git.meshlink.io Git - utcp/blob - utcp.c

   1 /*
   2     utcp.c -- Userspace TCP
   3     Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
   4
   5     This program is free software; you can redistribute it and/or modify
   6     it under the terms of the GNU General Public License as published by
   7     the Free Software Foundation; either version 2 of the License, or
   8     (at your option) any later version.
   9
  10     This program is distributed in the hope that it will be useful,
  11     but WITHOUT ANY WARRANTY; without even the implied warranty of
  12     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13     GNU General Public License for more details.
  14
  15     You should have received a copy of the GNU General Public License along
  16     with this program; if not, write to the Free Software Foundation, Inc.,
  17     51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18 */
  19
  20 #define _GNU_SOURCE
  21
  22 #include <assert.h>
  23 #include <errno.h>
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <stdint.h>
  27 #include <stdbool.h>
  28 #include <string.h>
  29 #include <unistd.h>
  30 #include <time.h>
  31
  32 #include "utcp_priv.h"
  33
  34 #ifndef EBADMSG
  35 #define EBADMSG         104
  36 #endif
  37
  38 #ifndef SHUT_RDWR
  39 #define SHUT_RDWR 2
  40 #endif
  41
  42 #ifdef poll
  43 #undef poll
  44 #endif
  45
  46 #ifndef UTCP_CLOCK
  47 #if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
  48 #define UTCP_CLOCK CLOCK_MONOTONIC_RAW
  49 #else
  50 #define UTCP_CLOCK CLOCK_MONOTONIC
  51 #endif
  52 #endif
  53
  54 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
  55         r->tv_sec = a->tv_sec - b->tv_sec;
  56         r->tv_nsec = a->tv_nsec - b->tv_nsec;
  57
  58         if(r->tv_nsec < 0) {
  59                 r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
  60         }
  61 }
  62
  63 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
  64         return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
  65 }
  66
  67 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
  68         if(a->tv_sec == b->tv_sec) {
  69                 return a->tv_nsec < b->tv_nsec;
  70         } else {
  71                 return a->tv_sec < b->tv_sec;
  72         }
  73 }
  74
  75 static void timespec_clear(struct timespec *a) {
  76         a->tv_sec = 0;
  77         a->tv_nsec = 0;
  78 }
  79
  80 static bool timespec_isset(const struct timespec *a) {
  81         return a->tv_sec;
  82 }
  83
  84 static long CLOCK_GRANULARITY; // usec
  85
  86 static inline size_t min(size_t a, size_t b) {
  87         return a < b ? a : b;
  88 }
  89
  90 static inline size_t max(size_t a, size_t b) {
  91         return a > b ? a : b;
  92 }
  93
  94 #ifdef UTCP_DEBUG
  95 #include <stdarg.h>
  96
  97 #ifndef UTCP_DEBUG_DATALEN
  98 #define UTCP_DEBUG_DATALEN 20
  99 #endif
 100
 101 static void debug(struct utcp_connection *c, const char *format, ...) {
 102         struct timespec tv;
 103         char buf[1024];
 104         int len;
 105
 106         clock_gettime(CLOCK_REALTIME, &tv);
 107         len = snprintf(buf, sizeof(buf), "%ld.%06lu %u:%u ", (long)tv.tv_sec, tv.tv_nsec / 1000, c ? c->src : 0, c ? c->dst : 0);
 108         va_list ap;
 109         va_start(ap, format);
 110         len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
 111         va_end(ap);
 112
 113         if(len > 0 && (size_t)len < sizeof(buf)) {
 114                 fwrite(buf, len, 1, stderr);
 115         }
 116 }
 117
 118 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
 119         struct hdr hdr;
 120
 121         if(len < sizeof(hdr)) {
 122                 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
 123                 return;
 124         }
 125
 126         memcpy(&hdr, pkt, sizeof(hdr));
 127
 128         uint32_t datalen;
 129
 130         if(len > sizeof(hdr)) {
 131                 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
 132         } else {
 133                 datalen = 0;
 134         }
 135
 136
 137         const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 138         char str[datalen * 2 + 1];
 139         char *p = str;
 140
 141         for(uint32_t i = 0; i < datalen; i++) {
 142                 *p++ = "0123456789ABCDEF"[data[i] >> 4];
 143                 *p++ = "0123456789ABCDEF"[data[i] & 15];
 144         }
 145
 146         *p = 0;
 147
 148         debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s%s data %s\n",
 149               dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
 150               hdr.ctl & SYN ? "SYN" : "",
 151               hdr.ctl & RST ? "RST" : "",
 152               hdr.ctl & FIN ? "FIN" : "",
 153               hdr.ctl & ACK ? "ACK" : "",
 154               hdr.ctl & MF ? "MF" : "",
 155               str
 156              );
 157 }
 158
 159 static void debug_cwnd(struct utcp_connection *c) {
 160         debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
 161 }
 162 #else
 163 #define debug(...) do {} while(0)
 164 #define print_packet(...) do {} while(0)
 165 #define debug_cwnd(...) do {} while(0)
 166 #endif
 167
 168 static void set_state(struct utcp_connection *c, enum state state) {
 169         c->state = state;
 170
 171         if(state == ESTABLISHED) {
 172                 timespec_clear(&c->conn_timeout);
 173         }
 174
 175         debug(c, "state %s\n", strstate[state]);
 176 }
 177
 178 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 179         if(seq != c->snd.last) {
 180                 return false;
 181         }
 182
 183         switch(c->state) {
 184         case FIN_WAIT_1:
 185         case CLOSING:
 186         case LAST_ACK:
 187                 return true;
 188
 189         default:
 190                 return false;
 191         }
 192 }
 193
 194 static bool is_reliable(struct utcp_connection *c) {
 195         return c->flags & UTCP_RELIABLE;
 196 }
 197
 198 static int32_t seqdiff(uint32_t a, uint32_t b) {
 199         return a - b;
 200 }
 201
 202 // Buffer functions
 203 static bool buffer_wraps(struct buffer *buf) {
 204         return buf->size - buf->offset < buf->used;
 205 }
 206
 207 static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
 208         char *newdata = realloc(buf->data, newsize);
 209
 210         if(!newdata) {
 211                 return false;
 212         }
 213
 214         buf->data = newdata;
 215
 216         if(buffer_wraps(buf)) {
 217                 // Shift the right part of the buffer until it hits the end of the new buffer.
 218                 // Old situation:
 219                 // [345......012]
 220                 // New situation:
 221                 // [345.........|........012]
 222                 uint32_t tailsize = buf->size - buf->offset;
 223                 uint32_t newoffset = newsize - tailsize;
 224                 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
 225                 buf->offset = newoffset;
 226         }
 227
 228         buf->size = newsize;
 229         return true;
 230 }
 231
 232 // Store data into the buffer
 233 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 234         debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 235
 236         // Ensure we don't store more than maxsize bytes in total
 237         size_t required = offset + len;
 238
 239         if(required > buf->maxsize) {
 240                 if(offset >= buf->maxsize) {
 241                         return 0;
 242                 }
 243
 244                 len = buf->maxsize - offset;
 245                 required = buf->maxsize;
 246         }
 247
 248         // Check if we need to resize the buffer
 249         if(required > buf->size) {
 250                 size_t newsize = buf->size;
 251
 252                 if(!newsize) {
 253                         newsize = 4096;
 254                 }
 255
 256                 do {
 257                         newsize *= 2;
 258                 } while(newsize < required);
 259
 260                 if(newsize > buf->maxsize) {
 261                         newsize = buf->maxsize;
 262                 }
 263
 264                 if(!buffer_resize(buf, newsize)) {
 265                         return -1;
 266                 }
 267         }
 268
 269         uint32_t realoffset = buf->offset + offset;
 270
 271         if(buf->size - buf->offset < offset) {
 272                 // The offset wrapped
 273                 realoffset -= buf->size;
 274         }
 275
 276         if(buf->size - realoffset < len) {
 277                 // The new chunk of data must be wrapped
 278                 memcpy(buf->data + realoffset, data, buf->size - realoffset);
 279                 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
 280         } else {
 281                 memcpy(buf->data + realoffset, data, len);
 282         }
 283
 284         if(required > buf->used) {
 285                 buf->used = required;
 286         }
 287
 288         return len;
 289 }
 290
 291 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 292         return buffer_put_at(buf, buf->used, data, len);
 293 }
 294
 295 // Copy data from the buffer without removing it.
 296 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 297         // Ensure we don't copy more than is actually stored in the buffer
 298         if(offset >= buf->used) {
 299                 return 0;
 300         }
 301
 302         if(buf->used - offset < len) {
 303                 len = buf->used - offset;
 304         }
 305
 306         uint32_t realoffset = buf->offset + offset;
 307
 308         if(buf->size - buf->offset < offset) {
 309                 // The offset wrapped
 310                 realoffset -= buf->size;
 311         }
 312
 313         if(buf->size - realoffset < len) {
 314                 // The data is wrapped
 315                 memcpy(data, buf->data + realoffset, buf->size - realoffset);
 316                 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
 317         } else {
 318                 memcpy(data, buf->data + realoffset, len);
 319         }
 320
 321         return len;
 322 }
 323
 324 // Copy data from the buffer without removing it.
 325 static ssize_t buffer_call(struct buffer *buf, utcp_recv_t cb, void *arg, size_t offset, size_t len) {
 326         // Ensure we don't copy more than is actually stored in the buffer
 327         if(offset >= buf->used) {
 328                 return 0;
 329         }
 330
 331         if(buf->used - offset < len) {
 332                 len = buf->used - offset;
 333         }
 334
 335         uint32_t realoffset = buf->offset + offset;
 336
 337         if(buf->size - buf->offset < offset) {
 338                 // The offset wrapped
 339                 realoffset -= buf->size;
 340         }
 341
 342         if(buf->size - realoffset < len) {
 343                 // The data is wrapped
 344                 ssize_t rx1 = cb(arg, buf->data + realoffset, buf->size - realoffset);
 345
 346                 if(rx1 < buf->size - realoffset) {
 347                         return rx1;
 348                 }
 349
 350                 ssize_t rx2 = cb(arg, buf->data, len - (buf->size - realoffset));
 351
 352                 if(rx2 < 0) {
 353                         return rx2;
 354                 } else {
 355                         return rx1 + rx2;
 356                 }
 357         } else {
 358                 return cb(arg, buf->data + realoffset, len);
 359         }
 360 }
 361
 362 // Discard data from the buffer.
 363 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
 364         if(buf->used < len) {
 365                 len = buf->used;
 366         }
 367
 368         if(buf->size - buf->offset < len) {
 369                 buf->offset -= buf->size;
 370         }
 371
 372         if(buf->used == len) {
 373                 buf->offset = 0;
 374         } else {
 375                 buf->offset += len;
 376         }
 377
 378         buf->used -= len;
 379
 380         return len;
 381 }
 382
 383 static void buffer_clear(struct buffer *buf) {
 384         buf->used = 0;
 385         buf->offset = 0;
 386 }
 387
 388 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
 389         if(maxsize < minsize) {
 390                 maxsize = minsize;
 391         }
 392
 393         buf->maxsize = maxsize;
 394
 395         return buf->size >= minsize || buffer_resize(buf, minsize);
 396 }
 397
 398 static void buffer_exit(struct buffer *buf) {
 399         free(buf->data);
 400         memset(buf, 0, sizeof(*buf));
 401 }
 402
 403 static uint32_t buffer_free(const struct buffer *buf) {
 404         return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
 405 }
 406
 407 // Connections are stored in a sorted list.
 408 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 409
 410 static int compare(const void *va, const void *vb) {
 411         assert(va && vb);
 412
 413         const struct utcp_connection *a = *(struct utcp_connection **)va;
 414         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 415
 416         assert(a && b);
 417         assert(a->src && b->src);
 418
 419         int c = (int)a->src - (int)b->src;
 420
 421         if(c) {
 422                 return c;
 423         }
 424
 425         c = (int)a->dst - (int)b->dst;
 426         return c;
 427 }
 428
 429 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 430         if(!utcp->nconnections) {
 431                 return NULL;
 432         }
 433
 434         struct utcp_connection key = {
 435                 .src = src,
 436                 .dst = dst,
 437         }, *keyp = &key;
 438         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 439         return match ? *match : NULL;
 440 }
 441
 442 static void free_connection(struct utcp_connection *c) {
 443         struct utcp *utcp = c->utcp;
 444         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 445
 446         assert(cp);
 447
 448         int i = cp - utcp->connections;
 449         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 450         utcp->nconnections--;
 451
 452         buffer_exit(&c->rcvbuf);
 453         buffer_exit(&c->sndbuf);
 454         free(c);
 455 }
 456
 457 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 458         // Check whether this combination of src and dst is free
 459
 460         if(src) {
 461                 if(find_connection(utcp, src, dst)) {
 462                         errno = EADDRINUSE;
 463                         return NULL;
 464                 }
 465         } else { // If src == 0, generate a random port number with the high bit set
 466                 if(utcp->nconnections >= 32767) {
 467                         errno = ENOMEM;
 468                         return NULL;
 469                 }
 470
 471                 src = rand() | 0x8000;
 472
 473                 while(find_connection(utcp, src, dst)) {
 474                         src++;
 475                 }
 476         }
 477
 478         // Allocate memory for the new connection
 479
 480         if(utcp->nconnections >= utcp->nallocated) {
 481                 if(!utcp->nallocated) {
 482                         utcp->nallocated = 4;
 483                 } else {
 484                         utcp->nallocated *= 2;
 485                 }
 486
 487                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 488
 489                 if(!new_array) {
 490                         return NULL;
 491                 }
 492
 493                 utcp->connections = new_array;
 494         }
 495
 496         struct utcp_connection *c = calloc(1, sizeof(*c));
 497
 498         if(!c) {
 499                 return NULL;
 500         }
 501
 502         if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 503                 free(c);
 504                 return NULL;
 505         }
 506
 507         if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 508                 buffer_exit(&c->sndbuf);
 509                 free(c);
 510                 return NULL;
 511         }
 512
 513         // Fill in the details
 514
 515         c->src = src;
 516         c->dst = dst;
 517 #ifdef UTCP_DEBUG
 518         c->snd.iss = 0;
 519 #else
 520         c->snd.iss = rand();
 521 #endif
 522         c->snd.una = c->snd.iss;
 523         c->snd.nxt = c->snd.iss + 1;
 524         c->snd.last = c->snd.nxt;
 525         c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
 526         c->snd.ssthresh = ~0;
 527         debug_cwnd(c);
 528         c->srtt = 0;
 529         c->rttvar = 0;
 530         c->rto = START_RTO;
 531         c->utcp = utcp;
 532
 533         // Add it to the sorted list of connections
 534
 535         utcp->connections[utcp->nconnections++] = c;
 536         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 537
 538         return c;
 539 }
 540
 541 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 542         if(a > b) {
 543                 return a - b;
 544         } else {
 545                 return b - a;
 546         }
 547 }
 548
 549 // Update RTT variables. See RFC 6298.
 550 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 551         if(!rtt) {
 552                 debug(c, "invalid rtt\n");
 553                 return;
 554         }
 555
 556         if(!c->srtt) {
 557                 c->srtt = rtt;
 558                 c->rttvar = rtt / 2;
 559         } else {
 560                 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
 561                 c->srtt = (c->srtt * 7 + rtt) / 8;
 562         }
 563
 564         c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
 565
 566         if(c->rto > MAX_RTO) {
 567                 c->rto = MAX_RTO;
 568         }
 569
 570         debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
 571 }
 572
 573 static void start_retransmit_timer(struct utcp_connection *c) {
 574         clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
 575
 576         uint32_t rto = c->rto;
 577
 578         while(rto > USEC_PER_SEC) {
 579                 c->rtrx_timeout.tv_sec++;
 580                 rto -= USEC_PER_SEC;
 581         }
 582
 583         c->rtrx_timeout.tv_nsec += rto * 1000;
 584
 585         if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
 586                 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
 587                 c->rtrx_timeout.tv_sec++;
 588         }
 589
 590         debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
 591 }
 592
 593 static void stop_retransmit_timer(struct utcp_connection *c) {
 594         timespec_clear(&c->rtrx_timeout);
 595         debug(c, "rtrx_timeout cleared\n");
 596 }
 597
 598 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 599         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 600
 601         if(!c) {
 602                 return NULL;
 603         }
 604
 605         assert((flags & ~0x1f) == 0);
 606
 607         c->flags = flags;
 608         c->recv = recv;
 609         c->priv = priv;
 610
 611         struct {
 612                 struct hdr hdr;
 613                 uint8_t init[4];
 614         } pkt;
 615
 616         pkt.hdr.src = c->src;
 617         pkt.hdr.dst = c->dst;
 618         pkt.hdr.seq = c->snd.iss;
 619         pkt.hdr.ack = 0;
 620         pkt.hdr.wnd = c->rcvbuf.maxsize;
 621         pkt.hdr.ctl = SYN;
 622         pkt.hdr.aux = 0x0101;
 623         pkt.init[0] = 1;
 624         pkt.init[1] = 0;
 625         pkt.init[2] = 0;
 626         pkt.init[3] = flags & 0x7;
 627
 628         set_state(c, SYN_SENT);
 629
 630         print_packet(c, "send", &pkt, sizeof(pkt));
 631         utcp->send(utcp, &pkt, sizeof(pkt));
 632
 633         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 634         c->conn_timeout.tv_sec += utcp->timeout;
 635
 636         start_retransmit_timer(c);
 637
 638         return c;
 639 }
 640
 641 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 642         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 643 }
 644
 645 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 646         if(c->reapable || c->state != SYN_RECEIVED) {
 647                 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
 648                 return;
 649         }
 650
 651         debug(c, "accepted %p %p\n", c, recv, priv);
 652         c->recv = recv;
 653         c->priv = priv;
 654         set_state(c, ESTABLISHED);
 655 }
 656
 657 static void ack(struct utcp_connection *c, bool sendatleastone) {
 658         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 659         int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
 660
 661         assert(left >= 0);
 662
 663         if(cwndleft <= 0) {
 664                 left = 0;
 665         } else if(cwndleft < left) {
 666                 left = cwndleft;
 667
 668                 if(!sendatleastone || cwndleft > c->utcp->mss) {
 669                         left -= left % c->utcp->mss;
 670                 }
 671         }
 672
 673         debug(c, "cwndleft %d left %d\n", cwndleft, left);
 674
 675         if(!left && !sendatleastone) {
 676                 return;
 677         }
 678
 679         struct {
 680                 struct hdr hdr;
 681                 uint8_t data[];
 682         } *pkt = c->utcp->pkt;
 683
 684         pkt->hdr.src = c->src;
 685         pkt->hdr.dst = c->dst;
 686         pkt->hdr.ack = c->rcv.nxt;
 687         pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
 688         pkt->hdr.ctl = ACK;
 689         pkt->hdr.aux = 0;
 690
 691         do {
 692                 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
 693                 pkt->hdr.seq = c->snd.nxt;
 694
 695                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 696
 697                 c->snd.nxt += seglen;
 698                 left -= seglen;
 699
 700                 if(!is_reliable(c)) {
 701                         if(left) {
 702                                 pkt->hdr.ctl |= MF;
 703                         } else {
 704                                 pkt->hdr.ctl &= ~MF;
 705                         }
 706                 }
 707
 708                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 709                         seglen--;
 710                         pkt->hdr.ctl |= FIN;
 711                 }
 712
 713                 if(!c->rtt_start.tv_sec) {
 714                         // Start RTT measurement
 715                         clock_gettime(UTCP_CLOCK, &c->rtt_start);
 716                         c->rtt_seq = pkt->hdr.seq + seglen;
 717                         debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 718                 }
 719
 720                 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 721                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 722
 723                 if(left && !is_reliable(c)) {
 724                         pkt->hdr.wnd += seglen;
 725                 }
 726         } while(left);
 727 }
 728
 729 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 730         if(c->reapable) {
 731                 debug(c, "send() called on closed connection\n");
 732                 errno = EBADF;
 733                 return -1;
 734         }
 735
 736         switch(c->state) {
 737         case CLOSED:
 738         case LISTEN:
 739                 debug(c, "send() called on unconnected connection\n");
 740                 errno = ENOTCONN;
 741                 return -1;
 742
 743         case SYN_SENT:
 744         case SYN_RECEIVED:
 745         case ESTABLISHED:
 746         case CLOSE_WAIT:
 747                 break;
 748
 749         case FIN_WAIT_1:
 750         case FIN_WAIT_2:
 751         case CLOSING:
 752         case LAST_ACK:
 753         case TIME_WAIT:
 754                 debug(c, "send() called on closed connection\n");
 755                 errno = EPIPE;
 756                 return -1;
 757         }
 758
 759         // Exit early if we have nothing to send.
 760
 761         if(!len) {
 762                 return 0;
 763         }
 764
 765         if(!data) {
 766                 errno = EFAULT;
 767                 return -1;
 768         }
 769
 770         // Check if we need to be able to buffer all data
 771
 772         if(c->flags & UTCP_NO_PARTIAL) {
 773                 if(len > buffer_free(&c->sndbuf)) {
 774                         if(len > c->sndbuf.maxsize) {
 775                                 errno = EMSGSIZE;
 776                                 return -1;
 777                         } else {
 778                                 errno = EWOULDBLOCK;
 779                                 return 0;
 780                         }
 781                 }
 782         }
 783
 784         // Add data to send buffer.
 785
 786         if(is_reliable(c)) {
 787                 len = buffer_put(&c->sndbuf, data, len);
 788         } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
 789                 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
 790                         errno = EMSGSIZE;
 791                         return -1;
 792                 }
 793         } else {
 794                 return 0;
 795         }
 796
 797         if(len <= 0) {
 798                 if(is_reliable(c)) {
 799                         errno = EWOULDBLOCK;
 800                         return 0;
 801                 } else {
 802                         return len;
 803                 }
 804         }
 805
 806         c->snd.last += len;
 807
 808         // Don't send anything yet if the connection has not fully established yet
 809
 810         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 811                 return len;
 812         }
 813
 814         ack(c, false);
 815
 816         if(!is_reliable(c)) {
 817                 c->snd.una = c->snd.nxt = c->snd.last;
 818                 buffer_discard(&c->sndbuf, c->sndbuf.used);
 819         }
 820
 821         if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
 822                 start_retransmit_timer(c);
 823         }
 824
 825         if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
 826                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 827                 c->conn_timeout.tv_sec += c->utcp->timeout;
 828         }
 829
 830         return len;
 831 }
 832
 833 static void swap_ports(struct hdr *hdr) {
 834         uint16_t tmp = hdr->src;
 835         hdr->src = hdr->dst;
 836         hdr->dst = tmp;
 837 }
 838
 839 static void fast_retransmit(struct utcp_connection *c) {
 840         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 841                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 842                 return;
 843         }
 844
 845         struct utcp *utcp = c->utcp;
 846
 847         struct {
 848                 struct hdr hdr;
 849                 uint8_t data[];
 850         } *pkt;
 851
 852         pkt = malloc(c->utcp->mtu);
 853
 854         if(!pkt) {
 855                 return;
 856         }
 857
 858         pkt->hdr.src = c->src;
 859         pkt->hdr.dst = c->dst;
 860         pkt->hdr.wnd = c->rcvbuf.maxsize;
 861         pkt->hdr.aux = 0;
 862
 863         switch(c->state) {
 864         case ESTABLISHED:
 865         case FIN_WAIT_1:
 866         case CLOSE_WAIT:
 867         case CLOSING:
 868         case LAST_ACK:
 869                 // Send unacked data again.
 870                 pkt->hdr.seq = c->snd.una;
 871                 pkt->hdr.ack = c->rcv.nxt;
 872                 pkt->hdr.ctl = ACK;
 873                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 874
 875                 if(fin_wanted(c, c->snd.una + len)) {
 876                         len--;
 877                         pkt->hdr.ctl |= FIN;
 878                 }
 879
 880                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 881                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 882                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 883                 break;
 884
 885         default:
 886                 break;
 887         }
 888
 889         free(pkt);
 890 }
 891
 892 static void retransmit(struct utcp_connection *c) {
 893         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 894                 debug(c, "retransmit() called but nothing to retransmit!\n");
 895                 stop_retransmit_timer(c);
 896                 return;
 897         }
 898
 899         struct utcp *utcp = c->utcp;
 900
 901         struct {
 902                 struct hdr hdr;
 903                 uint8_t data[];
 904         } *pkt = c->utcp->pkt;
 905
 906         pkt->hdr.src = c->src;
 907         pkt->hdr.dst = c->dst;
 908         pkt->hdr.wnd = c->rcvbuf.maxsize;
 909         pkt->hdr.aux = 0;
 910
 911         switch(c->state) {
 912         case SYN_SENT:
 913                 // Send our SYN again
 914                 pkt->hdr.seq = c->snd.iss;
 915                 pkt->hdr.ack = 0;
 916                 pkt->hdr.ctl = SYN;
 917                 pkt->hdr.aux = 0x0101;
 918                 pkt->data[0] = 1;
 919                 pkt->data[1] = 0;
 920                 pkt->data[2] = 0;
 921                 pkt->data[3] = c->flags & 0x7;
 922                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 923                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 924                 break;
 925
 926         case SYN_RECEIVED:
 927                 // Send SYNACK again
 928                 pkt->hdr.seq = c->snd.nxt;
 929                 pkt->hdr.ack = c->rcv.nxt;
 930                 pkt->hdr.ctl = SYN | ACK;
 931                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
 932                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 933                 break;
 934
 935         case ESTABLISHED:
 936         case FIN_WAIT_1:
 937         case CLOSE_WAIT:
 938         case CLOSING:
 939         case LAST_ACK:
 940                 // Send unacked data again.
 941                 pkt->hdr.seq = c->snd.una;
 942                 pkt->hdr.ack = c->rcv.nxt;
 943                 pkt->hdr.ctl = ACK;
 944                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 945
 946                 if(fin_wanted(c, c->snd.una + len)) {
 947                         len--;
 948                         pkt->hdr.ctl |= FIN;
 949                 }
 950
 951                 // RFC 5681 slow start after timeout
 952                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
 953                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
 954                 c->snd.cwnd = utcp->mss;
 955                 debug_cwnd(c);
 956
 957                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 958                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 959                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 960
 961                 c->snd.nxt = c->snd.una + len;
 962                 break;
 963
 964         case CLOSED:
 965         case LISTEN:
 966         case TIME_WAIT:
 967         case FIN_WAIT_2:
 968                 // We shouldn't need to retransmit anything in this state.
 969 #ifdef UTCP_DEBUG
 970                 abort();
 971 #endif
 972                 stop_retransmit_timer(c);
 973                 goto cleanup;
 974         }
 975
 976         start_retransmit_timer(c);
 977         c->rto *= 2;
 978
 979         if(c->rto > MAX_RTO) {
 980                 c->rto = MAX_RTO;
 981         }
 982
 983         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 984         c->dupack = 0; // cancel any ongoing fast recovery
 985
 986 cleanup:
 987         return;
 988 }
 989
 990 /* Update receive buffer and SACK entries after consuming data.
 991  *
 992  * Situation:
 993  *
 994  * |.....0000..1111111111.....22222......3333|
 995  * |---------------^
 996  *
 997  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 998  * to remove data from the receive buffer. The idea is to substract "len"
 999  * from the offset of all the SACK entries, and then remove/cut down entries
1000  * that are shifted to before the start of the receive buffer.
1001  *
1002  * There are three cases:
1003  * - the SACK entry is after ^, in that case just change the offset.
1004  * - the SACK entry starts before and ends after ^, so we have to
1005  *   change both its offset and size.
1006  * - the SACK entry is completely before ^, in that case delete it.
1007  */
1008 static void sack_consume(struct utcp_connection *c, size_t len) {
1009         debug(c, "sack_consume %lu\n", (unsigned long)len);
1010
1011         if(len > c->rcvbuf.used) {
1012                 debug(c, "all SACK entries consumed\n");
1013                 c->sacks[0].len = 0;
1014                 return;
1015         }
1016
1017         buffer_discard(&c->rcvbuf, len);
1018
1019         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1020                 if(len < c->sacks[i].offset) {
1021                         c->sacks[i].offset -= len;
1022                         i++;
1023                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1024                         c->sacks[i].len -= len - c->sacks[i].offset;
1025                         c->sacks[i].offset = 0;
1026                         i++;
1027                 } else {
1028                         if(i < NSACKS - 1) {
1029                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1030                                 c->sacks[NSACKS - 1].len = 0;
1031                         } else {
1032                                 c->sacks[i].len = 0;
1033                                 break;
1034                         }
1035                 }
1036         }
1037
1038         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1039                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1040         }
1041 }
1042
1043 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1044         debug(c, "out of order packet, offset %u\n", offset);
1045         // Packet loss or reordering occured. Store the data in the buffer.
1046         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1047
1048         if(rxd <= 0) {
1049                 debug(c, "packet outside receive buffer, dropping\n");
1050                 return;
1051         }
1052
1053         if((size_t)rxd < len) {
1054                 debug(c, "packet partially outside receive buffer\n");
1055                 len = rxd;
1056         }
1057
1058         // Make note of where we put it.
1059         for(int i = 0; i < NSACKS; i++) {
1060                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1061                         debug(c, "new SACK entry %d\n", i);
1062                         c->sacks[i].offset = offset;
1063                         c->sacks[i].len = rxd;
1064                         break;
1065                 } else if(offset < c->sacks[i].offset) {
1066                         if(offset + rxd < c->sacks[i].offset) { // insert before
1067                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1068                                         debug(c, "insert SACK entry at %d\n", i);
1069                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1070                                         c->sacks[i].offset = offset;
1071                                         c->sacks[i].len = rxd;
1072                                 } else {
1073                                         debug(c, "SACK entries full, dropping packet\n");
1074                                 }
1075
1076                                 break;
1077                         } else { // merge
1078                                 debug(c, "merge with start of SACK entry at %d\n", i);
1079                                 c->sacks[i].offset = offset;
1080                                 break;
1081                         }
1082                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1083                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1084                                 debug(c, "merge with end of SACK entry at %d\n", i);
1085                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1086                                 // TODO: handle potential merge with next entry
1087                         }
1088
1089                         break;
1090                 }
1091         }
1092
1093         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1094                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1095         }
1096 }
1097
1098 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1099         if(c->recv) {
1100                 ssize_t rxd = c->recv(c, data, len);
1101
1102                 if(rxd != (ssize_t)len) {
1103                         // TODO: handle the application not accepting all data.
1104                         abort();
1105                 }
1106         }
1107
1108         // Check if we can process out-of-order data now.
1109         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1110                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1111
1112                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1113                         size_t offset = len;
1114                         len = c->sacks[0].offset + c->sacks[0].len;
1115                         size_t remainder = len - offset;
1116
1117                         if(c->recv) {
1118                                 ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
1119
1120                                 if(rxd != (ssize_t)remainder) {
1121                                         // TODO: handle the application not accepting all data.
1122                                         abort();
1123                                 }
1124                         }
1125                 }
1126         }
1127
1128         if(c->rcvbuf.used) {
1129                 sack_consume(c, len);
1130         }
1131
1132         c->rcv.nxt += len;
1133 }
1134
1135 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1136         // Fast path for unfragmented packets
1137         if(!hdr->wnd && !(hdr->ctl & MF)) {
1138                 if(c->recv) {
1139                         c->recv(c, data, len);
1140                 }
1141
1142                 c->rcv.nxt = hdr->seq + len;
1143                 return;
1144         }
1145
1146         // Ensure reassembled packet are not larger than 64 kiB
1147         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1148                 return;
1149         }
1150
1151         // Don't accept out of order fragments
1152         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1153                 return;
1154         }
1155
1156         // Reset the receive buffer for the first fragment
1157         if(!hdr->wnd) {
1158                 buffer_clear(&c->rcvbuf);
1159         }
1160
1161         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1162
1163         if(rxd != (ssize_t)len) {
1164                 return;
1165         }
1166
1167         // Send the packet if it's the final fragment
1168         if(!(hdr->ctl & MF) && c->recv) {
1169                 buffer_call(&c->rcvbuf, c->recv, c, 0, hdr->wnd + len);
1170         }
1171
1172         c->rcv.nxt = hdr->seq + len;
1173 }
1174
1175 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1176         if(!is_reliable(c)) {
1177                 handle_unreliable(c, hdr, data, len);
1178                 return;
1179         }
1180
1181         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1182
1183         if(offset) {
1184                 handle_out_of_order(c, offset, data, len);
1185         } else {
1186                 handle_in_order(c, data, len);
1187         }
1188 }
1189
1190
1191 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1192         const uint8_t *ptr = data;
1193
1194         if(!utcp) {
1195                 errno = EFAULT;
1196                 return -1;
1197         }
1198
1199         if(!len) {
1200                 return 0;
1201         }
1202
1203         if(!data) {
1204                 errno = EFAULT;
1205                 return -1;
1206         }
1207
1208         // Drop packets smaller than the header
1209
1210         struct hdr hdr;
1211
1212         if(len < sizeof(hdr)) {
1213                 print_packet(NULL, "recv", data, len);
1214                 errno = EBADMSG;
1215                 return -1;
1216         }
1217
1218         // Make a copy from the potentially unaligned data to a struct hdr
1219
1220         memcpy(&hdr, ptr, sizeof(hdr));
1221
1222         // Try to match the packet to an existing connection
1223
1224         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1225         print_packet(c, "recv", data, len);
1226
1227         // Process the header
1228
1229         ptr += sizeof(hdr);
1230         len -= sizeof(hdr);
1231
1232         // Drop packets with an unknown CTL flag
1233
1234         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1235                 print_packet(NULL, "recv", data, len);
1236                 errno = EBADMSG;
1237                 return -1;
1238         }
1239
1240         // Check for auxiliary headers
1241
1242         const uint8_t *init = NULL;
1243
1244         uint16_t aux = hdr.aux;
1245
1246         while(aux) {
1247                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1248                 uint8_t auxtype = aux & 0xff;
1249
1250                 if(len < auxlen) {
1251                         errno = EBADMSG;
1252                         return -1;
1253                 }
1254
1255                 switch(auxtype) {
1256                 case AUX_INIT:
1257                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1258                                 errno = EBADMSG;
1259                                 return -1;
1260                         }
1261
1262                         init = ptr;
1263                         break;
1264
1265                 default:
1266                         errno = EBADMSG;
1267                         return -1;
1268                 }
1269
1270                 len -= auxlen;
1271                 ptr += auxlen;
1272
1273                 if(!(aux & 0x800)) {
1274                         break;
1275                 }
1276
1277                 if(len < 2) {
1278                         errno = EBADMSG;
1279                         return -1;
1280                 }
1281
1282                 memcpy(&aux, ptr, 2);
1283                 len -= 2;
1284                 ptr += 2;
1285         }
1286
1287         bool has_data = len || (hdr.ctl & (SYN | FIN));
1288
1289         // Is it for a new connection?
1290
1291         if(!c) {
1292                 // Ignore RST packets
1293
1294                 if(hdr.ctl & RST) {
1295                         return 0;
1296                 }
1297
1298                 // Is it a SYN packet and are we LISTENing?
1299
1300                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1301                         // If we don't want to accept it, send a RST back
1302                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1303                                 len = 1;
1304                                 goto reset;
1305                         }
1306
1307                         // Try to allocate memory, otherwise send a RST back
1308                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1309
1310                         if(!c) {
1311                                 len = 1;
1312                                 goto reset;
1313                         }
1314
1315                         // Parse auxilliary information
1316                         if(init) {
1317                                 if(init[0] < 1) {
1318                                         len = 1;
1319                                         goto reset;
1320                                 }
1321
1322                                 c->flags = init[3] & 0x7;
1323                         } else {
1324                                 c->flags = UTCP_TCP;
1325                         }
1326
1327 synack:
1328                         // Return SYN+ACK, go to SYN_RECEIVED state
1329                         c->snd.wnd = hdr.wnd;
1330                         c->rcv.irs = hdr.seq;
1331                         c->rcv.nxt = c->rcv.irs + 1;
1332                         set_state(c, SYN_RECEIVED);
1333
1334                         struct {
1335                                 struct hdr hdr;
1336                                 uint8_t data[4];
1337                         } pkt;
1338
1339                         pkt.hdr.src = c->src;
1340                         pkt.hdr.dst = c->dst;
1341                         pkt.hdr.ack = c->rcv.irs + 1;
1342                         pkt.hdr.seq = c->snd.iss;
1343                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1344                         pkt.hdr.ctl = SYN | ACK;
1345
1346                         if(init) {
1347                                 pkt.hdr.aux = 0x0101;
1348                                 pkt.data[0] = 1;
1349                                 pkt.data[1] = 0;
1350                                 pkt.data[2] = 0;
1351                                 pkt.data[3] = c->flags & 0x7;
1352                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1353                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1354                         } else {
1355                                 pkt.hdr.aux = 0;
1356                                 print_packet(c, "send", &pkt, sizeof(hdr));
1357                                 utcp->send(utcp, &pkt, sizeof(hdr));
1358                         }
1359                 } else {
1360                         // No, we don't want your packets, send a RST back
1361                         len = 1;
1362                         goto reset;
1363                 }
1364
1365                 return 0;
1366         }
1367
1368         debug(c, "state %s\n", strstate[c->state]);
1369
1370         // In case this is for a CLOSED connection, ignore the packet.
1371         // TODO: make it so incoming packets can never match a CLOSED connection.
1372
1373         if(c->state == CLOSED) {
1374                 debug(c, "got packet for closed connection\n");
1375                 return 0;
1376         }
1377
1378         // It is for an existing connection.
1379
1380         // 1. Drop invalid packets.
1381
1382         // 1a. Drop packets that should not happen in our current state.
1383
1384         switch(c->state) {
1385         case SYN_SENT:
1386         case SYN_RECEIVED:
1387         case ESTABLISHED:
1388         case FIN_WAIT_1:
1389         case FIN_WAIT_2:
1390         case CLOSE_WAIT:
1391         case CLOSING:
1392         case LAST_ACK:
1393         case TIME_WAIT:
1394                 break;
1395
1396         default:
1397 #ifdef UTCP_DEBUG
1398                 abort();
1399 #endif
1400                 break;
1401         }
1402
1403         // 1b. Discard data that is not in our receive window.
1404
1405         if(is_reliable(c)) {
1406                 bool acceptable;
1407
1408                 if(c->state == SYN_SENT) {
1409                         acceptable = true;
1410                 } else if(len == 0) {
1411                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1412                 } else {
1413                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1414
1415                         // cut already accepted front overlapping
1416                         if(rcv_offset < 0) {
1417                                 acceptable = len > (size_t) - rcv_offset;
1418
1419                                 if(acceptable) {
1420                                         ptr -= rcv_offset;
1421                                         len += rcv_offset;
1422                                         hdr.seq -= rcv_offset;
1423                                 }
1424                         } else {
1425                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1426                         }
1427                 }
1428
1429                 if(!acceptable) {
1430                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1431
1432                         // Ignore unacceptable RST packets.
1433                         if(hdr.ctl & RST) {
1434                                 return 0;
1435                         }
1436
1437                         // Otherwise, continue processing.
1438                         len = 0;
1439                 }
1440         } else {
1441 #if UTCP_DEBUG
1442                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1443
1444                 if(rcv_offset) {
1445                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1446                 }
1447
1448 #endif
1449         }
1450
1451         c->snd.wnd = hdr.wnd; // TODO: move below
1452
1453         // 1c. Drop packets with an invalid ACK.
1454         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1455         // (= snd.una + c->sndbuf.used).
1456
1457         if(!is_reliable(c)) {
1458                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1459                         hdr.ack = c->snd.una;
1460                 }
1461         }
1462
1463         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1464                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1465
1466                 // Ignore unacceptable RST packets.
1467                 if(hdr.ctl & RST) {
1468                         return 0;
1469                 }
1470
1471                 goto reset;
1472         }
1473
1474         // 2. Handle RST packets
1475
1476         if(hdr.ctl & RST) {
1477                 switch(c->state) {
1478                 case SYN_SENT:
1479                         if(!(hdr.ctl & ACK)) {
1480                                 return 0;
1481                         }
1482
1483                         // The peer has refused our connection.
1484                         set_state(c, CLOSED);
1485                         errno = ECONNREFUSED;
1486
1487                         if(c->recv) {
1488                                 c->recv(c, NULL, 0);
1489                         }
1490
1491                         if(c->poll && !c->reapable) {
1492                                 c->poll(c, 0);
1493                         }
1494
1495                         return 0;
1496
1497                 case SYN_RECEIVED:
1498                         if(hdr.ctl & ACK) {
1499                                 return 0;
1500                         }
1501
1502                         // We haven't told the application about this connection yet. Silently delete.
1503                         free_connection(c);
1504                         return 0;
1505
1506                 case ESTABLISHED:
1507                 case FIN_WAIT_1:
1508                 case FIN_WAIT_2:
1509                 case CLOSE_WAIT:
1510                         if(hdr.ctl & ACK) {
1511                                 return 0;
1512                         }
1513
1514                         // The peer has aborted our connection.
1515                         set_state(c, CLOSED);
1516                         errno = ECONNRESET;
1517
1518                         if(c->recv) {
1519                                 c->recv(c, NULL, 0);
1520                         }
1521
1522                         if(c->poll && !c->reapable) {
1523                                 c->poll(c, 0);
1524                         }
1525
1526                         return 0;
1527
1528                 case CLOSING:
1529                 case LAST_ACK:
1530                 case TIME_WAIT:
1531                         if(hdr.ctl & ACK) {
1532                                 return 0;
1533                         }
1534
1535                         // As far as the application is concerned, the connection has already been closed.
1536                         // If it has called utcp_close() already, we can immediately free this connection.
1537                         if(c->reapable) {
1538                                 free_connection(c);
1539                                 return 0;
1540                         }
1541
1542                         // Otherwise, immediately move to the CLOSED state.
1543                         set_state(c, CLOSED);
1544                         return 0;
1545
1546                 default:
1547 #ifdef UTCP_DEBUG
1548                         abort();
1549 #endif
1550                         break;
1551                 }
1552         }
1553
1554         uint32_t advanced;
1555
1556         if(!(hdr.ctl & ACK)) {
1557                 advanced = 0;
1558                 goto skip_ack;
1559         }
1560
1561         // 3. Advance snd.una
1562
1563         advanced = seqdiff(hdr.ack, c->snd.una);
1564
1565         if(advanced) {
1566                 // RTT measurement
1567                 if(c->rtt_start.tv_sec) {
1568                         if(c->rtt_seq == hdr.ack) {
1569                                 struct timespec now;
1570                                 clock_gettime(UTCP_CLOCK, &now);
1571                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1572                                 update_rtt(c, diff);
1573                                 c->rtt_start.tv_sec = 0;
1574                         } else if(c->rtt_seq < hdr.ack) {
1575                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1576                                 c->rtt_start.tv_sec = 0;
1577                         }
1578                 }
1579
1580                 int32_t data_acked = advanced;
1581
1582                 switch(c->state) {
1583                 case SYN_SENT:
1584                 case SYN_RECEIVED:
1585                         data_acked--;
1586                         break;
1587
1588                 // TODO: handle FIN as well.
1589                 default:
1590                         break;
1591                 }
1592
1593                 assert(data_acked >= 0);
1594
1595 #ifndef NDEBUG
1596                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1597                 assert(data_acked <= bufused);
1598 #endif
1599
1600                 if(data_acked) {
1601                         buffer_discard(&c->sndbuf, data_acked);
1602
1603                         if(is_reliable(c)) {
1604                                 c->do_poll = true;
1605                         }
1606                 }
1607
1608                 // Also advance snd.nxt if possible
1609                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1610                         c->snd.nxt = hdr.ack;
1611                 }
1612
1613                 c->snd.una = hdr.ack;
1614
1615                 if(c->dupack) {
1616                         if(c->dupack >= 3) {
1617                                 debug(c, "fast recovery ended\n");
1618                                 c->snd.cwnd = c->snd.ssthresh;
1619                         }
1620
1621                         c->dupack = 0;
1622                 }
1623
1624                 // Increase the congestion window according to RFC 5681
1625                 if(c->snd.cwnd < c->snd.ssthresh) {
1626                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1627                 } else {
1628                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1629                 }
1630
1631                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1632                         c->snd.cwnd = c->sndbuf.maxsize;
1633                 }
1634
1635                 debug_cwnd(c);
1636
1637                 // Check if we have sent a FIN that is now ACKed.
1638                 switch(c->state) {
1639                 case FIN_WAIT_1:
1640                         if(c->snd.una == c->snd.last) {
1641                                 set_state(c, FIN_WAIT_2);
1642                         }
1643
1644                         break;
1645
1646                 case CLOSING:
1647                         if(c->snd.una == c->snd.last) {
1648                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1649                                 c->conn_timeout.tv_sec += utcp->timeout;
1650                                 set_state(c, TIME_WAIT);
1651                         }
1652
1653                         break;
1654
1655                 default:
1656                         break;
1657                 }
1658         } else {
1659                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1660                         c->dupack++;
1661                         debug(c, "duplicate ACK %d\n", c->dupack);
1662
1663                         if(c->dupack == 3) {
1664                                 // RFC 5681 fast recovery
1665                                 debug(c, "fast recovery started\n", c->dupack);
1666                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1667                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1668                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1669
1670                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1671                                         c->snd.cwnd = c->sndbuf.maxsize;
1672                                 }
1673
1674                                 debug_cwnd(c);
1675
1676                                 fast_retransmit(c);
1677                         } else if(c->dupack > 3) {
1678                                 c->snd.cwnd += utcp->mss;
1679
1680                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1681                                         c->snd.cwnd = c->sndbuf.maxsize;
1682                                 }
1683
1684                                 debug_cwnd(c);
1685                         }
1686
1687                         // We got an ACK which indicates the other side did get one of our packets.
1688                         // Reset the retransmission timer to avoid going to slow start,
1689                         // but don't touch the connection timeout.
1690                         start_retransmit_timer(c);
1691                 }
1692         }
1693
1694         // 4. Update timers
1695
1696         if(advanced) {
1697                 if(c->snd.una == c->snd.last) {
1698                         stop_retransmit_timer(c);
1699                         timespec_clear(&c->conn_timeout);
1700                 } else if(is_reliable(c)) {
1701                         start_retransmit_timer(c);
1702                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1703                         c->conn_timeout.tv_sec += utcp->timeout;
1704                 }
1705         }
1706
1707 skip_ack:
1708         // 5. Process SYN stuff
1709
1710         if(hdr.ctl & SYN) {
1711                 switch(c->state) {
1712                 case SYN_SENT:
1713
1714                         // This is a SYNACK. It should always have ACKed the SYN.
1715                         if(!advanced) {
1716                                 goto reset;
1717                         }
1718
1719                         c->rcv.irs = hdr.seq;
1720                         c->rcv.nxt = hdr.seq;
1721
1722                         if(c->shut_wr) {
1723                                 c->snd.last++;
1724                                 set_state(c, FIN_WAIT_1);
1725                         } else {
1726                                 set_state(c, ESTABLISHED);
1727                         }
1728
1729                         // TODO: notify application of this somehow.
1730                         break;
1731
1732                 case SYN_RECEIVED:
1733                         // This is a retransmit of a SYN, send back the SYNACK.
1734                         goto synack;
1735
1736                 case ESTABLISHED:
1737                 case FIN_WAIT_1:
1738                 case FIN_WAIT_2:
1739                 case CLOSE_WAIT:
1740                 case CLOSING:
1741                 case LAST_ACK:
1742                 case TIME_WAIT:
1743                         // Ehm, no. We should never receive a second SYN.
1744                         return 0;
1745
1746                 default:
1747 #ifdef UTCP_DEBUG
1748                         abort();
1749 #endif
1750                         return 0;
1751                 }
1752
1753                 // SYN counts as one sequence number
1754                 c->rcv.nxt++;
1755         }
1756
1757         // 6. Process new data
1758
1759         if(c->state == SYN_RECEIVED) {
1760                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1761                 if(!advanced) {
1762                         goto reset;
1763                 }
1764
1765                 // Are we still LISTENing?
1766                 if(utcp->accept) {
1767                         utcp->accept(c, c->src);
1768                 }
1769
1770                 if(c->state != ESTABLISHED) {
1771                         set_state(c, CLOSED);
1772                         c->reapable = true;
1773                         goto reset;
1774                 }
1775         }
1776
1777         if(len) {
1778                 switch(c->state) {
1779                 case SYN_SENT:
1780                 case SYN_RECEIVED:
1781                         // This should never happen.
1782 #ifdef UTCP_DEBUG
1783                         abort();
1784 #endif
1785                         return 0;
1786
1787                 case ESTABLISHED:
1788                 case FIN_WAIT_1:
1789                 case FIN_WAIT_2:
1790                         break;
1791
1792                 case CLOSE_WAIT:
1793                 case CLOSING:
1794                 case LAST_ACK:
1795                 case TIME_WAIT:
1796                         // Ehm no, We should never receive more data after a FIN.
1797                         goto reset;
1798
1799                 default:
1800 #ifdef UTCP_DEBUG
1801                         abort();
1802 #endif
1803                         return 0;
1804                 }
1805
1806                 handle_incoming_data(c, &hdr, ptr, len);
1807         }
1808
1809         // 7. Process FIN stuff
1810
1811         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1812                 switch(c->state) {
1813                 case SYN_SENT:
1814                 case SYN_RECEIVED:
1815                         // This should never happen.
1816 #ifdef UTCP_DEBUG
1817                         abort();
1818 #endif
1819                         break;
1820
1821                 case ESTABLISHED:
1822                         set_state(c, CLOSE_WAIT);
1823                         break;
1824
1825                 case FIN_WAIT_1:
1826                         set_state(c, CLOSING);
1827                         break;
1828
1829                 case FIN_WAIT_2:
1830                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1831                         c->conn_timeout.tv_sec += utcp->timeout;
1832                         set_state(c, TIME_WAIT);
1833                         break;
1834
1835                 case CLOSE_WAIT:
1836                 case CLOSING:
1837                 case LAST_ACK:
1838                 case TIME_WAIT:
1839                         // Ehm, no. We should never receive a second FIN.
1840                         goto reset;
1841
1842                 default:
1843 #ifdef UTCP_DEBUG
1844                         abort();
1845 #endif
1846                         break;
1847                 }
1848
1849                 // FIN counts as one sequence number
1850                 c->rcv.nxt++;
1851                 len++;
1852
1853                 // Inform the application that the peer closed its end of the connection.
1854                 if(c->recv) {
1855                         errno = 0;
1856                         c->recv(c, NULL, 0);
1857                 }
1858         }
1859
1860         // Now we send something back if:
1861         // - we received data, so we have to send back an ACK
1862         //   -> sendatleastone = true
1863         // - or we got an ack, so we should maybe send a bit more data
1864         //   -> sendatleastone = false
1865
1866         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1867                 ack(c, has_data);
1868         }
1869
1870         return 0;
1871
1872 reset:
1873         swap_ports(&hdr);
1874         hdr.wnd = 0;
1875         hdr.aux = 0;
1876
1877         if(hdr.ctl & ACK) {
1878                 hdr.seq = hdr.ack;
1879                 hdr.ctl = RST;
1880         } else {
1881                 hdr.ack = hdr.seq + len;
1882                 hdr.seq = 0;
1883                 hdr.ctl = RST | ACK;
1884         }
1885
1886         print_packet(c, "send", &hdr, sizeof(hdr));
1887         utcp->send(utcp, &hdr, sizeof(hdr));
1888         return 0;
1889
1890 }
1891
1892 int utcp_shutdown(struct utcp_connection *c, int dir) {
1893         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1894
1895         if(!c) {
1896                 errno = EFAULT;
1897                 return -1;
1898         }
1899
1900         if(c->reapable) {
1901                 debug(c, "shutdown() called on closed connection\n");
1902                 errno = EBADF;
1903                 return -1;
1904         }
1905
1906         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1907                 errno = EINVAL;
1908                 return -1;
1909         }
1910
1911         // TCP does not have a provision for stopping incoming packets.
1912         // The best we can do is to just ignore them.
1913         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1914                 c->recv = NULL;
1915         }
1916
1917         // The rest of the code deals with shutting down writes.
1918         if(dir == UTCP_SHUT_RD) {
1919                 return 0;
1920         }
1921
1922         // Only process shutting down writes once.
1923         if(c->shut_wr) {
1924                 return 0;
1925         }
1926
1927         c->shut_wr = true;
1928
1929         switch(c->state) {
1930         case CLOSED:
1931         case LISTEN:
1932                 errno = ENOTCONN;
1933                 return -1;
1934
1935         case SYN_SENT:
1936                 return 0;
1937
1938         case SYN_RECEIVED:
1939         case ESTABLISHED:
1940                 set_state(c, FIN_WAIT_1);
1941                 break;
1942
1943         case FIN_WAIT_1:
1944         case FIN_WAIT_2:
1945                 return 0;
1946
1947         case CLOSE_WAIT:
1948                 set_state(c, CLOSING);
1949                 break;
1950
1951         case CLOSING:
1952         case LAST_ACK:
1953         case TIME_WAIT:
1954                 return 0;
1955         }
1956
1957         c->snd.last++;
1958
1959         ack(c, false);
1960
1961         if(!timespec_isset(&c->rtrx_timeout)) {
1962                 start_retransmit_timer(c);
1963         }
1964
1965         return 0;
1966 }
1967
1968 static bool reset_connection(struct utcp_connection *c) {
1969         if(!c) {
1970                 errno = EFAULT;
1971                 return false;
1972         }
1973
1974         if(c->reapable) {
1975                 debug(c, "abort() called on closed connection\n");
1976                 errno = EBADF;
1977                 return false;
1978         }
1979
1980         c->recv = NULL;
1981         c->poll = NULL;
1982
1983         switch(c->state) {
1984         case CLOSED:
1985                 return true;
1986
1987         case LISTEN:
1988         case SYN_SENT:
1989         case CLOSING:
1990         case LAST_ACK:
1991         case TIME_WAIT:
1992                 set_state(c, CLOSED);
1993                 return true;
1994
1995         case SYN_RECEIVED:
1996         case ESTABLISHED:
1997         case FIN_WAIT_1:
1998         case FIN_WAIT_2:
1999         case CLOSE_WAIT:
2000                 set_state(c, CLOSED);
2001                 break;
2002         }
2003
2004         // Send RST
2005
2006         struct hdr hdr;
2007
2008         hdr.src = c->src;
2009         hdr.dst = c->dst;
2010         hdr.seq = c->snd.nxt;
2011         hdr.ack = 0;
2012         hdr.wnd = 0;
2013         hdr.ctl = RST;
2014
2015         print_packet(c, "send", &hdr, sizeof(hdr));
2016         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2017         return true;
2018 }
2019
2020 // Closes all the opened connections
2021 void utcp_abort_all_connections(struct utcp *utcp) {
2022         if(!utcp) {
2023                 errno = EINVAL;
2024                 return;
2025         }
2026
2027         for(int i = 0; i < utcp->nconnections; i++) {
2028                 struct utcp_connection *c = utcp->connections[i];
2029
2030                 if(c->reapable || c->state == CLOSED) {
2031                         continue;
2032                 }
2033
2034                 utcp_recv_t old_recv = c->recv;
2035                 utcp_poll_t old_poll = c->poll;
2036
2037                 reset_connection(c);
2038
2039                 if(old_recv) {
2040                         errno = 0;
2041                         old_recv(c, NULL, 0);
2042                 }
2043
2044                 if(old_poll && !c->reapable) {
2045                         errno = 0;
2046                         old_poll(c, 0);
2047                 }
2048         }
2049
2050         return;
2051 }
2052
2053 int utcp_close(struct utcp_connection *c) {
2054         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2055                 return -1;
2056         }
2057
2058         c->recv = NULL;
2059         c->poll = NULL;
2060         c->reapable = true;
2061         return 0;
2062 }
2063
2064 int utcp_abort(struct utcp_connection *c) {
2065         if(!reset_connection(c)) {
2066                 return -1;
2067         }
2068
2069         c->reapable = true;
2070         return 0;
2071 }
2072
2073 /* Handle timeouts.
2074  * One call to this function will loop through all connections,
2075  * checking if something needs to be resent or not.
2076  * The return value is the time to the next timeout in milliseconds,
2077  * or maybe a negative value if the timeout is infinite.
2078  */
2079 struct timespec utcp_timeout(struct utcp *utcp) {
2080         struct timespec now;
2081         clock_gettime(UTCP_CLOCK, &now);
2082         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2083
2084         for(int i = 0; i < utcp->nconnections; i++) {
2085                 struct utcp_connection *c = utcp->connections[i];
2086
2087                 if(!c) {
2088                         continue;
2089                 }
2090
2091                 // delete connections that have been utcp_close()d.
2092                 if(c->state == CLOSED) {
2093                         if(c->reapable) {
2094                                 debug(c, "reaping\n");
2095                                 free_connection(c);
2096                                 i--;
2097                         }
2098
2099                         continue;
2100                 }
2101
2102                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2103                         errno = ETIMEDOUT;
2104                         c->state = CLOSED;
2105
2106                         if(c->recv) {
2107                                 c->recv(c, NULL, 0);
2108                         }
2109
2110                         if(c->poll && !c->reapable) {
2111                                 c->poll(c, 0);
2112                         }
2113
2114                         continue;
2115                 }
2116
2117                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2118                         debug(c, "retransmitting after timeout\n");
2119                         retransmit(c);
2120                 }
2121
2122                 if(c->poll) {
2123                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2124                                 c->do_poll = false;
2125                                 uint32_t len = buffer_free(&c->sndbuf);
2126
2127                                 if(len) {
2128                                         c->poll(c, len);
2129                                 }
2130                         } else if(c->state == CLOSED) {
2131                                 c->poll(c, 0);
2132                         }
2133                 }
2134
2135                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2136                         next = c->conn_timeout;
2137                 }
2138
2139                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2140                         next = c->rtrx_timeout;
2141                 }
2142         }
2143
2144         struct timespec diff;
2145
2146         timespec_sub(&next, &now, &diff);
2147
2148         return diff;
2149 }
2150
2151 bool utcp_is_active(struct utcp *utcp) {
2152         if(!utcp) {
2153                 return false;
2154         }
2155
2156         for(int i = 0; i < utcp->nconnections; i++)
2157                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2158                         return true;
2159                 }
2160
2161         return false;
2162 }
2163
2164 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2165         if(!send) {
2166                 errno = EFAULT;
2167                 return NULL;
2168         }
2169
2170         struct utcp *utcp = calloc(1, sizeof(*utcp));
2171
2172         if(!utcp) {
2173                 return NULL;
2174         }
2175
2176         if(!CLOCK_GRANULARITY) {
2177                 struct timespec res;
2178                 clock_getres(UTCP_CLOCK, &res);
2179                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2180         }
2181
2182         utcp->accept = accept;
2183         utcp->pre_accept = pre_accept;
2184         utcp->send = send;
2185         utcp->priv = priv;
2186         utcp_set_mtu(utcp, DEFAULT_MTU);
2187         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2188
2189         return utcp;
2190 }
2191
2192 void utcp_exit(struct utcp *utcp) {
2193         if(!utcp) {
2194                 return;
2195         }
2196
2197         for(int i = 0; i < utcp->nconnections; i++) {
2198                 struct utcp_connection *c = utcp->connections[i];
2199
2200                 if(!c->reapable) {
2201                         if(c->recv) {
2202                                 c->recv(c, NULL, 0);
2203                         }
2204
2205                         if(c->poll && !c->reapable) {
2206                                 c->poll(c, 0);
2207                         }
2208                 }
2209
2210                 buffer_exit(&c->rcvbuf);
2211                 buffer_exit(&c->sndbuf);
2212                 free(c);
2213         }
2214
2215         free(utcp->connections);
2216         free(utcp->pkt);
2217         free(utcp);
2218 }
2219
2220 uint16_t utcp_get_mtu(struct utcp *utcp) {
2221         return utcp ? utcp->mtu : 0;
2222 }
2223
2224 uint16_t utcp_get_mss(struct utcp *utcp) {
2225         return utcp ? utcp->mss : 0;
2226 }
2227
2228 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2229         if(!utcp) {
2230                 return;
2231         }
2232
2233         if(mtu <= sizeof(struct hdr)) {
2234                 return;
2235         }
2236
2237         if(mtu > utcp->mtu) {
2238                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2239
2240                 if(!new) {
2241                         return;
2242                 }
2243
2244                 utcp->pkt = new;
2245         }
2246
2247         utcp->mtu = mtu;
2248         utcp->mss = mtu - sizeof(struct hdr);
2249 }
2250
2251 void utcp_reset_timers(struct utcp *utcp) {
2252         if(!utcp) {
2253                 return;
2254         }
2255
2256         struct timespec now, then;
2257
2258         clock_gettime(UTCP_CLOCK, &now);
2259
2260         then = now;
2261
2262         then.tv_sec += utcp->timeout;
2263
2264         for(int i = 0; i < utcp->nconnections; i++) {
2265                 struct utcp_connection *c = utcp->connections[i];
2266
2267                 if(c->reapable) {
2268                         continue;
2269                 }
2270
2271                 if(timespec_isset(&c->rtrx_timeout)) {
2272                         c->rtrx_timeout = now;
2273                 }
2274
2275                 if(timespec_isset(&c->conn_timeout)) {
2276                         c->conn_timeout = then;
2277                 }
2278
2279                 c->rtt_start.tv_sec = 0;
2280
2281                 if(c->rto > START_RTO) {
2282                         c->rto = START_RTO;
2283                 }
2284         }
2285 }
2286
2287 int utcp_get_user_timeout(struct utcp *u) {
2288         return u ? u->timeout : 0;
2289 }
2290
2291 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2292         if(u) {
2293                 u->timeout = timeout;
2294         }
2295 }
2296
2297 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2298         return c ? c->sndbuf.maxsize : 0;
2299 }
2300
2301 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2302         if(!c) {
2303                 return 0;
2304         }
2305
2306         switch(c->state) {
2307         case SYN_SENT:
2308         case SYN_RECEIVED:
2309         case ESTABLISHED:
2310         case CLOSE_WAIT:
2311                 return buffer_free(&c->sndbuf);
2312
2313         default:
2314                 return 0;
2315         }
2316 }
2317
2318 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2319         if(!c) {
2320                 return;
2321         }
2322
2323         c->sndbuf.maxsize = size;
2324
2325         if(c->sndbuf.maxsize != size) {
2326                 c->sndbuf.maxsize = -1;
2327         }
2328
2329         c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2330 }
2331
2332 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2333         return c ? c->rcvbuf.maxsize : 0;
2334 }
2335
2336 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2337         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2338                 return buffer_free(&c->rcvbuf);
2339         } else {
2340                 return 0;
2341         }
2342 }
2343
2344 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2345         if(!c) {
2346                 return;
2347         }
2348
2349         c->rcvbuf.maxsize = size;
2350
2351         if(c->rcvbuf.maxsize != size) {
2352                 c->rcvbuf.maxsize = -1;
2353         }
2354 }
2355
2356 size_t utcp_get_sendq(struct utcp_connection *c) {
2357         return c->sndbuf.used;
2358 }
2359
2360 size_t utcp_get_recvq(struct utcp_connection *c) {
2361         return c->rcvbuf.used;
2362 }
2363
2364 bool utcp_get_nodelay(struct utcp_connection *c) {
2365         return c ? c->nodelay : false;
2366 }
2367
2368 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2369         if(c) {
2370                 c->nodelay = nodelay;
2371         }
2372 }
2373
2374 bool utcp_get_keepalive(struct utcp_connection *c) {
2375         return c ? c->keepalive : false;
2376 }
2377
2378 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2379         if(c) {
2380                 c->keepalive = keepalive;
2381         }
2382 }
2383
2384 size_t utcp_get_outq(struct utcp_connection *c) {
2385         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2386 }
2387
2388 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2389         if(c) {
2390                 c->recv = recv;
2391         }
2392 }
2393
2394 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2395         if(c) {
2396                 c->poll = poll;
2397                 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2398         }
2399 }
2400
2401 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2402         if(utcp) {
2403                 utcp->accept = accept;
2404                 utcp->pre_accept = pre_accept;
2405         }
2406 }
2407
2408 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2409         if(!c || c->reapable) {
2410                 return;
2411         }
2412
2413         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2414                 return;
2415         }
2416
2417         if(expect) {
2418                 // If we expect data, start the connection timer.
2419                 if(!timespec_isset(&c->conn_timeout)) {
2420                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2421                         c->conn_timeout.tv_sec += c->utcp->timeout;
2422                 }
2423         } else {
2424                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2425                 if(c->snd.una == c->snd.last) {
2426                         timespec_clear(&c->conn_timeout);
2427                 }
2428         }
2429 }
2430
2431 void utcp_offline(struct utcp *utcp, bool offline) {
2432         struct timespec now;
2433         clock_gettime(UTCP_CLOCK, &now);
2434
2435         for(int i = 0; i < utcp->nconnections; i++) {
2436                 struct utcp_connection *c = utcp->connections[i];
2437
2438                 if(c->reapable) {
2439                         continue;
2440                 }
2441
2442                 utcp_expect_data(c, offline);
2443
2444                 if(!offline) {
2445                         if(timespec_isset(&c->rtrx_timeout)) {
2446                                 c->rtrx_timeout = now;
2447                         }
2448
2449                         utcp->connections[i]->rtt_start.tv_sec = 0;
2450
2451                         if(c->rto > START_RTO) {
2452                                 c->rto = START_RTO;
2453                         }
2454                 }
2455         }
2456 }
2457
2458 void utcp_set_clock_granularity(long granularity) {
2459         CLOCK_GRANULARITY = granularity;
2460 }