git.meshlink.io Git - meshlink/blob - src/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 #include "system.h"
  21 #include <time.h>
  22
  23 #include "utcp_priv.h"
  24
  25 #ifndef EBADMSG
  26 #define EBADMSG         104
  27 #endif
  28
  29 #ifndef SHUT_RDWR
  30 #define SHUT_RDWR 2
  31 #endif
  32
  33 #ifdef poll
  34 #undef poll
  35 #endif
  36
  37 #ifndef UTCP_CLOCK
  38 #if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
  39 #define UTCP_CLOCK CLOCK_MONOTONIC_RAW
  40 #else
  41 #define UTCP_CLOCK CLOCK_MONOTONIC
  42 #endif
  43 #endif
  44
  45 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
  46         r->tv_sec = a->tv_sec - b->tv_sec;
  47         r->tv_nsec = a->tv_nsec - b->tv_nsec;
  48
  49         if(r->tv_nsec < 0) {
  50                 r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
  51         }
  52 }
  53
  54 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
  55         return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
  56 }
  57
  58 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
  59         if(a->tv_sec == b->tv_sec) {
  60                 return a->tv_nsec < b->tv_nsec;
  61         } else {
  62                 return a->tv_sec < b->tv_sec;
  63         }
  64 }
  65
  66 static void timespec_clear(struct timespec *a) {
  67         a->tv_sec = 0;
  68         a->tv_nsec = 0;
  69 }
  70
  71 static bool timespec_isset(const struct timespec *a) {
  72         return a->tv_sec;
  73 }
  74
  75 static long CLOCK_GRANULARITY; // usec
  76
  77 static inline size_t min(size_t a, size_t b) {
  78         return a < b ? a : b;
  79 }
  80
  81 static inline size_t max(size_t a, size_t b) {
  82         return a > b ? a : b;
  83 }
  84
  85 #ifdef UTCP_DEBUG
  86 #include <stdarg.h>
  87
  88 #ifndef UTCP_DEBUG_DATALEN
  89 #define UTCP_DEBUG_DATALEN 20
  90 #endif
  91
  92 static void debug(struct utcp_connection *c, const char *format, ...) {
  93         struct timespec tv;
  94         char buf[1024];
  95         int len;
  96
  97         clock_gettime(CLOCK_REALTIME, &tv);
  98         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);
  99         va_list ap;
 100         va_start(ap, format);
 101         len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
 102         va_end(ap);
 103
 104         if(len > 0 && (size_t)len < sizeof(buf)) {
 105                 fwrite(buf, len, 1, stderr);
 106         }
 107 }
 108
 109 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
 110         struct hdr hdr;
 111
 112         if(len < sizeof(hdr)) {
 113                 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
 114                 return;
 115         }
 116
 117         memcpy(&hdr, pkt, sizeof(hdr));
 118
 119         uint32_t datalen;
 120
 121         if(len > sizeof(hdr)) {
 122                 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
 123         } else {
 124                 datalen = 0;
 125         }
 126
 127
 128         const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 129         char str[datalen * 2 + 1];
 130         char *p = str;
 131
 132         for(uint32_t i = 0; i < datalen; i++) {
 133                 *p++ = "0123456789ABCDEF"[data[i] >> 4];
 134                 *p++ = "0123456789ABCDEF"[data[i] & 15];
 135         }
 136
 137         *p = 0;
 138
 139         debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s%s data %s\n",
 140               dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
 141               hdr.ctl & SYN ? "SYN" : "",
 142               hdr.ctl & RST ? "RST" : "",
 143               hdr.ctl & FIN ? "FIN" : "",
 144               hdr.ctl & ACK ? "ACK" : "",
 145               hdr.ctl & MF ? "MF" : "",
 146               str
 147              );
 148 }
 149
 150 static void debug_cwnd(struct utcp_connection *c) {
 151         debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
 152 }
 153 #else
 154 #define debug(...) do {} while(0)
 155 #define print_packet(...) do {} while(0)
 156 #define debug_cwnd(...) do {} while(0)
 157 #endif
 158
 159 static void set_state(struct utcp_connection *c, enum state state) {
 160         c->state = state;
 161
 162         if(state == ESTABLISHED) {
 163                 timespec_clear(&c->conn_timeout);
 164         }
 165
 166         debug(c, "state %s\n", strstate[state]);
 167 }
 168
 169 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 170         if(seq != c->snd.last) {
 171                 return false;
 172         }
 173
 174         switch(c->state) {
 175         case FIN_WAIT_1:
 176         case CLOSING:
 177         case LAST_ACK:
 178                 return true;
 179
 180         default:
 181                 return false;
 182         }
 183 }
 184
 185 static bool is_reliable(struct utcp_connection *c) {
 186         return c->flags & UTCP_RELIABLE;
 187 }
 188
 189 static int32_t seqdiff(uint32_t a, uint32_t b) {
 190         return a - b;
 191 }
 192
 193 // Buffer functions
 194 static bool buffer_wraps(struct buffer *buf) {
 195         return buf->size - buf->offset < buf->used;
 196 }
 197
 198 static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
 199         char *newdata = realloc(buf->data, newsize);
 200
 201         if(!newdata) {
 202                 return false;
 203         }
 204
 205         buf->data = newdata;
 206
 207         if(buffer_wraps(buf)) {
 208                 // Shift the right part of the buffer until it hits the end of the new buffer.
 209                 // Old situation:
 210                 // [345......012]
 211                 // New situation:
 212                 // [345.........|........012]
 213                 uint32_t tailsize = buf->size - buf->offset;
 214                 uint32_t newoffset = newsize - tailsize;
 215                 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
 216                 buf->offset = newoffset;
 217         }
 218
 219         buf->size = newsize;
 220         return true;
 221 }
 222
 223 // Store data into the buffer
 224 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 225         debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 226
 227         // Ensure we don't store more than maxsize bytes in total
 228         size_t required = offset + len;
 229
 230         if(required > buf->maxsize) {
 231                 if(offset >= buf->maxsize) {
 232                         return 0;
 233                 }
 234
 235                 len = buf->maxsize - offset;
 236                 required = buf->maxsize;
 237         }
 238
 239         // Check if we need to resize the buffer
 240         if(required > buf->size) {
 241                 size_t newsize = buf->size;
 242
 243                 if(!newsize) {
 244                         newsize = 4096;
 245                 }
 246
 247                 do {
 248                         newsize *= 2;
 249                 } while(newsize < required);
 250
 251                 if(newsize > buf->maxsize) {
 252                         newsize = buf->maxsize;
 253                 }
 254
 255                 if(!buffer_resize(buf, newsize)) {
 256                         return -1;
 257                 }
 258         }
 259
 260         uint32_t realoffset = buf->offset + offset;
 261
 262         if(buf->size - buf->offset <= offset) {
 263                 // The offset wrapped
 264                 realoffset -= buf->size;
 265         }
 266
 267         if(buf->size - realoffset < len) {
 268                 // The new chunk of data must be wrapped
 269                 memcpy(buf->data + realoffset, data, buf->size - realoffset);
 270                 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
 271         } else {
 272                 memcpy(buf->data + realoffset, data, len);
 273         }
 274
 275         if(required > buf->used) {
 276                 buf->used = required;
 277         }
 278
 279         return len;
 280 }
 281
 282 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 283         return buffer_put_at(buf, buf->used, data, len);
 284 }
 285
 286 // Copy data from the buffer without removing it.
 287 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 288         // Ensure we don't copy more than is actually stored in the buffer
 289         if(offset >= buf->used) {
 290                 return 0;
 291         }
 292
 293         if(buf->used - offset < len) {
 294                 len = buf->used - offset;
 295         }
 296
 297         uint32_t realoffset = buf->offset + offset;
 298
 299         if(buf->size - buf->offset <= offset) {
 300                 // The offset wrapped
 301                 realoffset -= buf->size;
 302         }
 303
 304         if(buf->size - realoffset < len) {
 305                 // The data is wrapped
 306                 memcpy(data, buf->data + realoffset, buf->size - realoffset);
 307                 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
 308         } else {
 309                 memcpy(data, buf->data + realoffset, len);
 310         }
 311
 312         return len;
 313 }
 314
 315 // Copy data from the buffer without removing it.
 316 static ssize_t buffer_call(struct utcp_connection *c, struct buffer *buf, size_t offset, size_t len) {
 317         if(!c->recv) {
 318                 return len;
 319         }
 320
 321         // Ensure we don't copy more than is actually stored in the buffer
 322         if(offset >= buf->used) {
 323                 return 0;
 324         }
 325
 326         if(buf->used - offset < len) {
 327                 len = buf->used - offset;
 328         }
 329
 330         uint32_t realoffset = buf->offset + offset;
 331
 332         if(buf->size - buf->offset <= offset) {
 333                 // The offset wrapped
 334                 realoffset -= buf->size;
 335         }
 336
 337         if(buf->size - realoffset < len) {
 338                 // The data is wrapped
 339                 ssize_t rx1 = c->recv(c, buf->data + realoffset, buf->size - realoffset);
 340
 341                 if(rx1 < buf->size - realoffset) {
 342                         return rx1;
 343                 }
 344
 345                 // The channel might have been closed by the previous callback
 346                 if(!c->recv) {
 347                         return len;
 348                 }
 349
 350                 ssize_t rx2 = c->recv(c, buf->data, len - (buf->size - realoffset));
 351
 352                 if(rx2 < 0) {
 353                         return rx2;
 354                 } else {
 355                         return rx1 + rx2;
 356                 }
 357         } else {
 358                 return c->recv(c, buf->data + realoffset, len);
 359         }
 360 }
 361
 362 // Discard data from the buffer.
 363 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
 364         if(buf->used < len) {
 365                 len = buf->used;
 366         }
 367
 368         if(buf->size - buf->offset <= len) {
 369                 buf->offset -= buf->size;
 370         }
 371
 372         if(buf->used == len) {
 373                 buf->offset = 0;
 374         } else {
 375                 buf->offset += len;
 376         }
 377
 378         buf->used -= len;
 379
 380         return len;
 381 }
 382
 383 static void buffer_clear(struct buffer *buf) {
 384         buf->used = 0;
 385         buf->offset = 0;
 386 }
 387
 388 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
 389         if(maxsize < minsize) {
 390                 maxsize = minsize;
 391         }
 392
 393         buf->maxsize = maxsize;
 394
 395         return buf->size >= minsize || buffer_resize(buf, minsize);
 396 }
 397
 398 static void buffer_exit(struct buffer *buf) {
 399         free(buf->data);
 400         memset(buf, 0, sizeof(*buf));
 401 }
 402
 403 static uint32_t buffer_free(const struct buffer *buf) {
 404         return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
 405 }
 406
 407 // Connections are stored in a sorted list.
 408 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 409
 410 static int compare(const void *va, const void *vb) {
 411         assert(va && vb);
 412
 413         const struct utcp_connection *a = *(struct utcp_connection **)va;
 414         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 415
 416         assert(a && b);
 417
 418         int c = (int)a->src - (int)b->src;
 419
 420         if(c) {
 421                 return c;
 422         }
 423
 424         c = (int)a->dst - (int)b->dst;
 425         return c;
 426 }
 427
 428 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 429         if(!utcp->nconnections) {
 430                 return NULL;
 431         }
 432
 433         struct utcp_connection key = {
 434                 .src = src,
 435                 .dst = dst,
 436         }, *keyp = &key;
 437         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 438         return match ? *match : NULL;
 439 }
 440
 441 static void free_connection(struct utcp_connection *c) {
 442         struct utcp *utcp = c->utcp;
 443         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 444
 445         assert(cp);
 446
 447         int i = cp - utcp->connections;
 448         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 449         utcp->nconnections--;
 450
 451         buffer_exit(&c->rcvbuf);
 452         buffer_exit(&c->sndbuf);
 453         free(c);
 454 }
 455
 456 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 457         // Check whether this combination of src and dst is free
 458
 459         if(src) {
 460                 if(find_connection(utcp, src, dst)) {
 461                         errno = EADDRINUSE;
 462                         return NULL;
 463                 }
 464         } else { // If src == 0, generate a random port number with the high bit set
 465                 if(utcp->nconnections >= 32767) {
 466                         errno = ENOMEM;
 467                         return NULL;
 468                 }
 469
 470                 src = rand() | 0x8000;
 471
 472                 while(find_connection(utcp, src, dst)) {
 473                         src++;
 474                 }
 475         }
 476
 477         // Allocate memory for the new connection
 478
 479         if(utcp->nconnections >= utcp->nallocated) {
 480                 if(!utcp->nallocated) {
 481                         utcp->nallocated = 4;
 482                 } else {
 483                         utcp->nallocated *= 2;
 484                 }
 485
 486                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 487
 488                 if(!new_array) {
 489                         return NULL;
 490                 }
 491
 492                 utcp->connections = new_array;
 493         }
 494
 495         struct utcp_connection *c = calloc(1, sizeof(*c));
 496
 497         if(!c) {
 498                 return NULL;
 499         }
 500
 501         if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 502                 free(c);
 503                 return NULL;
 504         }
 505
 506         if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 507                 buffer_exit(&c->sndbuf);
 508                 free(c);
 509                 return NULL;
 510         }
 511
 512         // Fill in the details
 513
 514         c->src = src;
 515         c->dst = dst;
 516 #ifdef UTCP_DEBUG
 517         c->snd.iss = 0;
 518 #else
 519         c->snd.iss = rand();
 520 #endif
 521         c->snd.una = c->snd.iss;
 522         c->snd.nxt = c->snd.iss + 1;
 523         c->snd.last = c->snd.nxt;
 524         c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
 525         c->snd.ssthresh = ~0;
 526         debug_cwnd(c);
 527         c->srtt = 0;
 528         c->rttvar = 0;
 529         c->rto = START_RTO;
 530         c->utcp = utcp;
 531
 532         // Add it to the sorted list of connections
 533
 534         utcp->connections[utcp->nconnections++] = c;
 535         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 536
 537         return c;
 538 }
 539
 540 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 541         if(a > b) {
 542                 return a - b;
 543         } else {
 544                 return b - a;
 545         }
 546 }
 547
 548 // Update RTT variables. See RFC 6298.
 549 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 550         if(!rtt) {
 551                 debug(c, "invalid rtt\n");
 552                 return;
 553         }
 554
 555         if(!c->srtt) {
 556                 c->srtt = rtt;
 557                 c->rttvar = rtt / 2;
 558         } else {
 559                 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
 560                 c->srtt = (c->srtt * 7 + rtt) / 8;
 561         }
 562
 563         c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
 564
 565         if(c->rto > MAX_RTO) {
 566                 c->rto = MAX_RTO;
 567         }
 568
 569         debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
 570 }
 571
 572 static void start_retransmit_timer(struct utcp_connection *c) {
 573         clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
 574
 575         uint32_t rto = c->rto;
 576
 577         while(rto > USEC_PER_SEC) {
 578                 c->rtrx_timeout.tv_sec++;
 579                 rto -= USEC_PER_SEC;
 580         }
 581
 582         c->rtrx_timeout.tv_nsec += rto * 1000;
 583
 584         if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
 585                 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
 586                 c->rtrx_timeout.tv_sec++;
 587         }
 588
 589         debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
 590 }
 591
 592 static void stop_retransmit_timer(struct utcp_connection *c) {
 593         timespec_clear(&c->rtrx_timeout);
 594         debug(c, "rtrx_timeout cleared\n");
 595 }
 596
 597 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 598         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 599
 600         if(!c) {
 601                 return NULL;
 602         }
 603
 604         assert((flags & ~0x1f) == 0);
 605
 606         c->flags = flags;
 607         c->recv = recv;
 608         c->priv = priv;
 609
 610         struct {
 611                 struct hdr hdr;
 612                 uint8_t init[4];
 613         } pkt;
 614
 615         pkt.hdr.src = c->src;
 616         pkt.hdr.dst = c->dst;
 617         pkt.hdr.seq = c->snd.iss;
 618         pkt.hdr.ack = 0;
 619         pkt.hdr.wnd = c->rcvbuf.maxsize;
 620         pkt.hdr.ctl = SYN;
 621         pkt.hdr.aux = 0x0101;
 622         pkt.init[0] = 1;
 623         pkt.init[1] = 0;
 624         pkt.init[2] = 0;
 625         pkt.init[3] = flags & 0x7;
 626
 627         set_state(c, SYN_SENT);
 628
 629         print_packet(c, "send", &pkt, sizeof(pkt));
 630         utcp->send(utcp, &pkt, sizeof(pkt));
 631
 632         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 633         c->conn_timeout.tv_sec += utcp->timeout;
 634
 635         start_retransmit_timer(c);
 636
 637         return c;
 638 }
 639
 640 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 641         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 642 }
 643
 644 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 645         if(c->reapable || c->state != SYN_RECEIVED) {
 646                 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
 647                 return;
 648         }
 649
 650         debug(c, "accepted %p %p\n", c, recv, priv);
 651         c->recv = recv;
 652         c->priv = priv;
 653         set_state(c, ESTABLISHED);
 654 }
 655
 656 static void ack(struct utcp_connection *c, bool sendatleastone) {
 657         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 658         int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
 659
 660         assert(left >= 0);
 661
 662         if(cwndleft <= 0) {
 663                 left = 0;
 664         } else if(cwndleft < left) {
 665                 left = cwndleft;
 666
 667                 if(!sendatleastone || cwndleft > c->utcp->mss) {
 668                         left -= left % c->utcp->mss;
 669                 }
 670         }
 671
 672         debug(c, "cwndleft %d left %d\n", cwndleft, left);
 673
 674         if(!left && !sendatleastone) {
 675                 return;
 676         }
 677
 678         struct {
 679                 struct hdr hdr;
 680                 uint8_t data[];
 681         } *pkt = c->utcp->pkt;
 682
 683         pkt->hdr.src = c->src;
 684         pkt->hdr.dst = c->dst;
 685         pkt->hdr.ack = c->rcv.nxt;
 686         pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
 687         pkt->hdr.ctl = ACK;
 688         pkt->hdr.aux = 0;
 689
 690         do {
 691                 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
 692                 pkt->hdr.seq = c->snd.nxt;
 693
 694                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 695
 696                 c->snd.nxt += seglen;
 697                 left -= seglen;
 698
 699                 if(!is_reliable(c)) {
 700                         if(left) {
 701                                 pkt->hdr.ctl |= MF;
 702                         } else {
 703                                 pkt->hdr.ctl &= ~MF;
 704                         }
 705                 }
 706
 707                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 708                         seglen--;
 709                         pkt->hdr.ctl |= FIN;
 710                 }
 711
 712                 if(!c->rtt_start.tv_sec) {
 713                         // Start RTT measurement
 714                         clock_gettime(UTCP_CLOCK, &c->rtt_start);
 715                         c->rtt_seq = pkt->hdr.seq + seglen;
 716                         debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 717                 }
 718
 719                 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 720                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 721
 722                 if(left && !is_reliable(c)) {
 723                         pkt->hdr.wnd += seglen;
 724                 }
 725         } while(left);
 726 }
 727
 728 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 729         if(c->reapable) {
 730                 debug(c, "send() called on closed connection\n");
 731                 errno = EBADF;
 732                 return -1;
 733         }
 734
 735         switch(c->state) {
 736         case CLOSED:
 737         case LISTEN:
 738                 debug(c, "send() called on unconnected connection\n");
 739                 errno = ENOTCONN;
 740                 return -1;
 741
 742         case SYN_SENT:
 743         case SYN_RECEIVED:
 744         case ESTABLISHED:
 745         case CLOSE_WAIT:
 746                 break;
 747
 748         case FIN_WAIT_1:
 749         case FIN_WAIT_2:
 750         case CLOSING:
 751         case LAST_ACK:
 752         case TIME_WAIT:
 753                 debug(c, "send() called on closed connection\n");
 754                 errno = EPIPE;
 755                 return -1;
 756         }
 757
 758         // Exit early if we have nothing to send.
 759
 760         if(!len) {
 761                 return 0;
 762         }
 763
 764         if(!data) {
 765                 errno = EFAULT;
 766                 return -1;
 767         }
 768
 769         // Check if we need to be able to buffer all data
 770
 771         if(c->flags & UTCP_NO_PARTIAL) {
 772                 if(len > buffer_free(&c->sndbuf)) {
 773                         if(len > c->sndbuf.maxsize) {
 774                                 errno = EMSGSIZE;
 775                                 return -1;
 776                         } else {
 777                                 errno = EWOULDBLOCK;
 778                                 return 0;
 779                         }
 780                 }
 781         }
 782
 783         // Add data to send buffer.
 784
 785         if(is_reliable(c)) {
 786                 len = buffer_put(&c->sndbuf, data, len);
 787         } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
 788                 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
 789                         errno = EMSGSIZE;
 790                         return -1;
 791                 }
 792         } else {
 793                 return 0;
 794         }
 795
 796         if(len <= 0) {
 797                 if(is_reliable(c)) {
 798                         errno = EWOULDBLOCK;
 799                         return 0;
 800                 } else {
 801                         return len;
 802                 }
 803         }
 804
 805         c->snd.last += len;
 806
 807         // Don't send anything yet if the connection has not fully established yet
 808
 809         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 810                 return len;
 811         }
 812
 813         ack(c, false);
 814
 815         if(!is_reliable(c)) {
 816                 c->snd.una = c->snd.nxt = c->snd.last;
 817                 buffer_discard(&c->sndbuf, c->sndbuf.used);
 818         }
 819
 820         if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
 821                 start_retransmit_timer(c);
 822         }
 823
 824         if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
 825                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 826                 c->conn_timeout.tv_sec += c->utcp->timeout;
 827         }
 828
 829         return len;
 830 }
 831
 832 static void swap_ports(struct hdr *hdr) {
 833         uint16_t tmp = hdr->src;
 834         hdr->src = hdr->dst;
 835         hdr->dst = tmp;
 836 }
 837
 838 static void fast_retransmit(struct utcp_connection *c) {
 839         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 840                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 841                 return;
 842         }
 843
 844         struct utcp *utcp = c->utcp;
 845
 846         struct {
 847                 struct hdr hdr;
 848                 uint8_t data[];
 849         } *pkt = c->utcp->pkt;
 850
 851         pkt->hdr.src = c->src;
 852         pkt->hdr.dst = c->dst;
 853         pkt->hdr.wnd = c->rcvbuf.maxsize;
 854         pkt->hdr.aux = 0;
 855
 856         switch(c->state) {
 857         case ESTABLISHED:
 858         case FIN_WAIT_1:
 859         case CLOSE_WAIT:
 860         case CLOSING:
 861         case LAST_ACK:
 862                 // Send unacked data again.
 863                 pkt->hdr.seq = c->snd.una;
 864                 pkt->hdr.ack = c->rcv.nxt;
 865                 pkt->hdr.ctl = ACK;
 866                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 867
 868                 if(fin_wanted(c, c->snd.una + len)) {
 869                         len--;
 870                         pkt->hdr.ctl |= FIN;
 871                 }
 872
 873                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 874                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 875                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 876                 break;
 877
 878         default:
 879                 break;
 880         }
 881 }
 882
 883 static void retransmit(struct utcp_connection *c) {
 884         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 885                 debug(c, "retransmit() called but nothing to retransmit!\n");
 886                 stop_retransmit_timer(c);
 887                 return;
 888         }
 889
 890         struct utcp *utcp = c->utcp;
 891
 892         if(utcp->retransmit) {
 893                 utcp->retransmit(c);
 894         }
 895
 896         struct {
 897                 struct hdr hdr;
 898                 uint8_t data[];
 899         } *pkt = c->utcp->pkt;
 900
 901         pkt->hdr.src = c->src;
 902         pkt->hdr.dst = c->dst;
 903         pkt->hdr.wnd = c->rcvbuf.maxsize;
 904         pkt->hdr.aux = 0;
 905
 906         switch(c->state) {
 907         case SYN_SENT:
 908                 // Send our SYN again
 909                 pkt->hdr.seq = c->snd.iss;
 910                 pkt->hdr.ack = 0;
 911                 pkt->hdr.ctl = SYN;
 912                 pkt->hdr.aux = 0x0101;
 913                 pkt->data[0] = 1;
 914                 pkt->data[1] = 0;
 915                 pkt->data[2] = 0;
 916                 pkt->data[3] = c->flags & 0x7;
 917                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 918                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 919                 break;
 920
 921         case SYN_RECEIVED:
 922                 // Send SYNACK again
 923                 pkt->hdr.seq = c->snd.nxt;
 924                 pkt->hdr.ack = c->rcv.nxt;
 925                 pkt->hdr.ctl = SYN | ACK;
 926                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
 927                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 928                 break;
 929
 930         case ESTABLISHED:
 931         case FIN_WAIT_1:
 932         case CLOSE_WAIT:
 933         case CLOSING:
 934         case LAST_ACK:
 935                 // Send unacked data again.
 936                 pkt->hdr.seq = c->snd.una;
 937                 pkt->hdr.ack = c->rcv.nxt;
 938                 pkt->hdr.ctl = ACK;
 939                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 940
 941                 if(fin_wanted(c, c->snd.una + len)) {
 942                         len--;
 943                         pkt->hdr.ctl |= FIN;
 944                 }
 945
 946                 // RFC 5681 slow start after timeout
 947                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
 948                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
 949                 c->snd.cwnd = utcp->mss;
 950                 debug_cwnd(c);
 951
 952                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 953                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 954                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 955
 956                 c->snd.nxt = c->snd.una + len;
 957                 break;
 958
 959         case CLOSED:
 960         case LISTEN:
 961         case TIME_WAIT:
 962         case FIN_WAIT_2:
 963                 // We shouldn't need to retransmit anything in this state.
 964 #ifdef UTCP_DEBUG
 965                 abort();
 966 #endif
 967                 stop_retransmit_timer(c);
 968                 goto cleanup;
 969         }
 970
 971         start_retransmit_timer(c);
 972         c->rto *= 2;
 973
 974         if(c->rto > MAX_RTO) {
 975                 c->rto = MAX_RTO;
 976         }
 977
 978         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 979         c->dupack = 0; // cancel any ongoing fast recovery
 980
 981 cleanup:
 982         return;
 983 }
 984
 985 /* Update receive buffer and SACK entries after consuming data.
 986  *
 987  * Situation:
 988  *
 989  * |.....0000..1111111111.....22222......3333|
 990  * |---------------^
 991  *
 992  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 993  * to remove data from the receive buffer. The idea is to substract "len"
 994  * from the offset of all the SACK entries, and then remove/cut down entries
 995  * that are shifted to before the start of the receive buffer.
 996  *
 997  * There are three cases:
 998  * - the SACK entry is after ^, in that case just change the offset.
 999  * - the SACK entry starts before and ends after ^, so we have to
1000  *   change both its offset and size.
1001  * - the SACK entry is completely before ^, in that case delete it.
1002  */
1003 static void sack_consume(struct utcp_connection *c, size_t len) {
1004         debug(c, "sack_consume %lu\n", (unsigned long)len);
1005
1006         if(len > c->rcvbuf.used) {
1007                 debug(c, "all SACK entries consumed\n");
1008                 c->sacks[0].len = 0;
1009                 return;
1010         }
1011
1012         buffer_discard(&c->rcvbuf, len);
1013
1014         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1015                 if(len < c->sacks[i].offset) {
1016                         c->sacks[i].offset -= len;
1017                         i++;
1018                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1019                         c->sacks[i].len -= len - c->sacks[i].offset;
1020                         c->sacks[i].offset = 0;
1021                         i++;
1022                 } else {
1023                         if(i < NSACKS - 1) {
1024                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1025                                 c->sacks[NSACKS - 1].len = 0;
1026                         } else {
1027                                 c->sacks[i].len = 0;
1028                                 break;
1029                         }
1030                 }
1031         }
1032
1033         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1034                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1035         }
1036 }
1037
1038 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1039         debug(c, "out of order packet, offset %u\n", offset);
1040         // Packet loss or reordering occured. Store the data in the buffer.
1041         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1042
1043         if(rxd <= 0) {
1044                 debug(c, "packet outside receive buffer, dropping\n");
1045                 return;
1046         }
1047
1048         if((size_t)rxd < len) {
1049                 debug(c, "packet partially outside receive buffer\n");
1050                 len = rxd;
1051         }
1052
1053         // Make note of where we put it.
1054         for(int i = 0; i < NSACKS; i++) {
1055                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1056                         debug(c, "new SACK entry %d\n", i);
1057                         c->sacks[i].offset = offset;
1058                         c->sacks[i].len = rxd;
1059                         break;
1060                 } else if(offset < c->sacks[i].offset) {
1061                         if(offset + rxd < c->sacks[i].offset) { // insert before
1062                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1063                                         debug(c, "insert SACK entry at %d\n", i);
1064                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1065                                         c->sacks[i].offset = offset;
1066                                         c->sacks[i].len = rxd;
1067                                 } else {
1068                                         debug(c, "SACK entries full, dropping packet\n");
1069                                 }
1070
1071                                 break;
1072                         } else { // merge
1073                                 debug(c, "merge with start of SACK entry at %d\n", i);
1074                                 c->sacks[i].offset = offset;
1075                                 break;
1076                         }
1077                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1078                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1079                                 debug(c, "merge with end of SACK entry at %d\n", i);
1080                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1081                                 // TODO: handle potential merge with next entry
1082                         }
1083
1084                         break;
1085                 }
1086         }
1087
1088         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1089                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1090         }
1091 }
1092
1093 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1094         if(c->recv) {
1095                 ssize_t rxd = c->recv(c, data, len);
1096
1097                 if(rxd != (ssize_t)len) {
1098                         // TODO: handle the application not accepting all data.
1099                         abort();
1100                 }
1101         }
1102
1103         // Check if we can process out-of-order data now.
1104         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1105                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1106
1107                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1108                         size_t offset = len;
1109                         len = c->sacks[0].offset + c->sacks[0].len;
1110                         size_t remainder = len - offset;
1111
1112                         ssize_t rxd = buffer_call(c, &c->rcvbuf, offset, remainder);
1113
1114                         if(rxd != (ssize_t)remainder) {
1115                                 // TODO: handle the application not accepting all data.
1116                                 abort();
1117                         }
1118                 }
1119         }
1120
1121         if(c->rcvbuf.used) {
1122                 sack_consume(c, len);
1123         }
1124
1125         c->rcv.nxt += len;
1126 }
1127
1128 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1129         // Fast path for unfragmented packets
1130         if(!hdr->wnd && !(hdr->ctl & MF)) {
1131                 if(c->recv) {
1132                         c->recv(c, data, len);
1133                 }
1134
1135                 c->rcv.nxt = hdr->seq + len;
1136                 return;
1137         }
1138
1139         // Ensure reassembled packet are not larger than 64 kiB
1140         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1141                 return;
1142         }
1143
1144         // Don't accept out of order fragments
1145         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1146                 return;
1147         }
1148
1149         // Reset the receive buffer for the first fragment
1150         if(!hdr->wnd) {
1151                 buffer_clear(&c->rcvbuf);
1152         }
1153
1154         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1155
1156         if(rxd != (ssize_t)len) {
1157                 return;
1158         }
1159
1160         // Send the packet if it's the final fragment
1161         if(!(hdr->ctl & MF)) {
1162                 buffer_call(c, &c->rcvbuf, 0, hdr->wnd + len);
1163         }
1164
1165         c->rcv.nxt = hdr->seq + len;
1166 }
1167
1168 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1169         if(!is_reliable(c)) {
1170                 handle_unreliable(c, hdr, data, len);
1171                 return;
1172         }
1173
1174         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1175
1176         if(offset) {
1177                 handle_out_of_order(c, offset, data, len);
1178         } else {
1179                 handle_in_order(c, data, len);
1180         }
1181 }
1182
1183
1184 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1185         const uint8_t *ptr = data;
1186
1187         if(!utcp) {
1188                 errno = EFAULT;
1189                 return -1;
1190         }
1191
1192         if(!len) {
1193                 return 0;
1194         }
1195
1196         if(!data) {
1197                 errno = EFAULT;
1198                 return -1;
1199         }
1200
1201         // Drop packets smaller than the header
1202
1203         struct hdr hdr;
1204
1205         if(len < sizeof(hdr)) {
1206                 print_packet(NULL, "recv", data, len);
1207                 errno = EBADMSG;
1208                 return -1;
1209         }
1210
1211         // Make a copy from the potentially unaligned data to a struct hdr
1212
1213         memcpy(&hdr, ptr, sizeof(hdr));
1214
1215         // Try to match the packet to an existing connection
1216
1217         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1218         print_packet(c, "recv", data, len);
1219
1220         // Process the header
1221
1222         ptr += sizeof(hdr);
1223         len -= sizeof(hdr);
1224
1225         // Drop packets with an unknown CTL flag
1226
1227         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1228                 print_packet(NULL, "recv", data, len);
1229                 errno = EBADMSG;
1230                 return -1;
1231         }
1232
1233         // Check for auxiliary headers
1234
1235         const uint8_t *init = NULL;
1236
1237         uint16_t aux = hdr.aux;
1238
1239         while(aux) {
1240                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1241                 uint8_t auxtype = aux & 0xff;
1242
1243                 if(len < auxlen) {
1244                         errno = EBADMSG;
1245                         return -1;
1246                 }
1247
1248                 switch(auxtype) {
1249                 case AUX_INIT:
1250                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1251                                 errno = EBADMSG;
1252                                 return -1;
1253                         }
1254
1255                         init = ptr;
1256                         break;
1257
1258                 default:
1259                         errno = EBADMSG;
1260                         return -1;
1261                 }
1262
1263                 len -= auxlen;
1264                 ptr += auxlen;
1265
1266                 if(!(aux & 0x800)) {
1267                         break;
1268                 }
1269
1270                 if(len < 2) {
1271                         errno = EBADMSG;
1272                         return -1;
1273                 }
1274
1275                 memcpy(&aux, ptr, 2);
1276                 len -= 2;
1277                 ptr += 2;
1278         }
1279
1280         bool has_data = len || (hdr.ctl & (SYN | FIN));
1281
1282         // Is it for a new connection?
1283
1284         if(!c) {
1285                 // Ignore RST packets
1286
1287                 if(hdr.ctl & RST) {
1288                         return 0;
1289                 }
1290
1291                 // Is it a SYN packet and are we LISTENing?
1292
1293                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1294                         // If we don't want to accept it, send a RST back
1295                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1296                                 len = 1;
1297                                 goto reset;
1298                         }
1299
1300                         // Try to allocate memory, otherwise send a RST back
1301                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1302
1303                         if(!c) {
1304                                 len = 1;
1305                                 goto reset;
1306                         }
1307
1308                         // Parse auxilliary information
1309                         if(init) {
1310                                 if(init[0] < 1) {
1311                                         len = 1;
1312                                         goto reset;
1313                                 }
1314
1315                                 c->flags = init[3] & 0x7;
1316                         } else {
1317                                 c->flags = UTCP_TCP;
1318                         }
1319
1320 synack:
1321                         // Return SYN+ACK, go to SYN_RECEIVED state
1322                         c->snd.wnd = hdr.wnd;
1323                         c->rcv.irs = hdr.seq;
1324                         c->rcv.nxt = c->rcv.irs + 1;
1325                         set_state(c, SYN_RECEIVED);
1326
1327                         struct {
1328                                 struct hdr hdr;
1329                                 uint8_t data[4];
1330                         } pkt;
1331
1332                         pkt.hdr.src = c->src;
1333                         pkt.hdr.dst = c->dst;
1334                         pkt.hdr.ack = c->rcv.irs + 1;
1335                         pkt.hdr.seq = c->snd.iss;
1336                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1337                         pkt.hdr.ctl = SYN | ACK;
1338
1339                         if(init) {
1340                                 pkt.hdr.aux = 0x0101;
1341                                 pkt.data[0] = 1;
1342                                 pkt.data[1] = 0;
1343                                 pkt.data[2] = 0;
1344                                 pkt.data[3] = c->flags & 0x7;
1345                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1346                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1347                         } else {
1348                                 pkt.hdr.aux = 0;
1349                                 print_packet(c, "send", &pkt, sizeof(hdr));
1350                                 utcp->send(utcp, &pkt, sizeof(hdr));
1351                         }
1352
1353                         start_retransmit_timer(c);
1354                 } else {
1355                         // No, we don't want your packets, send a RST back
1356                         len = 1;
1357                         goto reset;
1358                 }
1359
1360                 return 0;
1361         }
1362
1363         debug(c, "state %s\n", strstate[c->state]);
1364
1365         // In case this is for a CLOSED connection, ignore the packet.
1366         // TODO: make it so incoming packets can never match a CLOSED connection.
1367
1368         if(c->state == CLOSED) {
1369                 debug(c, "got packet for closed connection\n");
1370                 return 0;
1371         }
1372
1373         // It is for an existing connection.
1374
1375         // 1. Drop invalid packets.
1376
1377         // 1a. Drop packets that should not happen in our current state.
1378
1379         switch(c->state) {
1380         case SYN_SENT:
1381         case SYN_RECEIVED:
1382         case ESTABLISHED:
1383         case FIN_WAIT_1:
1384         case FIN_WAIT_2:
1385         case CLOSE_WAIT:
1386         case CLOSING:
1387         case LAST_ACK:
1388         case TIME_WAIT:
1389                 break;
1390
1391         default:
1392 #ifdef UTCP_DEBUG
1393                 abort();
1394 #endif
1395                 break;
1396         }
1397
1398         // 1b. Discard data that is not in our receive window.
1399
1400         if(is_reliable(c)) {
1401                 bool acceptable;
1402
1403                 if(c->state == SYN_SENT) {
1404                         acceptable = true;
1405                 } else if(len == 0) {
1406                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1407                 } else {
1408                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1409
1410                         // cut already accepted front overlapping
1411                         if(rcv_offset < 0) {
1412                                 acceptable = len > (size_t) - rcv_offset;
1413
1414                                 if(acceptable) {
1415                                         ptr -= rcv_offset;
1416                                         len += rcv_offset;
1417                                         hdr.seq -= rcv_offset;
1418                                 }
1419                         } else {
1420                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1421                         }
1422                 }
1423
1424                 if(!acceptable) {
1425                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1426
1427                         // Ignore unacceptable RST packets.
1428                         if(hdr.ctl & RST) {
1429                                 return 0;
1430                         }
1431
1432                         // Otherwise, continue processing.
1433                         len = 0;
1434                 }
1435         } else {
1436 #if UTCP_DEBUG
1437                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1438
1439                 if(rcv_offset) {
1440                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1441                 }
1442
1443 #endif
1444         }
1445
1446         c->snd.wnd = hdr.wnd; // TODO: move below
1447
1448         // 1c. Drop packets with an invalid ACK.
1449         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1450         // (= snd.una + c->sndbuf.used).
1451
1452         if(!is_reliable(c)) {
1453                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1454                         hdr.ack = c->snd.una;
1455                 }
1456         }
1457
1458         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1459                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1460
1461                 // Ignore unacceptable RST packets.
1462                 if(hdr.ctl & RST) {
1463                         return 0;
1464                 }
1465
1466                 goto reset;
1467         }
1468
1469         // 2. Handle RST packets
1470
1471         if(hdr.ctl & RST) {
1472                 switch(c->state) {
1473                 case SYN_SENT:
1474                         if(!(hdr.ctl & ACK)) {
1475                                 return 0;
1476                         }
1477
1478                         // The peer has refused our connection.
1479                         set_state(c, CLOSED);
1480                         errno = ECONNREFUSED;
1481
1482                         if(c->recv) {
1483                                 c->recv(c, NULL, 0);
1484                         }
1485
1486                         if(c->poll && !c->reapable) {
1487                                 c->poll(c, 0);
1488                         }
1489
1490                         return 0;
1491
1492                 case SYN_RECEIVED:
1493                         if(hdr.ctl & ACK) {
1494                                 return 0;
1495                         }
1496
1497                         // We haven't told the application about this connection yet. Silently delete.
1498                         free_connection(c);
1499                         return 0;
1500
1501                 case ESTABLISHED:
1502                 case FIN_WAIT_1:
1503                 case FIN_WAIT_2:
1504                 case CLOSE_WAIT:
1505                         if(hdr.ctl & ACK) {
1506                                 return 0;
1507                         }
1508
1509                         // The peer has aborted our connection.
1510                         set_state(c, CLOSED);
1511                         errno = ECONNRESET;
1512
1513                         if(c->recv) {
1514                                 c->recv(c, NULL, 0);
1515                         }
1516
1517                         if(c->poll && !c->reapable) {
1518                                 c->poll(c, 0);
1519                         }
1520
1521                         return 0;
1522
1523                 case CLOSING:
1524                 case LAST_ACK:
1525                 case TIME_WAIT:
1526                         if(hdr.ctl & ACK) {
1527                                 return 0;
1528                         }
1529
1530                         // As far as the application is concerned, the connection has already been closed.
1531                         // If it has called utcp_close() already, we can immediately free this connection.
1532                         if(c->reapable) {
1533                                 free_connection(c);
1534                                 return 0;
1535                         }
1536
1537                         // Otherwise, immediately move to the CLOSED state.
1538                         set_state(c, CLOSED);
1539                         return 0;
1540
1541                 default:
1542 #ifdef UTCP_DEBUG
1543                         abort();
1544 #endif
1545                         break;
1546                 }
1547         }
1548
1549         uint32_t advanced;
1550
1551         if(!(hdr.ctl & ACK)) {
1552                 advanced = 0;
1553                 goto skip_ack;
1554         }
1555
1556         // 3. Advance snd.una
1557
1558         advanced = seqdiff(hdr.ack, c->snd.una);
1559
1560         if(advanced) {
1561                 // RTT measurement
1562                 if(c->rtt_start.tv_sec) {
1563                         if(c->rtt_seq == hdr.ack) {
1564                                 struct timespec now;
1565                                 clock_gettime(UTCP_CLOCK, &now);
1566                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1567                                 update_rtt(c, diff);
1568                                 c->rtt_start.tv_sec = 0;
1569                         } else if(c->rtt_seq < hdr.ack) {
1570                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1571                                 c->rtt_start.tv_sec = 0;
1572                         }
1573                 }
1574
1575                 int32_t data_acked = advanced;
1576
1577                 switch(c->state) {
1578                 case SYN_SENT:
1579                 case SYN_RECEIVED:
1580                         data_acked--;
1581                         break;
1582
1583                 // TODO: handle FIN as well.
1584                 default:
1585                         break;
1586                 }
1587
1588                 assert(data_acked >= 0);
1589
1590 #ifndef NDEBUG
1591                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1592                 assert(data_acked <= bufused);
1593 #endif
1594
1595                 if(data_acked) {
1596                         buffer_discard(&c->sndbuf, data_acked);
1597
1598                         if(is_reliable(c)) {
1599                                 c->do_poll = true;
1600                         }
1601                 }
1602
1603                 // Also advance snd.nxt if possible
1604                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1605                         c->snd.nxt = hdr.ack;
1606                 }
1607
1608                 c->snd.una = hdr.ack;
1609
1610                 if(c->dupack) {
1611                         if(c->dupack >= 3) {
1612                                 debug(c, "fast recovery ended\n");
1613                                 c->snd.cwnd = c->snd.ssthresh;
1614                         }
1615
1616                         c->dupack = 0;
1617                 }
1618
1619                 // Increase the congestion window according to RFC 5681
1620                 if(c->snd.cwnd < c->snd.ssthresh) {
1621                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1622                 } else {
1623                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1624                 }
1625
1626                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1627                         c->snd.cwnd = c->sndbuf.maxsize;
1628                 }
1629
1630                 debug_cwnd(c);
1631
1632                 // Check if we have sent a FIN that is now ACKed.
1633                 switch(c->state) {
1634                 case FIN_WAIT_1:
1635                         if(c->snd.una == c->snd.last) {
1636                                 set_state(c, FIN_WAIT_2);
1637                         }
1638
1639                         break;
1640
1641                 case CLOSING:
1642                         if(c->snd.una == c->snd.last) {
1643                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1644                                 c->conn_timeout.tv_sec += utcp->timeout;
1645                                 set_state(c, TIME_WAIT);
1646                         }
1647
1648                         break;
1649
1650                 default:
1651                         break;
1652                 }
1653         } else {
1654                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1655                         c->dupack++;
1656                         debug(c, "duplicate ACK %d\n", c->dupack);
1657
1658                         if(c->dupack == 3) {
1659                                 // RFC 5681 fast recovery
1660                                 debug(c, "fast recovery started\n", c->dupack);
1661                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1662                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1663                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1664
1665                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1666                                         c->snd.cwnd = c->sndbuf.maxsize;
1667                                 }
1668
1669                                 debug_cwnd(c);
1670
1671                                 fast_retransmit(c);
1672                         } else if(c->dupack > 3) {
1673                                 c->snd.cwnd += utcp->mss;
1674
1675                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1676                                         c->snd.cwnd = c->sndbuf.maxsize;
1677                                 }
1678
1679                                 debug_cwnd(c);
1680                         }
1681
1682                         // We got an ACK which indicates the other side did get one of our packets.
1683                         // Reset the retransmission timer to avoid going to slow start,
1684                         // but don't touch the connection timeout.
1685                         start_retransmit_timer(c);
1686                 }
1687         }
1688
1689         // 4. Update timers
1690
1691         if(advanced) {
1692                 if(c->snd.una == c->snd.last) {
1693                         stop_retransmit_timer(c);
1694                         timespec_clear(&c->conn_timeout);
1695                 } else if(is_reliable(c)) {
1696                         start_retransmit_timer(c);
1697                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1698                         c->conn_timeout.tv_sec += utcp->timeout;
1699                 }
1700         }
1701
1702 skip_ack:
1703         // 5. Process SYN stuff
1704
1705         if(hdr.ctl & SYN) {
1706                 switch(c->state) {
1707                 case SYN_SENT:
1708
1709                         // This is a SYNACK. It should always have ACKed the SYN.
1710                         if(!advanced) {
1711                                 goto reset;
1712                         }
1713
1714                         c->rcv.irs = hdr.seq;
1715                         c->rcv.nxt = hdr.seq + 1;
1716
1717                         if(c->shut_wr) {
1718                                 c->snd.last++;
1719                                 set_state(c, FIN_WAIT_1);
1720                         } else {
1721                                 set_state(c, ESTABLISHED);
1722                         }
1723
1724                         break;
1725
1726                 case SYN_RECEIVED:
1727                         // This is a retransmit of a SYN, send back the SYNACK.
1728                         goto synack;
1729
1730                 case ESTABLISHED:
1731                 case FIN_WAIT_1:
1732                 case FIN_WAIT_2:
1733                 case CLOSE_WAIT:
1734                 case CLOSING:
1735                 case LAST_ACK:
1736                 case TIME_WAIT:
1737                         // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
1738                         break;
1739
1740                 default:
1741 #ifdef UTCP_DEBUG
1742                         abort();
1743 #endif
1744                         return 0;
1745                 }
1746         }
1747
1748         // 6. Process new data
1749
1750         if(c->state == SYN_RECEIVED) {
1751                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1752                 if(!advanced) {
1753                         goto reset;
1754                 }
1755
1756                 // Are we still LISTENing?
1757                 if(utcp->accept) {
1758                         utcp->accept(c, c->src);
1759                 }
1760
1761                 if(c->state != ESTABLISHED) {
1762                         set_state(c, CLOSED);
1763                         c->reapable = true;
1764                         goto reset;
1765                 }
1766         }
1767
1768         if(len) {
1769                 switch(c->state) {
1770                 case SYN_SENT:
1771                 case SYN_RECEIVED:
1772                         // This should never happen.
1773 #ifdef UTCP_DEBUG
1774                         abort();
1775 #endif
1776                         return 0;
1777
1778                 case ESTABLISHED:
1779                 case FIN_WAIT_1:
1780                 case FIN_WAIT_2:
1781                         break;
1782
1783                 case CLOSE_WAIT:
1784                 case CLOSING:
1785                 case LAST_ACK:
1786                 case TIME_WAIT:
1787                         // Ehm no, We should never receive more data after a FIN.
1788                         goto reset;
1789
1790                 default:
1791 #ifdef UTCP_DEBUG
1792                         abort();
1793 #endif
1794                         return 0;
1795                 }
1796
1797                 handle_incoming_data(c, &hdr, ptr, len);
1798         }
1799
1800         // 7. Process FIN stuff
1801
1802         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1803                 switch(c->state) {
1804                 case SYN_SENT:
1805                 case SYN_RECEIVED:
1806                         // This should never happen.
1807 #ifdef UTCP_DEBUG
1808                         abort();
1809 #endif
1810                         break;
1811
1812                 case ESTABLISHED:
1813                         set_state(c, CLOSE_WAIT);
1814                         break;
1815
1816                 case FIN_WAIT_1:
1817                         set_state(c, CLOSING);
1818                         break;
1819
1820                 case FIN_WAIT_2:
1821                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1822                         c->conn_timeout.tv_sec += utcp->timeout;
1823                         set_state(c, TIME_WAIT);
1824                         break;
1825
1826                 case CLOSE_WAIT:
1827                 case CLOSING:
1828                 case LAST_ACK:
1829                 case TIME_WAIT:
1830                         // Ehm, no. We should never receive a second FIN.
1831                         goto reset;
1832
1833                 default:
1834 #ifdef UTCP_DEBUG
1835                         abort();
1836 #endif
1837                         break;
1838                 }
1839
1840                 // FIN counts as one sequence number
1841                 c->rcv.nxt++;
1842                 len++;
1843
1844                 // Inform the application that the peer closed its end of the connection.
1845                 if(c->recv) {
1846                         errno = 0;
1847                         c->recv(c, NULL, 0);
1848                 }
1849         }
1850
1851         // Now we send something back if:
1852         // - we received data, so we have to send back an ACK
1853         //   -> sendatleastone = true
1854         // - or we got an ack, so we should maybe send a bit more data
1855         //   -> sendatleastone = false
1856
1857         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1858                 ack(c, has_data);
1859         }
1860
1861         return 0;
1862
1863 reset:
1864         swap_ports(&hdr);
1865         hdr.wnd = 0;
1866         hdr.aux = 0;
1867
1868         if(hdr.ctl & ACK) {
1869                 hdr.seq = hdr.ack;
1870                 hdr.ctl = RST;
1871         } else {
1872                 hdr.ack = hdr.seq + len;
1873                 hdr.seq = 0;
1874                 hdr.ctl = RST | ACK;
1875         }
1876
1877         print_packet(c, "send", &hdr, sizeof(hdr));
1878         utcp->send(utcp, &hdr, sizeof(hdr));
1879         return 0;
1880
1881 }
1882
1883 int utcp_shutdown(struct utcp_connection *c, int dir) {
1884         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1885
1886         if(!c) {
1887                 errno = EFAULT;
1888                 return -1;
1889         }
1890
1891         if(c->reapable) {
1892                 debug(c, "shutdown() called on closed connection\n");
1893                 errno = EBADF;
1894                 return -1;
1895         }
1896
1897         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1898                 errno = EINVAL;
1899                 return -1;
1900         }
1901
1902         // TCP does not have a provision for stopping incoming packets.
1903         // The best we can do is to just ignore them.
1904         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1905                 c->recv = NULL;
1906         }
1907
1908         // The rest of the code deals with shutting down writes.
1909         if(dir == UTCP_SHUT_RD) {
1910                 return 0;
1911         }
1912
1913         // Only process shutting down writes once.
1914         if(c->shut_wr) {
1915                 return 0;
1916         }
1917
1918         c->shut_wr = true;
1919
1920         switch(c->state) {
1921         case CLOSED:
1922         case LISTEN:
1923                 errno = ENOTCONN;
1924                 return -1;
1925
1926         case SYN_SENT:
1927                 return 0;
1928
1929         case SYN_RECEIVED:
1930         case ESTABLISHED:
1931                 set_state(c, FIN_WAIT_1);
1932                 break;
1933
1934         case FIN_WAIT_1:
1935         case FIN_WAIT_2:
1936                 return 0;
1937
1938         case CLOSE_WAIT:
1939                 set_state(c, CLOSING);
1940                 break;
1941
1942         case CLOSING:
1943         case LAST_ACK:
1944         case TIME_WAIT:
1945                 return 0;
1946         }
1947
1948         c->snd.last++;
1949
1950         ack(c, false);
1951
1952         if(!timespec_isset(&c->rtrx_timeout)) {
1953                 start_retransmit_timer(c);
1954         }
1955
1956         return 0;
1957 }
1958
1959 static bool reset_connection(struct utcp_connection *c) {
1960         if(!c) {
1961                 errno = EFAULT;
1962                 return false;
1963         }
1964
1965         if(c->reapable) {
1966                 debug(c, "abort() called on closed connection\n");
1967                 errno = EBADF;
1968                 return false;
1969         }
1970
1971         c->recv = NULL;
1972         c->poll = NULL;
1973
1974         switch(c->state) {
1975         case CLOSED:
1976                 return true;
1977
1978         case LISTEN:
1979         case SYN_SENT:
1980         case CLOSING:
1981         case LAST_ACK:
1982         case TIME_WAIT:
1983                 set_state(c, CLOSED);
1984                 return true;
1985
1986         case SYN_RECEIVED:
1987         case ESTABLISHED:
1988         case FIN_WAIT_1:
1989         case FIN_WAIT_2:
1990         case CLOSE_WAIT:
1991                 set_state(c, CLOSED);
1992                 break;
1993         }
1994
1995         // Send RST
1996
1997         struct hdr hdr;
1998
1999         hdr.src = c->src;
2000         hdr.dst = c->dst;
2001         hdr.seq = c->snd.nxt;
2002         hdr.ack = 0;
2003         hdr.wnd = 0;
2004         hdr.ctl = RST;
2005
2006         print_packet(c, "send", &hdr, sizeof(hdr));
2007         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2008         return true;
2009 }
2010
2011 // Closes all the opened connections
2012 void utcp_abort_all_connections(struct utcp *utcp) {
2013         if(!utcp) {
2014                 errno = EINVAL;
2015                 return;
2016         }
2017
2018         for(int i = 0; i < utcp->nconnections; i++) {
2019                 struct utcp_connection *c = utcp->connections[i];
2020
2021                 if(c->reapable || c->state == CLOSED) {
2022                         continue;
2023                 }
2024
2025                 utcp_recv_t old_recv = c->recv;
2026                 utcp_poll_t old_poll = c->poll;
2027
2028                 reset_connection(c);
2029
2030                 if(old_recv) {
2031                         errno = 0;
2032                         old_recv(c, NULL, 0);
2033                 }
2034
2035                 if(old_poll && !c->reapable) {
2036                         errno = 0;
2037                         old_poll(c, 0);
2038                 }
2039         }
2040
2041         return;
2042 }
2043
2044 int utcp_close(struct utcp_connection *c) {
2045         if(c->rcvbuf.used) {
2046                 fprintf(stderr, "UTCP channel closed with stuff in receive buffer\n");
2047                 return reset_connection(c) ? 0 : -1;
2048         }
2049
2050         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2051                 return -1;
2052         }
2053
2054         c->recv = NULL;
2055         c->poll = NULL;
2056         c->reapable = true;
2057         return 0;
2058 }
2059
2060 int utcp_abort(struct utcp_connection *c) {
2061         if(!reset_connection(c)) {
2062                 return -1;
2063         }
2064
2065         c->reapable = true;
2066         return 0;
2067 }
2068
2069 /* Handle timeouts.
2070  * One call to this function will loop through all connections,
2071  * checking if something needs to be resent or not.
2072  * The return value is the time to the next timeout in milliseconds,
2073  * or maybe a negative value if the timeout is infinite.
2074  */
2075 struct timespec utcp_timeout(struct utcp *utcp) {
2076         struct timespec now;
2077         clock_gettime(UTCP_CLOCK, &now);
2078         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2079
2080         for(int i = 0; i < utcp->nconnections; i++) {
2081                 struct utcp_connection *c = utcp->connections[i];
2082
2083                 if(!c) {
2084                         continue;
2085                 }
2086
2087                 // delete connections that have been utcp_close()d.
2088                 if(c->state == CLOSED) {
2089                         if(c->reapable) {
2090                                 debug(c, "reaping\n");
2091                                 free_connection(c);
2092                                 i--;
2093                         }
2094
2095                         continue;
2096                 }
2097
2098                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2099                         errno = ETIMEDOUT;
2100                         c->state = CLOSED;
2101
2102                         if(c->recv) {
2103                                 c->recv(c, NULL, 0);
2104                         }
2105
2106                         if(c->poll && !c->reapable) {
2107                                 c->poll(c, 0);
2108                         }
2109
2110                         continue;
2111                 }
2112
2113                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2114                         debug(c, "retransmitting after timeout\n");
2115                         retransmit(c);
2116                 }
2117
2118                 if(c->poll) {
2119                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2120                                 c->do_poll = false;
2121                                 uint32_t len = buffer_free(&c->sndbuf);
2122
2123                                 if(len) {
2124                                         c->poll(c, len);
2125                                 }
2126                         } else if(c->state == CLOSED) {
2127                                 c->poll(c, 0);
2128                         }
2129                 }
2130
2131                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2132                         next = c->conn_timeout;
2133                 }
2134
2135                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2136                         next = c->rtrx_timeout;
2137                 }
2138         }
2139
2140         struct timespec diff;
2141
2142         timespec_sub(&next, &now, &diff);
2143
2144         return diff;
2145 }
2146
2147 bool utcp_is_active(struct utcp *utcp) {
2148         if(!utcp) {
2149                 return false;
2150         }
2151
2152         for(int i = 0; i < utcp->nconnections; i++)
2153                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2154                         return true;
2155                 }
2156
2157         return false;
2158 }
2159
2160 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2161         if(!send) {
2162                 errno = EFAULT;
2163                 return NULL;
2164         }
2165
2166         struct utcp *utcp = calloc(1, sizeof(*utcp));
2167
2168         if(!utcp) {
2169                 return NULL;
2170         }
2171
2172         utcp_set_mtu(utcp, DEFAULT_MTU);
2173
2174         if(!utcp->pkt) {
2175                 free(utcp);
2176                 return NULL;
2177         }
2178
2179         if(!CLOCK_GRANULARITY) {
2180                 struct timespec res;
2181                 clock_getres(UTCP_CLOCK, &res);
2182                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2183         }
2184
2185         utcp->accept = accept;
2186         utcp->pre_accept = pre_accept;
2187         utcp->send = send;
2188         utcp->priv = priv;
2189         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2190
2191         return utcp;
2192 }
2193
2194 void utcp_exit(struct utcp *utcp) {
2195         if(!utcp) {
2196                 return;
2197         }
2198
2199         for(int i = 0; i < utcp->nconnections; i++) {
2200                 struct utcp_connection *c = utcp->connections[i];
2201
2202                 if(!c->reapable) {
2203                         if(c->recv) {
2204                                 c->recv(c, NULL, 0);
2205                         }
2206
2207                         if(c->poll && !c->reapable) {
2208                                 c->poll(c, 0);
2209                         }
2210                 }
2211
2212                 buffer_exit(&c->rcvbuf);
2213                 buffer_exit(&c->sndbuf);
2214                 free(c);
2215         }
2216
2217         free(utcp->connections);
2218         free(utcp->pkt);
2219         free(utcp);
2220 }
2221
2222 uint16_t utcp_get_mtu(struct utcp *utcp) {
2223         return utcp ? utcp->mtu : 0;
2224 }
2225
2226 uint16_t utcp_get_mss(struct utcp *utcp) {
2227         return utcp ? utcp->mss : 0;
2228 }
2229
2230 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2231         if(!utcp) {
2232                 return;
2233         }
2234
2235         if(mtu <= sizeof(struct hdr)) {
2236                 return;
2237         }
2238
2239         if(mtu > utcp->mtu) {
2240                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2241
2242                 if(!new) {
2243                         return;
2244                 }
2245
2246                 utcp->pkt = new;
2247         }
2248
2249         utcp->mtu = mtu;
2250         utcp->mss = mtu - sizeof(struct hdr);
2251 }
2252
2253 void utcp_reset_timers(struct utcp *utcp) {
2254         if(!utcp) {
2255                 return;
2256         }
2257
2258         struct timespec now, then;
2259
2260         clock_gettime(UTCP_CLOCK, &now);
2261
2262         then = now;
2263
2264         then.tv_sec += utcp->timeout;
2265
2266         for(int i = 0; i < utcp->nconnections; i++) {
2267                 struct utcp_connection *c = utcp->connections[i];
2268
2269                 if(c->reapable) {
2270                         continue;
2271                 }
2272
2273                 if(timespec_isset(&c->rtrx_timeout)) {
2274                         c->rtrx_timeout = now;
2275                 }
2276
2277                 if(timespec_isset(&c->conn_timeout)) {
2278                         c->conn_timeout = then;
2279                 }
2280
2281                 c->rtt_start.tv_sec = 0;
2282
2283                 if(c->rto > START_RTO) {
2284                         c->rto = START_RTO;
2285                 }
2286         }
2287 }
2288
2289 int utcp_get_user_timeout(struct utcp *u) {
2290         return u ? u->timeout : 0;
2291 }
2292
2293 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2294         if(u) {
2295                 u->timeout = timeout;
2296         }
2297 }
2298
2299 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2300         return c ? c->sndbuf.maxsize : 0;
2301 }
2302
2303 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2304         if(!c) {
2305                 return 0;
2306         }
2307
2308         switch(c->state) {
2309         case SYN_SENT:
2310         case SYN_RECEIVED:
2311         case ESTABLISHED:
2312         case CLOSE_WAIT:
2313                 return buffer_free(&c->sndbuf);
2314
2315         default:
2316                 return 0;
2317         }
2318 }
2319
2320 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2321         if(!c) {
2322                 return;
2323         }
2324
2325         c->sndbuf.maxsize = size;
2326
2327         if(c->sndbuf.maxsize != size) {
2328                 c->sndbuf.maxsize = -1;
2329         }
2330
2331         c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2332 }
2333
2334 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2335         return c ? c->rcvbuf.maxsize : 0;
2336 }
2337
2338 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2339         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2340                 return buffer_free(&c->rcvbuf);
2341         } else {
2342                 return 0;
2343         }
2344 }
2345
2346 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2347         if(!c) {
2348                 return;
2349         }
2350
2351         c->rcvbuf.maxsize = size;
2352
2353         if(c->rcvbuf.maxsize != size) {
2354                 c->rcvbuf.maxsize = -1;
2355         }
2356 }
2357
2358 size_t utcp_get_sendq(struct utcp_connection *c) {
2359         return c->sndbuf.used;
2360 }
2361
2362 size_t utcp_get_recvq(struct utcp_connection *c) {
2363         return c->rcvbuf.used;
2364 }
2365
2366 bool utcp_get_nodelay(struct utcp_connection *c) {
2367         return c ? c->nodelay : false;
2368 }
2369
2370 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2371         if(c) {
2372                 c->nodelay = nodelay;
2373         }
2374 }
2375
2376 bool utcp_get_keepalive(struct utcp_connection *c) {
2377         return c ? c->keepalive : false;
2378 }
2379
2380 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2381         if(c) {
2382                 c->keepalive = keepalive;
2383         }
2384 }
2385
2386 size_t utcp_get_outq(struct utcp_connection *c) {
2387         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2388 }
2389
2390 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2391         if(c) {
2392                 c->recv = recv;
2393         }
2394 }
2395
2396 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2397         if(c) {
2398                 c->poll = poll;
2399                 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2400         }
2401 }
2402
2403 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2404         if(utcp) {
2405                 utcp->accept = accept;
2406                 utcp->pre_accept = pre_accept;
2407         }
2408 }
2409
2410 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2411         if(!c || c->reapable) {
2412                 return;
2413         }
2414
2415         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2416                 return;
2417         }
2418
2419         if(expect) {
2420                 // If we expect data, start the connection timer.
2421                 if(!timespec_isset(&c->conn_timeout)) {
2422                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2423                         c->conn_timeout.tv_sec += c->utcp->timeout;
2424                 }
2425         } else {
2426                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2427                 if(c->snd.una == c->snd.last) {
2428                         timespec_clear(&c->conn_timeout);
2429                 }
2430         }
2431 }
2432
2433 void utcp_offline(struct utcp *utcp, bool offline) {
2434         struct timespec now;
2435         clock_gettime(UTCP_CLOCK, &now);
2436
2437         for(int i = 0; i < utcp->nconnections; i++) {
2438                 struct utcp_connection *c = utcp->connections[i];
2439
2440                 if(c->reapable) {
2441                         continue;
2442                 }
2443
2444                 utcp_expect_data(c, offline);
2445
2446                 if(!offline) {
2447                         if(timespec_isset(&c->rtrx_timeout)) {
2448                                 c->rtrx_timeout = now;
2449                         }
2450
2451                         utcp->connections[i]->rtt_start.tv_sec = 0;
2452
2453                         if(c->rto > START_RTO) {
2454                                 c->rto = START_RTO;
2455                         }
2456                 }
2457         }
2458 }
2459
2460 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t cb) {
2461         utcp->retransmit = cb;
2462 }
2463
2464 void utcp_set_clock_granularity(long granularity) {
2465         CLOCK_GRANULARITY = granularity;
2466 }