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         assert(!buf->external);
 200
 201         if(!newsize) {
 202                 free(buf->data);
 203                 buf->data = NULL;
 204                 buf->size = 0;
 205                 buf->offset = 0;
 206                 return true;
 207         }
 208
 209         char *newdata = realloc(buf->data, newsize);
 210
 211         if(!newdata) {
 212                 return false;
 213         }
 214
 215         buf->data = newdata;
 216
 217         if(buffer_wraps(buf)) {
 218                 // Shift the right part of the buffer until it hits the end of the new buffer.
 219                 // Old situation:
 220                 // [345......012]
 221                 // New situation:
 222                 // [345.........|........012]
 223                 uint32_t tailsize = buf->size - buf->offset;
 224                 uint32_t newoffset = newsize - tailsize;
 225                 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
 226                 buf->offset = newoffset;
 227         }
 228
 229         buf->size = newsize;
 230         return true;
 231 }
 232
 233 // Store data into the buffer
 234 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 235         debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 236
 237         // Ensure we don't store more than maxsize bytes in total
 238         size_t required = offset + len;
 239
 240         if(required > buf->maxsize) {
 241                 if(offset >= buf->maxsize) {
 242                         return 0;
 243                 }
 244
 245                 len = buf->maxsize - offset;
 246                 required = buf->maxsize;
 247         }
 248
 249         // Check if we need to resize the buffer
 250         if(required > buf->size) {
 251                 size_t newsize = buf->size;
 252
 253                 if(!newsize) {
 254                         newsize = 4096;
 255                 }
 256
 257                 do {
 258                         newsize *= 2;
 259                 } while(newsize < required);
 260
 261                 if(newsize > buf->maxsize) {
 262                         newsize = buf->maxsize;
 263                 }
 264
 265                 if(!buffer_resize(buf, newsize)) {
 266                         return -1;
 267                 }
 268         }
 269
 270         uint32_t realoffset = buf->offset + offset;
 271
 272         if(buf->size - buf->offset <= offset) {
 273                 // The offset wrapped
 274                 realoffset -= buf->size;
 275         }
 276
 277         if(buf->size - realoffset < len) {
 278                 // The new chunk of data must be wrapped
 279                 memcpy(buf->data + realoffset, data, buf->size - realoffset);
 280                 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
 281         } else {
 282                 memcpy(buf->data + realoffset, data, len);
 283         }
 284
 285         if(required > buf->used) {
 286                 buf->used = required;
 287         }
 288
 289         return len;
 290 }
 291
 292 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 293         return buffer_put_at(buf, buf->used, data, len);
 294 }
 295
 296 // Copy data from the buffer without removing it.
 297 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 298         // Ensure we don't copy more than is actually stored in the buffer
 299         if(offset >= buf->used) {
 300                 return 0;
 301         }
 302
 303         if(buf->used - offset < len) {
 304                 len = buf->used - offset;
 305         }
 306
 307         uint32_t realoffset = buf->offset + offset;
 308
 309         if(buf->size - buf->offset <= offset) {
 310                 // The offset wrapped
 311                 realoffset -= buf->size;
 312         }
 313
 314         if(buf->size - realoffset < len) {
 315                 // The data is wrapped
 316                 memcpy(data, buf->data + realoffset, buf->size - realoffset);
 317                 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
 318         } else {
 319                 memcpy(data, buf->data + realoffset, len);
 320         }
 321
 322         return len;
 323 }
 324
 325 // Copy data from the buffer without removing it.
 326 static ssize_t buffer_call(struct utcp_connection *c, struct buffer *buf, size_t offset, size_t len) {
 327         if(!c->recv) {
 328                 return len;
 329         }
 330
 331         // Ensure we don't copy more than is actually stored in the buffer
 332         if(offset >= buf->used) {
 333                 return 0;
 334         }
 335
 336         if(buf->used - offset < len) {
 337                 len = buf->used - offset;
 338         }
 339
 340         uint32_t realoffset = buf->offset + offset;
 341
 342         if(buf->size - buf->offset <= offset) {
 343                 // The offset wrapped
 344                 realoffset -= buf->size;
 345         }
 346
 347         if(buf->size - realoffset < len) {
 348                 // The data is wrapped
 349                 ssize_t rx1 = c->recv(c, buf->data + realoffset, buf->size - realoffset);
 350
 351                 if(rx1 < buf->size - realoffset) {
 352                         return rx1;
 353                 }
 354
 355                 // The channel might have been closed by the previous callback
 356                 if(!c->recv) {
 357                         return len;
 358                 }
 359
 360                 ssize_t rx2 = c->recv(c, buf->data, len - (buf->size - realoffset));
 361
 362                 if(rx2 < 0) {
 363                         return rx2;
 364                 } else {
 365                         return rx1 + rx2;
 366                 }
 367         } else {
 368                 return c->recv(c, buf->data + realoffset, len);
 369         }
 370 }
 371
 372 // Discard data from the buffer.
 373 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
 374         if(buf->used < len) {
 375                 len = buf->used;
 376         }
 377
 378         if(buf->size - buf->offset <= len) {
 379                 buf->offset -= buf->size;
 380         }
 381
 382         if(buf->used == len) {
 383                 buf->offset = 0;
 384         } else {
 385                 buf->offset += len;
 386         }
 387
 388         buf->used -= len;
 389
 390         return len;
 391 }
 392
 393 static void buffer_clear(struct buffer *buf) {
 394         buf->used = 0;
 395         buf->offset = 0;
 396 }
 397
 398 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
 399         if(maxsize < minsize) {
 400                 maxsize = minsize;
 401         }
 402
 403         buf->maxsize = maxsize;
 404
 405         return buf->size >= minsize || buffer_resize(buf, minsize);
 406 }
 407
 408 static void buffer_transfer(struct buffer *buf, char *newdata, size_t newsize) {
 409         if(buffer_wraps(buf)) {
 410                 // Old situation:
 411                 // [345......012]
 412                 // New situation:
 413                 // [012345......]
 414                 uint32_t tailsize = buf->size - buf->offset;
 415                 memcpy(newdata, buf->data + buf->offset, tailsize);
 416                 memcpy(newdata + tailsize, buf->data, buf->used - tailsize);
 417         } else {
 418                 // Old situation:
 419                 // [....012345..]
 420                 // New situation:
 421                 // [012345......]
 422                 memcpy(newdata, buf->data + buf->offset, buf->used);
 423         }
 424
 425         buf->offset = 0;
 426         buf->size = newsize;
 427 }
 428
 429 static void set_buffer_storage(struct buffer *buf, char *data, size_t size) {
 430         if(size > UINT32_MAX) {
 431                 size = UINT32_MAX;
 432         }
 433
 434         buf->maxsize = size;
 435
 436         if(data) {
 437                 if(buf->external) {
 438                         // Don't allow resizing an external buffer
 439                         abort();
 440                 }
 441
 442                 if(size < buf->used) {
 443                         // Ignore requests for an external buffer if we are already using more than it can store
 444                         return;
 445                 }
 446
 447                 // Transition from internal to external buffer
 448                 buffer_transfer(buf, data, size);
 449                 free(buf->data);
 450                 buf->data = data;
 451                 buf->external = true;
 452         } else if(buf->external) {
 453                 // Transition from external to internal buf
 454                 size_t minsize = buf->used <= DEFAULT_SNDBUFSIZE ? DEFAULT_SNDBUFSIZE : buf->used;
 455
 456                 if(minsize) {
 457                         data = malloc(minsize);
 458
 459                         if(!data) {
 460                                 // Cannot handle this
 461                                 abort();
 462                         }
 463
 464                         buffer_transfer(buf, data, minsize);
 465                         buf->data = data;
 466                 } else {
 467                         buf->data = NULL;
 468                         buf->size = 0;
 469                 }
 470
 471                 buf->external = false;
 472         } else {
 473                 // Don't do anything if the buffer wraps
 474                 if(buffer_wraps(buf)) {
 475                         return;
 476                 }
 477
 478                 // Realloc internal storage
 479                 size_t minsize = max(DEFAULT_SNDBUFSIZE, buf->offset + buf->used);
 480
 481                 if(minsize) {
 482                         data = realloc(buf->data, minsize);
 483
 484                         if(data) {
 485                                 buf->data = data;
 486                                 buf->size = minsize;
 487                         }
 488                 } else {
 489                         free(buf->data);
 490                         buf->data = NULL;
 491                         buf->size = 0;
 492                 }
 493         }
 494 }
 495
 496 static void buffer_exit(struct buffer *buf) {
 497         if(!buf->external) {
 498                 free(buf->data);
 499         }
 500
 501         memset(buf, 0, sizeof(*buf));
 502 }
 503
 504 static uint32_t buffer_free(const struct buffer *buf) {
 505         return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
 506 }
 507
 508 // Connections are stored in a sorted list.
 509 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 510
 511 static int compare(const void *va, const void *vb) {
 512         assert(va && vb);
 513
 514         const struct utcp_connection *a = *(struct utcp_connection **)va;
 515         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 516
 517         assert(a && b);
 518
 519         int c = (int)a->src - (int)b->src;
 520
 521         if(c) {
 522                 return c;
 523         }
 524
 525         c = (int)a->dst - (int)b->dst;
 526         return c;
 527 }
 528
 529 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 530         if(!utcp->nconnections) {
 531                 return NULL;
 532         }
 533
 534         struct utcp_connection key = {
 535                 .src = src,
 536                 .dst = dst,
 537         }, *keyp = &key;
 538         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 539         return match ? *match : NULL;
 540 }
 541
 542 static void free_connection(struct utcp_connection *c) {
 543         struct utcp *utcp = c->utcp;
 544         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 545
 546         assert(cp);
 547
 548         int i = cp - utcp->connections;
 549         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 550         utcp->nconnections--;
 551
 552         buffer_exit(&c->rcvbuf);
 553         buffer_exit(&c->sndbuf);
 554         free(c);
 555 }
 556
 557 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 558         // Check whether this combination of src and dst is free
 559
 560         if(src) {
 561                 if(find_connection(utcp, src, dst)) {
 562                         errno = EADDRINUSE;
 563                         return NULL;
 564                 }
 565         } else { // If src == 0, generate a random port number with the high bit set
 566                 if(utcp->nconnections >= 32767) {
 567                         errno = ENOMEM;
 568                         return NULL;
 569                 }
 570
 571                 src = rand() | 0x8000;
 572
 573                 while(find_connection(utcp, src, dst)) {
 574                         src++;
 575                 }
 576         }
 577
 578         // Allocate memory for the new connection
 579
 580         if(utcp->nconnections >= utcp->nallocated) {
 581                 if(!utcp->nallocated) {
 582                         utcp->nallocated = 4;
 583                 } else {
 584                         utcp->nallocated *= 2;
 585                 }
 586
 587                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 588
 589                 if(!new_array) {
 590                         return NULL;
 591                 }
 592
 593                 utcp->connections = new_array;
 594         }
 595
 596         struct utcp_connection *c = calloc(1, sizeof(*c));
 597
 598         if(!c) {
 599                 return NULL;
 600         }
 601
 602         if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 603                 free(c);
 604                 return NULL;
 605         }
 606
 607         if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 608                 buffer_exit(&c->sndbuf);
 609                 free(c);
 610                 return NULL;
 611         }
 612
 613         // Fill in the details
 614
 615         c->src = src;
 616         c->dst = dst;
 617 #ifdef UTCP_DEBUG
 618         c->snd.iss = 0;
 619 #else
 620         c->snd.iss = rand();
 621 #endif
 622         c->snd.una = c->snd.iss;
 623         c->snd.nxt = c->snd.iss + 1;
 624         c->snd.last = c->snd.nxt;
 625         c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
 626         c->snd.ssthresh = ~0;
 627         debug_cwnd(c);
 628         c->srtt = 0;
 629         c->rttvar = 0;
 630         c->rto = START_RTO;
 631         c->utcp = utcp;
 632
 633         // Add it to the sorted list of connections
 634
 635         utcp->connections[utcp->nconnections++] = c;
 636         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 637
 638         return c;
 639 }
 640
 641 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 642         if(a > b) {
 643                 return a - b;
 644         } else {
 645                 return b - a;
 646         }
 647 }
 648
 649 // Update RTT variables. See RFC 6298.
 650 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 651         if(!rtt) {
 652                 debug(c, "invalid rtt\n");
 653                 return;
 654         }
 655
 656         if(!c->srtt) {
 657                 c->srtt = rtt;
 658                 c->rttvar = rtt / 2;
 659         } else {
 660                 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
 661                 c->srtt = (c->srtt * 7 + rtt) / 8;
 662         }
 663
 664         c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
 665
 666         if(c->rto > MAX_RTO) {
 667                 c->rto = MAX_RTO;
 668         }
 669
 670         debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
 671 }
 672
 673 static void start_retransmit_timer(struct utcp_connection *c) {
 674         clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
 675
 676         uint32_t rto = c->rto;
 677
 678         while(rto > USEC_PER_SEC) {
 679                 c->rtrx_timeout.tv_sec++;
 680                 rto -= USEC_PER_SEC;
 681         }
 682
 683         c->rtrx_timeout.tv_nsec += rto * 1000;
 684
 685         if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
 686                 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
 687                 c->rtrx_timeout.tv_sec++;
 688         }
 689
 690         debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
 691 }
 692
 693 static void stop_retransmit_timer(struct utcp_connection *c) {
 694         timespec_clear(&c->rtrx_timeout);
 695         debug(c, "rtrx_timeout cleared\n");
 696 }
 697
 698 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 699         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 700
 701         if(!c) {
 702                 return NULL;
 703         }
 704
 705         assert((flags & ~0x1f) == 0);
 706
 707         c->flags = flags;
 708         c->recv = recv;
 709         c->priv = priv;
 710
 711         struct {
 712                 struct hdr hdr;
 713                 uint8_t init[4];
 714         } pkt;
 715
 716         pkt.hdr.src = c->src;
 717         pkt.hdr.dst = c->dst;
 718         pkt.hdr.seq = c->snd.iss;
 719         pkt.hdr.ack = 0;
 720         pkt.hdr.wnd = c->rcvbuf.maxsize;
 721         pkt.hdr.ctl = SYN;
 722         pkt.hdr.aux = 0x0101;
 723         pkt.init[0] = 1;
 724         pkt.init[1] = 0;
 725         pkt.init[2] = 0;
 726         pkt.init[3] = flags & 0x7;
 727
 728         set_state(c, SYN_SENT);
 729
 730         print_packet(c, "send", &pkt, sizeof(pkt));
 731         utcp->send(utcp, &pkt, sizeof(pkt));
 732
 733         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 734         c->conn_timeout.tv_sec += utcp->timeout;
 735
 736         start_retransmit_timer(c);
 737
 738         return c;
 739 }
 740
 741 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 742         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 743 }
 744
 745 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 746         if(c->reapable || c->state != SYN_RECEIVED) {
 747                 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
 748                 return;
 749         }
 750
 751         debug(c, "accepted %p %p\n", c, recv, priv);
 752         c->recv = recv;
 753         c->priv = priv;
 754         c->do_poll = true;
 755         set_state(c, ESTABLISHED);
 756 }
 757
 758 static void ack(struct utcp_connection *c, bool sendatleastone) {
 759         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 760         int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
 761
 762         assert(left >= 0);
 763
 764         if(cwndleft <= 0) {
 765                 left = 0;
 766         } else if(cwndleft < left) {
 767                 left = cwndleft;
 768
 769                 if(!sendatleastone || cwndleft > c->utcp->mss) {
 770                         left -= left % c->utcp->mss;
 771                 }
 772         }
 773
 774         debug(c, "cwndleft %d left %d\n", cwndleft, left);
 775
 776         if(!left && !sendatleastone) {
 777                 return;
 778         }
 779
 780         struct {
 781                 struct hdr hdr;
 782                 uint8_t data[];
 783         } *pkt = c->utcp->pkt;
 784
 785         pkt->hdr.src = c->src;
 786         pkt->hdr.dst = c->dst;
 787         pkt->hdr.ack = c->rcv.nxt;
 788         pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
 789         pkt->hdr.ctl = ACK;
 790         pkt->hdr.aux = 0;
 791
 792         do {
 793                 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
 794                 pkt->hdr.seq = c->snd.nxt;
 795
 796                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 797
 798                 c->snd.nxt += seglen;
 799                 left -= seglen;
 800
 801                 if(!is_reliable(c)) {
 802                         if(left) {
 803                                 pkt->hdr.ctl |= MF;
 804                         } else {
 805                                 pkt->hdr.ctl &= ~MF;
 806                         }
 807                 }
 808
 809                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 810                         seglen--;
 811                         pkt->hdr.ctl |= FIN;
 812                 }
 813
 814                 if(!c->rtt_start.tv_sec) {
 815                         // Start RTT measurement
 816                         clock_gettime(UTCP_CLOCK, &c->rtt_start);
 817                         c->rtt_seq = pkt->hdr.seq + seglen;
 818                         debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 819                 }
 820
 821                 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 822                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 823
 824                 if(left && !is_reliable(c)) {
 825                         pkt->hdr.wnd += seglen;
 826                 }
 827         } while(left);
 828 }
 829
 830 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 831         if(c->reapable) {
 832                 debug(c, "send() called on closed connection\n");
 833                 errno = EBADF;
 834                 return -1;
 835         }
 836
 837         switch(c->state) {
 838         case CLOSED:
 839         case LISTEN:
 840                 debug(c, "send() called on unconnected connection\n");
 841                 errno = ENOTCONN;
 842                 return -1;
 843
 844         case SYN_SENT:
 845         case SYN_RECEIVED:
 846         case ESTABLISHED:
 847         case CLOSE_WAIT:
 848                 break;
 849
 850         case FIN_WAIT_1:
 851         case FIN_WAIT_2:
 852         case CLOSING:
 853         case LAST_ACK:
 854         case TIME_WAIT:
 855                 debug(c, "send() called on closed connection\n");
 856                 errno = EPIPE;
 857                 return -1;
 858         }
 859
 860         // Exit early if we have nothing to send.
 861
 862         if(!len) {
 863                 return 0;
 864         }
 865
 866         if(!data) {
 867                 errno = EFAULT;
 868                 return -1;
 869         }
 870
 871         // Check if we need to be able to buffer all data
 872
 873         if(c->flags & UTCP_NO_PARTIAL) {
 874                 if(len > buffer_free(&c->sndbuf)) {
 875                         if(len > c->sndbuf.maxsize) {
 876                                 errno = EMSGSIZE;
 877                                 return -1;
 878                         } else {
 879                                 errno = EWOULDBLOCK;
 880                                 return 0;
 881                         }
 882                 }
 883         }
 884
 885         // Add data to send buffer.
 886
 887         if(is_reliable(c)) {
 888                 len = buffer_put(&c->sndbuf, data, len);
 889         } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
 890                 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
 891                         errno = EMSGSIZE;
 892                         return -1;
 893                 }
 894         } else {
 895                 return 0;
 896         }
 897
 898         if(len <= 0) {
 899                 if(is_reliable(c)) {
 900                         errno = EWOULDBLOCK;
 901                         return 0;
 902                 } else {
 903                         return len;
 904                 }
 905         }
 906
 907         c->snd.last += len;
 908
 909         // Don't send anything yet if the connection has not fully established yet
 910
 911         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 912                 return len;
 913         }
 914
 915         ack(c, false);
 916
 917         if(!is_reliable(c)) {
 918                 c->snd.una = c->snd.nxt = c->snd.last;
 919                 buffer_discard(&c->sndbuf, c->sndbuf.used);
 920         }
 921
 922         if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
 923                 start_retransmit_timer(c);
 924         }
 925
 926         if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
 927                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 928                 c->conn_timeout.tv_sec += c->utcp->timeout;
 929         }
 930
 931         return len;
 932 }
 933
 934 static void swap_ports(struct hdr *hdr) {
 935         uint16_t tmp = hdr->src;
 936         hdr->src = hdr->dst;
 937         hdr->dst = tmp;
 938 }
 939
 940 static void fast_retransmit(struct utcp_connection *c) {
 941         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 942                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 943                 return;
 944         }
 945
 946         struct utcp *utcp = c->utcp;
 947
 948         struct {
 949                 struct hdr hdr;
 950                 uint8_t data[];
 951         } *pkt = c->utcp->pkt;
 952
 953         pkt->hdr.src = c->src;
 954         pkt->hdr.dst = c->dst;
 955         pkt->hdr.wnd = c->rcvbuf.maxsize;
 956         pkt->hdr.aux = 0;
 957
 958         switch(c->state) {
 959         case ESTABLISHED:
 960         case FIN_WAIT_1:
 961         case CLOSE_WAIT:
 962         case CLOSING:
 963         case LAST_ACK:
 964                 // Send unacked data again.
 965                 pkt->hdr.seq = c->snd.una;
 966                 pkt->hdr.ack = c->rcv.nxt;
 967                 pkt->hdr.ctl = ACK;
 968                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 969
 970                 if(fin_wanted(c, c->snd.una + len)) {
 971                         len--;
 972                         pkt->hdr.ctl |= FIN;
 973                 }
 974
 975                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 976                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 977                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 978                 break;
 979
 980         default:
 981                 break;
 982         }
 983 }
 984
 985 static void retransmit(struct utcp_connection *c) {
 986         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 987                 debug(c, "retransmit() called but nothing to retransmit!\n");
 988                 stop_retransmit_timer(c);
 989                 return;
 990         }
 991
 992         struct utcp *utcp = c->utcp;
 993
 994         if(utcp->retransmit) {
 995                 utcp->retransmit(c);
 996         }
 997
 998         struct {
 999                 struct hdr hdr;
1000                 uint8_t data[];
1001         } *pkt = c->utcp->pkt;
1002
1003         pkt->hdr.src = c->src;
1004         pkt->hdr.dst = c->dst;
1005         pkt->hdr.wnd = c->rcvbuf.maxsize;
1006         pkt->hdr.aux = 0;
1007
1008         switch(c->state) {
1009         case SYN_SENT:
1010                 // Send our SYN again
1011                 pkt->hdr.seq = c->snd.iss;
1012                 pkt->hdr.ack = 0;
1013                 pkt->hdr.ctl = SYN;
1014                 pkt->hdr.aux = 0x0101;
1015                 pkt->data[0] = 1;
1016                 pkt->data[1] = 0;
1017                 pkt->data[2] = 0;
1018                 pkt->data[3] = c->flags & 0x7;
1019                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
1020                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
1021                 break;
1022
1023         case SYN_RECEIVED:
1024                 // Send SYNACK again
1025                 pkt->hdr.seq = c->snd.nxt;
1026                 pkt->hdr.ack = c->rcv.nxt;
1027                 pkt->hdr.ctl = SYN | ACK;
1028                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
1029                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
1030                 break;
1031
1032         case ESTABLISHED:
1033         case FIN_WAIT_1:
1034         case CLOSE_WAIT:
1035         case CLOSING:
1036         case LAST_ACK:
1037                 // Send unacked data again.
1038                 pkt->hdr.seq = c->snd.una;
1039                 pkt->hdr.ack = c->rcv.nxt;
1040                 pkt->hdr.ctl = ACK;
1041                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
1042
1043                 if(fin_wanted(c, c->snd.una + len)) {
1044                         len--;
1045                         pkt->hdr.ctl |= FIN;
1046                 }
1047
1048                 // RFC 5681 slow start after timeout
1049                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1050                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1051                 c->snd.cwnd = utcp->mss;
1052                 debug_cwnd(c);
1053
1054                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
1055                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
1056                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
1057
1058                 c->snd.nxt = c->snd.una + len;
1059                 break;
1060
1061         case CLOSED:
1062         case LISTEN:
1063         case TIME_WAIT:
1064         case FIN_WAIT_2:
1065                 // We shouldn't need to retransmit anything in this state.
1066 #ifdef UTCP_DEBUG
1067                 abort();
1068 #endif
1069                 stop_retransmit_timer(c);
1070                 goto cleanup;
1071         }
1072
1073         start_retransmit_timer(c);
1074         c->rto *= 2;
1075
1076         if(c->rto > MAX_RTO) {
1077                 c->rto = MAX_RTO;
1078         }
1079
1080         c->rtt_start.tv_sec = 0; // invalidate RTT timer
1081         c->dupack = 0; // cancel any ongoing fast recovery
1082
1083 cleanup:
1084         return;
1085 }
1086
1087 /* Update receive buffer and SACK entries after consuming data.
1088  *
1089  * Situation:
1090  *
1091  * |.....0000..1111111111.....22222......3333|
1092  * |---------------^
1093  *
1094  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
1095  * to remove data from the receive buffer. The idea is to substract "len"
1096  * from the offset of all the SACK entries, and then remove/cut down entries
1097  * that are shifted to before the start of the receive buffer.
1098  *
1099  * There are three cases:
1100  * - the SACK entry is after ^, in that case just change the offset.
1101  * - the SACK entry starts before and ends after ^, so we have to
1102  *   change both its offset and size.
1103  * - the SACK entry is completely before ^, in that case delete it.
1104  */
1105 static void sack_consume(struct utcp_connection *c, size_t len) {
1106         debug(c, "sack_consume %lu\n", (unsigned long)len);
1107
1108         if(len > c->rcvbuf.used) {
1109                 debug(c, "all SACK entries consumed\n");
1110                 c->sacks[0].len = 0;
1111                 return;
1112         }
1113
1114         buffer_discard(&c->rcvbuf, len);
1115
1116         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1117                 if(len < c->sacks[i].offset) {
1118                         c->sacks[i].offset -= len;
1119                         i++;
1120                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1121                         c->sacks[i].len -= len - c->sacks[i].offset;
1122                         c->sacks[i].offset = 0;
1123                         i++;
1124                 } else {
1125                         if(i < NSACKS - 1) {
1126                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1127                                 c->sacks[NSACKS - 1].len = 0;
1128                         } else {
1129                                 c->sacks[i].len = 0;
1130                                 break;
1131                         }
1132                 }
1133         }
1134
1135         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1136                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1137         }
1138 }
1139
1140 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1141         debug(c, "out of order packet, offset %u\n", offset);
1142         // Packet loss or reordering occured. Store the data in the buffer.
1143         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1144
1145         if(rxd <= 0) {
1146                 debug(c, "packet outside receive buffer, dropping\n");
1147                 return;
1148         }
1149
1150         if((size_t)rxd < len) {
1151                 debug(c, "packet partially outside receive buffer\n");
1152                 len = rxd;
1153         }
1154
1155         // Make note of where we put it.
1156         for(int i = 0; i < NSACKS; i++) {
1157                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1158                         debug(c, "new SACK entry %d\n", i);
1159                         c->sacks[i].offset = offset;
1160                         c->sacks[i].len = rxd;
1161                         break;
1162                 } else if(offset < c->sacks[i].offset) {
1163                         if(offset + rxd < c->sacks[i].offset) { // insert before
1164                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1165                                         debug(c, "insert SACK entry at %d\n", i);
1166                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1167                                         c->sacks[i].offset = offset;
1168                                         c->sacks[i].len = rxd;
1169                                 } else {
1170                                         debug(c, "SACK entries full, dropping packet\n");
1171                                 }
1172
1173                                 break;
1174                         } else { // merge
1175                                 debug(c, "merge with start of SACK entry at %d\n", i);
1176                                 c->sacks[i].offset = offset;
1177                                 break;
1178                         }
1179                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1180                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1181                                 debug(c, "merge with end of SACK entry at %d\n", i);
1182                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1183                                 // TODO: handle potential merge with next entry
1184                         }
1185
1186                         break;
1187                 }
1188         }
1189
1190         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1191                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1192         }
1193 }
1194
1195 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1196         if(c->recv) {
1197                 ssize_t rxd = c->recv(c, data, len);
1198
1199                 if(rxd != (ssize_t)len) {
1200                         // TODO: handle the application not accepting all data.
1201                         abort();
1202                 }
1203         }
1204
1205         // Check if we can process out-of-order data now.
1206         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1207                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1208
1209                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1210                         size_t offset = len;
1211                         len = c->sacks[0].offset + c->sacks[0].len;
1212                         size_t remainder = len - offset;
1213
1214                         ssize_t rxd = buffer_call(c, &c->rcvbuf, offset, remainder);
1215
1216                         if(rxd != (ssize_t)remainder) {
1217                                 // TODO: handle the application not accepting all data.
1218                                 abort();
1219                         }
1220                 }
1221         }
1222
1223         if(c->rcvbuf.used) {
1224                 sack_consume(c, len);
1225         }
1226
1227         c->rcv.nxt += len;
1228 }
1229
1230 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1231         // Fast path for unfragmented packets
1232         if(!hdr->wnd && !(hdr->ctl & MF)) {
1233                 if(c->recv) {
1234                         c->recv(c, data, len);
1235                 }
1236
1237                 c->rcv.nxt = hdr->seq + len;
1238                 return;
1239         }
1240
1241         // Ensure reassembled packet are not larger than 64 kiB
1242         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1243                 return;
1244         }
1245
1246         // Don't accept out of order fragments
1247         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1248                 return;
1249         }
1250
1251         // Reset the receive buffer for the first fragment
1252         if(!hdr->wnd) {
1253                 buffer_clear(&c->rcvbuf);
1254         }
1255
1256         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1257
1258         if(rxd != (ssize_t)len) {
1259                 return;
1260         }
1261
1262         // Send the packet if it's the final fragment
1263         if(!(hdr->ctl & MF)) {
1264                 buffer_call(c, &c->rcvbuf, 0, hdr->wnd + len);
1265         }
1266
1267         c->rcv.nxt = hdr->seq + len;
1268 }
1269
1270 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1271         if(!is_reliable(c)) {
1272                 handle_unreliable(c, hdr, data, len);
1273                 return;
1274         }
1275
1276         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1277
1278         if(offset) {
1279                 handle_out_of_order(c, offset, data, len);
1280         } else {
1281                 handle_in_order(c, data, len);
1282         }
1283 }
1284
1285
1286 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1287         const uint8_t *ptr = data;
1288
1289         if(!utcp) {
1290                 errno = EFAULT;
1291                 return -1;
1292         }
1293
1294         if(!len) {
1295                 return 0;
1296         }
1297
1298         if(!data) {
1299                 errno = EFAULT;
1300                 return -1;
1301         }
1302
1303         // Drop packets smaller than the header
1304
1305         struct hdr hdr;
1306
1307         if(len < sizeof(hdr)) {
1308                 print_packet(NULL, "recv", data, len);
1309                 errno = EBADMSG;
1310                 return -1;
1311         }
1312
1313         // Make a copy from the potentially unaligned data to a struct hdr
1314
1315         memcpy(&hdr, ptr, sizeof(hdr));
1316
1317         // Try to match the packet to an existing connection
1318
1319         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1320         print_packet(c, "recv", data, len);
1321
1322         // Process the header
1323
1324         ptr += sizeof(hdr);
1325         len -= sizeof(hdr);
1326
1327         // Drop packets with an unknown CTL flag
1328
1329         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1330                 print_packet(NULL, "recv", data, len);
1331                 errno = EBADMSG;
1332                 return -1;
1333         }
1334
1335         // Check for auxiliary headers
1336
1337         const uint8_t *init = NULL;
1338
1339         uint16_t aux = hdr.aux;
1340
1341         while(aux) {
1342                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1343                 uint8_t auxtype = aux & 0xff;
1344
1345                 if(len < auxlen) {
1346                         errno = EBADMSG;
1347                         return -1;
1348                 }
1349
1350                 switch(auxtype) {
1351                 case AUX_INIT:
1352                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1353                                 errno = EBADMSG;
1354                                 return -1;
1355                         }
1356
1357                         init = ptr;
1358                         break;
1359
1360                 default:
1361                         errno = EBADMSG;
1362                         return -1;
1363                 }
1364
1365                 len -= auxlen;
1366                 ptr += auxlen;
1367
1368                 if(!(aux & 0x800)) {
1369                         break;
1370                 }
1371
1372                 if(len < 2) {
1373                         errno = EBADMSG;
1374                         return -1;
1375                 }
1376
1377                 memcpy(&aux, ptr, 2);
1378                 len -= 2;
1379                 ptr += 2;
1380         }
1381
1382         bool has_data = len || (hdr.ctl & (SYN | FIN));
1383
1384         // Is it for a new connection?
1385
1386         if(!c) {
1387                 // Ignore RST packets
1388
1389                 if(hdr.ctl & RST) {
1390                         return 0;
1391                 }
1392
1393                 // Is it a SYN packet and are we LISTENing?
1394
1395                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1396                         // If we don't want to accept it, send a RST back
1397                         if((utcp->listen && !utcp->listen(utcp, hdr.dst))) {
1398                                 len = 1;
1399                                 goto reset;
1400                         }
1401
1402                         // Try to allocate memory, otherwise send a RST back
1403                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1404
1405                         if(!c) {
1406                                 len = 1;
1407                                 goto reset;
1408                         }
1409
1410                         // Parse auxilliary information
1411                         if(init) {
1412                                 if(init[0] < 1) {
1413                                         len = 1;
1414                                         goto reset;
1415                                 }
1416
1417                                 c->flags = init[3] & 0x7;
1418                         } else {
1419                                 c->flags = UTCP_TCP;
1420                         }
1421
1422 synack:
1423                         // Return SYN+ACK, go to SYN_RECEIVED state
1424                         c->snd.wnd = hdr.wnd;
1425                         c->rcv.irs = hdr.seq;
1426                         c->rcv.nxt = c->rcv.irs + 1;
1427                         set_state(c, SYN_RECEIVED);
1428
1429                         struct {
1430                                 struct hdr hdr;
1431                                 uint8_t data[4];
1432                         } pkt;
1433
1434                         pkt.hdr.src = c->src;
1435                         pkt.hdr.dst = c->dst;
1436                         pkt.hdr.ack = c->rcv.irs + 1;
1437                         pkt.hdr.seq = c->snd.iss;
1438                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1439                         pkt.hdr.ctl = SYN | ACK;
1440
1441                         if(init) {
1442                                 pkt.hdr.aux = 0x0101;
1443                                 pkt.data[0] = 1;
1444                                 pkt.data[1] = 0;
1445                                 pkt.data[2] = 0;
1446                                 pkt.data[3] = c->flags & 0x7;
1447                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1448                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1449                         } else {
1450                                 pkt.hdr.aux = 0;
1451                                 print_packet(c, "send", &pkt, sizeof(hdr));
1452                                 utcp->send(utcp, &pkt, sizeof(hdr));
1453                         }
1454
1455                         start_retransmit_timer(c);
1456                 } else {
1457                         // No, we don't want your packets, send a RST back
1458                         len = 1;
1459                         goto reset;
1460                 }
1461
1462                 return 0;
1463         }
1464
1465         debug(c, "state %s\n", strstate[c->state]);
1466
1467         // In case this is for a CLOSED connection, ignore the packet.
1468         // TODO: make it so incoming packets can never match a CLOSED connection.
1469
1470         if(c->state == CLOSED) {
1471                 debug(c, "got packet for closed connection\n");
1472                 goto reset;
1473         }
1474
1475         // It is for an existing connection.
1476
1477         // 1. Drop invalid packets.
1478
1479         // 1a. Drop packets that should not happen in our current state.
1480
1481         switch(c->state) {
1482         case SYN_SENT:
1483         case SYN_RECEIVED:
1484         case ESTABLISHED:
1485         case FIN_WAIT_1:
1486         case FIN_WAIT_2:
1487         case CLOSE_WAIT:
1488         case CLOSING:
1489         case LAST_ACK:
1490         case TIME_WAIT:
1491                 break;
1492
1493         default:
1494 #ifdef UTCP_DEBUG
1495                 abort();
1496 #endif
1497                 break;
1498         }
1499
1500         // 1b. Discard data that is not in our receive window.
1501
1502         if(is_reliable(c)) {
1503                 bool acceptable;
1504
1505                 if(c->state == SYN_SENT) {
1506                         acceptable = true;
1507                 } else if(len == 0) {
1508                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1509                 } else {
1510                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1511
1512                         // cut already accepted front overlapping
1513                         if(rcv_offset < 0) {
1514                                 acceptable = len > (size_t) - rcv_offset;
1515
1516                                 if(acceptable) {
1517                                         ptr -= rcv_offset;
1518                                         len += rcv_offset;
1519                                         hdr.seq -= rcv_offset;
1520                                 }
1521                         } else {
1522                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1523                         }
1524                 }
1525
1526                 if(!acceptable) {
1527                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1528
1529                         // Ignore unacceptable RST packets.
1530                         if(hdr.ctl & RST) {
1531                                 return 0;
1532                         }
1533
1534                         // Otherwise, continue processing.
1535                         len = 0;
1536                 }
1537         } else {
1538 #if UTCP_DEBUG
1539                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1540
1541                 if(rcv_offset) {
1542                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1543                 }
1544
1545 #endif
1546         }
1547
1548         c->snd.wnd = hdr.wnd; // TODO: move below
1549
1550         // 1c. Drop packets with an invalid ACK.
1551         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1552         // (= snd.una + c->sndbuf.used).
1553
1554         if(!is_reliable(c)) {
1555                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1556                         hdr.ack = c->snd.una;
1557                 }
1558         }
1559
1560         // 2. Handle RST packets
1561
1562         if(hdr.ctl & RST) {
1563                 switch(c->state) {
1564                 case SYN_SENT:
1565                         if(!(hdr.ctl & ACK)) {
1566                                 return 0;
1567                         }
1568
1569                         // The peer has refused our connection.
1570                         set_state(c, CLOSED);
1571                         errno = ECONNREFUSED;
1572
1573                         if(c->recv) {
1574                                 c->recv(c, NULL, 0);
1575                         }
1576
1577                         if(c->poll && !c->reapable) {
1578                                 c->poll(c, 0);
1579                         }
1580
1581                         return 0;
1582
1583                 case SYN_RECEIVED:
1584                         if(hdr.ctl & ACK) {
1585                                 return 0;
1586                         }
1587
1588                         // We haven't told the application about this connection yet. Silently delete.
1589                         free_connection(c);
1590                         return 0;
1591
1592                 case ESTABLISHED:
1593                 case FIN_WAIT_1:
1594                 case FIN_WAIT_2:
1595                 case CLOSE_WAIT:
1596                         if(hdr.ctl & ACK) {
1597                                 return 0;
1598                         }
1599
1600                         // The peer has aborted our connection.
1601                         set_state(c, CLOSED);
1602                         errno = ECONNRESET;
1603                         buffer_clear(&c->sndbuf);
1604                         buffer_clear(&c->rcvbuf);
1605
1606                         if(c->recv) {
1607                                 c->recv(c, NULL, 0);
1608                         }
1609
1610                         if(c->poll && !c->reapable) {
1611                                 c->poll(c, 0);
1612                         }
1613
1614                         return 0;
1615
1616                 case CLOSING:
1617                 case LAST_ACK:
1618                 case TIME_WAIT:
1619                         if(hdr.ctl & ACK) {
1620                                 return 0;
1621                         }
1622
1623                         // As far as the application is concerned, the connection has already been closed.
1624                         // If it has called utcp_close() already, we can immediately free this connection.
1625                         if(c->reapable) {
1626                                 free_connection(c);
1627                                 return 0;
1628                         }
1629
1630                         // Otherwise, immediately move to the CLOSED state.
1631                         set_state(c, CLOSED);
1632                         return 0;
1633
1634                 default:
1635 #ifdef UTCP_DEBUG
1636                         abort();
1637 #endif
1638                         break;
1639                 }
1640         }
1641
1642         uint32_t advanced;
1643
1644         if(!(hdr.ctl & ACK)) {
1645                 advanced = 0;
1646                 goto skip_ack;
1647         }
1648
1649         // 3. Advance snd.una
1650
1651         if(seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0) {
1652                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1653                 goto reset;
1654         }
1655
1656         advanced = seqdiff(hdr.ack, c->snd.una);
1657
1658         if(advanced) {
1659                 // RTT measurement
1660                 if(c->rtt_start.tv_sec) {
1661                         if(c->rtt_seq == hdr.ack) {
1662                                 struct timespec now;
1663                                 clock_gettime(UTCP_CLOCK, &now);
1664                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1665                                 update_rtt(c, diff);
1666                                 c->rtt_start.tv_sec = 0;
1667                         } else if(c->rtt_seq < hdr.ack) {
1668                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1669                                 c->rtt_start.tv_sec = 0;
1670                         }
1671                 }
1672
1673                 int32_t data_acked = advanced;
1674
1675                 switch(c->state) {
1676                 case SYN_SENT:
1677                 case SYN_RECEIVED:
1678                         data_acked--;
1679                         break;
1680
1681                 // TODO: handle FIN as well.
1682                 default:
1683                         break;
1684                 }
1685
1686                 assert(data_acked >= 0);
1687
1688 #ifndef NDEBUG
1689                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1690                 assert(data_acked <= bufused);
1691 #endif
1692
1693                 if(data_acked) {
1694                         buffer_discard(&c->sndbuf, data_acked);
1695
1696                         if(is_reliable(c)) {
1697                                 c->do_poll = true;
1698                         }
1699                 }
1700
1701                 // Also advance snd.nxt if possible
1702                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1703                         c->snd.nxt = hdr.ack;
1704                 }
1705
1706                 c->snd.una = hdr.ack;
1707
1708                 if(c->dupack) {
1709                         if(c->dupack >= 3) {
1710                                 debug(c, "fast recovery ended\n");
1711                                 c->snd.cwnd = c->snd.ssthresh;
1712                         }
1713
1714                         c->dupack = 0;
1715                 }
1716
1717                 // Increase the congestion window according to RFC 5681
1718                 if(c->snd.cwnd < c->snd.ssthresh) {
1719                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1720                 } else {
1721                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1722                 }
1723
1724                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1725                         c->snd.cwnd = c->sndbuf.maxsize;
1726                 }
1727
1728                 debug_cwnd(c);
1729
1730                 // Check if we have sent a FIN that is now ACKed.
1731                 switch(c->state) {
1732                 case FIN_WAIT_1:
1733                         if(c->snd.una == c->snd.last) {
1734                                 set_state(c, FIN_WAIT_2);
1735                         }
1736
1737                         break;
1738
1739                 case CLOSING:
1740                         if(c->snd.una == c->snd.last) {
1741                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1742                                 c->conn_timeout.tv_sec += utcp->timeout;
1743                                 set_state(c, TIME_WAIT);
1744                         }
1745
1746                         break;
1747
1748                 default:
1749                         break;
1750                 }
1751         } else {
1752                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1753                         c->dupack++;
1754                         debug(c, "duplicate ACK %d\n", c->dupack);
1755
1756                         if(c->dupack == 3) {
1757                                 // RFC 5681 fast recovery
1758                                 debug(c, "fast recovery started\n", c->dupack);
1759                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1760                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1761                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1762
1763                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1764                                         c->snd.cwnd = c->sndbuf.maxsize;
1765                                 }
1766
1767                                 debug_cwnd(c);
1768
1769                                 fast_retransmit(c);
1770                         } else if(c->dupack > 3) {
1771                                 c->snd.cwnd += utcp->mss;
1772
1773                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1774                                         c->snd.cwnd = c->sndbuf.maxsize;
1775                                 }
1776
1777                                 debug_cwnd(c);
1778                         }
1779
1780                         // We got an ACK which indicates the other side did get one of our packets.
1781                         // Reset the retransmission timer to avoid going to slow start,
1782                         // but don't touch the connection timeout.
1783                         start_retransmit_timer(c);
1784                 }
1785         }
1786
1787         // 4. Update timers
1788
1789         if(advanced) {
1790                 if(c->snd.una == c->snd.last) {
1791                         stop_retransmit_timer(c);
1792                         timespec_clear(&c->conn_timeout);
1793                 } else if(is_reliable(c)) {
1794                         start_retransmit_timer(c);
1795                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1796                         c->conn_timeout.tv_sec += utcp->timeout;
1797                 }
1798         }
1799
1800 skip_ack:
1801         // 5. Process SYN stuff
1802
1803         if(hdr.ctl & SYN) {
1804                 switch(c->state) {
1805                 case SYN_SENT:
1806
1807                         // This is a SYNACK. It should always have ACKed the SYN.
1808                         if(!advanced) {
1809                                 goto reset;
1810                         }
1811
1812                         c->rcv.irs = hdr.seq;
1813                         c->rcv.nxt = hdr.seq + 1;
1814
1815                         if(c->shut_wr) {
1816                                 c->snd.last++;
1817                                 set_state(c, FIN_WAIT_1);
1818                         } else {
1819                                 c->do_poll = true;
1820                                 set_state(c, ESTABLISHED);
1821                         }
1822
1823                         break;
1824
1825                 case SYN_RECEIVED:
1826                         // This is a retransmit of a SYN, send back the SYNACK.
1827                         goto synack;
1828
1829                 case ESTABLISHED:
1830                 case FIN_WAIT_1:
1831                 case FIN_WAIT_2:
1832                 case CLOSE_WAIT:
1833                 case CLOSING:
1834                 case LAST_ACK:
1835                 case TIME_WAIT:
1836                         // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
1837                         break;
1838
1839                 default:
1840 #ifdef UTCP_DEBUG
1841                         abort();
1842 #endif
1843                         return 0;
1844                 }
1845         }
1846
1847         // 6. Process new data
1848
1849         if(c->state == SYN_RECEIVED) {
1850                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1851                 if(!advanced) {
1852                         goto reset;
1853                 }
1854
1855                 // Are we still LISTENing?
1856                 if(utcp->accept) {
1857                         utcp->accept(c, c->src);
1858                 }
1859
1860                 if(c->state != ESTABLISHED) {
1861                         set_state(c, CLOSED);
1862                         c->reapable = true;
1863                         goto reset;
1864                 }
1865         }
1866
1867         if(len) {
1868                 switch(c->state) {
1869                 case SYN_SENT:
1870                 case SYN_RECEIVED:
1871                         // This should never happen.
1872 #ifdef UTCP_DEBUG
1873                         abort();
1874 #endif
1875                         return 0;
1876
1877                 case ESTABLISHED:
1878                         break;
1879
1880                 case FIN_WAIT_1:
1881                 case FIN_WAIT_2:
1882                         if(c->reapable) {
1883                                 // We already closed the connection and are not interested in more data.
1884                                 goto reset;
1885                         }
1886
1887                         break;
1888
1889                 case CLOSE_WAIT:
1890                 case CLOSING:
1891                 case LAST_ACK:
1892                 case TIME_WAIT:
1893                         // Ehm no, We should never receive more data after a FIN.
1894                         goto reset;
1895
1896                 default:
1897 #ifdef UTCP_DEBUG
1898                         abort();
1899 #endif
1900                         return 0;
1901                 }
1902
1903                 handle_incoming_data(c, &hdr, ptr, len);
1904         }
1905
1906         // 7. Process FIN stuff
1907
1908         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1909                 switch(c->state) {
1910                 case SYN_SENT:
1911                 case SYN_RECEIVED:
1912                         // This should never happen.
1913 #ifdef UTCP_DEBUG
1914                         abort();
1915 #endif
1916                         break;
1917
1918                 case ESTABLISHED:
1919                         set_state(c, CLOSE_WAIT);
1920                         break;
1921
1922                 case FIN_WAIT_1:
1923                         set_state(c, CLOSING);
1924                         break;
1925
1926                 case FIN_WAIT_2:
1927                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1928                         c->conn_timeout.tv_sec += utcp->timeout;
1929                         set_state(c, TIME_WAIT);
1930                         break;
1931
1932                 case CLOSE_WAIT:
1933                 case CLOSING:
1934                 case LAST_ACK:
1935                 case TIME_WAIT:
1936                         // Ehm, no. We should never receive a second FIN.
1937                         goto reset;
1938
1939                 default:
1940 #ifdef UTCP_DEBUG
1941                         abort();
1942 #endif
1943                         break;
1944                 }
1945
1946                 // FIN counts as one sequence number
1947                 c->rcv.nxt++;
1948                 len++;
1949
1950                 // Inform the application that the peer closed its end of the connection.
1951                 if(c->recv) {
1952                         errno = 0;
1953                         c->recv(c, NULL, 0);
1954                 }
1955         }
1956
1957         // Now we send something back if:
1958         // - we received data, so we have to send back an ACK
1959         //   -> sendatleastone = true
1960         // - or we got an ack, so we should maybe send a bit more data
1961         //   -> sendatleastone = false
1962
1963         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1964                 ack(c, has_data);
1965         }
1966
1967         return 0;
1968
1969 reset:
1970         swap_ports(&hdr);
1971         hdr.wnd = 0;
1972         hdr.aux = 0;
1973
1974         if(hdr.ctl & ACK) {
1975                 hdr.seq = hdr.ack;
1976                 hdr.ctl = RST;
1977         } else {
1978                 hdr.ack = hdr.seq + len;
1979                 hdr.seq = 0;
1980                 hdr.ctl = RST | ACK;
1981         }
1982
1983         print_packet(c, "send", &hdr, sizeof(hdr));
1984         utcp->send(utcp, &hdr, sizeof(hdr));
1985         return 0;
1986
1987 }
1988
1989 int utcp_shutdown(struct utcp_connection *c, int dir) {
1990         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1991
1992         if(!c) {
1993                 errno = EFAULT;
1994                 return -1;
1995         }
1996
1997         if(c->reapable) {
1998                 debug(c, "shutdown() called on closed connection\n");
1999                 errno = EBADF;
2000                 return -1;
2001         }
2002
2003         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
2004                 errno = EINVAL;
2005                 return -1;
2006         }
2007
2008         // TCP does not have a provision for stopping incoming packets.
2009         // The best we can do is to just ignore them.
2010         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
2011                 c->recv = NULL;
2012         }
2013
2014         // The rest of the code deals with shutting down writes.
2015         if(dir == UTCP_SHUT_RD) {
2016                 return 0;
2017         }
2018
2019         // Only process shutting down writes once.
2020         if(c->shut_wr) {
2021                 return 0;
2022         }
2023
2024         c->shut_wr = true;
2025
2026         switch(c->state) {
2027         case CLOSED:
2028         case LISTEN:
2029                 errno = ENOTCONN;
2030                 return -1;
2031
2032         case SYN_SENT:
2033                 return 0;
2034
2035         case SYN_RECEIVED:
2036         case ESTABLISHED:
2037                 set_state(c, FIN_WAIT_1);
2038                 break;
2039
2040         case FIN_WAIT_1:
2041         case FIN_WAIT_2:
2042                 return 0;
2043
2044         case CLOSE_WAIT:
2045                 set_state(c, CLOSING);
2046                 break;
2047
2048         case CLOSING:
2049         case LAST_ACK:
2050         case TIME_WAIT:
2051                 return 0;
2052         }
2053
2054         c->snd.last++;
2055
2056         ack(c, false);
2057
2058         if(!timespec_isset(&c->rtrx_timeout)) {
2059                 start_retransmit_timer(c);
2060         }
2061
2062         return 0;
2063 }
2064
2065 static bool reset_connection(struct utcp_connection *c) {
2066         if(!c) {
2067                 errno = EFAULT;
2068                 return false;
2069         }
2070
2071         if(c->reapable) {
2072                 debug(c, "abort() called on closed connection\n");
2073                 errno = EBADF;
2074                 return false;
2075         }
2076
2077         buffer_clear(&c->sndbuf);
2078         buffer_clear(&c->rcvbuf);
2079
2080         switch(c->state) {
2081         case CLOSED:
2082                 return true;
2083
2084         case LISTEN:
2085         case SYN_SENT:
2086         case CLOSING:
2087         case LAST_ACK:
2088         case TIME_WAIT:
2089                 set_state(c, CLOSED);
2090                 return true;
2091
2092         case SYN_RECEIVED:
2093         case ESTABLISHED:
2094         case FIN_WAIT_1:
2095         case FIN_WAIT_2:
2096         case CLOSE_WAIT:
2097                 set_state(c, CLOSED);
2098                 break;
2099         }
2100
2101         // Send RST
2102
2103         struct hdr hdr;
2104
2105         hdr.src = c->src;
2106         hdr.dst = c->dst;
2107         hdr.seq = c->snd.nxt;
2108         hdr.ack = c->rcv.nxt;
2109         hdr.wnd = 0;
2110         hdr.ctl = RST;
2111         hdr.aux = 0;
2112
2113         print_packet(c, "send", &hdr, sizeof(hdr));
2114         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2115         return true;
2116 }
2117
2118 static void set_reapable(struct utcp_connection *c) {
2119         set_buffer_storage(&c->sndbuf, NULL, min(c->sndbuf.maxsize, DEFAULT_MAXSNDBUFSIZE));
2120         set_buffer_storage(&c->rcvbuf, NULL, min(c->rcvbuf.maxsize, DEFAULT_MAXRCVBUFSIZE));
2121
2122         c->recv = NULL;
2123         c->poll = NULL;
2124         c->reapable = true;
2125 }
2126
2127 // Resets all connections, but does not invalidate connection handles
2128 void utcp_reset_all_connections(struct utcp *utcp) {
2129         if(!utcp) {
2130                 errno = EINVAL;
2131                 return;
2132         }
2133
2134         for(int i = 0; i < utcp->nconnections; i++) {
2135                 struct utcp_connection *c = utcp->connections[i];
2136
2137                 if(c->reapable || c->state == CLOSED) {
2138                         continue;
2139                 }
2140
2141                 reset_connection(c);
2142
2143                 if(c->recv) {
2144                         errno = 0;
2145                         c->recv(c, NULL, 0);
2146                 }
2147
2148                 if(c->poll && !c->reapable) {
2149                         errno = 0;
2150                         c->poll(c, 0);
2151                 }
2152         }
2153
2154         return;
2155 }
2156
2157 int utcp_close(struct utcp_connection *c) {
2158         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2159                 return -1;
2160         }
2161
2162         set_reapable(c);
2163         return 0;
2164 }
2165
2166 int utcp_abort(struct utcp_connection *c) {
2167         if(!reset_connection(c)) {
2168                 return -1;
2169         }
2170
2171         set_reapable(c);
2172         return 0;
2173 }
2174
2175 /* Handle timeouts.
2176  * One call to this function will loop through all connections,
2177  * checking if something needs to be resent or not.
2178  * The return value is the time to the next timeout in milliseconds,
2179  * or maybe a negative value if the timeout is infinite.
2180  */
2181 struct timespec utcp_timeout(struct utcp *utcp) {
2182         struct timespec now;
2183         clock_gettime(UTCP_CLOCK, &now);
2184         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2185
2186         for(int i = 0; i < utcp->nconnections; i++) {
2187                 struct utcp_connection *c = utcp->connections[i];
2188
2189                 if(!c) {
2190                         continue;
2191                 }
2192
2193                 // delete connections that have been utcp_close()d.
2194                 if(c->state == CLOSED) {
2195                         if(c->reapable) {
2196                                 debug(c, "reaping\n");
2197                                 free_connection(c);
2198                                 i--;
2199                         }
2200
2201                         continue;
2202                 }
2203
2204                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2205                         errno = ETIMEDOUT;
2206                         c->state = CLOSED;
2207                         buffer_clear(&c->sndbuf);
2208                         buffer_clear(&c->rcvbuf);
2209
2210                         if(c->recv) {
2211                                 c->recv(c, NULL, 0);
2212                         }
2213
2214                         if(c->poll && !c->reapable) {
2215                                 c->poll(c, 0);
2216                         }
2217
2218                         continue;
2219                 }
2220
2221                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2222                         debug(c, "retransmitting after timeout\n");
2223                         retransmit(c);
2224                 }
2225
2226                 if(c->poll) {
2227                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2228                                 c->do_poll = false;
2229                                 uint32_t len = buffer_free(&c->sndbuf);
2230
2231                                 if(len) {
2232                                         c->poll(c, len);
2233                                 }
2234                         } else if(c->state == CLOSED) {
2235                                 c->poll(c, 0);
2236                         }
2237                 }
2238
2239                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2240                         next = c->conn_timeout;
2241                 }
2242
2243                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2244                         next = c->rtrx_timeout;
2245                 }
2246         }
2247
2248         struct timespec diff;
2249
2250         timespec_sub(&next, &now, &diff);
2251
2252         return diff;
2253 }
2254
2255 bool utcp_is_active(struct utcp *utcp) {
2256         if(!utcp) {
2257                 return false;
2258         }
2259
2260         for(int i = 0; i < utcp->nconnections; i++)
2261                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2262                         return true;
2263                 }
2264
2265         return false;
2266 }
2267
2268 struct utcp *utcp_init(utcp_accept_t accept, utcp_listen_t listen, utcp_send_t send, void *priv) {
2269         if(!send) {
2270                 errno = EFAULT;
2271                 return NULL;
2272         }
2273
2274         struct utcp *utcp = calloc(1, sizeof(*utcp));
2275
2276         if(!utcp) {
2277                 return NULL;
2278         }
2279
2280         utcp_set_mtu(utcp, DEFAULT_MTU);
2281
2282         if(!utcp->pkt) {
2283                 free(utcp);
2284                 return NULL;
2285         }
2286
2287         if(!CLOCK_GRANULARITY) {
2288                 struct timespec res;
2289                 clock_getres(UTCP_CLOCK, &res);
2290                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2291         }
2292
2293         utcp->accept = accept;
2294         utcp->listen = listen;
2295         utcp->send = send;
2296         utcp->priv = priv;
2297         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2298
2299         return utcp;
2300 }
2301
2302 void utcp_exit(struct utcp *utcp) {
2303         if(!utcp) {
2304                 return;
2305         }
2306
2307         for(int i = 0; i < utcp->nconnections; i++) {
2308                 struct utcp_connection *c = utcp->connections[i];
2309
2310                 if(!c->reapable) {
2311                         buffer_clear(&c->sndbuf);
2312                         buffer_clear(&c->rcvbuf);
2313
2314                         if(c->recv) {
2315                                 c->recv(c, NULL, 0);
2316                         }
2317
2318                         if(c->poll && !c->reapable) {
2319                                 c->poll(c, 0);
2320                         }
2321                 }
2322
2323                 buffer_exit(&c->rcvbuf);
2324                 buffer_exit(&c->sndbuf);
2325                 free(c);
2326         }
2327
2328         free(utcp->connections);
2329         free(utcp->pkt);
2330         free(utcp);
2331 }
2332
2333 uint16_t utcp_get_mtu(struct utcp *utcp) {
2334         return utcp ? utcp->mtu : 0;
2335 }
2336
2337 uint16_t utcp_get_mss(struct utcp *utcp) {
2338         return utcp ? utcp->mss : 0;
2339 }
2340
2341 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2342         if(!utcp) {
2343                 return;
2344         }
2345
2346         if(mtu <= sizeof(struct hdr)) {
2347                 return;
2348         }
2349
2350         if(mtu > utcp->mtu) {
2351                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2352
2353                 if(!new) {
2354                         return;
2355                 }
2356
2357                 utcp->pkt = new;
2358         }
2359
2360         utcp->mtu = mtu;
2361         utcp->mss = mtu - sizeof(struct hdr);
2362 }
2363
2364 void utcp_reset_timers(struct utcp *utcp) {
2365         if(!utcp) {
2366                 return;
2367         }
2368
2369         struct timespec now, then;
2370
2371         clock_gettime(UTCP_CLOCK, &now);
2372
2373         then = now;
2374
2375         then.tv_sec += utcp->timeout;
2376
2377         for(int i = 0; i < utcp->nconnections; i++) {
2378                 struct utcp_connection *c = utcp->connections[i];
2379
2380                 if(c->reapable) {
2381                         continue;
2382                 }
2383
2384                 if(timespec_isset(&c->rtrx_timeout)) {
2385                         c->rtrx_timeout = now;
2386                 }
2387
2388                 if(timespec_isset(&c->conn_timeout)) {
2389                         c->conn_timeout = then;
2390                 }
2391
2392                 c->rtt_start.tv_sec = 0;
2393
2394                 if(c->rto > START_RTO) {
2395                         c->rto = START_RTO;
2396                 }
2397         }
2398 }
2399
2400 int utcp_get_user_timeout(struct utcp *u) {
2401         return u ? u->timeout : 0;
2402 }
2403
2404 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2405         if(u) {
2406                 u->timeout = timeout;
2407         }
2408 }
2409
2410 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2411         return c ? c->sndbuf.maxsize : 0;
2412 }
2413
2414 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2415         if(!c) {
2416                 return 0;
2417         }
2418
2419         switch(c->state) {
2420         case SYN_SENT:
2421         case SYN_RECEIVED:
2422         case ESTABLISHED:
2423         case CLOSE_WAIT:
2424                 return buffer_free(&c->sndbuf);
2425
2426         default:
2427                 return 0;
2428         }
2429 }
2430
2431 void utcp_set_sndbuf(struct utcp_connection *c, void *data, size_t size) {
2432         if(!c) {
2433                 return;
2434         }
2435
2436         set_buffer_storage(&c->sndbuf, data, size);
2437
2438         c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2439 }
2440
2441 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2442         return c ? c->rcvbuf.maxsize : 0;
2443 }
2444
2445 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2446         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2447                 return buffer_free(&c->rcvbuf);
2448         } else {
2449                 return 0;
2450         }
2451 }
2452
2453 void utcp_set_rcvbuf(struct utcp_connection *c, void *data, size_t size) {
2454         if(!c) {
2455                 return;
2456         }
2457
2458         set_buffer_storage(&c->rcvbuf, data, size);
2459 }
2460
2461 size_t utcp_get_sendq(struct utcp_connection *c) {
2462         return c->sndbuf.used;
2463 }
2464
2465 size_t utcp_get_recvq(struct utcp_connection *c) {
2466         return c->rcvbuf.used;
2467 }
2468
2469 bool utcp_get_nodelay(struct utcp_connection *c) {
2470         return c ? c->nodelay : false;
2471 }
2472
2473 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2474         if(c) {
2475                 c->nodelay = nodelay;
2476         }
2477 }
2478
2479 bool utcp_get_keepalive(struct utcp_connection *c) {
2480         return c ? c->keepalive : false;
2481 }
2482
2483 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2484         if(c) {
2485                 c->keepalive = keepalive;
2486         }
2487 }
2488
2489 size_t utcp_get_outq(struct utcp_connection *c) {
2490         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2491 }
2492
2493 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2494         if(c) {
2495                 c->recv = recv;
2496         }
2497 }
2498
2499 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2500         if(c) {
2501                 c->poll = poll;
2502                 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2503         }
2504 }
2505
2506 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_listen_t listen) {
2507         if(utcp) {
2508                 utcp->accept = accept;
2509                 utcp->listen = listen;
2510         }
2511 }
2512
2513 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2514         if(!c || c->reapable) {
2515                 return;
2516         }
2517
2518         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2519                 return;
2520         }
2521
2522         if(expect) {
2523                 // If we expect data, start the connection timer.
2524                 if(!timespec_isset(&c->conn_timeout)) {
2525                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2526                         c->conn_timeout.tv_sec += c->utcp->timeout;
2527                 }
2528         } else {
2529                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2530                 if(c->snd.una == c->snd.last) {
2531                         timespec_clear(&c->conn_timeout);
2532                 }
2533         }
2534 }
2535
2536 void utcp_set_flags(struct utcp_connection *c, uint32_t flags) {
2537         c->flags &= ~UTCP_CHANGEABLE_FLAGS;
2538         c->flags |= flags & UTCP_CHANGEABLE_FLAGS;
2539 }
2540
2541 void utcp_offline(struct utcp *utcp, bool offline) {
2542         struct timespec now;
2543         clock_gettime(UTCP_CLOCK, &now);
2544
2545         for(int i = 0; i < utcp->nconnections; i++) {
2546                 struct utcp_connection *c = utcp->connections[i];
2547
2548                 if(c->reapable) {
2549                         continue;
2550                 }
2551
2552                 utcp_expect_data(c, offline);
2553
2554                 if(!offline) {
2555                         if(timespec_isset(&c->rtrx_timeout)) {
2556                                 c->rtrx_timeout = now;
2557                         }
2558
2559                         utcp->connections[i]->rtt_start.tv_sec = 0;
2560
2561                         if(c->rto > START_RTO) {
2562                                 c->rto = START_RTO;
2563                         }
2564                 }
2565         }
2566 }
2567
2568 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t cb) {
2569         utcp->retransmit = cb;
2570 }
2571
2572 void utcp_set_clock_granularity(long granularity) {
2573         CLOCK_GRANULARITY = granularity;
2574 }