842 }
1843
1844 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1845         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1846                 return buffer_free(&c->rcvbuf);
1847         } else {
1848                 return 0;
1849         }
1850 }
1851
1852 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1853         if(!c) {
1854                 return;
1855         }
1856
1857         c->rcvbuf.maxsize = size;
1858
1859         if(c->rcvbuf.maxsize != size) {
1860                 c->rcvbuf.maxsize = -1;
1861         }
1862 }
1863
1864 bool utcp_get_nodelay(struct utcp_connection *c) {
1865         return c ? c->nodelay : false;
1866 }
1867
1868 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1869         if(c) {
1870                 c->nodelay = nodelay;
1871         }
1872 }
1873
1874 bool utcp_get_keepalive(struct utcp_connection *c) {
1875         return c ? c->keepalive : false;
1876 }
1877
1878 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1879         if(c) {
1880                 c->keepalive = keepalive;
1881         }
1882 }
1883
1884 size_t utcp_get_outq(struct utcp_connection *c) {
1885         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1886 }
1887
1888 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1889         if(c) {
1890                 c->recv = recv;
1891         }
1892 }
1893
1894 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1895         if(c) {
1896                 c->poll = poll;
1897         }
1898 }
1899
1900 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1901         if(utcp) {
1902                 utcp->accept = accept;
1903                 utcp->pre_accept = pre_accept;
1904         }
1905 }