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