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
 895                 c->do_poll = true;
 896         } else {
 897                 return 0;
 898         }
 899
 900         if(len <= 0) {
 901                 if(is_reliable(c)) {
 902                         errno = EWOULDBLOCK;
 903                         return 0;
 904                 } else {
 905                         return len;
 906                 }
 907         }
 908
 909         c->snd.last += len;
 910
 911         // Don't send anything yet if the connection has not fully established yet
 912
 913         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 914                 return len;
 915         }
 916
 917         ack(c, false);
 918
 919         if(!is_reliable(c)) {
 920                 c->snd.una = c->snd.nxt = c->snd.last;
 921                 buffer_discard(&c->sndbuf, c->sndbuf.used);
 922         }
 923
 924         if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
 925                 start_retransmit_timer(c);
 926         }
 927
 928         if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
 929                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 930                 c->conn_timeout.tv_sec += c->utcp->timeout;
 931         }
 932
 933         return len;
 934 }
 935
 936 static void swap_ports(struct hdr *hdr) {
 937         uint16_t tmp = hdr->src;
 938         hdr->src = hdr->dst;
 939         hdr->dst = tmp;
 940 }
 941
 942 static void fast_retransmit(struct utcp_connection *c) {
 943         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 944                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 945                 return;
 946         }
 947
 948         struct utcp *utcp = c->utcp;
 949
 950         struct {
 951                 struct hdr hdr;
 952                 uint8_t data[];
 953         } *pkt = c->utcp->pkt;
 954
 955         pkt->hdr.src = c->src;
 956         pkt->hdr.dst = c->dst;
 957         pkt->hdr.wnd = c->rcvbuf.maxsize;
 958         pkt->hdr.aux = 0;
 959
 960         switch(c->state) {
 961         case ESTABLISHED:
 962         case FIN_WAIT_1:
 963         case CLOSE_WAIT:
 964         case CLOSING:
 965         case LAST_ACK:
 966                 // Send unacked data again.
 967                 pkt->hdr.seq = c->snd.una;
 968                 pkt->hdr.ack = c->rcv.nxt;
 969                 pkt->hdr.ctl = ACK;
 970                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 971
 972                 if(fin_wanted(c, c->snd.una + len)) {
 973                         len--;
 974                         pkt->hdr.ctl |= FIN;
 975                 }
 976
 977                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 978                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 979                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 980                 break;
 981
 982         default:
 983                 break;
 984         }
 985 }
 986
 987 static void retransmit(struct utcp_connection *c) {
 988         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 989                 debug(c, "retransmit() called but nothing to retransmit!\n");
 990                 stop_retransmit_timer(c);
 991                 return;
 992         }
 993
 994         struct utcp *utcp = c->utcp;
 995
 996         if(utcp->retransmit) {
 997                 utcp->retransmit(c);
 998         }
 999
1000         struct {
1001                 struct hdr hdr;
1002                 uint8_t data[];
1003         } *pkt = c->utcp->pkt;
1004
1005         pkt->hdr.src = c->src;
1006         pkt->hdr.dst = c->dst;
1007         pkt->hdr.wnd = c->rcvbuf.maxsize;
1008         pkt->hdr.aux = 0;
1009
1010         switch(c->state) {
1011         case SYN_SENT:
1012                 // Send our SYN again
1013                 pkt->hdr.seq = c->snd.iss;
1014                 pkt->hdr.ack = 0;
1015                 pkt->hdr.ctl = SYN;
1016                 pkt->hdr.aux = 0x0101;
1017                 pkt->data[0] = 1;
1018                 pkt->data[1] = 0;
1019                 pkt->data[2] = 0;
1020                 pkt->data[3] = c->flags & 0x7;
1021                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
1022                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
1023                 break;
1024
1025         case SYN_RECEIVED:
1026                 // Send SYNACK again
1027                 pkt->hdr.seq = c->snd.nxt;
1028                 pkt->hdr.ack = c->rcv.nxt;
1029                 pkt->hdr.ctl = SYN | ACK;
1030                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
1031                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
1032                 break;
1033
1034         case ESTABLISHED:
1035         case FIN_WAIT_1:
1036         case CLOSE_WAIT:
1037         case CLOSING:
1038         case LAST_ACK:
1039                 // Send unacked data again.
1040                 pkt->hdr.seq = c->snd.una;
1041                 pkt->hdr.ack = c->rcv.nxt;
1042                 pkt->hdr.ctl = ACK;
1043                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
1044
1045                 if(fin_wanted(c, c->snd.una + len)) {
1046                         len--;
1047                         pkt->hdr.ctl |= FIN;
1048                 }
1049
1050                 // RFC 5681 slow start after timeout
1051                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1052                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1053                 c->snd.cwnd = utcp->mss;
1054                 debug_cwnd(c);
1055
1056                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
1057                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
1058                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
1059
1060                 c->snd.nxt = c->snd.una + len;
1061                 break;
1062
1063         case CLOSED:
1064         case LISTEN:
1065         case TIME_WAIT:
1066         case FIN_WAIT_2:
1067                 // We shouldn't need to retransmit anything in this state.
1068 #ifdef UTCP_DEBUG
1069                 abort();
1070 #endif
1071                 stop_retransmit_timer(c);
1072                 goto cleanup;
1073         }
1074
1075         start_retransmit_timer(c);
1076         c->rto *= 2;
1077
1078         if(c->rto > MAX_RTO) {
1079                 c->rto = MAX_RTO;
1080         }
1081
1082         c->rtt_start.tv_sec = 0; // invalidate RTT timer
1083         c->dupack = 0; // cancel any ongoing fast recovery
1084
1085 cleanup:
1086         return;
1087 }
1088
1089 /* Update receive buffer and SACK entries after consuming data.
1090  *
1091  * Situation:
1092  *
1093  * |.....0000..1111111111.....22222......3333|
1094  * |---------------^
1095  *
1096  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
1097  * to remove data from the receive buffer. The idea is to substract "len"
1098  * from the offset of all the SACK entries, and then remove/cut down entries
1099  * that are shifted to before the start of the receive buffer.
1100  *
1101  * There are three cases:
1102  * - the SACK entry is after ^, in that case just change the offset.
1103  * - the SACK entry starts before and ends after ^, so we have to
1104  *   change both its offset and size.
1105  * - the SACK entry is completely before ^, in that case delete it.
1106  */
1107 static void sack_consume(struct utcp_connection *c, size_t len) {
1108         debug(c, "sack_consume %lu\n", (unsigned long)len);
1109
1110         if(len > c->rcvbuf.used) {
1111                 debug(c, "all SACK entries consumed\n");
1112                 c->sacks[0].len = 0;
1113                 return;
1114         }
1115
1116         buffer_discard(&c->rcvbuf, len);
1117
1118         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1119                 if(len < c->sacks[i].offset) {
1120                         c->sacks[i].offset -= len;
1121                         i++;
1122                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1123                         c->sacks[i].len -= len - c->sacks[i].offset;
1124                         c->sacks[i].offset = 0;
1125                         i++;
1126                 } else {
1127                         if(i < NSACKS - 1) {
1128                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1129                                 c->sacks[NSACKS - 1].len = 0;
1130                         } else {
1131                                 c->sacks[i].len = 0;
1132                                 break;
1133                         }
1134                 }
1135         }
1136
1137         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1138                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1139         }
1140 }
1141
1142 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1143         debug(c, "out of order packet, offset %u\n", offset);
1144         // Packet loss or reordering occured. Store the data in the buffer.
1145         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1146
1147         if(rxd <= 0) {
1148                 debug(c, "packet outside receive buffer, dropping\n");
1149                 return;
1150         }
1151
1152         if((size_t)rxd < len) {
1153                 debug(c, "packet partially outside receive buffer\n");
1154                 len = rxd;
1155         }
1156
1157         // Make note of where we put it.
1158         for(int i = 0; i < NSACKS; i++) {
1159                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1160                         debug(c, "new SACK entry %d\n", i);
1161                         c->sacks[i].offset = offset;
1162                         c->sacks[i].len = rxd;
1163                         break;
1164                 } else if(offset < c->sacks[i].offset) {
1165                         if(offset + rxd < c->sacks[i].offset) { // insert before
1166                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1167                                         debug(c, "insert SACK entry at %d\n", i);
1168                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1169                                         c->sacks[i].offset = offset;
1170                                         c->sacks[i].len = rxd;
1171                                 } else {
1172                                         debug(c, "SACK entries full, dropping packet\n");
1173                                 }
1174
1175                                 break;
1176                         } else { // merge
1177                                 debug(c, "merge with start of SACK entry at %d\n", i);
1178                                 c->sacks[i].offset = offset;
1179                                 break;
1180                         }
1181                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1182                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1183                                 debug(c, "merge with end of SACK entry at %d\n", i);
1184                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1185                                 // TODO: handle potential merge with next entry
1186                         }
1187
1188                         break;
1189                 }
1190         }
1191
1192         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1193                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1194         }
1195 }
1196
1197 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1198         if(c->recv) {
1199                 ssize_t rxd = c->recv(c, data, len);
1200
1201                 if(rxd != (ssize_t)len) {
1202                         // TODO: handle the application not accepting all data.
1203                         abort();
1204                 }
1205         }
1206
1207         // Check if we can process out-of-order data now.
1208         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1209                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1210
1211                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1212                         size_t offset = len;
1213                         len = c->sacks[0].offset + c->sacks[0].len;
1214                         size_t remainder = len - offset;
1215
1216                         ssize_t rxd = buffer_call(c, &c->rcvbuf, offset, remainder);
1217
1218                         if(rxd != (ssize_t)remainder) {
1219                                 // TODO: handle the application not accepting all data.
1220                                 abort();
1221                         }
1222                 }
1223         }
1224
1225         if(c->rcvbuf.used) {
1226                 sack_consume(c, len);
1227         }
1228
1229         c->rcv.nxt += len;
1230 }
1231
1232 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1233         // Fast path for unfragmented packets
1234         if(!hdr->wnd && !(hdr->ctl & MF)) {
1235                 if(c->recv) {
1236                         c->recv(c, data, len);
1237                 }
1238
1239                 c->rcv.nxt = hdr->seq + len;
1240                 return;
1241         }
1242
1243         // Ensure reassembled packet are not larger than 64 kiB
1244         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1245                 return;
1246         }
1247
1248         // Don't accept out of order fragments
1249         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1250                 return;
1251         }
1252
1253         // Reset the receive buffer for the first fragment
1254         if(!hdr->wnd) {
1255                 buffer_clear(&c->rcvbuf);
1256         }
1257
1258         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1259
1260         if(rxd != (ssize_t)len) {
1261                 return;
1262         }
1263
1264         // Send the packet if it's the final fragment
1265         if(!(hdr->ctl & MF)) {
1266                 buffer_call(c, &c->rcvbuf, 0, hdr->wnd + len);
1267         }
1268
1269         c->rcv.nxt = hdr->seq + len;
1270 }
1271
1272 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1273         if(!is_reliable(c)) {
1274                 handle_unreliable(c, hdr, data, len);
1275                 return;
1276         }
1277
1278         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1279
1280         if(offset) {
1281                 handle_out_of_order(c, offset, data, len);
1282         } else {
1283                 handle_in_order(c, data, len);
1284         }
1285 }
1286
1287
1288 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1289         const uint8_t *ptr = data;
1290
1291         if(!utcp) {
1292                 errno = EFAULT;
1293                 return -1;
1294         }
1295
1296         if(!len) {
1297                 return 0;
1298         }
1299
1300         if(!data) {
1301                 errno = EFAULT;
1302                 return -1;
1303         }
1304
1305         // Drop packets smaller than the header
1306
1307         struct hdr hdr;
1308
1309         if(len < sizeof(hdr)) {
1310                 print_packet(NULL, "recv", data, len);
1311                 errno = EBADMSG;
1312                 return -1;
1313         }
1314
1315         // Make a copy from the potentially unaligned data to a struct hdr
1316
1317         memcpy(&hdr, ptr, sizeof(hdr));
1318
1319         // Try to match the packet to an existing connection
1320
1321         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1322         print_packet(c, "recv", data, len);
1323
1324         // Process the header
1325
1326         ptr += sizeof(hdr);
1327         len -= sizeof(hdr);
1328
1329         // Drop packets with an unknown CTL flag
1330
1331         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1332                 print_packet(NULL, "recv", data, len);
1333                 errno = EBADMSG;
1334                 return -1;
1335         }
1336
1337         // Check for auxiliary headers
1338
1339         const uint8_t *init = NULL;
1340
1341         uint16_t aux = hdr.aux;
1342
1343         while(aux) {
1344                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1345                 uint8_t auxtype = aux & 0xff;
1346
1347                 if(len < auxlen) {
1348                         errno = EBADMSG;
1349                         return -1;
1350                 }
1351
1352                 switch(auxtype) {
1353                 case AUX_INIT:
1354                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1355                                 errno = EBADMSG;
1356                                 return -1;
1357                         }
1358
1359                         init = ptr;
1360                         break;
1361
1362                 default:
1363                         errno = EBADMSG;
1364                         return -1;
1365                 }
1366
1367                 len -= auxlen;
1368                 ptr += auxlen;
1369
1370                 if(!(aux & 0x800)) {
1371                         break;
1372                 }
1373
1374                 if(len < 2) {
1375                         errno = EBADMSG;
1376                         return -1;
1377                 }
1378
1379                 memcpy(&aux, ptr, 2);
1380                 len -= 2;
1381                 ptr += 2;
1382         }
1383
1384         bool has_data = len || (hdr.ctl & (SYN | FIN));
1385
1386         // Is it for a new connection?
1387
1388         if(!c) {
1389                 // Ignore RST packets
1390
1391                 if(hdr.ctl & RST) {
1392                         return 0;
1393                 }
1394
1395                 // Is it a SYN packet and are we LISTENing?
1396
1397                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1398                         // If we don't want to accept it, send a RST back
1399                         if((utcp->listen && !utcp->listen(utcp, hdr.dst))) {
1400                                 len = 1;
1401                                 goto reset;
1402                         }
1403
1404                         // Try to allocate memory, otherwise send a RST back
1405                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1406
1407                         if(!c) {
1408                                 len = 1;
1409                                 goto reset;
1410                         }
1411
1412                         // Parse auxilliary information
1413                         if(init) {
1414                                 if(init[0] < 1) {
1415                                         len = 1;
1416                                         goto reset;
1417                                 }
1418
1419                                 c->flags = init[3] & 0x7;
1420                         } else {
1421                                 c->flags = UTCP_TCP;
1422                         }
1423
1424 synack:
1425                         // Return SYN+ACK, go to SYN_RECEIVED state
1426                         c->snd.wnd = hdr.wnd;
1427                         c->rcv.irs = hdr.seq;
1428                         c->rcv.nxt = c->rcv.irs + 1;
1429                         set_state(c, SYN_RECEIVED);
1430
1431                         struct {
1432                                 struct hdr hdr;
1433                                 uint8_t data[4];
1434                         } pkt;
1435
1436                         pkt.hdr.src = c->src;
1437                         pkt.hdr.dst = c->dst;
1438                         pkt.hdr.ack = c->rcv.irs + 1;
1439                         pkt.hdr.seq = c->snd.iss;
1440                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1441                         pkt.hdr.ctl = SYN | ACK;
1442
1443                         if(init) {
1444                                 pkt.hdr.aux = 0x0101;
1445                                 pkt.data[0] = 1;
1446                                 pkt.data[1] = 0;
1447                                 pkt.data[2] = 0;
1448                                 pkt.data[3] = c->flags & 0x7;
1449                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1450                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1451                         } else {
1452                                 pkt.hdr.aux = 0;
1453                                 print_packet(c, "send", &pkt, sizeof(hdr));
1454                                 utcp->send(utcp, &pkt, sizeof(hdr));
1455                         }
1456
1457                         start_retransmit_timer(c);
1458                 } else {
1459                         // No, we don't want your packets, send a RST back
1460                         len = 1;
1461                         goto reset;
1462                 }
1463
1464                 return 0;
1465         }
1466
1467         debug(c, "state %s\n", strstate[c->state]);
1468
1469         // In case this is for a CLOSED connection, ignore the packet.
1470         // TODO: make it so incoming packets can never match a CLOSED connection.
1471
1472         if(c->state == CLOSED) {
1473                 debug(c, "got packet for closed connection\n");
1474                 goto reset;
1475         }
1476
1477         // It is for an existing connection.
1478
1479         // 1. Drop invalid packets.
1480
1481         // 1a. Drop packets that should not happen in our current state.
1482
1483         switch(c->state) {
1484         case SYN_SENT:
1485         case SYN_RECEIVED:
1486         case ESTABLISHED:
1487         case FIN_WAIT_1:
1488         case FIN_WAIT_2:
1489         case CLOSE_WAIT:
1490         case CLOSING:
1491         case LAST_ACK:
1492         case TIME_WAIT:
1493                 break;
1494
1495         default:
1496 #ifdef UTCP_DEBUG
1497                 abort();
1498 #endif
1499                 break;
1500         }
1501
1502         // 1b. Discard data that is not in our receive window.
1503
1504         if(is_reliable(c)) {
1505                 bool acceptable;
1506
1507                 if(c->state == SYN_SENT) {
1508                         acceptable = true;
1509                 } else if(len == 0) {
1510                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1511                 } else {
1512                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1513
1514                         // cut already accepted front overlapping
1515                         if(rcv_offset < 0) {
1516                                 acceptable = len > (size_t) - rcv_offset;
1517
1518                                 if(acceptable) {
1519                                         ptr -= rcv_offset;
1520                                         len += rcv_offset;
1521                                         hdr.seq -= rcv_offset;
1522                                 }
1523                         } else {
1524                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1525                         }
1526                 }
1527
1528                 if(!acceptable) {
1529                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1530
1531                         // Ignore unacceptable RST packets.
1532                         if(hdr.ctl & RST) {
1533                                 return 0;
1534                         }
1535
1536                         // Otherwise, continue processing.
1537                         len = 0;
1538                 }
1539         } else {
1540 #if UTCP_DEBUG
1541                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1542
1543                 if(rcv_offset) {
1544                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1545                 }
1546
1547 #endif
1548         }
1549
1550         c->snd.wnd = hdr.wnd; // TODO: move below
1551
1552         // 1c. Drop packets with an invalid ACK.
1553         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1554         // (= snd.una + c->sndbuf.used).
1555
1556         if(!is_reliable(c)) {
1557                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1558                         hdr.ack = c->snd.una;
1559                 }
1560         }
1561
1562         // 2. Handle RST packets
1563
1564         if(hdr.ctl & RST) {
1565                 switch(c->state) {
1566                 case SYN_SENT:
1567                         if(!(hdr.ctl & ACK)) {
1568                                 return 0;
1569                         }
1570
1571                         // The peer has refused our connection.
1572                         set_state(c, CLOSED);
1573                         errno = ECONNREFUSED;
1574
1575                         if(c->recv) {
1576                                 c->recv(c, NULL, 0);
1577                         }
1578
1579                         if(c->poll && !c->reapable) {
1580                                 c->poll(c, 0);
1581                         }
1582
1583                         return 0;
1584
1585                 case SYN_RECEIVED:
1586                         if(hdr.ctl & ACK) {
1587                                 return 0;
1588                         }
1589
1590                         // We haven't told the application about this connection yet. Silently delete.
1591                         free_connection(c);
1592                         return 0;
1593
1594                 case ESTABLISHED:
1595                 case FIN_WAIT_1:
1596                 case FIN_WAIT_2:
1597                 case CLOSE_WAIT:
1598                         if(hdr.ctl & ACK) {
1599                                 return 0;
1600                         }
1601
1602                         // The peer has aborted our connection.
1603                         set_state(c, CLOSED);
1604                         errno = ECONNRESET;
1605                         buffer_clear(&c->sndbuf);
1606                         buffer_clear(&c->rcvbuf);
1607
1608                         if(c->recv) {
1609                                 c->recv(c, NULL, 0);
1610                         }
1611
1612                         if(c->poll && !c->reapable) {
1613                                 c->poll(c, 0);
1614                         }
1615
1616                         return 0;
1617
1618                 case CLOSING:
1619                 case LAST_ACK:
1620                 case TIME_WAIT:
1621                         if(hdr.ctl & ACK) {
1622                                 return 0;
1623                         }
1624
1625                         // As far as the application is concerned, the connection has already been closed.
1626                         // If it has called utcp_close() already, we can immediately free this connection.
1627                         if(c->reapable) {
1628                                 free_connection(c);
1629                                 return 0;
1630                         }
1631
1632                         // Otherwise, immediately move to the CLOSED state.
1633                         set_state(c, CLOSED);
1634                         return 0;
1635
1636                 default:
1637 #ifdef UTCP_DEBUG
1638                         abort();
1639 #endif
1640                         break;
1641                 }
1642         }
1643
1644         uint32_t advanced;
1645
1646         if(!(hdr.ctl & ACK)) {
1647                 advanced = 0;
1648                 goto skip_ack;
1649         }
1650
1651         // 3. Advance snd.una
1652
1653         if(seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0) {
1654                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1655                 goto reset;
1656         }
1657
1658         advanced = seqdiff(hdr.ack, c->snd.una);
1659
1660         if(advanced) {
1661                 // RTT measurement
1662                 if(c->rtt_start.tv_sec) {
1663                         if(c->rtt_seq == hdr.ack) {
1664                                 struct timespec now;
1665                                 clock_gettime(UTCP_CLOCK, &now);
1666                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1667                                 update_rtt(c, diff);
1668                                 c->rtt_start.tv_sec = 0;
1669                         } else if(c->rtt_seq < hdr.ack) {
1670                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1671                                 c->rtt_start.tv_sec = 0;
1672                         }
1673                 }
1674
1675                 int32_t data_acked = advanced;
1676
1677                 switch(c->state) {
1678                 case SYN_SENT:
1679                 case SYN_RECEIVED:
1680                         data_acked--;
1681                         break;
1682
1683                 // TODO: handle FIN as well.
1684                 default:
1685                         break;
1686                 }
1687
1688                 assert(data_acked >= 0);
1689
1690 #ifndef NDEBUG
1691                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1692                 assert(data_acked <= bufused);
1693 #endif
1694
1695                 if(data_acked) {
1696                         buffer_discard(&c->sndbuf, data_acked);
1697
1698                         if(is_reliable(c)) {
1699                                 c->do_poll = true;
1700                         }
1701                 }
1702
1703                 // Also advance snd.nxt if possible
1704                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1705                         c->snd.nxt = hdr.ack;
1706                 }
1707
1708                 c->snd.una = hdr.ack;
1709
1710                 if(c->dupack) {
1711                         if(c->dupack >= 3) {
1712                                 debug(c, "fast recovery ended\n");
1713                                 c->snd.cwnd = c->snd.ssthresh;
1714                         }
1715
1716                         c->dupack = 0;
1717                 }
1718
1719                 // Increase the congestion window according to RFC 5681
1720                 if(c->snd.cwnd < c->snd.ssthresh) {
1721                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1722                 } else {
1723                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1724                 }
1725
1726                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1727                         c->snd.cwnd = c->sndbuf.maxsize;
1728                 }
1729
1730                 debug_cwnd(c);
1731
1732                 // Check if we have sent a FIN that is now ACKed.
1733                 switch(c->state) {
1734                 case FIN_WAIT_1:
1735                         if(c->snd.una == c->snd.last) {
1736                                 set_state(c, FIN_WAIT_2);
1737                         }
1738
1739                         break;
1740
1741                 case CLOSING:
1742                         if(c->snd.una == c->snd.last) {
1743                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1744                                 c->conn_timeout.tv_sec += utcp->timeout;
1745                                 set_state(c, TIME_WAIT);
1746                         }
1747
1748                         break;
1749
1750                 default:
1751                         break;
1752                 }
1753         } else {
1754                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1755                         c->dupack++;
1756                         debug(c, "duplicate ACK %d\n", c->dupack);
1757
1758                         if(c->dupack == 3) {
1759                                 // RFC 5681 fast recovery
1760                                 debug(c, "fast recovery started\n", c->dupack);
1761                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1762                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1763                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1764
1765                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1766                                         c->snd.cwnd = c->sndbuf.maxsize;
1767                                 }
1768
1769                                 debug_cwnd(c);
1770
1771                                 fast_retransmit(c);
1772                         } else if(c->dupack > 3) {
1773                                 c->snd.cwnd += utcp->mss;
1774
1775                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1776                                         c->snd.cwnd = c->sndbuf.maxsize;
1777                                 }
1778
1779                                 debug_cwnd(c);
1780                         }
1781
1782                         // We got an ACK which indicates the other side did get one of our packets.
1783                         // Reset the retransmission timer to avoid going to slow start,
1784                         // but don't touch the connection timeout.
1785                         start_retransmit_timer(c);
1786                 }
1787         }
1788
1789         // 4. Update timers
1790
1791         if(advanced) {
1792                 if(c->snd.una == c->snd.last) {
1793                         stop_retransmit_timer(c);
1794                         timespec_clear(&c->conn_timeout);
1795                 } else if(is_reliable(c)) {
1796                         start_retransmit_timer(c);
1797                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1798                         c->conn_timeout.tv_sec += utcp->timeout;
1799                 }
1800         }
1801
1802 skip_ack:
1803         // 5. Process SYN stuff
1804
1805         if(hdr.ctl & SYN) {
1806                 switch(c->state) {
1807                 case SYN_SENT:
1808
1809                         // This is a SYNACK. It should always have ACKed the SYN.
1810                         if(!advanced) {
1811                                 goto reset;
1812                         }
1813
1814                         c->rcv.irs = hdr.seq;
1815                         c->rcv.nxt = hdr.seq + 1;
1816
1817                         if(c->shut_wr) {
1818                                 c->snd.last++;
1819                                 set_state(c, FIN_WAIT_1);
1820                         } else {
1821                                 c->do_poll = true;
1822                                 set_state(c, ESTABLISHED);
1823                         }
1824
1825                         break;
1826
1827                 case SYN_RECEIVED:
1828                         // This is a retransmit of a SYN, send back the SYNACK.
1829                         goto synack;
1830
1831                 case ESTABLISHED:
1832                 case FIN_WAIT_1:
1833                 case FIN_WAIT_2:
1834                 case CLOSE_WAIT:
1835                 case CLOSING:
1836                 case LAST_ACK:
1837                 case TIME_WAIT:
1838                         // This could be a retransmission. Ignore the SYN flag, but send an ACK back.
1839                         break;
1840
1841                 default:
1842 #ifdef UTCP_DEBUG
1843                         abort();
1844 #endif
1845                         return 0;
1846                 }
1847         }
1848
1849         // 6. Process new data
1850
1851         if(c->state == SYN_RECEIVED) {
1852                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1853                 if(!advanced) {
1854                         goto reset;
1855                 }
1856
1857                 // Are we still LISTENing?
1858                 if(utcp->accept) {
1859                         utcp->accept(c, c->src);
1860                 }
1861
1862                 if(c->state != ESTABLISHED) {
1863                         set_state(c, CLOSED);
1864                         c->reapable = true;
1865                         goto reset;
1866                 }
1867         }
1868
1869         if(len) {
1870                 switch(c->state) {
1871                 case SYN_SENT:
1872                 case SYN_RECEIVED:
1873                         // This should never happen.
1874 #ifdef UTCP_DEBUG
1875                         abort();
1876 #endif
1877                         return 0;
1878
1879                 case ESTABLISHED:
1880                         break;
1881
1882                 case FIN_WAIT_1:
1883                 case FIN_WAIT_2:
1884                         if(c->reapable) {
1885                                 // We already closed the connection and are not interested in more data.
1886                                 goto reset;
1887                         }
1888
1889                         break;
1890
1891                 case CLOSE_WAIT:
1892                 case CLOSING:
1893                 case LAST_ACK:
1894                 case TIME_WAIT:
1895                         // Ehm no, We should never receive more data after a FIN.
1896                         goto reset;
1897
1898                 default:
1899 #ifdef UTCP_DEBUG
1900                         abort();
1901 #endif
1902                         return 0;
1903                 }
1904
1905                 handle_incoming_data(c, &hdr, ptr, len);
1906         }
1907
1908         // 7. Process FIN stuff
1909
1910         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1911                 switch(c->state) {
1912                 case SYN_SENT:
1913                 case SYN_RECEIVED:
1914                         // This should never happen.
1915 #ifdef UTCP_DEBUG
1916                         abort();
1917 #endif
1918                         break;
1919
1920                 case ESTABLISHED:
1921                         set_state(c, CLOSE_WAIT);
1922                         break;
1923
1924                 case FIN_WAIT_1:
1925                         set_state(c, CLOSING);
1926                         break;
1927
1928                 case FIN_WAIT_2:
1929                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1930                         c->conn_timeout.tv_sec += utcp->timeout;
1931                         set_state(c, TIME_WAIT);
1932                         break;
1933
1934                 case CLOSE_WAIT:
1935                 case CLOSING:
1936                 case LAST_ACK:
1937                 case TIME_WAIT:
1938                         // Ehm, no. We should never receive a second FIN.
1939                         goto reset;
1940
1941                 default:
1942 #ifdef UTCP_DEBUG
1943                         abort();
1944 #endif
1945                         break;
1946                 }
1947
1948                 // FIN counts as one sequence number
1949                 c->rcv.nxt++;
1950                 len++;
1951
1952                 // Inform the application that the peer closed its end of the connection.
1953                 if(c->recv) {
1954                         errno = 0;
1955                         c->recv(c, NULL, 0);
1956                 }
1957         }
1958
1959         // Now we send something back if:
1960         // - we received data, so we have to send back an ACK
1961         //   -> sendatleastone = true
1962         // - or we got an ack, so we should maybe send a bit more data
1963         //   -> sendatleastone = false
1964
1965         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1966                 ack(c, has_data);
1967         }
1968
1969         return 0;
1970
1971 reset:
1972         swap_ports(&hdr);
1973         hdr.wnd = 0;
1974         hdr.aux = 0;
1975
1976         if(hdr.ctl & ACK) {
1977                 hdr.seq = hdr.ack;
1978                 hdr.ctl = RST;
1979         } else {
1980                 hdr.ack = hdr.seq + len;
1981                 hdr.seq = 0;
1982                 hdr.ctl = RST | ACK;
1983         }
1984
1985         print_packet(c, "send", &hdr, sizeof(hdr));
1986         utcp->send(utcp, &hdr, sizeof(hdr));
1987         return 0;
1988
1989 }
1990
1991 int utcp_shutdown(struct utcp_connection *c, int dir) {
1992         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1993
1994         if(!c) {
1995                 errno = EFAULT;
1996                 return -1;
1997         }
1998
1999         if(c->reapable) {
2000                 debug(c, "shutdown() called on closed connection\n");
2001                 errno = EBADF;
2002                 return -1;
2003         }
2004
2005         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
2006                 errno = EINVAL;
2007                 return -1;
2008         }
2009
2010         // TCP does not have a provision for stopping incoming packets.
2011         // The best we can do is to just ignore them.
2012         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
2013                 c->recv = NULL;
2014         }
2015
2016         // The rest of the code deals with shutting down writes.
2017         if(dir == UTCP_SHUT_RD) {
2018                 return 0;
2019         }
2020
2021         // Only process shutting down writes once.
2022         if(c->shut_wr) {
2023                 return 0;
2024         }
2025
2026         c->shut_wr = true;
2027
2028         switch(c->state) {
2029         case CLOSED:
2030         case LISTEN:
2031                 errno = ENOTCONN;
2032                 return -1;
2033
2034         case SYN_SENT:
2035                 return 0;
2036
2037         case SYN_RECEIVED:
2038         case ESTABLISHED:
2039                 set_state(c, FIN_WAIT_1);
2040                 break;
2041
2042         case FIN_WAIT_1:
2043         case FIN_WAIT_2:
2044                 return 0;
2045
2046         case CLOSE_WAIT:
2047                 set_state(c, CLOSING);
2048                 break;
2049
2050         case CLOSING:
2051         case LAST_ACK:
2052         case TIME_WAIT:
2053                 return 0;
2054         }
2055
2056         c->snd.last++;
2057
2058         ack(c, false);
2059
2060         if(!timespec_isset(&c->rtrx_timeout)) {
2061                 start_retransmit_timer(c);
2062         }
2063
2064         return 0;
2065 }
2066
2067 static bool reset_connection(struct utcp_connection *c) {
2068         if(!c) {
2069                 errno = EFAULT;
2070                 return false;
2071         }
2072
2073         if(c->reapable) {
2074                 debug(c, "abort() called on closed connection\n");
2075                 errno = EBADF;
2076                 return false;
2077         }
2078
2079         buffer_clear(&c->sndbuf);
2080         buffer_clear(&c->rcvbuf);
2081
2082         switch(c->state) {
2083         case CLOSED:
2084                 return true;
2085
2086         case LISTEN:
2087         case SYN_SENT:
2088         case CLOSING:
2089         case LAST_ACK:
2090         case TIME_WAIT:
2091                 set_state(c, CLOSED);
2092                 return true;
2093
2094         case SYN_RECEIVED:
2095         case ESTABLISHED:
2096         case FIN_WAIT_1:
2097         case FIN_WAIT_2:
2098         case CLOSE_WAIT:
2099                 set_state(c, CLOSED);
2100                 break;
2101         }
2102
2103         // Send RST
2104
2105         struct hdr hdr;
2106
2107         hdr.src = c->src;
2108         hdr.dst = c->dst;
2109         hdr.seq = c->snd.nxt;
2110         hdr.ack = c->rcv.nxt;
2111         hdr.wnd = 0;
2112         hdr.ctl = RST;
2113         hdr.aux = 0;
2114
2115         print_packet(c, "send", &hdr, sizeof(hdr));
2116         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2117         return true;
2118 }
2119
2120 static void set_reapable(struct utcp_connection *c) {
2121         set_buffer_storage(&c->sndbuf, NULL, min(c->sndbuf.maxsize, DEFAULT_MAXSNDBUFSIZE));
2122         set_buffer_storage(&c->rcvbuf, NULL, min(c->rcvbuf.maxsize, DEFAULT_MAXRCVBUFSIZE));
2123
2124         c->recv = NULL;
2125         c->poll = NULL;
2126         c->reapable = true;
2127 }
2128
2129 // Resets all connections, but does not invalidate connection handles
2130 void utcp_reset_all_connections(struct utcp *utcp) {
2131         if(!utcp) {
2132                 errno = EINVAL;
2133                 return;
2134         }
2135
2136         for(int i = 0; i < utcp->nconnections; i++) {
2137                 struct utcp_connection *c = utcp->connections[i];
2138
2139                 if(c->reapable || c->state == CLOSED) {
2140                         continue;
2141                 }
2142
2143                 reset_connection(c);
2144
2145                 if(c->recv) {
2146                         errno = 0;
2147                         c->recv(c, NULL, 0);
2148                 }
2149
2150                 if(c->poll && !c->reapable) {
2151                         errno = 0;
2152                         c->poll(c, 0);
2153                 }
2154         }
2155
2156         return;
2157 }
2158
2159 int utcp_close(struct utcp_connection *c) {
2160         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2161                 return -1;
2162         }
2163
2164         set_reapable(c);
2165         return 0;
2166 }
2167
2168 int utcp_abort(struct utcp_connection *c) {
2169         if(!reset_connection(c)) {
2170                 return -1;
2171         }
2172
2173         set_reapable(c);
2174         return 0;
2175 }
2176
2177 /* Handle timeouts.
2178  * One call to this function will loop through all connections,
2179  * checking if something needs to be resent or not.
2180  * The return value is the time to the next timeout in milliseconds,
2181  * or maybe a negative value if the timeout is infinite.
2182  */
2183 struct timespec utcp_timeout(struct utcp *utcp) {
2184         struct timespec now;
2185         clock_gettime(UTCP_CLOCK, &now);
2186         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2187
2188         for(int i = 0; i < utcp->nconnections; i++) {
2189                 struct utcp_connection *c = utcp->connections[i];
2190
2191                 if(!c) {
2192                         continue;
2193                 }
2194
2195                 // delete connections that have been utcp_close()d.
2196                 if(c->state == CLOSED) {
2197                         if(c->reapable) {
2198                                 debug(c, "reaping\n");
2199                                 free_connection(c);
2200                                 i--;
2201                         }
2202
2203                         continue;
2204                 }
2205
2206                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2207                         errno = ETIMEDOUT;
2208                         c->state = CLOSED;
2209                         buffer_clear(&c->sndbuf);
2210                         buffer_clear(&c->rcvbuf);
2211
2212                         if(c->recv) {
2213                                 c->recv(c, NULL, 0);
2214                         }
2215
2216                         if(c->poll && !c->reapable) {
2217                                 c->poll(c, 0);
2218                         }
2219
2220                         continue;
2221                 }
2222
2223                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2224                         debug(c, "retransmitting after timeout\n");
2225                         retransmit(c);
2226                 }
2227
2228                 if(c->poll) {
2229                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2230                                 c->do_poll = false;
2231                                 uint32_t len = buffer_free(&c->sndbuf);
2232
2233                                 if(len) {
2234                                         c->poll(c, len);
2235                                 }
2236                         } else if(c->state == CLOSED) {
2237                                 c->poll(c, 0);
2238                         }
2239                 }
2240
2241                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2242                         next = c->conn_timeout;
2243                 }
2244
2245                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2246                         next = c->rtrx_timeout;
2247                 }
2248         }
2249
2250         struct timespec diff;
2251
2252         timespec_sub(&next, &now, &diff);
2253
2254         return diff;
2255 }
2256
2257 bool utcp_is_active(struct utcp *utcp) {
2258         if(!utcp) {
2259                 return false;
2260         }
2261
2262         for(int i = 0; i < utcp->nconnections; i++)
2263                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2264                         return true;
2265                 }
2266
2267         return false;
2268 }
2269
2270 struct utcp *utcp_init(utcp_accept_t accept, utcp_listen_t listen, utcp_send_t send, void *priv) {
2271         if(!send) {
2272                 errno = EFAULT;
2273                 return NULL;
2274         }
2275
2276         struct utcp *utcp = calloc(1, sizeof(*utcp));
2277
2278         if(!utcp) {
2279                 return NULL;
2280         }
2281
2282         utcp_set_mtu(utcp, DEFAULT_MTU);
2283
2284         if(!utcp->pkt) {
2285                 free(utcp);
2286                 return NULL;
2287         }
2288
2289         if(!CLOCK_GRANULARITY) {
2290                 struct timespec res;
2291                 clock_getres(UTCP_CLOCK, &res);
2292                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2293         }
2294
2295         utcp->accept = accept;
2296         utcp->listen = listen;
2297         utcp->send = send;
2298         utcp->priv = priv;
2299         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2300
2301         return utcp;
2302 }
2303
2304 void utcp_exit(struct utcp *utcp) {
2305         if(!utcp) {
2306                 return;
2307         }
2308
2309         for(int i = 0; i < utcp->nconnections; i++) {
2310                 struct utcp_connection *c = utcp->connections[i];
2311
2312                 if(!c->reapable) {
2313                         buffer_clear(&c->sndbuf);
2314                         buffer_clear(&c->rcvbuf);
2315
2316                         if(c->recv) {
2317                                 c->recv(c, NULL, 0);
2318                         }
2319
2320                         if(c->poll && !c->reapable) {
2321                                 c->poll(c, 0);
2322                         }
2323                 }
2324
2325                 buffer_exit(&c->rcvbuf);
2326                 buffer_exit(&c->sndbuf);
2327                 free(c);
2328         }
2329
2330         free(utcp->connections);
2331         free(utcp->pkt);
2332         free(utcp);
2333 }
2334
2335 uint16_t utcp_get_mtu(struct utcp *utcp) {
2336         return utcp ? utcp->mtu : 0;
2337 }
2338
2339 uint16_t utcp_get_mss(struct utcp *utcp) {
2340         return utcp ? utcp->mss : 0;
2341 }
2342
2343 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2344         if(!utcp) {
2345                 return;
2346         }
2347
2348         if(mtu <= sizeof(struct hdr)) {
2349                 return;
2350         }
2351
2352         if(mtu > utcp->mtu) {
2353                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2354
2355                 if(!new) {
2356                         return;
2357                 }
2358
2359                 utcp->pkt = new;
2360         }
2361
2362         utcp->mtu = mtu;
2363         utcp->mss = mtu - sizeof(struct hdr);
2364 }
2365
2366 void utcp_reset_timers(struct utcp *utcp) {
2367         if(!utcp) {
2368                 return;
2369         }
2370
2371         struct timespec now, then;
2372
2373         clock_gettime(UTCP_CLOCK, &now);
2374
2375         then = now;
2376
2377         then.tv_sec += utcp->timeout;
2378
2379         for(int i = 0; i < utcp->nconnections; i++) {
2380                 struct utcp_connection *c = utcp->connections[i];
2381
2382                 if(c->reapable) {
2383                         continue;
2384                 }
2385
2386                 if(timespec_isset(&c->rtrx_timeout)) {
2387                         c->rtrx_timeout = now;
2388                 }
2389
2390                 if(timespec_isset(&c->conn_timeout)) {
2391                         c->conn_timeout = then;
2392                 }
2393
2394                 c->rtt_start.tv_sec = 0;
2395
2396                 if(c->rto > START_RTO) {
2397                         c->rto = START_RTO;
2398                 }
2399         }
2400 }
2401
2402 int utcp_get_user_timeout(struct utcp *u) {
2403         return u ? u->timeout : 0;
2404 }
2405
2406 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2407         if(u) {
2408                 u->timeout = timeout;
2409         }
2410 }
2411
2412 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2413         return c ? c->sndbuf.maxsize : 0;
2414 }
2415
2416 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2417         if(!c) {
2418                 return 0;
2419         }
2420
2421         switch(c->state) {
2422         case SYN_SENT:
2423         case SYN_RECEIVED:
2424         case ESTABLISHED:
2425         case CLOSE_WAIT:
2426                 return buffer_free(&c->sndbuf);
2427
2428         default:
2429                 return 0;
2430         }
2431 }
2432
2433 void utcp_set_sndbuf(struct utcp_connection *c, void *data, size_t size) {
2434         if(!c) {
2435                 return;
2436         }
2437
2438         set_buffer_storage(&c->sndbuf, data, size);
2439
2440         c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2441 }
2442
2443 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2444         return c ? c->rcvbuf.maxsize : 0;
2445 }
2446
2447 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2448         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2449                 return buffer_free(&c->rcvbuf);
2450         } else {
2451                 return 0;
2452         }
2453 }
2454
2455 void utcp_set_rcvbuf(struct utcp_connection *c, void *data, size_t size) {
2456         if(!c) {
2457                 return;
2458         }
2459
2460         set_buffer_storage(&c->rcvbuf, data, size);
2461 }
2462
2463 size_t utcp_get_sendq(struct utcp_connection *c) {
2464         return c->sndbuf.used;
2465 }
2466
2467 size_t utcp_get_recvq(struct utcp_connection *c) {
2468         return c->rcvbuf.used;
2469 }
2470
2471 bool utcp_get_nodelay(struct utcp_connection *c) {
2472         return c ? c->nodelay : false;
2473 }
2474
2475 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2476         if(c) {
2477                 c->nodelay = nodelay;
2478         }
2479 }
2480
2481 bool utcp_get_keepalive(struct utcp_connection *c) {
2482         return c ? c->keepalive : false;
2483 }
2484
2485 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2486         if(c) {
2487                 c->keepalive = keepalive;
2488         }
2489 }
2490
2491 size_t utcp_get_outq(struct utcp_connection *c) {
2492         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2493 }
2494
2495 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2496         if(c) {
2497                 c->recv = recv;
2498         }
2499 }
2500
2501 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2502         if(c) {
2503                 c->poll = poll;
2504                 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2505         }
2506 }
2507
2508 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_listen_t listen) {
2509         if(utcp) {
2510                 utcp->accept = accept;
2511                 utcp->listen = listen;
2512         }
2513 }
2514
2515 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2516         if(!c || c->reapable) {
2517                 return;
2518         }
2519
2520         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2521                 return;
2522         }
2523
2524         if(expect) {
2525                 // If we expect data, start the connection timer.
2526                 if(!timespec_isset(&c->conn_timeout)) {
2527                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2528                         c->conn_timeout.tv_sec += c->utcp->timeout;
2529                 }
2530         } else {
2531                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2532                 if(c->snd.una == c->snd.last) {
2533                         timespec_clear(&c->conn_timeout);
2534                 }
2535         }
2536 }
2537
2538 void utcp_set_flags(struct utcp_connection *c, uint32_t flags) {
2539         c->flags &= ~UTCP_CHANGEABLE_FLAGS;
2540         c->flags |= flags & UTCP_CHANGEABLE_FLAGS;
2541 }
2542
2543 void utcp_offline(struct utcp *utcp, bool offline) {
2544         struct timespec now;
2545         clock_gettime(UTCP_CLOCK, &now);
2546
2547         for(int i = 0; i < utcp->nconnections; i++) {
2548                 struct utcp_connection *c = utcp->connections[i];
2549
2550                 if(c->reapable) {
2551                         continue;
2552                 }
2553
2554                 utcp_expect_data(c, offline);
2555
2556                 if(!offline) {
2557                         if(timespec_isset(&c->rtrx_timeout)) {
2558                                 c->rtrx_timeout = now;
2559                         }
2560
2561                         utcp->connections[i]->rtt_start.tv_sec = 0;
2562
2563                         if(c->rto > START_RTO) {
2564                                 c->rto = START_RTO;
2565                         }
2566                 }
2567         }
2568 }
2569
2570 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t cb) {
2571         utcp->retransmit = cb;
2572 }
2573
2574 void utcp_set_clock_granularity(long granularity) {
2575         CLOCK_GRANULARITY = granularity;
2576 }