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                 c->do_poll = true;
 820         }
 821
 822         if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
 823                 start_retransmit_timer(c);
 824         }
 825
 826         if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
 827                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 828                 c->conn_timeout.tv_sec += c->utcp->timeout;
 829         }
 830
 831         return len;
 832 }
 833
 834 static void swap_ports(struct hdr *hdr) {
 835         uint16_t tmp = hdr->src;
 836         hdr->src = hdr->dst;
 837         hdr->dst = tmp;
 838 }
 839
 840 static void fast_retransmit(struct utcp_connection *c) {
 841         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 842                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 843                 return;
 844         }
 845
 846         struct utcp *utcp = c->utcp;
 847
 848         struct {
 849                 struct hdr hdr;
 850                 uint8_t data[];
 851         } *pkt;
 852
 853         pkt = malloc(c->utcp->mtu);
 854
 855         if(!pkt) {
 856                 return;
 857         }
 858
 859         pkt->hdr.src = c->src;
 860         pkt->hdr.dst = c->dst;
 861         pkt->hdr.wnd = c->rcvbuf.maxsize;
 862         pkt->hdr.aux = 0;
 863
 864         switch(c->state) {
 865         case ESTABLISHED:
 866         case FIN_WAIT_1:
 867         case CLOSE_WAIT:
 868         case CLOSING:
 869         case LAST_ACK:
 870                 // Send unacked data again.
 871                 pkt->hdr.seq = c->snd.una;
 872                 pkt->hdr.ack = c->rcv.nxt;
 873                 pkt->hdr.ctl = ACK;
 874                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 875
 876                 if(fin_wanted(c, c->snd.una + len)) {
 877                         len--;
 878                         pkt->hdr.ctl |= FIN;
 879                 }
 880
 881                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 882                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 883                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 884                 break;
 885
 886         default:
 887                 break;
 888         }
 889
 890         free(pkt);
 891 }
 892
 893 static void retransmit(struct utcp_connection *c) {
 894         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 895                 debug(c, "retransmit() called but nothing to retransmit!\n");
 896                 stop_retransmit_timer(c);
 897                 return;
 898         }
 899
 900         struct utcp *utcp = c->utcp;
 901
 902         struct {
 903                 struct hdr hdr;
 904                 uint8_t data[];
 905         } *pkt = c->utcp->pkt;
 906
 907         pkt->hdr.src = c->src;
 908         pkt->hdr.dst = c->dst;
 909         pkt->hdr.wnd = c->rcvbuf.maxsize;
 910         pkt->hdr.aux = 0;
 911
 912         switch(c->state) {
 913         case SYN_SENT:
 914                 // Send our SYN again
 915                 pkt->hdr.seq = c->snd.iss;
 916                 pkt->hdr.ack = 0;
 917                 pkt->hdr.ctl = SYN;
 918                 pkt->hdr.aux = 0x0101;
 919                 pkt->data[0] = 1;
 920                 pkt->data[1] = 0;
 921                 pkt->data[2] = 0;
 922                 pkt->data[3] = c->flags & 0x7;
 923                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 924                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 925                 break;
 926
 927         case SYN_RECEIVED:
 928                 // Send SYNACK again
 929                 pkt->hdr.seq = c->snd.nxt;
 930                 pkt->hdr.ack = c->rcv.nxt;
 931                 pkt->hdr.ctl = SYN | ACK;
 932                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
 933                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 934                 break;
 935
 936         case ESTABLISHED:
 937         case FIN_WAIT_1:
 938         case CLOSE_WAIT:
 939         case CLOSING:
 940         case LAST_ACK:
 941                 // Send unacked data again.
 942                 pkt->hdr.seq = c->snd.una;
 943                 pkt->hdr.ack = c->rcv.nxt;
 944                 pkt->hdr.ctl = ACK;
 945                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 946
 947                 if(fin_wanted(c, c->snd.una + len)) {
 948                         len--;
 949                         pkt->hdr.ctl |= FIN;
 950                 }
 951
 952                 // RFC 5681 slow start after timeout
 953                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
 954                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
 955                 c->snd.cwnd = utcp->mss;
 956                 debug_cwnd(c);
 957
 958                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 959                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 960                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 961
 962                 c->snd.nxt = c->snd.una + len;
 963                 break;
 964
 965         case CLOSED:
 966         case LISTEN:
 967         case TIME_WAIT:
 968         case FIN_WAIT_2:
 969                 // We shouldn't need to retransmit anything in this state.
 970 #ifdef UTCP_DEBUG
 971                 abort();
 972 #endif
 973                 stop_retransmit_timer(c);
 974                 goto cleanup;
 975         }
 976
 977         start_retransmit_timer(c);
 978         c->rto *= 2;
 979
 980         if(c->rto > MAX_RTO) {
 981                 c->rto = MAX_RTO;
 982         }
 983
 984         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 985         c->dupack = 0; // cancel any ongoing fast recovery
 986
 987 cleanup:
 988         return;
 989 }
 990
 991 /* Update receive buffer and SACK entries after consuming data.
 992  *
 993  * Situation:
 994  *
 995  * |.....0000..1111111111.....22222......3333|
 996  * |---------------^
 997  *
 998  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 999  * to remove data from the receive buffer. The idea is to substract "len"
1000  * from the offset of all the SACK entries, and then remove/cut down entries
1001  * that are shifted to before the start of the receive buffer.
1002  *
1003  * There are three cases:
1004  * - the SACK entry is after ^, in that case just change the offset.
1005  * - the SACK entry starts before and ends after ^, so we have to
1006  *   change both its offset and size.
1007  * - the SACK entry is completely before ^, in that case delete it.
1008  */
1009 static void sack_consume(struct utcp_connection *c, size_t len) {
1010         debug(c, "sack_consume %lu\n", (unsigned long)len);
1011
1012         if(len > c->rcvbuf.used) {
1013                 debug(c, "all SACK entries consumed\n");
1014                 c->sacks[0].len = 0;
1015                 return;
1016         }
1017
1018         buffer_discard(&c->rcvbuf, len);
1019
1020         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1021                 if(len < c->sacks[i].offset) {
1022                         c->sacks[i].offset -= len;
1023                         i++;
1024                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1025                         c->sacks[i].len -= len - c->sacks[i].offset;
1026                         c->sacks[i].offset = 0;
1027                         i++;
1028                 } else {
1029                         if(i < NSACKS - 1) {
1030                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1031                                 c->sacks[NSACKS - 1].len = 0;
1032                         } else {
1033                                 c->sacks[i].len = 0;
1034                                 break;
1035                         }
1036                 }
1037         }
1038
1039         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1040                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1041         }
1042 }
1043
1044 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1045         debug(c, "out of order packet, offset %u\n", offset);
1046         // Packet loss or reordering occured. Store the data in the buffer.
1047         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1048
1049         if(rxd <= 0) {
1050                 debug(c, "packet outside receive buffer, dropping\n");
1051                 return;
1052         }
1053
1054         if((size_t)rxd < len) {
1055                 debug(c, "packet partially outside receive buffer\n");
1056                 len = rxd;
1057         }
1058
1059         // Make note of where we put it.
1060         for(int i = 0; i < NSACKS; i++) {
1061                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1062                         debug(c, "new SACK entry %d\n", i);
1063                         c->sacks[i].offset = offset;
1064                         c->sacks[i].len = rxd;
1065                         break;
1066                 } else if(offset < c->sacks[i].offset) {
1067                         if(offset + rxd < c->sacks[i].offset) { // insert before
1068                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1069                                         debug(c, "insert SACK entry at %d\n", i);
1070                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1071                                         c->sacks[i].offset = offset;
1072                                         c->sacks[i].len = rxd;
1073                                 } else {
1074                                         debug(c, "SACK entries full, dropping packet\n");
1075                                 }
1076
1077                                 break;
1078                         } else { // merge
1079                                 debug(c, "merge with start of SACK entry at %d\n", i);
1080                                 c->sacks[i].offset = offset;
1081                                 break;
1082                         }
1083                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1084                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1085                                 debug(c, "merge with end of SACK entry at %d\n", i);
1086                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1087                                 // TODO: handle potential merge with next entry
1088                         }
1089
1090                         break;
1091                 }
1092         }
1093
1094         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1095                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1096         }
1097 }
1098
1099 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1100         if(c->recv) {
1101                 ssize_t rxd = c->recv(c, data, len);
1102
1103                 if(rxd != (ssize_t)len) {
1104                         // TODO: handle the application not accepting all data.
1105                         abort();
1106                 }
1107         }
1108
1109         // Check if we can process out-of-order data now.
1110         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1111                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1112
1113                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1114                         size_t offset = len;
1115                         len = c->sacks[0].offset + c->sacks[0].len;
1116                         size_t remainder = len - offset;
1117                         ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
1118
1119                         if(rxd != (ssize_t)remainder) {
1120                                 // TODO: handle the application not accepting all data.
1121                                 abort();
1122                         }
1123                 }
1124         }
1125
1126         if(c->rcvbuf.used) {
1127                 sack_consume(c, len);
1128         }
1129
1130         c->rcv.nxt += len;
1131 }
1132
1133 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1134         // Fast path for unfragmented packets
1135         if(!hdr->wnd && !(hdr->ctl & MF)) {
1136                 c->recv(c, data, len);
1137                 c->rcv.nxt = hdr->seq + len;
1138                 return;
1139         }
1140
1141         // Ensure reassembled packet are not larger than 64 kiB
1142         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1143                 return;
1144         }
1145
1146         // Don't accept out of order fragments
1147         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1148                 return;
1149         }
1150
1151         // Reset the receive buffer for the first fragment
1152         if(!hdr->wnd) {
1153                 buffer_clear(&c->rcvbuf);
1154         }
1155
1156         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1157
1158         if(rxd != (ssize_t)len) {
1159                 return;
1160         }
1161
1162         // Send the packet if it's the final fragment
1163         if(!(hdr->ctl & MF)) {
1164                 buffer_call(&c->rcvbuf, c->recv, c, 0, hdr->wnd + len);
1165         }
1166
1167         c->rcv.nxt = hdr->seq + len;
1168 }
1169
1170 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1171         if(!is_reliable(c)) {
1172                 handle_unreliable(c, hdr, data, len);
1173                 return;
1174         }
1175
1176         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1177
1178         if(offset) {
1179                 handle_out_of_order(c, offset, data, len);
1180         } else {
1181                 handle_in_order(c, data, len);
1182         }
1183 }
1184
1185
1186 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1187         const uint8_t *ptr = data;
1188
1189         if(!utcp) {
1190                 errno = EFAULT;
1191                 return -1;
1192         }
1193
1194         if(!len) {
1195                 return 0;
1196         }
1197
1198         if(!data) {
1199                 errno = EFAULT;
1200                 return -1;
1201         }
1202
1203         // Drop packets smaller than the header
1204
1205         struct hdr hdr;
1206
1207         if(len < sizeof(hdr)) {
1208                 print_packet(NULL, "recv", data, len);
1209                 errno = EBADMSG;
1210                 return -1;
1211         }
1212
1213         // Make a copy from the potentially unaligned data to a struct hdr
1214
1215         memcpy(&hdr, ptr, sizeof(hdr));
1216
1217         // Try to match the packet to an existing connection
1218
1219         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1220         print_packet(c, "recv", data, len);
1221
1222         // Process the header
1223
1224         ptr += sizeof(hdr);
1225         len -= sizeof(hdr);
1226
1227         // Drop packets with an unknown CTL flag
1228
1229         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1230                 print_packet(NULL, "recv", data, len);
1231                 errno = EBADMSG;
1232                 return -1;
1233         }
1234
1235         // Check for auxiliary headers
1236
1237         const uint8_t *init = NULL;
1238
1239         uint16_t aux = hdr.aux;
1240
1241         while(aux) {
1242                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1243                 uint8_t auxtype = aux & 0xff;
1244
1245                 if(len < auxlen) {
1246                         errno = EBADMSG;
1247                         return -1;
1248                 }
1249
1250                 switch(auxtype) {
1251                 case AUX_INIT:
1252                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1253                                 errno = EBADMSG;
1254                                 return -1;
1255                         }
1256
1257                         init = ptr;
1258                         break;
1259
1260                 default:
1261                         errno = EBADMSG;
1262                         return -1;
1263                 }
1264
1265                 len -= auxlen;
1266                 ptr += auxlen;
1267
1268                 if(!(aux & 0x800)) {
1269                         break;
1270                 }
1271
1272                 if(len < 2) {
1273                         errno = EBADMSG;
1274                         return -1;
1275                 }
1276
1277                 memcpy(&aux, ptr, 2);
1278                 len -= 2;
1279                 ptr += 2;
1280         }
1281
1282         bool has_data = len || (hdr.ctl & (SYN | FIN));
1283
1284         // Is it for a new connection?
1285
1286         if(!c) {
1287                 // Ignore RST packets
1288
1289                 if(hdr.ctl & RST) {
1290                         return 0;
1291                 }
1292
1293                 // Is it a SYN packet and are we LISTENing?
1294
1295                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1296                         // If we don't want to accept it, send a RST back
1297                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1298                                 len = 1;
1299                                 goto reset;
1300                         }
1301
1302                         // Try to allocate memory, otherwise send a RST back
1303                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1304
1305                         if(!c) {
1306                                 len = 1;
1307                                 goto reset;
1308                         }
1309
1310                         // Parse auxilliary information
1311                         if(init) {
1312                                 if(init[0] < 1) {
1313                                         len = 1;
1314                                         goto reset;
1315                                 }
1316
1317                                 c->flags = init[3] & 0x7;
1318                         } else {
1319                                 c->flags = UTCP_TCP;
1320                         }
1321
1322 synack:
1323                         // Return SYN+ACK, go to SYN_RECEIVED state
1324                         c->snd.wnd = hdr.wnd;
1325                         c->rcv.irs = hdr.seq;
1326                         c->rcv.nxt = c->rcv.irs + 1;
1327                         set_state(c, SYN_RECEIVED);
1328
1329                         struct {
1330                                 struct hdr hdr;
1331                                 uint8_t data[4];
1332                         } pkt;
1333
1334                         pkt.hdr.src = c->src;
1335                         pkt.hdr.dst = c->dst;
1336                         pkt.hdr.ack = c->rcv.irs + 1;
1337                         pkt.hdr.seq = c->snd.iss;
1338                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1339                         pkt.hdr.ctl = SYN | ACK;
1340
1341                         if(init) {
1342                                 pkt.hdr.aux = 0x0101;
1343                                 pkt.data[0] = 1;
1344                                 pkt.data[1] = 0;
1345                                 pkt.data[2] = 0;
1346                                 pkt.data[3] = c->flags & 0x7;
1347                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1348                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1349                         } else {
1350                                 pkt.hdr.aux = 0;
1351                                 print_packet(c, "send", &pkt, sizeof(hdr));
1352                                 utcp->send(utcp, &pkt, sizeof(hdr));
1353                         }
1354                 } else {
1355                         // No, we don't want your packets, send a RST back
1356                         len = 1;
1357                         goto reset;
1358                 }
1359
1360                 return 0;
1361         }
1362
1363         debug(c, "state %s\n", strstate[c->state]);
1364
1365         // In case this is for a CLOSED connection, ignore the packet.
1366         // TODO: make it so incoming packets can never match a CLOSED connection.
1367
1368         if(c->state == CLOSED) {
1369                 debug(c, "got packet for closed connection\n");
1370                 return 0;
1371         }
1372
1373         // It is for an existing connection.
1374
1375         // 1. Drop invalid packets.
1376
1377         // 1a. Drop packets that should not happen in our current state.
1378
1379         switch(c->state) {
1380         case SYN_SENT:
1381         case SYN_RECEIVED:
1382         case ESTABLISHED:
1383         case FIN_WAIT_1:
1384         case FIN_WAIT_2:
1385         case CLOSE_WAIT:
1386         case CLOSING:
1387         case LAST_ACK:
1388         case TIME_WAIT:
1389                 break;
1390
1391         default:
1392 #ifdef UTCP_DEBUG
1393                 abort();
1394 #endif
1395                 break;
1396         }
1397
1398         // 1b. Discard data that is not in our receive window.
1399
1400         if(is_reliable(c)) {
1401                 bool acceptable;
1402
1403                 if(c->state == SYN_SENT) {
1404                         acceptable = true;
1405                 } else if(len == 0) {
1406                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1407                 } else {
1408                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1409
1410                         // cut already accepted front overlapping
1411                         if(rcv_offset < 0) {
1412                                 acceptable = len > (size_t) - rcv_offset;
1413
1414                                 if(acceptable) {
1415                                         ptr -= rcv_offset;
1416                                         len += rcv_offset;
1417                                         hdr.seq -= rcv_offset;
1418                                 }
1419                         } else {
1420                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1421                         }
1422                 }
1423
1424                 if(!acceptable) {
1425                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1426
1427                         // Ignore unacceptable RST packets.
1428                         if(hdr.ctl & RST) {
1429                                 return 0;
1430                         }
1431
1432                         // Otherwise, continue processing.
1433                         len = 0;
1434                 }
1435         } else {
1436 #if UTCP_DEBUG
1437                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1438
1439                 if(rcv_offset) {
1440                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1441                 }
1442
1443 #endif
1444         }
1445
1446         c->snd.wnd = hdr.wnd; // TODO: move below
1447
1448         // 1c. Drop packets with an invalid ACK.
1449         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1450         // (= snd.una + c->sndbuf.used).
1451
1452         if(!is_reliable(c)) {
1453                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1454                         hdr.ack = c->snd.una;
1455                 }
1456         }
1457
1458         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1459                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1460
1461                 // Ignore unacceptable RST packets.
1462                 if(hdr.ctl & RST) {
1463                         return 0;
1464                 }
1465
1466                 goto reset;
1467         }
1468
1469         // 2. Handle RST packets
1470
1471         if(hdr.ctl & RST) {
1472                 switch(c->state) {
1473                 case SYN_SENT:
1474                         if(!(hdr.ctl & ACK)) {
1475                                 return 0;
1476                         }
1477
1478                         // The peer has refused our connection.
1479                         set_state(c, CLOSED);
1480                         errno = ECONNREFUSED;
1481
1482                         if(c->recv) {
1483                                 c->recv(c, NULL, 0);
1484                         }
1485
1486                         if(c->poll && !c->reapable) {
1487                                 c->poll(c, 0);
1488                         }
1489
1490                         return 0;
1491
1492                 case SYN_RECEIVED:
1493                         if(hdr.ctl & ACK) {
1494                                 return 0;
1495                         }
1496
1497                         // We haven't told the application about this connection yet. Silently delete.
1498                         free_connection(c);
1499                         return 0;
1500
1501                 case ESTABLISHED:
1502                 case FIN_WAIT_1:
1503                 case FIN_WAIT_2:
1504                 case CLOSE_WAIT:
1505                         if(hdr.ctl & ACK) {
1506                                 return 0;
1507                         }
1508
1509                         // The peer has aborted our connection.
1510                         set_state(c, CLOSED);
1511                         errno = ECONNRESET;
1512
1513                         if(c->recv) {
1514                                 c->recv(c, NULL, 0);
1515                         }
1516
1517                         if(c->poll && !c->reapable) {
1518                                 c->poll(c, 0);
1519                         }
1520
1521                         return 0;
1522
1523                 case CLOSING:
1524                 case LAST_ACK:
1525                 case TIME_WAIT:
1526                         if(hdr.ctl & ACK) {
1527                                 return 0;
1528                         }
1529
1530                         // As far as the application is concerned, the connection has already been closed.
1531                         // If it has called utcp_close() already, we can immediately free this connection.
1532                         if(c->reapable) {
1533                                 free_connection(c);
1534                                 return 0;
1535                         }
1536
1537                         // Otherwise, immediately move to the CLOSED state.
1538                         set_state(c, CLOSED);
1539                         return 0;
1540
1541                 default:
1542 #ifdef UTCP_DEBUG
1543                         abort();
1544 #endif
1545                         break;
1546                 }
1547         }
1548
1549         uint32_t advanced;
1550
1551         if(!(hdr.ctl & ACK)) {
1552                 advanced = 0;
1553                 goto skip_ack;
1554         }
1555
1556         // 3. Advance snd.una
1557
1558         advanced = seqdiff(hdr.ack, c->snd.una);
1559
1560         if(advanced) {
1561                 // RTT measurement
1562                 if(c->rtt_start.tv_sec) {
1563                         if(c->rtt_seq == hdr.ack) {
1564                                 struct timespec now;
1565                                 clock_gettime(UTCP_CLOCK, &now);
1566                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1567                                 update_rtt(c, diff);
1568                                 c->rtt_start.tv_sec = 0;
1569                         } else if(c->rtt_seq < hdr.ack) {
1570                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1571                                 c->rtt_start.tv_sec = 0;
1572                         }
1573                 }
1574
1575                 int32_t data_acked = advanced;
1576
1577                 switch(c->state) {
1578                 case SYN_SENT:
1579                 case SYN_RECEIVED:
1580                         data_acked--;
1581                         break;
1582
1583                 // TODO: handle FIN as well.
1584                 default:
1585                         break;
1586                 }
1587
1588                 assert(data_acked >= 0);
1589
1590 #ifndef NDEBUG
1591                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1592                 assert(data_acked <= bufused);
1593 #endif
1594
1595                 if(data_acked) {
1596                         buffer_discard(&c->sndbuf, data_acked);
1597                         c->do_poll = true;
1598                 }
1599
1600                 // Also advance snd.nxt if possible
1601                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1602                         c->snd.nxt = hdr.ack;
1603                 }
1604
1605                 c->snd.una = hdr.ack;
1606
1607                 if(c->dupack) {
1608                         if(c->dupack >= 3) {
1609                                 debug(c, "fast recovery ended\n");
1610                                 c->snd.cwnd = c->snd.ssthresh;
1611                         }
1612
1613                         c->dupack = 0;
1614                 }
1615
1616                 // Increase the congestion window according to RFC 5681
1617                 if(c->snd.cwnd < c->snd.ssthresh) {
1618                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1619                 } else {
1620                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1621                 }
1622
1623                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1624                         c->snd.cwnd = c->sndbuf.maxsize;
1625                 }
1626
1627                 debug_cwnd(c);
1628
1629                 // Check if we have sent a FIN that is now ACKed.
1630                 switch(c->state) {
1631                 case FIN_WAIT_1:
1632                         if(c->snd.una == c->snd.last) {
1633                                 set_state(c, FIN_WAIT_2);
1634                         }
1635
1636                         break;
1637
1638                 case CLOSING:
1639                         if(c->snd.una == c->snd.last) {
1640                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1641                                 c->conn_timeout.tv_sec += utcp->timeout;
1642                                 set_state(c, TIME_WAIT);
1643                         }
1644
1645                         break;
1646
1647                 default:
1648                         break;
1649                 }
1650         } else {
1651                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1652                         c->dupack++;
1653                         debug(c, "duplicate ACK %d\n", c->dupack);
1654
1655                         if(c->dupack == 3) {
1656                                 // RFC 5681 fast recovery
1657                                 debug(c, "fast recovery started\n", c->dupack);
1658                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1659                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1660                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1661
1662                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1663                                         c->snd.cwnd = c->sndbuf.maxsize;
1664                                 }
1665
1666                                 debug_cwnd(c);
1667
1668                                 fast_retransmit(c);
1669                         } else if(c->dupack > 3) {
1670                                 c->snd.cwnd += utcp->mss;
1671
1672                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1673                                         c->snd.cwnd = c->sndbuf.maxsize;
1674                                 }
1675
1676                                 debug_cwnd(c);
1677                         }
1678
1679                         // We got an ACK which indicates the other side did get one of our packets.
1680                         // Reset the retransmission timer to avoid going to slow start,
1681                         // but don't touch the connection timeout.
1682                         start_retransmit_timer(c);
1683                 }
1684         }
1685
1686         // 4. Update timers
1687
1688         if(advanced) {
1689                 if(c->snd.una == c->snd.last) {
1690                         stop_retransmit_timer(c);
1691                         timespec_clear(&c->conn_timeout);
1692                 } else if(is_reliable(c)) {
1693                         start_retransmit_timer(c);
1694                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1695                         c->conn_timeout.tv_sec += utcp->timeout;
1696                 }
1697         }
1698
1699 skip_ack:
1700         // 5. Process SYN stuff
1701
1702         if(hdr.ctl & SYN) {
1703                 switch(c->state) {
1704                 case SYN_SENT:
1705
1706                         // This is a SYNACK. It should always have ACKed the SYN.
1707                         if(!advanced) {
1708                                 goto reset;
1709                         }
1710
1711                         c->rcv.irs = hdr.seq;
1712                         c->rcv.nxt = hdr.seq;
1713
1714                         if(c->shut_wr) {
1715                                 c->snd.last++;
1716                                 set_state(c, FIN_WAIT_1);
1717                         } else {
1718                                 set_state(c, ESTABLISHED);
1719                         }
1720
1721                         // TODO: notify application of this somehow.
1722                         break;
1723
1724                 case SYN_RECEIVED:
1725                         // This is a retransmit of a SYN, send back the SYNACK.
1726                         goto synack;
1727
1728                 case ESTABLISHED:
1729                 case FIN_WAIT_1:
1730                 case FIN_WAIT_2:
1731                 case CLOSE_WAIT:
1732                 case CLOSING:
1733                 case LAST_ACK:
1734                 case TIME_WAIT:
1735                         // Ehm, no. We should never receive a second SYN.
1736                         return 0;
1737
1738                 default:
1739 #ifdef UTCP_DEBUG
1740                         abort();
1741 #endif
1742                         return 0;
1743                 }
1744
1745                 // SYN counts as one sequence number
1746                 c->rcv.nxt++;
1747         }
1748
1749         // 6. Process new data
1750
1751         if(c->state == SYN_RECEIVED) {
1752                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1753                 if(!advanced) {
1754                         goto reset;
1755                 }
1756
1757                 // Are we still LISTENing?
1758                 if(utcp->accept) {
1759                         utcp->accept(c, c->src);
1760                 }
1761
1762                 if(c->state != ESTABLISHED) {
1763                         set_state(c, CLOSED);
1764                         c->reapable = true;
1765                         goto reset;
1766                 }
1767         }
1768
1769         if(len) {
1770                 switch(c->state) {
1771                 case SYN_SENT:
1772                 case SYN_RECEIVED:
1773                         // This should never happen.
1774 #ifdef UTCP_DEBUG
1775                         abort();
1776 #endif
1777                         return 0;
1778
1779                 case ESTABLISHED:
1780                 case FIN_WAIT_1:
1781                 case FIN_WAIT_2:
1782                         break;
1783
1784                 case CLOSE_WAIT:
1785                 case CLOSING:
1786                 case LAST_ACK:
1787                 case TIME_WAIT:
1788                         // Ehm no, We should never receive more data after a FIN.
1789                         goto reset;
1790
1791                 default:
1792 #ifdef UTCP_DEBUG
1793                         abort();
1794 #endif
1795                         return 0;
1796                 }
1797
1798                 handle_incoming_data(c, &hdr, ptr, len);
1799         }
1800
1801         // 7. Process FIN stuff
1802
1803         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1804                 switch(c->state) {
1805                 case SYN_SENT:
1806                 case SYN_RECEIVED:
1807                         // This should never happen.
1808 #ifdef UTCP_DEBUG
1809                         abort();
1810 #endif
1811                         break;
1812
1813                 case ESTABLISHED:
1814                         set_state(c, CLOSE_WAIT);
1815                         break;
1816
1817                 case FIN_WAIT_1:
1818                         set_state(c, CLOSING);
1819                         break;
1820
1821                 case FIN_WAIT_2:
1822                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1823                         c->conn_timeout.tv_sec += utcp->timeout;
1824                         set_state(c, TIME_WAIT);
1825                         break;
1826
1827                 case CLOSE_WAIT:
1828                 case CLOSING:
1829                 case LAST_ACK:
1830                 case TIME_WAIT:
1831                         // Ehm, no. We should never receive a second FIN.
1832                         goto reset;
1833
1834                 default:
1835 #ifdef UTCP_DEBUG
1836                         abort();
1837 #endif
1838                         break;
1839                 }
1840
1841                 // FIN counts as one sequence number
1842                 c->rcv.nxt++;
1843                 len++;
1844
1845                 // Inform the application that the peer closed its end of the connection.
1846                 if(c->recv) {
1847                         errno = 0;
1848                         c->recv(c, NULL, 0);
1849                 }
1850         }
1851
1852         // Now we send something back if:
1853         // - we received data, so we have to send back an ACK
1854         //   -> sendatleastone = true
1855         // - or we got an ack, so we should maybe send a bit more data
1856         //   -> sendatleastone = false
1857
1858         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1859                 ack(c, has_data);
1860         }
1861
1862         return 0;
1863
1864 reset:
1865         swap_ports(&hdr);
1866         hdr.wnd = 0;
1867         hdr.aux = 0;
1868
1869         if(hdr.ctl & ACK) {
1870                 hdr.seq = hdr.ack;
1871                 hdr.ctl = RST;
1872         } else {
1873                 hdr.ack = hdr.seq + len;
1874                 hdr.seq = 0;
1875                 hdr.ctl = RST | ACK;
1876         }
1877
1878         print_packet(c, "send", &hdr, sizeof(hdr));
1879         utcp->send(utcp, &hdr, sizeof(hdr));
1880         return 0;
1881
1882 }
1883
1884 int utcp_shutdown(struct utcp_connection *c, int dir) {
1885         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1886
1887         if(!c) {
1888                 errno = EFAULT;
1889                 return -1;
1890         }
1891
1892         if(c->reapable) {
1893                 debug(c, "shutdown() called on closed connection\n");
1894                 errno = EBADF;
1895                 return -1;
1896         }
1897
1898         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1899                 errno = EINVAL;
1900                 return -1;
1901         }
1902
1903         // TCP does not have a provision for stopping incoming packets.
1904         // The best we can do is to just ignore them.
1905         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1906                 c->recv = NULL;
1907         }
1908
1909         // The rest of the code deals with shutting down writes.
1910         if(dir == UTCP_SHUT_RD) {
1911                 return 0;
1912         }
1913
1914         // Only process shutting down writes once.
1915         if(c->shut_wr) {
1916                 return 0;
1917         }
1918
1919         c->shut_wr = true;
1920
1921         switch(c->state) {
1922         case CLOSED:
1923         case LISTEN:
1924                 errno = ENOTCONN;
1925                 return -1;
1926
1927         case SYN_SENT:
1928                 return 0;
1929
1930         case SYN_RECEIVED:
1931         case ESTABLISHED:
1932                 set_state(c, FIN_WAIT_1);
1933                 break;
1934
1935         case FIN_WAIT_1:
1936         case FIN_WAIT_2:
1937                 return 0;
1938
1939         case CLOSE_WAIT:
1940                 set_state(c, CLOSING);
1941                 break;
1942
1943         case CLOSING:
1944         case LAST_ACK:
1945         case TIME_WAIT:
1946                 return 0;
1947         }
1948
1949         c->snd.last++;
1950
1951         ack(c, false);
1952
1953         if(!timespec_isset(&c->rtrx_timeout)) {
1954                 start_retransmit_timer(c);
1955         }
1956
1957         return 0;
1958 }
1959
1960 static bool reset_connection(struct utcp_connection *c) {
1961         if(!c) {
1962                 errno = EFAULT;
1963                 return false;
1964         }
1965
1966         if(c->reapable) {
1967                 debug(c, "abort() called on closed connection\n");
1968                 errno = EBADF;
1969                 return false;
1970         }
1971
1972         c->recv = NULL;
1973         c->poll = NULL;
1974
1975         switch(c->state) {
1976         case CLOSED:
1977                 return true;
1978
1979         case LISTEN:
1980         case SYN_SENT:
1981         case CLOSING:
1982         case LAST_ACK:
1983         case TIME_WAIT:
1984                 set_state(c, CLOSED);
1985                 return true;
1986
1987         case SYN_RECEIVED:
1988         case ESTABLISHED:
1989         case FIN_WAIT_1:
1990         case FIN_WAIT_2:
1991         case CLOSE_WAIT:
1992                 set_state(c, CLOSED);
1993                 break;
1994         }
1995
1996         // Send RST
1997
1998         struct hdr hdr;
1999
2000         hdr.src = c->src;
2001         hdr.dst = c->dst;
2002         hdr.seq = c->snd.nxt;
2003         hdr.ack = 0;
2004         hdr.wnd = 0;
2005         hdr.ctl = RST;
2006
2007         print_packet(c, "send", &hdr, sizeof(hdr));
2008         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2009         return true;
2010 }
2011
2012 // Closes all the opened connections
2013 void utcp_abort_all_connections(struct utcp *utcp) {
2014         if(!utcp) {
2015                 errno = EINVAL;
2016                 return;
2017         }
2018
2019         for(int i = 0; i < utcp->nconnections; i++) {
2020                 struct utcp_connection *c = utcp->connections[i];
2021
2022                 if(c->reapable || c->state == CLOSED) {
2023                         continue;
2024                 }
2025
2026                 utcp_recv_t old_recv = c->recv;
2027                 utcp_poll_t old_poll = c->poll;
2028
2029                 reset_connection(c);
2030
2031                 if(old_recv) {
2032                         errno = 0;
2033                         old_recv(c, NULL, 0);
2034                 }
2035
2036                 if(old_poll && !c->reapable) {
2037                         errno = 0;
2038                         old_poll(c, 0);
2039                 }
2040         }
2041
2042         return;
2043 }
2044
2045 int utcp_close(struct utcp_connection *c) {
2046         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2047                 return -1;
2048         }
2049
2050         c->recv = NULL;
2051         c->poll = NULL;
2052         c->reapable = true;
2053         return 0;
2054 }
2055
2056 int utcp_abort(struct utcp_connection *c) {
2057         if(!reset_connection(c)) {
2058                 return -1;
2059         }
2060
2061         c->reapable = true;
2062         return 0;
2063 }
2064
2065 /* Handle timeouts.
2066  * One call to this function will loop through all connections,
2067  * checking if something needs to be resent or not.
2068  * The return value is the time to the next timeout in milliseconds,
2069  * or maybe a negative value if the timeout is infinite.
2070  */
2071 struct timespec utcp_timeout(struct utcp *utcp) {
2072         struct timespec now;
2073         clock_gettime(UTCP_CLOCK, &now);
2074         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2075
2076         for(int i = 0; i < utcp->nconnections; i++) {
2077                 struct utcp_connection *c = utcp->connections[i];
2078
2079                 if(!c) {
2080                         continue;
2081                 }
2082
2083                 // delete connections that have been utcp_close()d.
2084                 if(c->state == CLOSED) {
2085                         if(c->reapable) {
2086                                 debug(c, "reaping\n");
2087                                 free_connection(c);
2088                                 i--;
2089                         }
2090
2091                         continue;
2092                 }
2093
2094                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2095                         errno = ETIMEDOUT;
2096                         c->state = CLOSED;
2097
2098                         if(c->recv) {
2099                                 c->recv(c, NULL, 0);
2100                         }
2101
2102                         if(c->poll && !c->reapable) {
2103                                 c->poll(c, 0);
2104                         }
2105
2106                         continue;
2107                 }
2108
2109                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2110                         debug(c, "retransmitting after timeout\n");
2111                         retransmit(c);
2112                 }
2113
2114                 if(c->poll) {
2115                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2116                                 c->do_poll = false;
2117                                 uint32_t len = buffer_free(&c->sndbuf);
2118
2119                                 if(len) {
2120                                         c->poll(c, len);
2121                                 }
2122                         } else if(c->state == CLOSED) {
2123                                 c->poll(c, 0);
2124                         }
2125                 }
2126
2127                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2128                         next = c->conn_timeout;
2129                 }
2130
2131                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2132                         next = c->rtrx_timeout;
2133                 }
2134         }
2135
2136         struct timespec diff;
2137
2138         timespec_sub(&next, &now, &diff);
2139
2140         return diff;
2141 }
2142
2143 bool utcp_is_active(struct utcp *utcp) {
2144         if(!utcp) {
2145                 return false;
2146         }
2147
2148         for(int i = 0; i < utcp->nconnections; i++)
2149                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2150                         return true;
2151                 }
2152
2153         return false;
2154 }
2155
2156 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2157         if(!send) {
2158                 errno = EFAULT;
2159                 return NULL;
2160         }
2161
2162         struct utcp *utcp = calloc(1, sizeof(*utcp));
2163
2164         if(!utcp) {
2165                 return NULL;
2166         }
2167
2168         if(!CLOCK_GRANULARITY) {
2169                 struct timespec res;
2170                 clock_getres(UTCP_CLOCK, &res);
2171                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2172         }
2173
2174         utcp->accept = accept;
2175         utcp->pre_accept = pre_accept;
2176         utcp->send = send;
2177         utcp->priv = priv;
2178         utcp_set_mtu(utcp, DEFAULT_MTU);
2179         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2180
2181         return utcp;
2182 }
2183
2184 void utcp_exit(struct utcp *utcp) {
2185         if(!utcp) {
2186                 return;
2187         }
2188
2189         for(int i = 0; i < utcp->nconnections; i++) {
2190                 struct utcp_connection *c = utcp->connections[i];
2191
2192                 if(!c->reapable) {
2193                         if(c->recv) {
2194                                 c->recv(c, NULL, 0);
2195                         }
2196
2197                         if(c->poll && !c->reapable) {
2198                                 c->poll(c, 0);
2199                         }
2200                 }
2201
2202                 buffer_exit(&c->rcvbuf);
2203                 buffer_exit(&c->sndbuf);
2204                 free(c);
2205         }
2206
2207         free(utcp->connections);
2208         free(utcp->pkt);
2209         free(utcp);
2210 }
2211
2212 uint16_t utcp_get_mtu(struct utcp *utcp) {
2213         return utcp ? utcp->mtu : 0;
2214 }
2215
2216 uint16_t utcp_get_mss(struct utcp *utcp) {
2217         return utcp ? utcp->mss : 0;
2218 }
2219
2220 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2221         if(!utcp) {
2222                 return;
2223         }
2224
2225         if(mtu <= sizeof(struct hdr)) {
2226                 return;
2227         }
2228
2229         if(mtu > utcp->mtu) {
2230                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2231
2232                 if(!new) {
2233                         return;
2234                 }
2235
2236                 utcp->pkt = new;
2237         }
2238
2239         utcp->mtu = mtu;
2240         utcp->mss = mtu - sizeof(struct hdr);
2241 }
2242
2243 void utcp_reset_timers(struct utcp *utcp) {
2244         if(!utcp) {
2245                 return;
2246         }
2247
2248         struct timespec now, then;
2249
2250         clock_gettime(UTCP_CLOCK, &now);
2251
2252         then = now;
2253
2254         then.tv_sec += utcp->timeout;
2255
2256         for(int i = 0; i < utcp->nconnections; i++) {
2257                 struct utcp_connection *c = utcp->connections[i];
2258
2259                 if(c->reapable) {
2260                         continue;
2261                 }
2262
2263                 if(timespec_isset(&c->rtrx_timeout)) {
2264                         c->rtrx_timeout = now;
2265                 }
2266
2267                 if(timespec_isset(&c->conn_timeout)) {
2268                         c->conn_timeout = then;
2269                 }
2270
2271                 c->rtt_start.tv_sec = 0;
2272
2273                 if(c->rto > START_RTO) {
2274                         c->rto = START_RTO;
2275                 }
2276         }
2277 }
2278
2279 int utcp_get_user_timeout(struct utcp *u) {
2280         return u ? u->timeout : 0;
2281 }
2282
2283 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2284         if(u) {
2285                 u->timeout = timeout;
2286         }
2287 }
2288
2289 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2290         return c ? c->sndbuf.maxsize : 0;
2291 }
2292
2293 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2294         if(!c) {
2295                 return 0;
2296         }
2297
2298         switch(c->state) {
2299         case SYN_SENT:
2300         case SYN_RECEIVED:
2301         case ESTABLISHED:
2302         case CLOSE_WAIT:
2303                 return buffer_free(&c->sndbuf);
2304
2305         default:
2306                 return 0;
2307         }
2308 }
2309
2310 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2311         if(!c) {
2312                 return;
2313         }
2314
2315         c->sndbuf.maxsize = size;
2316
2317         if(c->sndbuf.maxsize != size) {
2318                 c->sndbuf.maxsize = -1;
2319         }
2320
2321         c->do_poll = buffer_free(&c->sndbuf);
2322 }
2323
2324 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2325         return c ? c->rcvbuf.maxsize : 0;
2326 }
2327
2328 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2329         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2330                 return buffer_free(&c->rcvbuf);
2331         } else {
2332                 return 0;
2333         }
2334 }
2335
2336 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2337         if(!c) {
2338                 return;
2339         }
2340
2341         c->rcvbuf.maxsize = size;
2342
2343         if(c->rcvbuf.maxsize != size) {
2344                 c->rcvbuf.maxsize = -1;
2345         }
2346 }
2347
2348 size_t utcp_get_sendq(struct utcp_connection *c) {
2349         return c->sndbuf.used;
2350 }
2351
2352 size_t utcp_get_recvq(struct utcp_connection *c) {
2353         return c->rcvbuf.used;
2354 }
2355
2356 bool utcp_get_nodelay(struct utcp_connection *c) {
2357         return c ? c->nodelay : false;
2358 }
2359
2360 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2361         if(c) {
2362                 c->nodelay = nodelay;
2363         }
2364 }
2365
2366 bool utcp_get_keepalive(struct utcp_connection *c) {
2367         return c ? c->keepalive : false;
2368 }
2369
2370 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2371         if(c) {
2372                 c->keepalive = keepalive;
2373         }
2374 }
2375
2376 size_t utcp_get_outq(struct utcp_connection *c) {
2377         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2378 }
2379
2380 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2381         if(c) {
2382                 c->recv = recv;
2383         }
2384 }
2385
2386 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2387         if(c) {
2388                 c->poll = poll;
2389                 c->do_poll = buffer_free(&c->sndbuf);
2390         }
2391 }
2392
2393 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2394         if(utcp) {
2395                 utcp->accept = accept;
2396                 utcp->pre_accept = pre_accept;
2397         }
2398 }
2399
2400 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2401         if(!c || c->reapable) {
2402                 return;
2403         }
2404
2405         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2406                 return;
2407         }
2408
2409         if(expect) {
2410                 // If we expect data, start the connection timer.
2411                 if(!timespec_isset(&c->conn_timeout)) {
2412                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2413                         c->conn_timeout.tv_sec += c->utcp->timeout;
2414                 }
2415         } else {
2416                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2417                 if(c->snd.una == c->snd.last) {
2418                         timespec_clear(&c->conn_timeout);
2419                 }
2420         }
2421 }
2422
2423 void utcp_offline(struct utcp *utcp, bool offline) {
2424         struct timespec now;
2425         clock_gettime(UTCP_CLOCK, &now);
2426
2427         for(int i = 0; i < utcp->nconnections; i++) {
2428                 struct utcp_connection *c = utcp->connections[i];
2429
2430                 if(c->reapable) {
2431                         continue;
2432                 }
2433
2434                 utcp_expect_data(c, offline);
2435
2436                 if(!offline) {
2437                         if(timespec_isset(&c->rtrx_timeout)) {
2438                                 c->rtrx_timeout = now;
2439                         }
2440
2441                         utcp->connections[i]->rtt_start.tv_sec = 0;
2442
2443                         if(c->rto > START_RTO) {
2444                                 c->rto = START_RTO;
2445                         }
2446                 }
2447         }
2448 }
2449
2450 void utcp_set_clock_granularity(long granularity) {
2451         CLOCK_GRANULARITY = granularity;
2452 }