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         } else {
 419                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 420                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 421         }
 422
 423         utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
 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         if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 631                 len = buffer_put(&c->sndbuf, data, len);
 632         } else {
 633                 return 0;
 634         }
 635
 636         if(len <= 0) {
 637                 if(is_reliable(c)) {
 638                         errno = EWOULDBLOCK;
 639                         return 0;
 640                 } else {
 641                         return len;
 642                 }
 643         }
 644
 645         c->snd.last += len;
 646
 647         // Don't send anything yet if the connection has not fully established yet
 648
 649         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 650                 return len;
 651         }
 652
 653         ack(c, false);
 654
 655         if(!is_reliable(c)) {
 656                 c->snd.una = c->snd.nxt = c->snd.last;
 657                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 658         }
 659
 660         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 661                 start_retransmit_timer(c);
 662         }
 663
 664         if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
 665                 gettimeofday(&c->conn_timeout, NULL);
 666                 c->conn_timeout.tv_sec += c->utcp->timeout;
 667         }
 668
 669         return len;
 670 }
 671
 672 static void swap_ports(struct hdr *hdr) {
 673         uint16_t tmp = hdr->src;
 674         hdr->src = hdr->dst;
 675         hdr->dst = tmp;
 676 }
 677
 678 static void retransmit(struct utcp_connection *c) {
 679         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 680                 debug("Retransmit() called but nothing to retransmit!\n");
 681                 stop_retransmit_timer(c);
 682                 return;
 683         }
 684
 685         struct utcp *utcp = c->utcp;
 686
 687         struct {
 688                 struct hdr hdr;
 689                 uint8_t data[];
 690         } *pkt;
 691
 692         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 693
 694         if(!pkt) {
 695                 return;
 696         }
 697
 698         pkt->hdr.src = c->src;
 699         pkt->hdr.dst = c->dst;
 700         pkt->hdr.wnd = c->rcv.wnd;
 701         pkt->hdr.aux = 0;
 702
 703         switch(c->state) {
 704         case SYN_SENT:
 705                 // Send our SYN again
 706                 pkt->hdr.seq = c->snd.iss;
 707                 pkt->hdr.ack = 0;
 708                 pkt->hdr.ctl = SYN;
 709                 pkt->hdr.aux = 0x0101;
 710                 pkt->data[0] = 1;
 711                 pkt->data[1] = 0;
 712                 pkt->data[2] = 0;
 713                 pkt->data[3] = c->flags & 0x7;
 714                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 715                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 716                 break;
 717
 718         case SYN_RECEIVED:
 719                 // Send SYNACK again
 720                 pkt->hdr.seq = c->snd.nxt;
 721                 pkt->hdr.ack = c->rcv.nxt;
 722                 pkt->hdr.ctl = SYN | ACK;
 723                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
 724                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 725                 break;
 726
 727         case ESTABLISHED:
 728         case FIN_WAIT_1:
 729         case CLOSE_WAIT:
 730         case CLOSING:
 731         case LAST_ACK:
 732                 // Send unacked data again.
 733                 pkt->hdr.seq = c->snd.una;
 734                 pkt->hdr.ack = c->rcv.nxt;
 735                 pkt->hdr.ctl = ACK;
 736                 uint32_t len = seqdiff(c->snd.last, c->snd.una);
 737
 738                 if(len > utcp->mtu) {
 739                         len = utcp->mtu;
 740                 }
 741
 742                 if(fin_wanted(c, c->snd.una + len)) {
 743                         len--;
 744                         pkt->hdr.ctl |= FIN;
 745                 }
 746
 747                 c->snd.nxt = c->snd.una + len;
 748                 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 749                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 750                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 751                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 752                 break;
 753
 754         case CLOSED:
 755         case LISTEN:
 756         case TIME_WAIT:
 757         case FIN_WAIT_2:
 758                 // We shouldn't need to retransmit anything in this state.
 759 #ifdef UTCP_DEBUG
 760                 abort();
 761 #endif
 762                 stop_retransmit_timer(c);
 763                 goto cleanup;
 764         }
 765
 766         start_retransmit_timer(c);
 767         utcp->rto *= 2;
 768
 769         if(utcp->rto > MAX_RTO) {
 770                 utcp->rto = MAX_RTO;
 771         }
 772
 773         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 774
 775 cleanup:
 776         free(pkt);
 777 }
 778
 779 /* Update receive buffer and SACK entries after consuming data.
 780  *
 781  * Situation:
 782  *
 783  * |.....0000..1111111111.....22222......3333|
 784  * |---------------^
 785  *
 786  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 787  * to remove data from the receive buffer. The idea is to substract "len"
 788  * from the offset of all the SACK entries, and then remove/cut down entries
 789  * that are shifted to before the start of the receive buffer.
 790  *
 791  * There are three cases:
 792  * - the SACK entry is after ^, in that case just change the offset.
 793  * - the SACK entry starts before and ends after ^, so we have to
 794  *   change both its offset and size.
 795  * - the SACK entry is completely before ^, in that case delete it.
 796  */
 797 static void sack_consume(struct utcp_connection *c, size_t len) {
 798         debug("sack_consume %lu\n", (unsigned long)len);
 799
 800         if(len > c->rcvbuf.used) {
 801                 debug("All SACK entries consumed");
 802                 c->sacks[0].len = 0;
 803                 return;
 804         }
 805
 806         buffer_get(&c->rcvbuf, NULL, len);
 807
 808         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 809                 if(len < c->sacks[i].offset) {
 810                         c->sacks[i].offset -= len;
 811                         i++;
 812                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 813                         c->sacks[i].len -= len - c->sacks[i].offset;
 814                         c->sacks[i].offset = 0;
 815                         i++;
 816                 } else {
 817                         if(i < NSACKS - 1) {
 818                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 819                                 c->sacks[NSACKS - 1].len = 0;
 820                         } else {
 821                                 c->sacks[i].len = 0;
 822                                 break;
 823                         }
 824                 }
 825         }
 826
 827         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 828                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 829         }
 830 }
 831
 832 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 833         debug("out of order packet, offset %u\n", offset);
 834         // Packet loss or reordering occured. Store the data in the buffer.
 835         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 836
 837         if(rxd < 0 || (size_t)rxd < len) {
 838                 abort();
 839         }
 840
 841         // Make note of where we put it.
 842         for(int i = 0; i < NSACKS; i++) {
 843                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 844                         debug("New SACK entry %d\n", i);
 845                         c->sacks[i].offset = offset;
 846                         c->sacks[i].len = rxd;
 847                         break;
 848                 } else if(offset < c->sacks[i].offset) {
 849                         if(offset + rxd < c->sacks[i].offset) { // insert before
 850                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 851                                         debug("Insert SACK entry at %d\n", i);
 852                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 853                                         c->sacks[i].offset = offset;
 854                                         c->sacks[i].len = rxd;
 855                                 } else {
 856                                         debug("SACK entries full, dropping packet\n");
 857                                 }
 858
 859                                 break;
 860                         } else { // merge
 861                                 debug("Merge with start of SACK entry at %d\n", i);
 862                                 c->sacks[i].offset = offset;
 863                                 break;
 864                         }
 865                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 866                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 867                                 debug("Merge with end of SACK entry at %d\n", i);
 868                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 869                                 // TODO: handle potential merge with next entry
 870                         }
 871
 872                         break;
 873                 }
 874         }
 875
 876         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 877                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 878         }
 879 }
 880
 881 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 882         // Check if we can process out-of-order data now.
 883         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 884                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 885                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 886                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 887                 data = c->rcvbuf.data;
 888         }
 889
 890         if(c->recv) {
 891                 ssize_t rxd = c->recv(c, data, len);
 892
 893                 if(rxd < 0 || (size_t)rxd != len) {
 894                         // TODO: handle the application not accepting all data.
 895                         abort();
 896                 }
 897         }
 898
 899         if(c->rcvbuf.used) {
 900                 sack_consume(c, len);
 901         }
 902
 903         c->rcv.nxt += len;
 904 }
 905
 906
 907 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 908         if(!is_reliable(c)) {
 909                 c->recv(c, data, len);
 910                 c->rcv.nxt = seq + len;
 911                 return;
 912         }
 913
 914         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 915
 916         if(offset + len > c->rcvbuf.maxsize) {
 917                 abort();
 918         }
 919
 920         if(offset) {
 921                 handle_out_of_order(c, offset, data, len);
 922         } else {
 923                 handle_in_order(c, data, len);
 924         }
 925 }
 926
 927
 928 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 929         const uint8_t *ptr = data;
 930
 931         if(!utcp) {
 932                 errno = EFAULT;
 933                 return -1;
 934         }
 935
 936         if(!len) {
 937                 return 0;
 938         }
 939
 940         if(!data) {
 941                 errno = EFAULT;
 942                 return -1;
 943         }
 944
 945         print_packet(utcp, "recv", data, len);
 946
 947         // Drop packets smaller than the header
 948
 949         struct hdr hdr;
 950
 951         if(len < sizeof(hdr)) {
 952                 errno = EBADMSG;
 953                 return -1;
 954         }
 955
 956         // Make a copy from the potentially unaligned data to a struct hdr
 957
 958         memcpy(&hdr, ptr, sizeof(hdr));
 959         ptr += sizeof(hdr);
 960         len -= sizeof(hdr);
 961
 962         // Drop packets with an unknown CTL flag
 963
 964         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 965                 errno = EBADMSG;
 966                 return -1;
 967         }
 968
 969         // Check for auxiliary headers
 970
 971         const uint8_t *init = NULL;
 972
 973         uint16_t aux = hdr.aux;
 974
 975         while(aux) {
 976                 size_t auxlen = 4 * (aux >> 8) & 0xf;
 977                 uint8_t auxtype = aux & 0xff;
 978
 979                 if(len < auxlen) {
 980                         errno = EBADMSG;
 981                         return -1;
 982                 }
 983
 984                 switch(auxtype) {
 985                 case AUX_INIT:
 986                         if(!(hdr.ctl & SYN) || auxlen != 4) {
 987                                 errno = EBADMSG;
 988                                 return -1;
 989                         }
 990
 991                         init = ptr;
 992                         break;
 993
 994                 default:
 995                         errno = EBADMSG;
 996                         return -1;
 997                 }
 998
 999                 len -= auxlen;
1000                 ptr += auxlen;
1001
1002                 if(!(aux & 0x800)) {
1003                         break;
1004                 }
1005
1006                 if(len < 2) {
1007                         errno = EBADMSG;
1008                         return -1;
1009                 }
1010
1011                 memcpy(&aux, ptr, 2);
1012                 len -= 2;
1013                 ptr += 2;
1014         }
1015
1016         bool has_data = len || (hdr.ctl & (SYN | FIN));
1017
1018         // Try to match the packet to an existing connection
1019
1020         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1021
1022         // Is it for a new connection?
1023
1024         if(!c) {
1025                 // Ignore RST packets
1026
1027                 if(hdr.ctl & RST) {
1028                         return 0;
1029                 }
1030
1031                 // Is it a SYN packet and are we LISTENing?
1032
1033                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1034                         // If we don't want to accept it, send a RST back
1035                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1036                                 len = 1;
1037                                 goto reset;
1038                         }
1039
1040                         // Try to allocate memory, otherwise send a RST back
1041                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1042
1043                         if(!c) {
1044                                 len = 1;
1045                                 goto reset;
1046                         }
1047
1048                         // Parse auxilliary information
1049                         if(init) {
1050                                 if(init[0] < 1) {
1051                                         len = 1;
1052                                         goto reset;
1053                                 }
1054
1055                                 c->flags = init[3] & 0x7;
1056                         } else {
1057                                 c->flags = UTCP_TCP;
1058                         }
1059
1060 synack:
1061                         // Return SYN+ACK, go to SYN_RECEIVED state
1062                         c->snd.wnd = hdr.wnd;
1063                         c->rcv.irs = hdr.seq;
1064                         c->rcv.nxt = c->rcv.irs + 1;
1065                         set_state(c, SYN_RECEIVED);
1066
1067                         struct {
1068                                 struct hdr hdr;
1069                                 uint8_t data[4];
1070                         } pkt;
1071
1072                         pkt.hdr.src = c->src;
1073                         pkt.hdr.dst = c->dst;
1074                         pkt.hdr.ack = c->rcv.irs + 1;
1075                         pkt.hdr.seq = c->snd.iss;
1076                         pkt.hdr.wnd = c->rcv.wnd;
1077                         pkt.hdr.ctl = SYN | ACK;
1078
1079                         if(init) {
1080                                 pkt.hdr.aux = 0x0101;
1081                                 pkt.data[0] = 1;
1082                                 pkt.data[1] = 0;
1083                                 pkt.data[2] = 0;
1084                                 pkt.data[3] = c->flags & 0x7;
1085                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1086                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1087                         } else {
1088                                 pkt.hdr.aux = 0;
1089                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1090                                 utcp->send(utcp, &pkt, sizeof(hdr));
1091                         }
1092                 } else {
1093                         // No, we don't want your packets, send a RST back
1094                         len = 1;
1095                         goto reset;
1096                 }
1097
1098                 return 0;
1099         }
1100
1101         debug("%p state %s\n", c->utcp, strstate[c->state]);
1102
1103         // In case this is for a CLOSED connection, ignore the packet.
1104         // TODO: make it so incoming packets can never match a CLOSED connection.
1105
1106         if(c->state == CLOSED) {
1107                 debug("Got packet for closed connection\n");
1108                 return 0;
1109         }
1110
1111         // It is for an existing connection.
1112
1113         // 1. Drop invalid packets.
1114
1115         // 1a. Drop packets that should not happen in our current state.
1116
1117         switch(c->state) {
1118         case SYN_SENT:
1119         case SYN_RECEIVED:
1120         case ESTABLISHED:
1121         case FIN_WAIT_1:
1122         case FIN_WAIT_2:
1123         case CLOSE_WAIT:
1124         case CLOSING:
1125         case LAST_ACK:
1126         case TIME_WAIT:
1127                 break;
1128
1129         default:
1130 #ifdef UTCP_DEBUG
1131                 abort();
1132 #endif
1133                 break;
1134         }
1135
1136         // 1b. Discard data that is not in our receive window.
1137
1138         if(is_reliable(c)) {
1139                 bool acceptable;
1140
1141                 if(c->state == SYN_SENT) {
1142                         acceptable = true;
1143                 } else if(len == 0) {
1144                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1145                 } else {
1146                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1147
1148                         // cut already accepted front overlapping
1149                         if(rcv_offset < 0) {
1150                                 acceptable = len > (size_t) - rcv_offset;
1151
1152                                 if(acceptable) {
1153                                         ptr -= rcv_offset;
1154                                         len += rcv_offset;
1155                                         hdr.seq -= rcv_offset;
1156                                 }
1157                         } else {
1158                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1159                         }
1160                 }
1161
1162                 if(!acceptable) {
1163                         debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1164
1165                         // Ignore unacceptable RST packets.
1166                         if(hdr.ctl & RST) {
1167                                 return 0;
1168                         }
1169
1170                         // Otherwise, continue processing.
1171                         len = 0;
1172                 }
1173         }
1174
1175         c->snd.wnd = hdr.wnd; // TODO: move below
1176
1177         // 1c. Drop packets with an invalid ACK.
1178         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1179         // (= snd.una + c->sndbuf.used).
1180
1181         if(!is_reliable(c)) {
1182                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1183                         hdr.ack = c->snd.una;
1184                 }
1185         }
1186
1187         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1188                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1189
1190                 // Ignore unacceptable RST packets.
1191                 if(hdr.ctl & RST) {
1192                         return 0;
1193                 }
1194
1195                 goto reset;
1196         }
1197
1198         // 2. Handle RST packets
1199
1200         if(hdr.ctl & RST) {
1201                 switch(c->state) {
1202                 case SYN_SENT:
1203                         if(!(hdr.ctl & ACK)) {
1204                                 return 0;
1205                         }
1206
1207                         // The peer has refused our connection.
1208                         set_state(c, CLOSED);
1209                         errno = ECONNREFUSED;
1210
1211                         if(c->recv) {
1212                                 c->recv(c, NULL, 0);
1213                         }
1214
1215                         if(c->poll && !c->reapable) {
1216                                 c->poll(c, 0);
1217                         }
1218
1219                         return 0;
1220
1221                 case SYN_RECEIVED:
1222                         if(hdr.ctl & ACK) {
1223                                 return 0;
1224                         }
1225
1226                         // We haven't told the application about this connection yet. Silently delete.
1227                         free_connection(c);
1228                         return 0;
1229
1230                 case ESTABLISHED:
1231                 case FIN_WAIT_1:
1232                 case FIN_WAIT_2:
1233                 case CLOSE_WAIT:
1234                         if(hdr.ctl & ACK) {
1235                                 return 0;
1236                         }
1237
1238                         // The peer has aborted our connection.
1239                         set_state(c, CLOSED);
1240                         errno = ECONNRESET;
1241
1242                         if(c->recv) {
1243                                 c->recv(c, NULL, 0);
1244                         }
1245
1246                         if(c->poll && !c->reapable) {
1247                                 c->poll(c, 0);
1248                         }
1249
1250                         return 0;
1251
1252                 case CLOSING:
1253                 case LAST_ACK:
1254                 case TIME_WAIT:
1255                         if(hdr.ctl & ACK) {
1256                                 return 0;
1257                         }
1258
1259                         // As far as the application is concerned, the connection has already been closed.
1260                         // If it has called utcp_close() already, we can immediately free this connection.
1261                         if(c->reapable) {
1262                                 free_connection(c);
1263                                 return 0;
1264                         }
1265
1266                         // Otherwise, immediately move to the CLOSED state.
1267                         set_state(c, CLOSED);
1268                         return 0;
1269
1270                 default:
1271 #ifdef UTCP_DEBUG
1272                         abort();
1273 #endif
1274                         break;
1275                 }
1276         }
1277
1278         uint32_t advanced;
1279
1280         if(!(hdr.ctl & ACK)) {
1281                 advanced = 0;
1282                 goto skip_ack;
1283         }
1284
1285         // 3. Advance snd.una
1286
1287         advanced = seqdiff(hdr.ack, c->snd.una);
1288
1289         if(advanced) {
1290                 // RTT measurement
1291                 if(c->rtt_start.tv_sec) {
1292                         if(c->rtt_seq == hdr.ack) {
1293                                 struct timeval now, diff;
1294                                 gettimeofday(&now, NULL);
1295                                 timersub(&now, &c->rtt_start, &diff);
1296                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1297                                 c->rtt_start.tv_sec = 0;
1298                         } else if(c->rtt_seq < hdr.ack) {
1299                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1300                                 c->rtt_start.tv_sec = 0;
1301                         }
1302                 }
1303
1304                 int32_t data_acked = advanced;
1305
1306                 switch(c->state) {
1307                 case SYN_SENT:
1308                 case SYN_RECEIVED:
1309                         data_acked--;
1310                         break;
1311
1312                 // TODO: handle FIN as well.
1313                 default:
1314                         break;
1315                 }
1316
1317                 assert(data_acked >= 0);
1318
1319 #ifndef NDEBUG
1320                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1321                 assert(data_acked <= bufused);
1322 #endif
1323
1324                 if(data_acked) {
1325                         buffer_get(&c->sndbuf, NULL, data_acked);
1326                 }
1327
1328                 // Also advance snd.nxt if possible
1329                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1330                         c->snd.nxt = hdr.ack;
1331                 }
1332
1333                 c->snd.una = hdr.ack;
1334
1335                 c->dupack = 0;
1336                 c->snd.cwnd += utcp->mtu;
1337
1338                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1339                         c->snd.cwnd = c->sndbuf.maxsize;
1340                 }
1341
1342                 // Check if we have sent a FIN that is now ACKed.
1343                 switch(c->state) {
1344                 case FIN_WAIT_1:
1345                         if(c->snd.una == c->snd.last) {
1346                                 set_state(c, FIN_WAIT_2);
1347                         }
1348
1349                         break;
1350
1351                 case CLOSING:
1352                         if(c->snd.una == c->snd.last) {
1353                                 gettimeofday(&c->conn_timeout, NULL);
1354                                 c->conn_timeout.tv_sec += utcp->timeout;
1355                                 set_state(c, TIME_WAIT);
1356                         }
1357
1358                         break;
1359
1360                 default:
1361                         break;
1362                 }
1363         } else {
1364                 if(!len && is_reliable(c)) {
1365                         c->dupack++;
1366
1367                         if(c->dupack == 3) {
1368                                 debug("Triplicate ACK\n");
1369                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1370                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1371                                 //Reset the congestion window so we wait for ACKs.
1372                                 c->snd.nxt = c->snd.una;
1373                                 c->snd.cwnd = utcp->mtu;
1374                                 start_retransmit_timer(c);
1375                         }
1376                 }
1377         }
1378
1379         // 4. Update timers
1380
1381         if(advanced) {
1382                 if(c->snd.una == c->snd.last) {
1383                         stop_retransmit_timer(c);
1384                         timerclear(&c->conn_timeout);
1385                 } else if(is_reliable(c)) {
1386                         start_retransmit_timer(c);
1387                         gettimeofday(&c->conn_timeout, NULL);
1388                         c->conn_timeout.tv_sec += utcp->timeout;
1389                 }
1390         }
1391
1392 skip_ack:
1393         // 5. Process SYN stuff
1394
1395         if(hdr.ctl & SYN) {
1396                 switch(c->state) {
1397                 case SYN_SENT:
1398
1399                         // This is a SYNACK. It should always have ACKed the SYN.
1400                         if(!advanced) {
1401                                 goto reset;
1402                         }
1403
1404                         c->rcv.irs = hdr.seq;
1405                         c->rcv.nxt = hdr.seq;
1406
1407                         if(c->shut_wr) {
1408                                 c->snd.last++;
1409                                 set_state(c, FIN_WAIT_1);
1410                         } else {
1411                                 set_state(c, ESTABLISHED);
1412                         }
1413
1414                         // TODO: notify application of this somehow.
1415                         break;
1416
1417                 case SYN_RECEIVED:
1418                         // This is a retransmit of a SYN, send back the SYNACK.
1419                         goto synack;
1420
1421                 case ESTABLISHED:
1422                 case FIN_WAIT_1:
1423                 case FIN_WAIT_2:
1424                 case CLOSE_WAIT:
1425                 case CLOSING:
1426                 case LAST_ACK:
1427                 case TIME_WAIT:
1428                         // Ehm, no. We should never receive a second SYN.
1429                         return 0;
1430
1431                 default:
1432 #ifdef UTCP_DEBUG
1433                         abort();
1434 #endif
1435                         return 0;
1436                 }
1437
1438                 // SYN counts as one sequence number
1439                 c->rcv.nxt++;
1440         }
1441
1442         // 6. Process new data
1443
1444         if(c->state == SYN_RECEIVED) {
1445                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1446                 if(!advanced) {
1447                         goto reset;
1448                 }
1449
1450                 // Are we still LISTENing?
1451                 if(utcp->accept) {
1452                         utcp->accept(c, c->src);
1453                 }
1454
1455                 if(c->state != ESTABLISHED) {
1456                         set_state(c, CLOSED);
1457                         c->reapable = true;
1458                         goto reset;
1459                 }
1460         }
1461
1462         if(len) {
1463                 switch(c->state) {
1464                 case SYN_SENT:
1465                 case SYN_RECEIVED:
1466                         // This should never happen.
1467 #ifdef UTCP_DEBUG
1468                         abort();
1469 #endif
1470                         return 0;
1471
1472                 case ESTABLISHED:
1473                 case FIN_WAIT_1:
1474                 case FIN_WAIT_2:
1475                         break;
1476
1477                 case CLOSE_WAIT:
1478                 case CLOSING:
1479                 case LAST_ACK:
1480                 case TIME_WAIT:
1481                         // Ehm no, We should never receive more data after a FIN.
1482                         goto reset;
1483
1484                 default:
1485 #ifdef UTCP_DEBUG
1486                         abort();
1487 #endif
1488                         return 0;
1489                 }
1490
1491                 handle_incoming_data(c, hdr.seq, ptr, len);
1492         }
1493
1494         // 7. Process FIN stuff
1495
1496         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1497                 switch(c->state) {
1498                 case SYN_SENT:
1499                 case SYN_RECEIVED:
1500                         // This should never happen.
1501 #ifdef UTCP_DEBUG
1502                         abort();
1503 #endif
1504                         break;
1505
1506                 case ESTABLISHED:
1507                         set_state(c, CLOSE_WAIT);
1508                         break;
1509
1510                 case FIN_WAIT_1:
1511                         set_state(c, CLOSING);
1512                         break;
1513
1514                 case FIN_WAIT_2:
1515                         gettimeofday(&c->conn_timeout, NULL);
1516                         c->conn_timeout.tv_sec += utcp->timeout;
1517                         set_state(c, TIME_WAIT);
1518                         break;
1519
1520                 case CLOSE_WAIT:
1521                 case CLOSING:
1522                 case LAST_ACK:
1523                 case TIME_WAIT:
1524                         // Ehm, no. We should never receive a second FIN.
1525                         goto reset;
1526
1527                 default:
1528 #ifdef UTCP_DEBUG
1529                         abort();
1530 #endif
1531                         break;
1532                 }
1533
1534                 // FIN counts as one sequence number
1535                 c->rcv.nxt++;
1536                 len++;
1537
1538                 // Inform the application that the peer closed its end of the connection.
1539                 if(c->recv) {
1540                         errno = 0;
1541                         c->recv(c, NULL, 0);
1542                 }
1543         }
1544
1545         // Now we send something back if:
1546         // - we received data, so we have to send back an ACK
1547         //   -> sendatleastone = true
1548         // - or we got an ack, so we should maybe send a bit more data
1549         //   -> sendatleastone = false
1550
1551         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1552                 ack(c, has_data);
1553         }
1554
1555         return 0;
1556
1557 reset:
1558         swap_ports(&hdr);
1559         hdr.wnd = 0;
1560         hdr.aux = 0;
1561
1562         if(hdr.ctl & ACK) {
1563                 hdr.seq = hdr.ack;
1564                 hdr.ctl = RST;
1565         } else {
1566                 hdr.ack = hdr.seq + len;
1567                 hdr.seq = 0;
1568                 hdr.ctl = RST | ACK;
1569         }
1570
1571         print_packet(utcp, "send", &hdr, sizeof(hdr));
1572         utcp->send(utcp, &hdr, sizeof(hdr));
1573         return 0;
1574
1575 }
1576
1577 int utcp_shutdown(struct utcp_connection *c, int dir) {
1578         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1579
1580         if(!c) {
1581                 errno = EFAULT;
1582                 return -1;
1583         }
1584
1585         if(c->reapable) {
1586                 debug("Error: shutdown() called on closed connection %p\n", c);
1587                 errno = EBADF;
1588                 return -1;
1589         }
1590
1591         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1592                 errno = EINVAL;
1593                 return -1;
1594         }
1595
1596         // TCP does not have a provision for stopping incoming packets.
1597         // The best we can do is to just ignore them.
1598         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1599                 c->recv = NULL;
1600         }
1601
1602         // The rest of the code deals with shutting down writes.
1603         if(dir == UTCP_SHUT_RD) {
1604                 return 0;
1605         }
1606
1607         // Only process shutting down writes once.
1608         if(c->shut_wr) {
1609                 return 0;
1610         }
1611
1612         c->shut_wr = true;
1613
1614         switch(c->state) {
1615         case CLOSED:
1616         case LISTEN:
1617                 errno = ENOTCONN;
1618                 return -1;
1619
1620         case SYN_SENT:
1621                 return 0;
1622
1623         case SYN_RECEIVED:
1624         case ESTABLISHED:
1625                 set_state(c, FIN_WAIT_1);
1626                 break;
1627
1628         case FIN_WAIT_1:
1629         case FIN_WAIT_2:
1630                 return 0;
1631
1632         case CLOSE_WAIT:
1633                 set_state(c, CLOSING);
1634                 break;
1635
1636         case CLOSING:
1637         case LAST_ACK:
1638         case TIME_WAIT:
1639                 return 0;
1640         }
1641
1642         c->snd.last++;
1643
1644         ack(c, false);
1645
1646         if(!timerisset(&c->rtrx_timeout)) {
1647                 start_retransmit_timer(c);
1648         }
1649
1650         return 0;
1651 }
1652
1653 static bool reset_connection(struct utcp_connection *c) {
1654         if(!c) {
1655                 errno = EFAULT;
1656                 return false;
1657         }
1658
1659         if(c->reapable) {
1660                 debug("Error: abort() called on closed connection %p\n", c);
1661                 errno = EBADF;
1662                 return false;
1663         }
1664
1665         c->recv = NULL;
1666         c->poll = NULL;
1667
1668         switch(c->state) {
1669         case CLOSED:
1670                 return true;
1671
1672         case LISTEN:
1673         case SYN_SENT:
1674         case CLOSING:
1675         case LAST_ACK:
1676         case TIME_WAIT:
1677                 set_state(c, CLOSED);
1678                 return true;
1679
1680         case SYN_RECEIVED:
1681         case ESTABLISHED:
1682         case FIN_WAIT_1:
1683         case FIN_WAIT_2:
1684         case CLOSE_WAIT:
1685                 set_state(c, CLOSED);
1686                 break;
1687         }
1688
1689         // Send RST
1690
1691         struct hdr hdr;
1692
1693         hdr.src = c->src;
1694         hdr.dst = c->dst;
1695         hdr.seq = c->snd.nxt;
1696         hdr.ack = 0;
1697         hdr.wnd = 0;
1698         hdr.ctl = RST;
1699
1700         print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1701         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1702         return true;
1703 }
1704
1705 // Closes all the opened connections
1706 void utcp_abort_all_connections(struct utcp *utcp) {
1707         if(!utcp) {
1708                 errno = EINVAL;
1709                 return;
1710         }
1711
1712         for(int i = 0; i < utcp->nconnections; i++) {
1713                 struct utcp_connection *c = utcp->connections[i];
1714
1715                 if(c->reapable || c->state == CLOSED) {
1716                         continue;
1717                 }
1718
1719                 utcp_recv_t old_recv = c->recv;
1720                 utcp_poll_t old_poll = c->poll;
1721
1722                 reset_connection(c);
1723
1724                 if(old_recv) {
1725                         errno = 0;
1726                         old_recv(c, NULL, 0);
1727                 }
1728
1729                 if(old_poll && !c->reapable) {
1730                         errno = 0;
1731                         old_poll(c, 0);
1732                 }
1733         }
1734
1735         return;
1736 }
1737
1738 int utcp_close(struct utcp_connection *c) {
1739         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1740                 return -1;
1741         }
1742
1743         c->recv = NULL;
1744         c->poll = NULL;
1745         c->reapable = true;
1746         return 0;
1747 }
1748
1749 int utcp_abort(struct utcp_connection *c) {
1750         if(!reset_connection(c)) {
1751                 return -1;
1752         }
1753
1754         c->reapable = true;
1755         return 0;
1756 }
1757
1758 /* Handle timeouts.
1759  * One call to this function will loop through all connections,
1760  * checking if something needs to be resent or not.
1761  * The return value is the time to the next timeout in milliseconds,
1762  * or maybe a negative value if the timeout is infinite.
1763  */
1764 struct timeval utcp_timeout(struct utcp *utcp) {
1765         struct timeval now;
1766         gettimeofday(&now, NULL);
1767         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1768
1769         for(int i = 0; i < utcp->nconnections; i++) {
1770                 struct utcp_connection *c = utcp->connections[i];
1771
1772                 if(!c) {
1773                         continue;
1774                 }
1775
1776                 // delete connections that have been utcp_close()d.
1777                 if(c->state == CLOSED) {
1778                         if(c->reapable) {
1779                                 debug("Reaping %p\n", c);
1780                                 free_connection(c);
1781                                 i--;
1782                         }
1783
1784                         continue;
1785                 }
1786
1787                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1788                         errno = ETIMEDOUT;
1789                         c->state = CLOSED;
1790
1791                         if(c->recv) {
1792                                 c->recv(c, NULL, 0);
1793                         }
1794
1795                         if(c->poll && !c->reapable) {
1796                                 c->poll(c, 0);
1797                         }
1798
1799                         continue;
1800                 }
1801
1802                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1803                         debug("retransmit()\n");
1804                         retransmit(c);
1805                 }
1806
1807                 if(c->poll) {
1808                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1809                                 uint32_t len =  buffer_free(&c->sndbuf);
1810
1811                                 if(len) {
1812                                         c->poll(c, len);
1813                                 }
1814                         } else if(c->state == CLOSED) {
1815                                 c->poll(c, 0);
1816                         }
1817                 }
1818
1819                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1820                         next = c->conn_timeout;
1821                 }
1822
1823                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1824                         next = c->rtrx_timeout;
1825                 }
1826         }
1827
1828         struct timeval diff;
1829
1830         timersub(&next, &now, &diff);
1831
1832         return diff;
1833 }
1834
1835 bool utcp_is_active(struct utcp *utcp) {
1836         if(!utcp) {
1837                 return false;
1838         }
1839
1840         for(int i = 0; i < utcp->nconnections; i++)
1841                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1842                         return true;
1843                 }
1844
1845         return false;
1846 }
1847
1848 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1849         if(!send) {
1850                 errno = EFAULT;
1851                 return NULL;
1852         }
1853
1854         struct utcp *utcp = calloc(1, sizeof(*utcp));
1855
1856         if(!utcp) {
1857                 return NULL;
1858         }
1859
1860         utcp->accept = accept;
1861         utcp->pre_accept = pre_accept;
1862         utcp->send = send;
1863         utcp->priv = priv;
1864         utcp->mtu = DEFAULT_MTU;
1865         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1866         utcp->rto = START_RTO; // usec
1867
1868         return utcp;
1869 }
1870
1871 void utcp_exit(struct utcp *utcp) {
1872         if(!utcp) {
1873                 return;
1874         }
1875
1876         for(int i = 0; i < utcp->nconnections; i++) {
1877                 struct utcp_connection *c = utcp->connections[i];
1878
1879                 if(!c->reapable) {
1880                         if(c->recv) {
1881                                 c->recv(c, NULL, 0);
1882                         }
1883
1884                         if(c->poll && !c->reapable) {
1885                                 c->poll(c, 0);
1886                         }
1887                 }
1888
1889                 buffer_exit(&c->rcvbuf);
1890                 buffer_exit(&c->sndbuf);
1891                 free(c);
1892         }
1893
1894         free(utcp->connections);
1895         free(utcp);
1896 }
1897
1898 uint16_t utcp_get_mtu(struct utcp *utcp) {
1899         return utcp ? utcp->mtu : 0;
1900 }
1901
1902 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1903         // TODO: handle overhead of the header
1904         if(utcp) {
1905                 utcp->mtu = mtu;
1906         }
1907 }
1908
1909 void utcp_reset_timers(struct utcp *utcp) {
1910         if(!utcp) {
1911                 return;
1912         }
1913
1914         struct timeval now, then;
1915
1916         gettimeofday(&now, NULL);
1917
1918         then = now;
1919
1920         then.tv_sec += utcp->timeout;
1921
1922         for(int i = 0; i < utcp->nconnections; i++) {
1923                 struct utcp_connection *c = utcp->connections[i];
1924
1925                 if(c->reapable) {
1926                         continue;
1927                 }
1928
1929                 if(timerisset(&c->rtrx_timeout)) {
1930                         c->rtrx_timeout = now;
1931                 }
1932
1933                 if(timerisset(&c->conn_timeout)) {
1934                         c->conn_timeout = then;
1935                 }
1936
1937                 c->rtt_start.tv_sec = 0;
1938         }
1939
1940         if(utcp->rto > START_RTO) {
1941                 utcp->rto = START_RTO;
1942         }
1943 }
1944
1945 int utcp_get_user_timeout(struct utcp *u) {
1946         return u ? u->timeout : 0;
1947 }
1948
1949 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1950         if(u) {
1951                 u->timeout = timeout;
1952         }
1953 }
1954
1955 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1956         return c ? c->sndbuf.maxsize : 0;
1957 }
1958
1959 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1960         if(!c) {
1961                 return 0;
1962         }
1963
1964         switch(c->state) {
1965         case SYN_SENT:
1966         case SYN_RECEIVED:
1967         case ESTABLISHED:
1968         case CLOSE_WAIT:
1969                 return buffer_free(&c->sndbuf);
1970
1971         default:
1972                 return 0;
1973         }
1974 }
1975
1976 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1977         if(!c) {
1978                 return;
1979         }
1980
1981         c->sndbuf.maxsize = size;
1982
1983         if(c->sndbuf.maxsize != size) {
1984                 c->sndbuf.maxsize = -1;
1985         }
1986 }
1987
1988 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1989         return c ? c->rcvbuf.maxsize : 0;
1990 }
1991
1992 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1993         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1994                 return buffer_free(&c->rcvbuf);
1995         } else {
1996                 return 0;
1997         }
1998 }
1999
2000 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2001         if(!c) {
2002                 return;
2003         }
2004
2005         c->rcvbuf.maxsize = size;
2006
2007         if(c->rcvbuf.maxsize != size) {
2008                 c->rcvbuf.maxsize = -1;
2009         }
2010 }
2011
2012 size_t utcp_get_sendq(struct utcp_connection *c) {
2013         return c->sndbuf.used;
2014 }
2015
2016 size_t utcp_get_recvq(struct utcp_connection *c) {
2017         return c->rcvbuf.used;
2018 }
2019
2020 bool utcp_get_nodelay(struct utcp_connection *c) {
2021         return c ? c->nodelay : false;
2022 }
2023
2024 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2025         if(c) {
2026                 c->nodelay = nodelay;
2027         }
2028 }
2029
2030 bool utcp_get_keepalive(struct utcp_connection *c) {
2031         return c ? c->keepalive : false;
2032 }
2033
2034 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2035         if(c) {
2036                 c->keepalive = keepalive;
2037         }
2038 }
2039
2040 size_t utcp_get_outq(struct utcp_connection *c) {
2041         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2042 }
2043
2044 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2045         if(c) {
2046                 c->recv = recv;
2047         }
2048 }
2049
2050 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2051         if(c) {
2052                 c->poll = poll;
2053         }
2054 }
2055
2056 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2057         if(utcp) {
2058                 utcp->accept = accept;
2059                 utcp->pre_accept = pre_accept;
2060         }
2061 }
2062
2063 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2064         if(!c || c->reapable) {
2065                 return;
2066         }
2067
2068         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2069                 return;
2070         }
2071
2072         if(expect) {
2073                 // If we expect data, start the connection timer.
2074                 if(!timerisset(&c->conn_timeout)) {
2075                         gettimeofday(&c->conn_timeout, NULL);
2076                         c->conn_timeout.tv_sec += c->utcp->timeout;
2077                 }
2078         } else {
2079                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2080                 if(c->snd.una == c->snd.last) {
2081                         timerclear(&c->conn_timeout);
2082                 }
2083         }
2084 }
2085
2086 void utcp_offline(struct utcp *utcp, bool offline) {
2087         struct timeval now;
2088         gettimeofday(&now, NULL);
2089
2090         for(int i = 0; i < utcp->nconnections; i++) {
2091                 struct utcp_connection *c = utcp->connections[i];
2092
2093                 if(c->reapable) {
2094                         continue;
2095                 }
2096
2097                 utcp_expect_data(c, offline);
2098
2099                 if(!offline) {
2100                         if(timerisset(&c->rtrx_timeout)) {
2101                                 c->rtrx_timeout = now;
2102                         }
2103
2104                         utcp->connections[i]->rtt_start.tv_sec = 0;
2105                 }
2106         }
2107
2108         if(!offline && utcp->rto > START_RTO) {
2109                 utcp->rto = START_RTO;
2110         }
2111 }