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 #include <time.h>
  33
  34 #include "utcp_priv.h"
  35
  36 #ifndef EBADMSG
  37 #define EBADMSG         104
  38 #endif
  39
  40 #ifndef SHUT_RDWR
  41 #define SHUT_RDWR 2
  42 #endif
  43
  44 #ifdef poll
  45 #undef poll
  46 #endif
  47
  48 #ifndef timersub
  49 #define timersub(a, b, r)\
  50         do {\
  51                 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
  52                 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
  53                 if((r)->tv_usec < 0)\
  54                         (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
  55         } while (0)
  56 #endif
  57
  58 static inline size_t min(size_t a, size_t b) {
  59         return a < b ? a : b;
  60 }
  61
  62 static inline size_t max(size_t a, size_t b) {
  63         return a > b ? a : b;
  64 }
  65
  66 #ifdef UTCP_DEBUG
  67 #include <stdarg.h>
  68
  69 #ifndef UTCP_DEBUG_DATALEN
  70 #define UTCP_DEBUG_DATALEN 20
  71 #endif
  72
  73 static void debug(struct utcp_connection *c, const char *format, ...) {
  74         struct timespec tv;
  75         char buf[1024];
  76         int len;
  77
  78         clock_gettime(CLOCK_REALTIME, &tv);
  79         len = snprintf(buf, sizeof(buf), "%ld.%06lu %u:%u ", (long)tv.tv_sec, tv.tv_nsec / 1000, c ? c->src : 0, c ? c->dst : 0);
  80         va_list ap;
  81         va_start(ap, format);
  82         len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
  83         va_end(ap);
  84
  85         if(len > 0 && (size_t)len < sizeof(buf)) {
  86                 fwrite(buf, len, 1, stderr);
  87         }
  88 }
  89
  90 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
  91         struct hdr hdr;
  92
  93         if(len < sizeof(hdr)) {
  94                 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
  95                 return;
  96         }
  97
  98         memcpy(&hdr, pkt, sizeof(hdr));
  99
 100         uint32_t datalen;
 101
 102         if(len > sizeof(hdr)) {
 103                 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
 104         } else {
 105                 datalen = 0;
 106         }
 107
 108
 109         const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 110         char str[datalen * 2 + 1];
 111         char *p = str;
 112
 113         for(uint32_t i = 0; i < datalen; i++) {
 114                 *p++ = "0123456789ABCDEF"[data[i] >> 4];
 115                 *p++ = "0123456789ABCDEF"[data[i] & 15];
 116         }
 117
 118         *p = 0;
 119
 120         debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s data %s\n",
 121               dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
 122               hdr.ctl & SYN ? "SYN" : "",
 123               hdr.ctl & RST ? "RST" : "",
 124               hdr.ctl & FIN ? "FIN" : "",
 125               hdr.ctl & ACK ? "ACK" : "",
 126               str
 127              );
 128 }
 129
 130 static void debug_cwnd(struct utcp_connection *c) {
 131         debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
 132 }
 133 #else
 134 #define debug(...) do {} while(0)
 135 #define print_packet(...) do {} while(0)
 136 #define debug_cwnd(...) do {} while(0)
 137 #endif
 138
 139 static void set_state(struct utcp_connection *c, enum state state) {
 140         c->state = state;
 141
 142         if(state == ESTABLISHED) {
 143                 timerclear(&c->conn_timeout);
 144         }
 145
 146         debug(c, "state %s\n", strstate[state]);
 147 }
 148
 149 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 150         if(seq != c->snd.last) {
 151                 return false;
 152         }
 153
 154         switch(c->state) {
 155         case FIN_WAIT_1:
 156         case CLOSING:
 157         case LAST_ACK:
 158                 return true;
 159
 160         default:
 161                 return false;
 162         }
 163 }
 164
 165 static bool is_reliable(struct utcp_connection *c) {
 166         return c->flags & UTCP_RELIABLE;
 167 }
 168
 169 static int32_t seqdiff(uint32_t a, uint32_t b) {
 170         return a - b;
 171 }
 172
 173 // Buffer functions
 174 // TODO: convert to ringbuffers to avoid memmove() operations.
 175
 176 // Store data into the buffer
 177 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 178         debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 179
 180         size_t required = offset + len;
 181
 182         if(required > buf->maxsize) {
 183                 if(offset >= buf->maxsize) {
 184                         return 0;
 185                 }
 186
 187                 len = buf->maxsize - offset;
 188                 required = buf->maxsize;
 189         }
 190
 191         if(required > buf->size) {
 192                 size_t newsize = buf->size;
 193
 194                 if(!newsize) {
 195                         newsize = required;
 196                 } else {
 197                         do {
 198                                 newsize *= 2;
 199                         } while(newsize < required);
 200                 }
 201
 202                 if(newsize > buf->maxsize) {
 203                         newsize = buf->maxsize;
 204                 }
 205
 206                 char *newdata = realloc(buf->data, newsize);
 207
 208                 if(!newdata) {
 209                         return -1;
 210                 }
 211
 212                 buf->data = newdata;
 213                 buf->size = newsize;
 214         }
 215
 216         memcpy(buf->data + offset, data, len);
 217
 218         if(required > buf->used) {
 219                 buf->used = required;
 220         }
 221
 222         return len;
 223 }
 224
 225 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 226         return buffer_put_at(buf, buf->used, data, len);
 227 }
 228
 229 // Get data from the buffer. data can be NULL.
 230 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 231         if(len > buf->used) {
 232                 len = buf->used;
 233         }
 234
 235         if(data) {
 236                 memcpy(data, buf->data, len);
 237         }
 238
 239         if(len < buf->used) {
 240                 memmove(buf->data, buf->data + len, buf->used - len);
 241         }
 242
 243         buf->used -= len;
 244         return len;
 245 }
 246
 247 // Copy data from the buffer without removing it.
 248 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 249         if(offset >= buf->used) {
 250                 return 0;
 251         }
 252
 253         if(offset + len > buf->used) {
 254                 len = buf->used - offset;
 255         }
 256
 257         memcpy(data, buf->data + offset, len);
 258         return len;
 259 }
 260
 261 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 262         memset(buf, 0, sizeof(*buf));
 263
 264         if(len) {
 265                 buf->data = malloc(len);
 266
 267                 if(!buf->data) {
 268                         return false;
 269                 }
 270         }
 271
 272         buf->size = len;
 273         buf->maxsize = maxlen;
 274         return true;
 275 }
 276
 277 static void buffer_exit(struct buffer *buf) {
 278         free(buf->data);
 279         memset(buf, 0, sizeof(*buf));
 280 }
 281
 282 static uint32_t buffer_free(const struct buffer *buf) {
 283         return buf->maxsize - buf->used;
 284 }
 285
 286 // Connections are stored in a sorted list.
 287 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 288
 289 static int compare(const void *va, const void *vb) {
 290         assert(va && vb);
 291
 292         const struct utcp_connection *a = *(struct utcp_connection **)va;
 293         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 294
 295         assert(a && b);
 296         assert(a->src && b->src);
 297
 298         int c = (int)a->src - (int)b->src;
 299
 300         if(c) {
 301                 return c;
 302         }
 303
 304         c = (int)a->dst - (int)b->dst;
 305         return c;
 306 }
 307
 308 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 309         if(!utcp->nconnections) {
 310                 return NULL;
 311         }
 312
 313         struct utcp_connection key = {
 314                 .src = src,
 315                 .dst = dst,
 316         }, *keyp = &key;
 317         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 318         return match ? *match : NULL;
 319 }
 320
 321 static void free_connection(struct utcp_connection *c) {
 322         struct utcp *utcp = c->utcp;
 323         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 324
 325         assert(cp);
 326
 327         int i = cp - utcp->connections;
 328         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 329         utcp->nconnections--;
 330
 331         buffer_exit(&c->rcvbuf);
 332         buffer_exit(&c->sndbuf);
 333         free(c);
 334 }
 335
 336 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 337         // Check whether this combination of src and dst is free
 338
 339         if(src) {
 340                 if(find_connection(utcp, src, dst)) {
 341                         errno = EADDRINUSE;
 342                         return NULL;
 343                 }
 344         } else { // If src == 0, generate a random port number with the high bit set
 345                 if(utcp->nconnections >= 32767) {
 346                         errno = ENOMEM;
 347                         return NULL;
 348                 }
 349
 350                 src = rand() | 0x8000;
 351
 352                 while(find_connection(utcp, src, dst)) {
 353                         src++;
 354                 }
 355         }
 356
 357         // Allocate memory for the new connection
 358
 359         if(utcp->nconnections >= utcp->nallocated) {
 360                 if(!utcp->nallocated) {
 361                         utcp->nallocated = 4;
 362                 } else {
 363                         utcp->nallocated *= 2;
 364                 }
 365
 366                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 367
 368                 if(!new_array) {
 369                         return NULL;
 370                 }
 371
 372                 utcp->connections = new_array;
 373         }
 374
 375         struct utcp_connection *c = calloc(1, sizeof(*c));
 376
 377         if(!c) {
 378                 return NULL;
 379         }
 380
 381         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 382                 free(c);
 383                 return NULL;
 384         }
 385
 386         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 387                 buffer_exit(&c->sndbuf);
 388                 free(c);
 389                 return NULL;
 390         }
 391
 392         // Fill in the details
 393
 394         c->src = src;
 395         c->dst = dst;
 396 #ifdef UTCP_DEBUG
 397         c->snd.iss = 0;
 398 #else
 399         c->snd.iss = rand();
 400 #endif
 401         c->snd.una = c->snd.iss;
 402         c->snd.nxt = c->snd.iss + 1;
 403         c->snd.last = c->snd.nxt;
 404         c->snd.cwnd = (utcp->mtu > 2190 ? 2 : utcp->mtu > 1095 ? 3 : 4) * utcp->mtu;
 405         c->snd.ssthresh = ~0;
 406         debug_cwnd(c);
 407         c->utcp = utcp;
 408
 409         // Add it to the sorted list of connections
 410
 411         utcp->connections[utcp->nconnections++] = c;
 412         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 413
 414         return c;
 415 }
 416
 417 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 418         if(a > b) {
 419                 return a - b;
 420         } else {
 421                 return b - a;
 422         }
 423 }
 424
 425 // Update RTT variables. See RFC 6298.
 426 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 427         if(!rtt) {
 428                 debug(c, "invalid rtt\n");
 429                 return;
 430         }
 431
 432         struct utcp *utcp = c->utcp;
 433
 434         if(!utcp->srtt) {
 435                 utcp->srtt = rtt;
 436                 utcp->rttvar = rtt / 2;
 437         } else {
 438                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 439                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 440         }
 441
 442         utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
 443
 444         if(utcp->rto > MAX_RTO) {
 445                 utcp->rto = MAX_RTO;
 446         }
 447
 448         debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 449 }
 450
 451 static void start_retransmit_timer(struct utcp_connection *c) {
 452         gettimeofday(&c->rtrx_timeout, NULL);
 453         c->rtrx_timeout.tv_usec += c->utcp->rto;
 454
 455         while(c->rtrx_timeout.tv_usec >= 1000000) {
 456                 c->rtrx_timeout.tv_usec -= 1000000;
 457                 c->rtrx_timeout.tv_sec++;
 458         }
 459
 460         debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec);
 461 }
 462
 463 static void stop_retransmit_timer(struct utcp_connection *c) {
 464         timerclear(&c->rtrx_timeout);
 465         debug(c, "rtrx_timeout cleared\n");
 466 }
 467
 468 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 469         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 470
 471         if(!c) {
 472                 return NULL;
 473         }
 474
 475         assert((flags & ~0x1f) == 0);
 476
 477         c->flags = flags;
 478         c->recv = recv;
 479         c->priv = priv;
 480
 481         struct {
 482                 struct hdr hdr;
 483                 uint8_t init[4];
 484         } pkt;
 485
 486         pkt.hdr.src = c->src;
 487         pkt.hdr.dst = c->dst;
 488         pkt.hdr.seq = c->snd.iss;
 489         pkt.hdr.ack = 0;
 490         pkt.hdr.wnd = c->rcvbuf.maxsize;
 491         pkt.hdr.ctl = SYN;
 492         pkt.hdr.aux = 0x0101;
 493         pkt.init[0] = 1;
 494         pkt.init[1] = 0;
 495         pkt.init[2] = 0;
 496         pkt.init[3] = flags & 0x7;
 497
 498         set_state(c, SYN_SENT);
 499
 500         print_packet(c, "send", &pkt, sizeof(pkt));
 501         utcp->send(utcp, &pkt, sizeof(pkt));
 502
 503         gettimeofday(&c->conn_timeout, NULL);
 504         c->conn_timeout.tv_sec += utcp->timeout;
 505
 506         start_retransmit_timer(c);
 507
 508         return c;
 509 }
 510
 511 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 512         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 513 }
 514
 515 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 516         if(c->reapable || c->state != SYN_RECEIVED) {
 517                 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
 518                 return;
 519         }
 520
 521         debug(c, "accepted %p %p\n", c, recv, priv);
 522         c->recv = recv;
 523         c->priv = priv;
 524         set_state(c, ESTABLISHED);
 525 }
 526
 527 static void ack(struct utcp_connection *c, bool sendatleastone) {
 528         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 529         int32_t cwndleft = min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una);
 530
 531         assert(left >= 0);
 532
 533         if(cwndleft <= 0) {
 534                 left = 0;
 535         } else if(cwndleft < left) {
 536                 left = cwndleft;
 537
 538                 if(!sendatleastone || cwndleft > c->utcp->mtu) {
 539                         left -= left % c->utcp->mtu;
 540                 }
 541         }
 542
 543         debug(c, "cwndleft %d left %d\n", cwndleft, left);
 544
 545         if(!left && !sendatleastone) {
 546                 return;
 547         }
 548
 549         struct {
 550                 struct hdr hdr;
 551                 uint8_t data[];
 552         } *pkt;
 553
 554         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 555
 556         if(!pkt) {
 557                 return;
 558         }
 559
 560         pkt->hdr.src = c->src;
 561         pkt->hdr.dst = c->dst;
 562         pkt->hdr.ack = c->rcv.nxt;
 563         pkt->hdr.wnd = c->rcvbuf.maxsize;
 564         pkt->hdr.ctl = ACK;
 565         pkt->hdr.aux = 0;
 566
 567         do {
 568                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 569                 pkt->hdr.seq = c->snd.nxt;
 570
 571                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 572
 573                 c->snd.nxt += seglen;
 574                 left -= seglen;
 575
 576                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 577                         seglen--;
 578                         pkt->hdr.ctl |= FIN;
 579                 }
 580
 581                 if(!c->rtt_start.tv_sec) {
 582                         // Start RTT measurement
 583                         gettimeofday(&c->rtt_start, NULL);
 584                         c->rtt_seq = pkt->hdr.seq + seglen;
 585                         debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 586                 }
 587
 588                 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 589                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 590         } while(left);
 591
 592         free(pkt);
 593 }
 594
 595 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 596         if(c->reapable) {
 597                 debug(c, "send() called on closed connection\n");
 598                 errno = EBADF;
 599                 return -1;
 600         }
 601
 602         switch(c->state) {
 603         case CLOSED:
 604         case LISTEN:
 605                 debug(c, "send() called on unconnected connection\n");
 606                 errno = ENOTCONN;
 607                 return -1;
 608
 609         case SYN_SENT:
 610         case SYN_RECEIVED:
 611         case ESTABLISHED:
 612         case CLOSE_WAIT:
 613                 break;
 614
 615         case FIN_WAIT_1:
 616         case FIN_WAIT_2:
 617         case CLOSING:
 618         case LAST_ACK:
 619         case TIME_WAIT:
 620                 debug(c, "send() called on closed connection\n");
 621                 errno = EPIPE;
 622                 return -1;
 623         }
 624
 625         // Exit early if we have nothing to send.
 626
 627         if(!len) {
 628                 return 0;
 629         }
 630
 631         if(!data) {
 632                 errno = EFAULT;
 633                 return -1;
 634         }
 635
 636         // Check if we need to be able to buffer all data
 637
 638         if(c->flags & UTCP_NO_PARTIAL) {
 639                 if(len > buffer_free(&c->sndbuf)) {
 640                         if(len > c->sndbuf.maxsize) {
 641                                 errno = EMSGSIZE;
 642                                 return -1;
 643                         } else {
 644                                 errno = EWOULDBLOCK;
 645                                 return 0;
 646                         }
 647                 }
 648         }
 649
 650         // Add data to send buffer.
 651
 652         if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 653                 len = buffer_put(&c->sndbuf, data, len);
 654         } else {
 655                 return 0;
 656         }
 657
 658         if(len <= 0) {
 659                 if(is_reliable(c)) {
 660                         errno = EWOULDBLOCK;
 661                         return 0;
 662                 } else {
 663                         return len;
 664                 }
 665         }
 666
 667         c->snd.last += len;
 668
 669         // Don't send anything yet if the connection has not fully established yet
 670
 671         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 672                 return len;
 673         }
 674
 675         ack(c, false);
 676
 677         if(!is_reliable(c)) {
 678                 c->snd.una = c->snd.nxt = c->snd.last;
 679                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 680         }
 681
 682         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 683                 start_retransmit_timer(c);
 684         }
 685
 686         if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
 687                 gettimeofday(&c->conn_timeout, NULL);
 688                 c->conn_timeout.tv_sec += c->utcp->timeout;
 689         }
 690
 691         return len;
 692 }
 693
 694 static void swap_ports(struct hdr *hdr) {
 695         uint16_t tmp = hdr->src;
 696         hdr->src = hdr->dst;
 697         hdr->dst = tmp;
 698 }
 699
 700 static void fast_retransmit(struct utcp_connection *c) {
 701         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 702                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 703                 return;
 704         }
 705
 706         struct utcp *utcp = c->utcp;
 707
 708         struct {
 709                 struct hdr hdr;
 710                 uint8_t data[];
 711         } *pkt;
 712
 713         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 714
 715         if(!pkt) {
 716                 return;
 717         }
 718
 719         pkt->hdr.src = c->src;
 720         pkt->hdr.dst = c->dst;
 721         pkt->hdr.wnd = c->rcvbuf.maxsize;
 722         pkt->hdr.aux = 0;
 723
 724         switch(c->state) {
 725         case ESTABLISHED:
 726         case FIN_WAIT_1:
 727         case CLOSE_WAIT:
 728         case CLOSING:
 729         case LAST_ACK:
 730                 // Send unacked data again.
 731                 pkt->hdr.seq = c->snd.una;
 732                 pkt->hdr.ack = c->rcv.nxt;
 733                 pkt->hdr.ctl = ACK;
 734                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mtu);
 735
 736                 if(fin_wanted(c, c->snd.una + len)) {
 737                         len--;
 738                         pkt->hdr.ctl |= FIN;
 739                 }
 740
 741                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 742                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 743                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 744                 break;
 745
 746         default:
 747                 break;
 748         }
 749
 750         free(pkt);
 751 }
 752
 753 static void retransmit(struct utcp_connection *c) {
 754         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 755                 debug(c, "retransmit() called but nothing to retransmit!\n");
 756                 stop_retransmit_timer(c);
 757                 return;
 758         }
 759
 760         struct utcp *utcp = c->utcp;
 761
 762         struct {
 763                 struct hdr hdr;
 764                 uint8_t data[];
 765         } *pkt;
 766
 767         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 768
 769         if(!pkt) {
 770                 return;
 771         }
 772
 773         pkt->hdr.src = c->src;
 774         pkt->hdr.dst = c->dst;
 775         pkt->hdr.wnd = c->rcvbuf.maxsize;
 776         pkt->hdr.aux = 0;
 777
 778         switch(c->state) {
 779         case SYN_SENT:
 780                 // Send our SYN again
 781                 pkt->hdr.seq = c->snd.iss;
 782                 pkt->hdr.ack = 0;
 783                 pkt->hdr.ctl = SYN;
 784                 pkt->hdr.aux = 0x0101;
 785                 pkt->data[0] = 1;
 786                 pkt->data[1] = 0;
 787                 pkt->data[2] = 0;
 788                 pkt->data[3] = c->flags & 0x7;
 789                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 790                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 791                 break;
 792
 793         case SYN_RECEIVED:
 794                 // Send SYNACK again
 795                 pkt->hdr.seq = c->snd.nxt;
 796                 pkt->hdr.ack = c->rcv.nxt;
 797                 pkt->hdr.ctl = SYN | ACK;
 798                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
 799                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 800                 break;
 801
 802         case ESTABLISHED:
 803         case FIN_WAIT_1:
 804         case CLOSE_WAIT:
 805         case CLOSING:
 806         case LAST_ACK:
 807                 // Send unacked data again.
 808                 pkt->hdr.seq = c->snd.una;
 809                 pkt->hdr.ack = c->rcv.nxt;
 810                 pkt->hdr.ctl = ACK;
 811                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mtu);
 812
 813                 if(fin_wanted(c, c->snd.una + len)) {
 814                         len--;
 815                         pkt->hdr.ctl |= FIN;
 816                 }
 817
 818                 // RFC 5681 slow start after timeout
 819                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
 820                 c->snd.ssthresh = max(flightsize / 2, utcp->mtu * 2); // eq. 4
 821                 c->snd.cwnd = utcp->mtu;
 822                 debug_cwnd(c);
 823
 824                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 825                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 826                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 827
 828                 c->snd.nxt = c->snd.una + len;
 829                 break;
 830
 831         case CLOSED:
 832         case LISTEN:
 833         case TIME_WAIT:
 834         case FIN_WAIT_2:
 835                 // We shouldn't need to retransmit anything in this state.
 836 #ifdef UTCP_DEBUG
 837                 abort();
 838 #endif
 839                 stop_retransmit_timer(c);
 840                 goto cleanup;
 841         }
 842
 843         start_retransmit_timer(c);
 844         utcp->rto *= 2;
 845
 846         if(utcp->rto > MAX_RTO) {
 847                 utcp->rto = MAX_RTO;
 848         }
 849
 850         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 851
 852 cleanup:
 853         free(pkt);
 854 }
 855
 856 /* Update receive buffer and SACK entries after consuming data.
 857  *
 858  * Situation:
 859  *
 860  * |.....0000..1111111111.....22222......3333|
 861  * |---------------^
 862  *
 863  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 864  * to remove data from the receive buffer. The idea is to substract "len"
 865  * from the offset of all the SACK entries, and then remove/cut down entries
 866  * that are shifted to before the start of the receive buffer.
 867  *
 868  * There are three cases:
 869  * - the SACK entry is after ^, in that case just change the offset.
 870  * - the SACK entry starts before and ends after ^, so we have to
 871  *   change both its offset and size.
 872  * - the SACK entry is completely before ^, in that case delete it.
 873  */
 874 static void sack_consume(struct utcp_connection *c, size_t len) {
 875         debug(c, "sack_consume %lu\n", (unsigned long)len);
 876
 877         if(len > c->rcvbuf.used) {
 878                 debug(c, "all SACK entries consumed\n");
 879                 c->sacks[0].len = 0;
 880                 return;
 881         }
 882
 883         buffer_get(&c->rcvbuf, NULL, len);
 884
 885         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 886                 if(len < c->sacks[i].offset) {
 887                         c->sacks[i].offset -= len;
 888                         i++;
 889                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 890                         c->sacks[i].len -= len - c->sacks[i].offset;
 891                         c->sacks[i].offset = 0;
 892                         i++;
 893                 } else {
 894                         if(i < NSACKS - 1) {
 895                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 896                                 c->sacks[NSACKS - 1].len = 0;
 897                         } else {
 898                                 c->sacks[i].len = 0;
 899                                 break;
 900                         }
 901                 }
 902         }
 903
 904         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 905                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 906         }
 907 }
 908
 909 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 910         debug(c, "out of order packet, offset %u\n", offset);
 911         // Packet loss or reordering occured. Store the data in the buffer.
 912         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 913
 914         if(rxd < 0 || (size_t)rxd < len) {
 915                 abort();
 916         }
 917
 918         // Make note of where we put it.
 919         for(int i = 0; i < NSACKS; i++) {
 920                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 921                         debug(c, "new SACK entry %d\n", i);
 922                         c->sacks[i].offset = offset;
 923                         c->sacks[i].len = rxd;
 924                         break;
 925                 } else if(offset < c->sacks[i].offset) {
 926                         if(offset + rxd < c->sacks[i].offset) { // insert before
 927                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 928                                         debug(c, "insert SACK entry at %d\n", i);
 929                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 930                                         c->sacks[i].offset = offset;
 931                                         c->sacks[i].len = rxd;
 932                                 } else {
 933                                         debug(c, "SACK entries full, dropping packet\n");
 934                                 }
 935
 936                                 break;
 937                         } else { // merge
 938                                 debug(c, "merge with start of SACK entry at %d\n", i);
 939                                 c->sacks[i].offset = offset;
 940                                 break;
 941                         }
 942                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 943                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 944                                 debug(c, "merge with end of SACK entry at %d\n", i);
 945                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 946                                 // TODO: handle potential merge with next entry
 947                         }
 948
 949                         break;
 950                 }
 951         }
 952
 953         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 954                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 955         }
 956 }
 957
 958 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 959         // Check if we can process out-of-order data now.
 960         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 961                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 962                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 963                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 964                 data = c->rcvbuf.data;
 965         }
 966
 967         if(c->recv) {
 968                 ssize_t rxd = c->recv(c, data, len);
 969
 970                 if(rxd < 0 || (size_t)rxd != len) {
 971                         // TODO: handle the application not accepting all data.
 972                         abort();
 973                 }
 974         }
 975
 976         if(c->rcvbuf.used) {
 977                 sack_consume(c, len);
 978         }
 979
 980         c->rcv.nxt += len;
 981 }
 982
 983
 984 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 985         if(!is_reliable(c)) {
 986                 c->recv(c, data, len);
 987                 c->rcv.nxt = seq + len;
 988                 return;
 989         }
 990
 991         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 992
 993         if(offset + len > c->rcvbuf.maxsize) {
 994                 abort();
 995         }
 996
 997         if(offset) {
 998                 handle_out_of_order(c, offset, data, len);
 999         } else {
1000                 handle_in_order(c, data, len);
1001         }
1002 }
1003
1004
1005 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1006         const uint8_t *ptr = data;
1007
1008         if(!utcp) {
1009                 errno = EFAULT;
1010                 return -1;
1011         }
1012
1013         if(!len) {
1014                 return 0;
1015         }
1016
1017         if(!data) {
1018                 errno = EFAULT;
1019                 return -1;
1020         }
1021
1022         // Drop packets smaller than the header
1023
1024         struct hdr hdr;
1025
1026         if(len < sizeof(hdr)) {
1027                 print_packet(NULL, "recv", data, len);
1028                 errno = EBADMSG;
1029                 return -1;
1030         }
1031
1032         // Make a copy from the potentially unaligned data to a struct hdr
1033
1034         memcpy(&hdr, ptr, sizeof(hdr));
1035
1036         // Try to match the packet to an existing connection
1037
1038         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1039         print_packet(c, "recv", data, len);
1040
1041         // Process the header
1042
1043         ptr += sizeof(hdr);
1044         len -= sizeof(hdr);
1045
1046         // Drop packets with an unknown CTL flag
1047
1048         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
1049                 print_packet(NULL, "recv", data, len);
1050                 errno = EBADMSG;
1051                 return -1;
1052         }
1053
1054         // Check for auxiliary headers
1055
1056         const uint8_t *init = NULL;
1057
1058         uint16_t aux = hdr.aux;
1059
1060         while(aux) {
1061                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1062                 uint8_t auxtype = aux & 0xff;
1063
1064                 if(len < auxlen) {
1065                         errno = EBADMSG;
1066                         return -1;
1067                 }
1068
1069                 switch(auxtype) {
1070                 case AUX_INIT:
1071                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1072                                 errno = EBADMSG;
1073                                 return -1;
1074                         }
1075
1076                         init = ptr;
1077                         break;
1078
1079                 default:
1080                         errno = EBADMSG;
1081                         return -1;
1082                 }
1083
1084                 len -= auxlen;
1085                 ptr += auxlen;
1086
1087                 if(!(aux & 0x800)) {
1088                         break;
1089                 }
1090
1091                 if(len < 2) {
1092                         errno = EBADMSG;
1093                         return -1;
1094                 }
1095
1096                 memcpy(&aux, ptr, 2);
1097                 len -= 2;
1098                 ptr += 2;
1099         }
1100
1101         bool has_data = len || (hdr.ctl & (SYN | FIN));
1102
1103         // Is it for a new connection?
1104
1105         if(!c) {
1106                 // Ignore RST packets
1107
1108                 if(hdr.ctl & RST) {
1109                         return 0;
1110                 }
1111
1112                 // Is it a SYN packet and are we LISTENing?
1113
1114                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1115                         // If we don't want to accept it, send a RST back
1116                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1117                                 len = 1;
1118                                 goto reset;
1119                         }
1120
1121                         // Try to allocate memory, otherwise send a RST back
1122                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1123
1124                         if(!c) {
1125                                 len = 1;
1126                                 goto reset;
1127                         }
1128
1129                         // Parse auxilliary information
1130                         if(init) {
1131                                 if(init[0] < 1) {
1132                                         len = 1;
1133                                         goto reset;
1134                                 }
1135
1136                                 c->flags = init[3] & 0x7;
1137                         } else {
1138                                 c->flags = UTCP_TCP;
1139                         }
1140
1141 synack:
1142                         // Return SYN+ACK, go to SYN_RECEIVED state
1143                         c->snd.wnd = hdr.wnd;
1144                         c->rcv.irs = hdr.seq;
1145                         c->rcv.nxt = c->rcv.irs + 1;
1146                         set_state(c, SYN_RECEIVED);
1147
1148                         struct {
1149                                 struct hdr hdr;
1150                                 uint8_t data[4];
1151                         } pkt;
1152
1153                         pkt.hdr.src = c->src;
1154                         pkt.hdr.dst = c->dst;
1155                         pkt.hdr.ack = c->rcv.irs + 1;
1156                         pkt.hdr.seq = c->snd.iss;
1157                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1158                         pkt.hdr.ctl = SYN | ACK;
1159
1160                         if(init) {
1161                                 pkt.hdr.aux = 0x0101;
1162                                 pkt.data[0] = 1;
1163                                 pkt.data[1] = 0;
1164                                 pkt.data[2] = 0;
1165                                 pkt.data[3] = c->flags & 0x7;
1166                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1167                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1168                         } else {
1169                                 pkt.hdr.aux = 0;
1170                                 print_packet(c, "send", &pkt, sizeof(hdr));
1171                                 utcp->send(utcp, &pkt, sizeof(hdr));
1172                         }
1173                 } else {
1174                         // No, we don't want your packets, send a RST back
1175                         len = 1;
1176                         goto reset;
1177                 }
1178
1179                 return 0;
1180         }
1181
1182         debug(c, "state %s\n", strstate[c->state]);
1183
1184         // In case this is for a CLOSED connection, ignore the packet.
1185         // TODO: make it so incoming packets can never match a CLOSED connection.
1186
1187         if(c->state == CLOSED) {
1188                 debug(c, "got packet for closed connection\n");
1189                 return 0;
1190         }
1191
1192         // It is for an existing connection.
1193
1194         // 1. Drop invalid packets.
1195
1196         // 1a. Drop packets that should not happen in our current state.
1197
1198         switch(c->state) {
1199         case SYN_SENT:
1200         case SYN_RECEIVED:
1201         case ESTABLISHED:
1202         case FIN_WAIT_1:
1203         case FIN_WAIT_2:
1204         case CLOSE_WAIT:
1205         case CLOSING:
1206         case LAST_ACK:
1207         case TIME_WAIT:
1208                 break;
1209
1210         default:
1211 #ifdef UTCP_DEBUG
1212                 abort();
1213 #endif
1214                 break;
1215         }
1216
1217         // 1b. Discard data that is not in our receive window.
1218
1219         if(is_reliable(c)) {
1220                 bool acceptable;
1221
1222                 if(c->state == SYN_SENT) {
1223                         acceptable = true;
1224                 } else if(len == 0) {
1225                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1226                 } else {
1227                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1228
1229                         // cut already accepted front overlapping
1230                         if(rcv_offset < 0) {
1231                                 acceptable = len > (size_t) - rcv_offset;
1232
1233                                 if(acceptable) {
1234                                         ptr -= rcv_offset;
1235                                         len += rcv_offset;
1236                                         hdr.seq -= rcv_offset;
1237                                 }
1238                         } else {
1239                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1240                         }
1241                 }
1242
1243                 if(!acceptable) {
1244                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1245
1246                         // Ignore unacceptable RST packets.
1247                         if(hdr.ctl & RST) {
1248                                 return 0;
1249                         }
1250
1251                         // Otherwise, continue processing.
1252                         len = 0;
1253                 }
1254         }
1255
1256         c->snd.wnd = hdr.wnd; // TODO: move below
1257
1258         // 1c. Drop packets with an invalid ACK.
1259         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1260         // (= snd.una + c->sndbuf.used).
1261
1262         if(!is_reliable(c)) {
1263                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1264                         hdr.ack = c->snd.una;
1265                 }
1266         }
1267
1268         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1269                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1270
1271                 // Ignore unacceptable RST packets.
1272                 if(hdr.ctl & RST) {
1273                         return 0;
1274                 }
1275
1276                 goto reset;
1277         }
1278
1279         // 2. Handle RST packets
1280
1281         if(hdr.ctl & RST) {
1282                 switch(c->state) {
1283                 case SYN_SENT:
1284                         if(!(hdr.ctl & ACK)) {
1285                                 return 0;
1286                         }
1287
1288                         // The peer has refused our connection.
1289                         set_state(c, CLOSED);
1290                         errno = ECONNREFUSED;
1291
1292                         if(c->recv) {
1293                                 c->recv(c, NULL, 0);
1294                         }
1295
1296                         if(c->poll && !c->reapable) {
1297                                 c->poll(c, 0);
1298                         }
1299
1300                         return 0;
1301
1302                 case SYN_RECEIVED:
1303                         if(hdr.ctl & ACK) {
1304                                 return 0;
1305                         }
1306
1307                         // We haven't told the application about this connection yet. Silently delete.
1308                         free_connection(c);
1309                         return 0;
1310
1311                 case ESTABLISHED:
1312                 case FIN_WAIT_1:
1313                 case FIN_WAIT_2:
1314                 case CLOSE_WAIT:
1315                         if(hdr.ctl & ACK) {
1316                                 return 0;
1317                         }
1318
1319                         // The peer has aborted our connection.
1320                         set_state(c, CLOSED);
1321                         errno = ECONNRESET;
1322
1323                         if(c->recv) {
1324                                 c->recv(c, NULL, 0);
1325                         }
1326
1327                         if(c->poll && !c->reapable) {
1328                                 c->poll(c, 0);
1329                         }
1330
1331                         return 0;
1332
1333                 case CLOSING:
1334                 case LAST_ACK:
1335                 case TIME_WAIT:
1336                         if(hdr.ctl & ACK) {
1337                                 return 0;
1338                         }
1339
1340                         // As far as the application is concerned, the connection has already been closed.
1341                         // If it has called utcp_close() already, we can immediately free this connection.
1342                         if(c->reapable) {
1343                                 free_connection(c);
1344                                 return 0;
1345                         }
1346
1347                         // Otherwise, immediately move to the CLOSED state.
1348                         set_state(c, CLOSED);
1349                         return 0;
1350
1351                 default:
1352 #ifdef UTCP_DEBUG
1353                         abort();
1354 #endif
1355                         break;
1356                 }
1357         }
1358
1359         uint32_t advanced;
1360
1361         if(!(hdr.ctl & ACK)) {
1362                 advanced = 0;
1363                 goto skip_ack;
1364         }
1365
1366         // 3. Advance snd.una
1367
1368         advanced = seqdiff(hdr.ack, c->snd.una);
1369
1370         if(advanced) {
1371                 // RTT measurement
1372                 if(c->rtt_start.tv_sec) {
1373                         if(c->rtt_seq == hdr.ack) {
1374                                 struct timeval now, diff;
1375                                 gettimeofday(&now, NULL);
1376                                 timersub(&now, &c->rtt_start, &diff);
1377                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1378                                 c->rtt_start.tv_sec = 0;
1379                         } else if(c->rtt_seq < hdr.ack) {
1380                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1381                                 c->rtt_start.tv_sec = 0;
1382                         }
1383                 }
1384
1385                 int32_t data_acked = advanced;
1386
1387                 switch(c->state) {
1388                 case SYN_SENT:
1389                 case SYN_RECEIVED:
1390                         data_acked--;
1391                         break;
1392
1393                 // TODO: handle FIN as well.
1394                 default:
1395                         break;
1396                 }
1397
1398                 assert(data_acked >= 0);
1399
1400 #ifndef NDEBUG
1401                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1402                 assert(data_acked <= bufused);
1403 #endif
1404
1405                 if(data_acked) {
1406                         buffer_get(&c->sndbuf, NULL, data_acked);
1407                 }
1408
1409                 // Also advance snd.nxt if possible
1410                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1411                         c->snd.nxt = hdr.ack;
1412                 }
1413
1414                 c->snd.una = hdr.ack;
1415
1416                 if(c->dupack) {
1417                         if(c->dupack >= 3) {
1418                                 debug(c, "fast recovery ended\n");
1419                                 c->snd.cwnd = c->snd.ssthresh;
1420                         }
1421
1422                         c->dupack = 0;
1423                 }
1424
1425                 // Increase the congestion window according to RFC 5681
1426                 if(c->snd.cwnd < c->snd.ssthresh) {
1427                         c->snd.cwnd += min(advanced, utcp->mtu); // eq. 2
1428                 } else {
1429                         c->snd.cwnd += max(1, (utcp->mtu * utcp->mtu) / c->snd.cwnd); // eq. 3
1430                 }
1431
1432                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1433                         c->snd.cwnd = c->sndbuf.maxsize;
1434                 }
1435
1436                 debug_cwnd(c);
1437
1438                 // Check if we have sent a FIN that is now ACKed.
1439                 switch(c->state) {
1440                 case FIN_WAIT_1:
1441                         if(c->snd.una == c->snd.last) {
1442                                 set_state(c, FIN_WAIT_2);
1443                         }
1444
1445                         break;
1446
1447                 case CLOSING:
1448                         if(c->snd.una == c->snd.last) {
1449                                 gettimeofday(&c->conn_timeout, NULL);
1450                                 c->conn_timeout.tv_sec += utcp->timeout;
1451                                 set_state(c, TIME_WAIT);
1452                         }
1453
1454                         break;
1455
1456                 default:
1457                         break;
1458                 }
1459         } else {
1460                 if(!len && is_reliable(c)) {
1461                         c->dupack++;
1462                         debug(c, "duplicate ACK %d\n", c->dupack);
1463
1464                         if(c->dupack == 3) {
1465                                 // RFC 5681 fast recovery
1466                                 debug(c, "fast recovery started\n", c->dupack);
1467                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1468                                 c->snd.ssthresh = max(flightsize / 2, utcp->mtu * 2); // eq. 4
1469                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mtu, c->sndbuf.maxsize);
1470
1471                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1472                                         c->snd.cwnd = c->sndbuf.maxsize;
1473                                 }
1474
1475                                 debug_cwnd(c);
1476
1477                                 fast_retransmit(c);
1478                         } else if(c->dupack > 3) {
1479                                 c->snd.cwnd += utcp->mtu;
1480
1481                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1482                                         c->snd.cwnd = c->sndbuf.maxsize;
1483                                 }
1484
1485                                 debug_cwnd(c);
1486                         }
1487                 }
1488         }
1489
1490         // 4. Update timers
1491
1492         if(advanced) {
1493                 if(c->snd.una == c->snd.last) {
1494                         stop_retransmit_timer(c);
1495                         timerclear(&c->conn_timeout);
1496                 } else if(is_reliable(c)) {
1497                         start_retransmit_timer(c);
1498                         gettimeofday(&c->conn_timeout, NULL);
1499                         c->conn_timeout.tv_sec += utcp->timeout;
1500                 }
1501         }
1502
1503 skip_ack:
1504         // 5. Process SYN stuff
1505
1506         if(hdr.ctl & SYN) {
1507                 switch(c->state) {
1508                 case SYN_SENT:
1509
1510                         // This is a SYNACK. It should always have ACKed the SYN.
1511                         if(!advanced) {
1512                                 goto reset;
1513                         }
1514
1515                         c->rcv.irs = hdr.seq;
1516                         c->rcv.nxt = hdr.seq;
1517
1518                         if(c->shut_wr) {
1519                                 c->snd.last++;
1520                                 set_state(c, FIN_WAIT_1);
1521                         } else {
1522                                 set_state(c, ESTABLISHED);
1523                         }
1524
1525                         // TODO: notify application of this somehow.
1526                         break;
1527
1528                 case SYN_RECEIVED:
1529                         // This is a retransmit of a SYN, send back the SYNACK.
1530                         goto synack;
1531
1532                 case ESTABLISHED:
1533                 case FIN_WAIT_1:
1534                 case FIN_WAIT_2:
1535                 case CLOSE_WAIT:
1536                 case CLOSING:
1537                 case LAST_ACK:
1538                 case TIME_WAIT:
1539                         // Ehm, no. We should never receive a second SYN.
1540                         return 0;
1541
1542                 default:
1543 #ifdef UTCP_DEBUG
1544                         abort();
1545 #endif
1546                         return 0;
1547                 }
1548
1549                 // SYN counts as one sequence number
1550                 c->rcv.nxt++;
1551         }
1552
1553         // 6. Process new data
1554
1555         if(c->state == SYN_RECEIVED) {
1556                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1557                 if(!advanced) {
1558                         goto reset;
1559                 }
1560
1561                 // Are we still LISTENing?
1562                 if(utcp->accept) {
1563                         utcp->accept(c, c->src);
1564                 }
1565
1566                 if(c->state != ESTABLISHED) {
1567                         set_state(c, CLOSED);
1568                         c->reapable = true;
1569                         goto reset;
1570                 }
1571         }
1572
1573         if(len) {
1574                 switch(c->state) {
1575                 case SYN_SENT:
1576                 case SYN_RECEIVED:
1577                         // This should never happen.
1578 #ifdef UTCP_DEBUG
1579                         abort();
1580 #endif
1581                         return 0;
1582
1583                 case ESTABLISHED:
1584                 case FIN_WAIT_1:
1585                 case FIN_WAIT_2:
1586                         break;
1587
1588                 case CLOSE_WAIT:
1589                 case CLOSING:
1590                 case LAST_ACK:
1591                 case TIME_WAIT:
1592                         // Ehm no, We should never receive more data after a FIN.
1593                         goto reset;
1594
1595                 default:
1596 #ifdef UTCP_DEBUG
1597                         abort();
1598 #endif
1599                         return 0;
1600                 }
1601
1602                 handle_incoming_data(c, hdr.seq, ptr, len);
1603         }
1604
1605         // 7. Process FIN stuff
1606
1607         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1608                 switch(c->state) {
1609                 case SYN_SENT:
1610                 case SYN_RECEIVED:
1611                         // This should never happen.
1612 #ifdef UTCP_DEBUG
1613                         abort();
1614 #endif
1615                         break;
1616
1617                 case ESTABLISHED:
1618                         set_state(c, CLOSE_WAIT);
1619                         break;
1620
1621                 case FIN_WAIT_1:
1622                         set_state(c, CLOSING);
1623                         break;
1624
1625                 case FIN_WAIT_2:
1626                         gettimeofday(&c->conn_timeout, NULL);
1627                         c->conn_timeout.tv_sec += utcp->timeout;
1628                         set_state(c, TIME_WAIT);
1629                         break;
1630
1631                 case CLOSE_WAIT:
1632                 case CLOSING:
1633                 case LAST_ACK:
1634                 case TIME_WAIT:
1635                         // Ehm, no. We should never receive a second FIN.
1636                         goto reset;
1637
1638                 default:
1639 #ifdef UTCP_DEBUG
1640                         abort();
1641 #endif
1642                         break;
1643                 }
1644
1645                 // FIN counts as one sequence number
1646                 c->rcv.nxt++;
1647                 len++;
1648
1649                 // Inform the application that the peer closed its end of the connection.
1650                 if(c->recv) {
1651                         errno = 0;
1652                         c->recv(c, NULL, 0);
1653                 }
1654         }
1655
1656         // Now we send something back if:
1657         // - we received data, so we have to send back an ACK
1658         //   -> sendatleastone = true
1659         // - or we got an ack, so we should maybe send a bit more data
1660         //   -> sendatleastone = false
1661
1662         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1663                 ack(c, has_data);
1664         }
1665
1666         return 0;
1667
1668 reset:
1669         swap_ports(&hdr);
1670         hdr.wnd = 0;
1671         hdr.aux = 0;
1672
1673         if(hdr.ctl & ACK) {
1674                 hdr.seq = hdr.ack;
1675                 hdr.ctl = RST;
1676         } else {
1677                 hdr.ack = hdr.seq + len;
1678                 hdr.seq = 0;
1679                 hdr.ctl = RST | ACK;
1680         }
1681
1682         print_packet(c, "send", &hdr, sizeof(hdr));
1683         utcp->send(utcp, &hdr, sizeof(hdr));
1684         return 0;
1685
1686 }
1687
1688 int utcp_shutdown(struct utcp_connection *c, int dir) {
1689         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1690
1691         if(!c) {
1692                 errno = EFAULT;
1693                 return -1;
1694         }
1695
1696         if(c->reapable) {
1697                 debug(c, "shutdown() called on closed connection\n");
1698                 errno = EBADF;
1699                 return -1;
1700         }
1701
1702         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1703                 errno = EINVAL;
1704                 return -1;
1705         }
1706
1707         // TCP does not have a provision for stopping incoming packets.
1708         // The best we can do is to just ignore them.
1709         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1710                 c->recv = NULL;
1711         }
1712
1713         // The rest of the code deals with shutting down writes.
1714         if(dir == UTCP_SHUT_RD) {
1715                 return 0;
1716         }
1717
1718         // Only process shutting down writes once.
1719         if(c->shut_wr) {
1720                 return 0;
1721         }
1722
1723         c->shut_wr = true;
1724
1725         switch(c->state) {
1726         case CLOSED:
1727         case LISTEN:
1728                 errno = ENOTCONN;
1729                 return -1;
1730
1731         case SYN_SENT:
1732                 return 0;
1733
1734         case SYN_RECEIVED:
1735         case ESTABLISHED:
1736                 set_state(c, FIN_WAIT_1);
1737                 break;
1738
1739         case FIN_WAIT_1:
1740         case FIN_WAIT_2:
1741                 return 0;
1742
1743         case CLOSE_WAIT:
1744                 set_state(c, CLOSING);
1745                 break;
1746
1747         case CLOSING:
1748         case LAST_ACK:
1749         case TIME_WAIT:
1750                 return 0;
1751         }
1752
1753         c->snd.last++;
1754
1755         ack(c, false);
1756
1757         if(!timerisset(&c->rtrx_timeout)) {
1758                 start_retransmit_timer(c);
1759         }
1760
1761         return 0;
1762 }
1763
1764 static bool reset_connection(struct utcp_connection *c) {
1765         if(!c) {
1766                 errno = EFAULT;
1767                 return false;
1768         }
1769
1770         if(c->reapable) {
1771                 debug(c, "abort() called on closed connection\n");
1772                 errno = EBADF;
1773                 return false;
1774         }
1775
1776         c->recv = NULL;
1777         c->poll = NULL;
1778
1779         switch(c->state) {
1780         case CLOSED:
1781                 return true;
1782
1783         case LISTEN:
1784         case SYN_SENT:
1785         case CLOSING:
1786         case LAST_ACK:
1787         case TIME_WAIT:
1788                 set_state(c, CLOSED);
1789                 return true;
1790
1791         case SYN_RECEIVED:
1792         case ESTABLISHED:
1793         case FIN_WAIT_1:
1794         case FIN_WAIT_2:
1795         case CLOSE_WAIT:
1796                 set_state(c, CLOSED);
1797                 break;
1798         }
1799
1800         // Send RST
1801
1802         struct hdr hdr;
1803
1804         hdr.src = c->src;
1805         hdr.dst = c->dst;
1806         hdr.seq = c->snd.nxt;
1807         hdr.ack = 0;
1808         hdr.wnd = 0;
1809         hdr.ctl = RST;
1810
1811         print_packet(c, "send", &hdr, sizeof(hdr));
1812         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1813         return true;
1814 }
1815
1816 // Closes all the opened connections
1817 void utcp_abort_all_connections(struct utcp *utcp) {
1818         if(!utcp) {
1819                 errno = EINVAL;
1820                 return;
1821         }
1822
1823         for(int i = 0; i < utcp->nconnections; i++) {
1824                 struct utcp_connection *c = utcp->connections[i];
1825
1826                 if(c->reapable || c->state == CLOSED) {
1827                         continue;
1828                 }
1829
1830                 utcp_recv_t old_recv = c->recv;
1831                 utcp_poll_t old_poll = c->poll;
1832
1833                 reset_connection(c);
1834
1835                 if(old_recv) {
1836                         errno = 0;
1837                         old_recv(c, NULL, 0);
1838                 }
1839
1840                 if(old_poll && !c->reapable) {
1841                         errno = 0;
1842                         old_poll(c, 0);
1843                 }
1844         }
1845
1846         return;
1847 }
1848
1849 int utcp_close(struct utcp_connection *c) {
1850         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1851                 return -1;
1852         }
1853
1854         c->recv = NULL;
1855         c->poll = NULL;
1856         c->reapable = true;
1857         return 0;
1858 }
1859
1860 int utcp_abort(struct utcp_connection *c) {
1861         if(!reset_connection(c)) {
1862                 return -1;
1863         }
1864
1865         c->reapable = true;
1866         return 0;
1867 }
1868
1869 /* Handle timeouts.
1870  * One call to this function will loop through all connections,
1871  * checking if something needs to be resent or not.
1872  * The return value is the time to the next timeout in milliseconds,
1873  * or maybe a negative value if the timeout is infinite.
1874  */
1875 struct timeval utcp_timeout(struct utcp *utcp) {
1876         struct timeval now;
1877         gettimeofday(&now, NULL);
1878         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1879
1880         for(int i = 0; i < utcp->nconnections; i++) {
1881                 struct utcp_connection *c = utcp->connections[i];
1882
1883                 if(!c) {
1884                         continue;
1885                 }
1886
1887                 // delete connections that have been utcp_close()d.
1888                 if(c->state == CLOSED) {
1889                         if(c->reapable) {
1890                                 debug(c, "reaping\n");
1891                                 free_connection(c);
1892                                 i--;
1893                         }
1894
1895                         continue;
1896                 }
1897
1898                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1899                         errno = ETIMEDOUT;
1900                         c->state = CLOSED;
1901
1902                         if(c->recv) {
1903                                 c->recv(c, NULL, 0);
1904                         }
1905
1906                         if(c->poll && !c->reapable) {
1907                                 c->poll(c, 0);
1908                         }
1909
1910                         continue;
1911                 }
1912
1913                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1914                         debug(c, "retransmitting after timeout\n");
1915                         retransmit(c);
1916                 }
1917
1918                 if(c->poll) {
1919                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1920                                 uint32_t len =  buffer_free(&c->sndbuf);
1921
1922                                 if(len) {
1923                                         c->poll(c, len);
1924                                 }
1925                         } else if(c->state == CLOSED) {
1926                                 c->poll(c, 0);
1927                         }
1928                 }
1929
1930                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1931                         next = c->conn_timeout;
1932                 }
1933
1934                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1935                         next = c->rtrx_timeout;
1936                 }
1937         }
1938
1939         struct timeval diff;
1940
1941         timersub(&next, &now, &diff);
1942
1943         return diff;
1944 }
1945
1946 bool utcp_is_active(struct utcp *utcp) {
1947         if(!utcp) {
1948                 return false;
1949         }
1950
1951         for(int i = 0; i < utcp->nconnections; i++)
1952                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1953                         return true;
1954                 }
1955
1956         return false;
1957 }
1958
1959 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1960         if(!send) {
1961                 errno = EFAULT;
1962                 return NULL;
1963         }
1964
1965         struct utcp *utcp = calloc(1, sizeof(*utcp));
1966
1967         if(!utcp) {
1968                 return NULL;
1969         }
1970
1971         utcp->accept = accept;
1972         utcp->pre_accept = pre_accept;
1973         utcp->send = send;
1974         utcp->priv = priv;
1975         utcp->mtu = DEFAULT_MTU;
1976         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1977         utcp->rto = START_RTO; // usec
1978
1979         return utcp;
1980 }
1981
1982 void utcp_exit(struct utcp *utcp) {
1983         if(!utcp) {
1984                 return;
1985         }
1986
1987         for(int i = 0; i < utcp->nconnections; i++) {
1988                 struct utcp_connection *c = utcp->connections[i];
1989
1990                 if(!c->reapable) {
1991                         if(c->recv) {
1992                                 c->recv(c, NULL, 0);
1993                         }
1994
1995                         if(c->poll && !c->reapable) {
1996                                 c->poll(c, 0);
1997                         }
1998                 }
1999
2000                 buffer_exit(&c->rcvbuf);
2001                 buffer_exit(&c->sndbuf);
2002                 free(c);
2003         }
2004
2005         free(utcp->connections);
2006         free(utcp);
2007 }
2008
2009 uint16_t utcp_get_mtu(struct utcp *utcp) {
2010         return utcp ? utcp->mtu : 0;
2011 }
2012
2013 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2014         // TODO: handle overhead of the header
2015         if(utcp) {
2016                 utcp->mtu = mtu;
2017         }
2018 }
2019
2020 void utcp_reset_timers(struct utcp *utcp) {
2021         if(!utcp) {
2022                 return;
2023         }
2024
2025         struct timeval now, then;
2026
2027         gettimeofday(&now, NULL);
2028
2029         then = now;
2030
2031         then.tv_sec += utcp->timeout;
2032
2033         for(int i = 0; i < utcp->nconnections; i++) {
2034                 struct utcp_connection *c = utcp->connections[i];
2035
2036                 if(c->reapable) {
2037                         continue;
2038                 }
2039
2040                 if(timerisset(&c->rtrx_timeout)) {
2041                         c->rtrx_timeout = now;
2042                 }
2043
2044                 if(timerisset(&c->conn_timeout)) {
2045                         c->conn_timeout = then;
2046                 }
2047
2048                 c->rtt_start.tv_sec = 0;
2049         }
2050
2051         if(utcp->rto > START_RTO) {
2052                 utcp->rto = START_RTO;
2053         }
2054 }
2055
2056 int utcp_get_user_timeout(struct utcp *u) {
2057         return u ? u->timeout : 0;
2058 }
2059
2060 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2061         if(u) {
2062                 u->timeout = timeout;
2063         }
2064 }
2065
2066 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2067         return c ? c->sndbuf.maxsize : 0;
2068 }
2069
2070 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2071         if(!c) {
2072                 return 0;
2073         }
2074
2075         switch(c->state) {
2076         case SYN_SENT:
2077         case SYN_RECEIVED:
2078         case ESTABLISHED:
2079         case CLOSE_WAIT:
2080                 return buffer_free(&c->sndbuf);
2081
2082         default:
2083                 return 0;
2084         }
2085 }
2086
2087 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2088         if(!c) {
2089                 return;
2090         }
2091
2092         c->sndbuf.maxsize = size;
2093
2094         if(c->sndbuf.maxsize != size) {
2095                 c->sndbuf.maxsize = -1;
2096         }
2097 }
2098
2099 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2100         return c ? c->rcvbuf.maxsize : 0;
2101 }
2102
2103 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2104         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2105                 return buffer_free(&c->rcvbuf);
2106         } else {
2107                 return 0;
2108         }
2109 }
2110
2111 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2112         if(!c) {
2113                 return;
2114         }
2115
2116         c->rcvbuf.maxsize = size;
2117
2118         if(c->rcvbuf.maxsize != size) {
2119                 c->rcvbuf.maxsize = -1;
2120         }
2121 }
2122
2123 size_t utcp_get_sendq(struct utcp_connection *c) {
2124         return c->sndbuf.used;
2125 }
2126
2127 size_t utcp_get_recvq(struct utcp_connection *c) {
2128         return c->rcvbuf.used;
2129 }
2130
2131 bool utcp_get_nodelay(struct utcp_connection *c) {
2132         return c ? c->nodelay : false;
2133 }
2134
2135 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2136         if(c) {
2137                 c->nodelay = nodelay;
2138         }
2139 }
2140
2141 bool utcp_get_keepalive(struct utcp_connection *c) {
2142         return c ? c->keepalive : false;
2143 }
2144
2145 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2146         if(c) {
2147                 c->keepalive = keepalive;
2148         }
2149 }
2150
2151 size_t utcp_get_outq(struct utcp_connection *c) {
2152         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2153 }
2154
2155 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2156         if(c) {
2157                 c->recv = recv;
2158         }
2159 }
2160
2161 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2162         if(c) {
2163                 c->poll = poll;
2164         }
2165 }
2166
2167 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2168         if(utcp) {
2169                 utcp->accept = accept;
2170                 utcp->pre_accept = pre_accept;
2171         }
2172 }
2173
2174 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2175         if(!c || c->reapable) {
2176                 return;
2177         }
2178
2179         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2180                 return;
2181         }
2182
2183         if(expect) {
2184                 // If we expect data, start the connection timer.
2185                 if(!timerisset(&c->conn_timeout)) {
2186                         gettimeofday(&c->conn_timeout, NULL);
2187                         c->conn_timeout.tv_sec += c->utcp->timeout;
2188                 }
2189         } else {
2190                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2191                 if(c->snd.una == c->snd.last) {
2192                         timerclear(&c->conn_timeout);
2193                 }
2194         }
2195 }
2196
2197 void utcp_offline(struct utcp *utcp, bool offline) {
2198         struct timeval now;
2199         gettimeofday(&now, NULL);
2200
2201         for(int i = 0; i < utcp->nconnections; i++) {
2202                 struct utcp_connection *c = utcp->connections[i];
2203
2204                 if(c->reapable) {
2205                         continue;
2206                 }
2207
2208                 utcp_expect_data(c, offline);
2209
2210                 if(!offline) {
2211                         if(timerisset(&c->rtrx_timeout)) {
2212                                 c->rtrx_timeout = now;
2213                         }
2214
2215                         utcp->connections[i]->rtt_start.tv_sec = 0;
2216                 }
2217         }
2218
2219         if(!offline && utcp->rto > START_RTO) {
2220                 utcp->rto = START_RTO;
2221         }
2222 }