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         int64_t diff = (a->tv_sec - b->tv_sec) * 1000000000 + a->tv_sec - b->tv_sec;
  65         return diff / 1000;
  66 }
  67
  68 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
  69         if(a->tv_sec == b->tv_sec) {
  70                 return a->tv_nsec < b->tv_nsec;
  71         } else {
  72                 return a->tv_sec < b->tv_sec;
  73         }
  74 }
  75
  76 static void timespec_clear(struct timespec *a) {
  77         a->tv_sec = 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;
 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->utcp = utcp;
 529
 530         // Add it to the sorted list of connections
 531
 532         utcp->connections[utcp->nconnections++] = c;
 533         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 534
 535         return c;
 536 }
 537
 538 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 539         if(a > b) {
 540                 return a - b;
 541         } else {
 542                 return b - a;
 543         }
 544 }
 545
 546 // Update RTT variables. See RFC 6298.
 547 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 548         if(!rtt) {
 549                 debug(c, "invalid rtt\n");
 550                 return;
 551         }
 552
 553         struct utcp *utcp = c->utcp;
 554
 555         if(!utcp->srtt) {
 556                 utcp->srtt = rtt;
 557                 utcp->rttvar = rtt / 2;
 558         } else {
 559                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 560                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 561         }
 562
 563         utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
 564
 565         if(utcp->rto > MAX_RTO) {
 566                 utcp->rto = MAX_RTO;
 567         }
 568
 569         debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 570 }
 571
 572 static void start_retransmit_timer(struct utcp_connection *c) {
 573         clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
 574
 575         uint32_t rto = c->utcp->rto;
 576
 577         while(rto > USEC_PER_SEC) {
 578                 c->rtrx_timeout.tv_sec++;
 579                 rto -= USEC_PER_SEC;
 580         }
 581
 582         c->rtrx_timeout.tv_nsec += c->utcp->rto * 1000;
 583
 584         if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
 585                 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
 586                 c->rtrx_timeout.tv_sec++;
 587         }
 588
 589         debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
 590 }
 591
 592 static void stop_retransmit_timer(struct utcp_connection *c) {
 593         timespec_clear(&c->rtrx_timeout);
 594         debug(c, "rtrx_timeout cleared\n");
 595 }
 596
 597 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 598         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 599
 600         if(!c) {
 601                 return NULL;
 602         }
 603
 604         assert((flags & ~0x1f) == 0);
 605
 606         c->flags = flags;
 607         c->recv = recv;
 608         c->priv = priv;
 609
 610         struct {
 611                 struct hdr hdr;
 612                 uint8_t init[4];
 613         } pkt;
 614
 615         pkt.hdr.src = c->src;
 616         pkt.hdr.dst = c->dst;
 617         pkt.hdr.seq = c->snd.iss;
 618         pkt.hdr.ack = 0;
 619         pkt.hdr.wnd = c->rcvbuf.maxsize;
 620         pkt.hdr.ctl = SYN;
 621         pkt.hdr.aux = 0x0101;
 622         pkt.init[0] = 1;
 623         pkt.init[1] = 0;
 624         pkt.init[2] = 0;
 625         pkt.init[3] = flags & 0x7;
 626
 627         set_state(c, SYN_SENT);
 628
 629         print_packet(c, "send", &pkt, sizeof(pkt));
 630         utcp->send(utcp, &pkt, sizeof(pkt));
 631
 632         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 633         c->conn_timeout.tv_sec += utcp->timeout;
 634
 635         start_retransmit_timer(c);
 636
 637         return c;
 638 }
 639
 640 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 641         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 642 }
 643
 644 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 645         if(c->reapable || c->state != SYN_RECEIVED) {
 646                 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
 647                 return;
 648         }
 649
 650         debug(c, "accepted %p %p\n", c, recv, priv);
 651         c->recv = recv;
 652         c->priv = priv;
 653         set_state(c, ESTABLISHED);
 654 }
 655
 656 static void ack(struct utcp_connection *c, bool sendatleastone) {
 657         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 658         int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
 659
 660         assert(left >= 0);
 661
 662         if(cwndleft <= 0) {
 663                 left = 0;
 664         } else if(cwndleft < left) {
 665                 left = cwndleft;
 666
 667                 if(!sendatleastone || cwndleft > c->utcp->mss) {
 668                         left -= left % c->utcp->mss;
 669                 }
 670         }
 671
 672         debug(c, "cwndleft %d left %d\n", cwndleft, left);
 673
 674         if(!left && !sendatleastone) {
 675                 return;
 676         }
 677
 678         struct {
 679                 struct hdr hdr;
 680                 uint8_t data[];
 681         } *pkt = c->utcp->pkt;
 682
 683         pkt->hdr.src = c->src;
 684         pkt->hdr.dst = c->dst;
 685         pkt->hdr.ack = c->rcv.nxt;
 686         pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
 687         pkt->hdr.ctl = ACK;
 688         pkt->hdr.aux = 0;
 689
 690         do {
 691                 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
 692                 pkt->hdr.seq = c->snd.nxt;
 693
 694                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 695
 696                 c->snd.nxt += seglen;
 697                 left -= seglen;
 698
 699                 if(!is_reliable(c)) {
 700                         if(left) {
 701                                 pkt->hdr.ctl |= MF;
 702                         } else {
 703                                 pkt->hdr.ctl &= ~MF;
 704                         }
 705                 }
 706
 707                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 708                         seglen--;
 709                         pkt->hdr.ctl |= FIN;
 710                 }
 711
 712                 if(!c->rtt_start.tv_sec) {
 713                         // Start RTT measurement
 714                         clock_gettime(UTCP_CLOCK, &c->rtt_start);
 715                         c->rtt_seq = pkt->hdr.seq + seglen;
 716                         debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 717                 }
 718
 719                 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 720                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 721
 722                 if(left && !is_reliable(c)) {
 723                         pkt->hdr.wnd += seglen;
 724                 }
 725         } while(left);
 726 }
 727
 728 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 729         if(c->reapable) {
 730                 debug(c, "send() called on closed connection\n");
 731                 errno = EBADF;
 732                 return -1;
 733         }
 734
 735         switch(c->state) {
 736         case CLOSED:
 737         case LISTEN:
 738                 debug(c, "send() called on unconnected connection\n");
 739                 errno = ENOTCONN;
 740                 return -1;
 741
 742         case SYN_SENT:
 743         case SYN_RECEIVED:
 744         case ESTABLISHED:
 745         case CLOSE_WAIT:
 746                 break;
 747
 748         case FIN_WAIT_1:
 749         case FIN_WAIT_2:
 750         case CLOSING:
 751         case LAST_ACK:
 752         case TIME_WAIT:
 753                 debug(c, "send() called on closed connection\n");
 754                 errno = EPIPE;
 755                 return -1;
 756         }
 757
 758         // Exit early if we have nothing to send.
 759
 760         if(!len) {
 761                 return 0;
 762         }
 763
 764         if(!data) {
 765                 errno = EFAULT;
 766                 return -1;
 767         }
 768
 769         // Check if we need to be able to buffer all data
 770
 771         if(c->flags & UTCP_NO_PARTIAL) {
 772                 if(len > buffer_free(&c->sndbuf)) {
 773                         if(len > c->sndbuf.maxsize) {
 774                                 errno = EMSGSIZE;
 775                                 return -1;
 776                         } else {
 777                                 errno = EWOULDBLOCK;
 778                                 return 0;
 779                         }
 780                 }
 781         }
 782
 783         // Add data to send buffer.
 784
 785         if(is_reliable(c)) {
 786                 len = buffer_put(&c->sndbuf, data, len);
 787         } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
 788                 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
 789                         errno = EMSGSIZE;
 790                         return -1;
 791                 }
 792         } else {
 793                 return 0;
 794         }
 795
 796         if(len <= 0) {
 797                 if(is_reliable(c)) {
 798                         errno = EWOULDBLOCK;
 799                         return 0;
 800                 } else {
 801                         return len;
 802                 }
 803         }
 804
 805         c->snd.last += len;
 806
 807         // Don't send anything yet if the connection has not fully established yet
 808
 809         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 810                 return len;
 811         }
 812
 813         ack(c, false);
 814
 815         if(!is_reliable(c)) {
 816                 c->snd.una = c->snd.nxt = c->snd.last;
 817                 buffer_discard(&c->sndbuf, c->sndbuf.used);
 818                 c->do_poll = true;
 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         utcp->rto *= 2;
 978
 979         if(utcp->rto > MAX_RTO) {
 980                 utcp->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 || (size_t)rxd < len) {
1049                 abort();
1050         }
1051
1052         // Make note of where we put it.
1053         for(int i = 0; i < NSACKS; i++) {
1054                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1055                         debug(c, "new SACK entry %d\n", i);
1056                         c->sacks[i].offset = offset;
1057                         c->sacks[i].len = rxd;
1058                         break;
1059                 } else if(offset < c->sacks[i].offset) {
1060                         if(offset + rxd < c->sacks[i].offset) { // insert before
1061                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1062                                         debug(c, "insert SACK entry at %d\n", i);
1063                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1064                                         c->sacks[i].offset = offset;
1065                                         c->sacks[i].len = rxd;
1066                                 } else {
1067                                         debug(c, "SACK entries full, dropping packet\n");
1068                                 }
1069
1070                                 break;
1071                         } else { // merge
1072                                 debug(c, "merge with start of SACK entry at %d\n", i);
1073                                 c->sacks[i].offset = offset;
1074                                 break;
1075                         }
1076                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1077                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1078                                 debug(c, "merge with end of SACK entry at %d\n", i);
1079                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1080                                 // TODO: handle potential merge with next entry
1081                         }
1082
1083                         break;
1084                 }
1085         }
1086
1087         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1088                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1089         }
1090 }
1091
1092 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1093         if(c->recv) {
1094                 ssize_t rxd = c->recv(c, data, len);
1095
1096                 if(rxd != (ssize_t)len) {
1097                         // TODO: handle the application not accepting all data.
1098                         abort();
1099                 }
1100         }
1101
1102         // Check if we can process out-of-order data now.
1103         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1104                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1105
1106                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1107                         size_t offset = len;
1108                         len = c->sacks[0].offset + c->sacks[0].len;
1109                         size_t remainder = len - offset;
1110                         ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
1111
1112                         if(rxd != (ssize_t)remainder) {
1113                                 // TODO: handle the application not accepting all data.
1114                                 abort();
1115                         }
1116                 }
1117         }
1118
1119         if(c->rcvbuf.used) {
1120                 sack_consume(c, len);
1121         }
1122
1123         c->rcv.nxt += len;
1124 }
1125
1126 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1127         // Fast path for unfragmented packets
1128         if(!hdr->wnd && !(hdr->ctl & MF)) {
1129                 c->recv(c, data, len);
1130                 c->rcv.nxt = hdr->seq + len;
1131                 return;
1132         }
1133
1134         // Ensure reassembled packet are not larger than 64 kiB
1135         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1136                 return;
1137         }
1138
1139         // Don't accept out of order fragments
1140         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1141                 return;
1142         }
1143
1144         // Reset the receive buffer for the first fragment
1145         if(!hdr->wnd) {
1146                 buffer_clear(&c->rcvbuf);
1147         }
1148
1149         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1150
1151         if(rxd != (ssize_t)len) {
1152                 return;
1153         }
1154
1155         // Send the packet if it's the final fragment
1156         if(!(hdr->ctl & MF)) {
1157                 buffer_call(&c->rcvbuf, c->recv, c, 0, hdr->wnd + len);
1158         }
1159
1160         c->rcv.nxt = hdr->seq + len;
1161 }
1162
1163 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1164         if(!is_reliable(c)) {
1165                 handle_unreliable(c, hdr, data, len);
1166                 return;
1167         }
1168
1169         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1170
1171         if(offset + len > c->rcvbuf.maxsize) {
1172                 abort();
1173         }
1174
1175         if(offset) {
1176                 handle_out_of_order(c, offset, data, len);
1177         } else {
1178                 handle_in_order(c, data, len);
1179         }
1180 }
1181
1182
1183 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1184         const uint8_t *ptr = data;
1185
1186         if(!utcp) {
1187                 errno = EFAULT;
1188                 return -1;
1189         }
1190
1191         if(!len) {
1192                 return 0;
1193         }
1194
1195         if(!data) {
1196                 errno = EFAULT;
1197                 return -1;
1198         }
1199
1200         // Drop packets smaller than the header
1201
1202         struct hdr hdr;
1203
1204         if(len < sizeof(hdr)) {
1205                 print_packet(NULL, "recv", data, len);
1206                 errno = EBADMSG;
1207                 return -1;
1208         }
1209
1210         // Make a copy from the potentially unaligned data to a struct hdr
1211
1212         memcpy(&hdr, ptr, sizeof(hdr));
1213
1214         // Try to match the packet to an existing connection
1215
1216         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1217         print_packet(c, "recv", data, len);
1218
1219         // Process the header
1220
1221         ptr += sizeof(hdr);
1222         len -= sizeof(hdr);
1223
1224         // Drop packets with an unknown CTL flag
1225
1226         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1227                 print_packet(NULL, "recv", data, len);
1228                 errno = EBADMSG;
1229                 return -1;
1230         }
1231
1232         // Check for auxiliary headers
1233
1234         const uint8_t *init = NULL;
1235
1236         uint16_t aux = hdr.aux;
1237
1238         while(aux) {
1239                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1240                 uint8_t auxtype = aux & 0xff;
1241
1242                 if(len < auxlen) {
1243                         errno = EBADMSG;
1244                         return -1;
1245                 }
1246
1247                 switch(auxtype) {
1248                 case AUX_INIT:
1249                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1250                                 errno = EBADMSG;
1251                                 return -1;
1252                         }
1253
1254                         init = ptr;
1255                         break;
1256
1257                 default:
1258                         errno = EBADMSG;
1259                         return -1;
1260                 }
1261
1262                 len -= auxlen;
1263                 ptr += auxlen;
1264
1265                 if(!(aux & 0x800)) {
1266                         break;
1267                 }
1268
1269                 if(len < 2) {
1270                         errno = EBADMSG;
1271                         return -1;
1272                 }
1273
1274                 memcpy(&aux, ptr, 2);
1275                 len -= 2;
1276                 ptr += 2;
1277         }
1278
1279         bool has_data = len || (hdr.ctl & (SYN | FIN));
1280
1281         // Is it for a new connection?
1282
1283         if(!c) {
1284                 // Ignore RST packets
1285
1286                 if(hdr.ctl & RST) {
1287                         return 0;
1288                 }
1289
1290                 // Is it a SYN packet and are we LISTENing?
1291
1292                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1293                         // If we don't want to accept it, send a RST back
1294                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1295                                 len = 1;
1296                                 goto reset;
1297                         }
1298
1299                         // Try to allocate memory, otherwise send a RST back
1300                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1301
1302                         if(!c) {
1303                                 len = 1;
1304                                 goto reset;
1305                         }
1306
1307                         // Parse auxilliary information
1308                         if(init) {
1309                                 if(init[0] < 1) {
1310                                         len = 1;
1311                                         goto reset;
1312                                 }
1313
1314                                 c->flags = init[3] & 0x7;
1315                         } else {
1316                                 c->flags = UTCP_TCP;
1317                         }
1318
1319 synack:
1320                         // Return SYN+ACK, go to SYN_RECEIVED state
1321                         c->snd.wnd = hdr.wnd;
1322                         c->rcv.irs = hdr.seq;
1323                         c->rcv.nxt = c->rcv.irs + 1;
1324                         set_state(c, SYN_RECEIVED);
1325
1326                         struct {
1327                                 struct hdr hdr;
1328                                 uint8_t data[4];
1329                         } pkt;
1330
1331                         pkt.hdr.src = c->src;
1332                         pkt.hdr.dst = c->dst;
1333                         pkt.hdr.ack = c->rcv.irs + 1;
1334                         pkt.hdr.seq = c->snd.iss;
1335                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1336                         pkt.hdr.ctl = SYN | ACK;
1337
1338                         if(init) {
1339                                 pkt.hdr.aux = 0x0101;
1340                                 pkt.data[0] = 1;
1341                                 pkt.data[1] = 0;
1342                                 pkt.data[2] = 0;
1343                                 pkt.data[3] = c->flags & 0x7;
1344                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1345                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1346                         } else {
1347                                 pkt.hdr.aux = 0;
1348                                 print_packet(c, "send", &pkt, sizeof(hdr));
1349                                 utcp->send(utcp, &pkt, sizeof(hdr));
1350                         }
1351                 } else {
1352                         // No, we don't want your packets, send a RST back
1353                         len = 1;
1354                         goto reset;
1355                 }
1356
1357                 return 0;
1358         }
1359
1360         debug(c, "state %s\n", strstate[c->state]);
1361
1362         // In case this is for a CLOSED connection, ignore the packet.
1363         // TODO: make it so incoming packets can never match a CLOSED connection.
1364
1365         if(c->state == CLOSED) {
1366                 debug(c, "got packet for closed connection\n");
1367                 return 0;
1368         }
1369
1370         // It is for an existing connection.
1371
1372         // 1. Drop invalid packets.
1373
1374         // 1a. Drop packets that should not happen in our current state.
1375
1376         switch(c->state) {
1377         case SYN_SENT:
1378         case SYN_RECEIVED:
1379         case ESTABLISHED:
1380         case FIN_WAIT_1:
1381         case FIN_WAIT_2:
1382         case CLOSE_WAIT:
1383         case CLOSING:
1384         case LAST_ACK:
1385         case TIME_WAIT:
1386                 break;
1387
1388         default:
1389 #ifdef UTCP_DEBUG
1390                 abort();
1391 #endif
1392                 break;
1393         }
1394
1395         // 1b. Discard data that is not in our receive window.
1396
1397         if(is_reliable(c)) {
1398                 bool acceptable;
1399
1400                 if(c->state == SYN_SENT) {
1401                         acceptable = true;
1402                 } else if(len == 0) {
1403                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1404                 } else {
1405                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1406
1407                         // cut already accepted front overlapping
1408                         if(rcv_offset < 0) {
1409                                 acceptable = len > (size_t) - rcv_offset;
1410
1411                                 if(acceptable) {
1412                                         ptr -= rcv_offset;
1413                                         len += rcv_offset;
1414                                         hdr.seq -= rcv_offset;
1415                                 }
1416                         } else {
1417                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1418                         }
1419                 }
1420
1421                 if(!acceptable) {
1422                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1423
1424                         // Ignore unacceptable RST packets.
1425                         if(hdr.ctl & RST) {
1426                                 return 0;
1427                         }
1428
1429                         // Otherwise, continue processing.
1430                         len = 0;
1431                 }
1432         } else {
1433 #if UTCP_DEBUG
1434                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1435
1436                 if(rcv_offset) {
1437                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1438                 }
1439
1440 #endif
1441         }
1442
1443         c->snd.wnd = hdr.wnd; // TODO: move below
1444
1445         // 1c. Drop packets with an invalid ACK.
1446         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1447         // (= snd.una + c->sndbuf.used).
1448
1449         if(!is_reliable(c)) {
1450                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1451                         hdr.ack = c->snd.una;
1452                 }
1453         }
1454
1455         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1456                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1457
1458                 // Ignore unacceptable RST packets.
1459                 if(hdr.ctl & RST) {
1460                         return 0;
1461                 }
1462
1463                 goto reset;
1464         }
1465
1466         // 2. Handle RST packets
1467
1468         if(hdr.ctl & RST) {
1469                 switch(c->state) {
1470                 case SYN_SENT:
1471                         if(!(hdr.ctl & ACK)) {
1472                                 return 0;
1473                         }
1474
1475                         // The peer has refused our connection.
1476                         set_state(c, CLOSED);
1477                         errno = ECONNREFUSED;
1478
1479                         if(c->recv) {
1480                                 c->recv(c, NULL, 0);
1481                         }
1482
1483                         if(c->poll && !c->reapable) {
1484                                 c->poll(c, 0);
1485                         }
1486
1487                         return 0;
1488
1489                 case SYN_RECEIVED:
1490                         if(hdr.ctl & ACK) {
1491                                 return 0;
1492                         }
1493
1494                         // We haven't told the application about this connection yet. Silently delete.
1495                         free_connection(c);
1496                         return 0;
1497
1498                 case ESTABLISHED:
1499                 case FIN_WAIT_1:
1500                 case FIN_WAIT_2:
1501                 case CLOSE_WAIT:
1502                         if(hdr.ctl & ACK) {
1503                                 return 0;
1504                         }
1505
1506                         // The peer has aborted our connection.
1507                         set_state(c, CLOSED);
1508                         errno = ECONNRESET;
1509
1510                         if(c->recv) {
1511                                 c->recv(c, NULL, 0);
1512                         }
1513
1514                         if(c->poll && !c->reapable) {
1515                                 c->poll(c, 0);
1516                         }
1517
1518                         return 0;
1519
1520                 case CLOSING:
1521                 case LAST_ACK:
1522                 case TIME_WAIT:
1523                         if(hdr.ctl & ACK) {
1524                                 return 0;
1525                         }
1526
1527                         // As far as the application is concerned, the connection has already been closed.
1528                         // If it has called utcp_close() already, we can immediately free this connection.
1529                         if(c->reapable) {
1530                                 free_connection(c);
1531                                 return 0;
1532                         }
1533
1534                         // Otherwise, immediately move to the CLOSED state.
1535                         set_state(c, CLOSED);
1536                         return 0;
1537
1538                 default:
1539 #ifdef UTCP_DEBUG
1540                         abort();
1541 #endif
1542                         break;
1543                 }
1544         }
1545
1546         uint32_t advanced;
1547
1548         if(!(hdr.ctl & ACK)) {
1549                 advanced = 0;
1550                 goto skip_ack;
1551         }
1552
1553         // 3. Advance snd.una
1554
1555         advanced = seqdiff(hdr.ack, c->snd.una);
1556
1557         if(advanced) {
1558                 // RTT measurement
1559                 if(c->rtt_start.tv_sec) {
1560                         if(c->rtt_seq == hdr.ack) {
1561                                 struct timespec now;
1562                                 clock_gettime(UTCP_CLOCK, &now);
1563                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1564                                 update_rtt(c, diff);
1565                                 c->rtt_start.tv_sec = 0;
1566                         } else if(c->rtt_seq < hdr.ack) {
1567                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1568                                 c->rtt_start.tv_sec = 0;
1569                         }
1570                 }
1571
1572                 int32_t data_acked = advanced;
1573
1574                 switch(c->state) {
1575                 case SYN_SENT:
1576                 case SYN_RECEIVED:
1577                         data_acked--;
1578                         break;
1579
1580                 // TODO: handle FIN as well.
1581                 default:
1582                         break;
1583                 }
1584
1585                 assert(data_acked >= 0);
1586
1587 #ifndef NDEBUG
1588                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1589                 assert(data_acked <= bufused);
1590 #endif
1591
1592                 if(data_acked) {
1593                         buffer_discard(&c->sndbuf, data_acked);
1594                         c->do_poll = true;
1595                 }
1596
1597                 // Also advance snd.nxt if possible
1598                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1599                         c->snd.nxt = hdr.ack;
1600                 }
1601
1602                 c->snd.una = hdr.ack;
1603
1604                 if(c->dupack) {
1605                         if(c->dupack >= 3) {
1606                                 debug(c, "fast recovery ended\n");
1607                                 c->snd.cwnd = c->snd.ssthresh;
1608                         }
1609
1610                         c->dupack = 0;
1611                 }
1612
1613                 // Increase the congestion window according to RFC 5681
1614                 if(c->snd.cwnd < c->snd.ssthresh) {
1615                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1616                 } else {
1617                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1618                 }
1619
1620                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1621                         c->snd.cwnd = c->sndbuf.maxsize;
1622                 }
1623
1624                 debug_cwnd(c);
1625
1626                 // Check if we have sent a FIN that is now ACKed.
1627                 switch(c->state) {
1628                 case FIN_WAIT_1:
1629                         if(c->snd.una == c->snd.last) {
1630                                 set_state(c, FIN_WAIT_2);
1631                         }
1632
1633                         break;
1634
1635                 case CLOSING:
1636                         if(c->snd.una == c->snd.last) {
1637                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1638                                 c->conn_timeout.tv_sec += utcp->timeout;
1639                                 set_state(c, TIME_WAIT);
1640                         }
1641
1642                         break;
1643
1644                 default:
1645                         break;
1646                 }
1647         } else {
1648                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1649                         c->dupack++;
1650                         debug(c, "duplicate ACK %d\n", c->dupack);
1651
1652                         if(c->dupack == 3) {
1653                                 // RFC 5681 fast recovery
1654                                 debug(c, "fast recovery started\n", c->dupack);
1655                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1656                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1657                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1658
1659                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1660                                         c->snd.cwnd = c->sndbuf.maxsize;
1661                                 }
1662
1663                                 debug_cwnd(c);
1664
1665                                 fast_retransmit(c);
1666                         } else if(c->dupack > 3) {
1667                                 c->snd.cwnd += utcp->mss;
1668
1669                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1670                                         c->snd.cwnd = c->sndbuf.maxsize;
1671                                 }
1672
1673                                 debug_cwnd(c);
1674                         }
1675
1676                         // We got an ACK which indicates the other side did get one of our packets.
1677                         // Reset the retransmission timer to avoid going to slow start,
1678                         // but don't touch the connection timeout.
1679                         start_retransmit_timer(c);
1680                 }
1681         }
1682
1683         // 4. Update timers
1684
1685         if(advanced) {
1686                 if(c->snd.una == c->snd.last) {
1687                         stop_retransmit_timer(c);
1688                         timespec_clear(&c->conn_timeout);
1689                 } else if(is_reliable(c)) {
1690                         start_retransmit_timer(c);
1691                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1692                         c->conn_timeout.tv_sec += utcp->timeout;
1693                 }
1694         }
1695
1696 skip_ack:
1697         // 5. Process SYN stuff
1698
1699         if(hdr.ctl & SYN) {
1700                 switch(c->state) {
1701                 case SYN_SENT:
1702
1703                         // This is a SYNACK. It should always have ACKed the SYN.
1704                         if(!advanced) {
1705                                 goto reset;
1706                         }
1707
1708                         c->rcv.irs = hdr.seq;
1709                         c->rcv.nxt = hdr.seq;
1710
1711                         if(c->shut_wr) {
1712                                 c->snd.last++;
1713                                 set_state(c, FIN_WAIT_1);
1714                         } else {
1715                                 set_state(c, ESTABLISHED);
1716                         }
1717
1718                         // TODO: notify application of this somehow.
1719                         break;
1720
1721                 case SYN_RECEIVED:
1722                         // This is a retransmit of a SYN, send back the SYNACK.
1723                         goto synack;
1724
1725                 case ESTABLISHED:
1726                 case FIN_WAIT_1:
1727                 case FIN_WAIT_2:
1728                 case CLOSE_WAIT:
1729                 case CLOSING:
1730                 case LAST_ACK:
1731                 case TIME_WAIT:
1732                         // Ehm, no. We should never receive a second SYN.
1733                         return 0;
1734
1735                 default:
1736 #ifdef UTCP_DEBUG
1737                         abort();
1738 #endif
1739                         return 0;
1740                 }
1741
1742                 // SYN counts as one sequence number
1743                 c->rcv.nxt++;
1744         }
1745
1746         // 6. Process new data
1747
1748         if(c->state == SYN_RECEIVED) {
1749                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1750                 if(!advanced) {
1751                         goto reset;
1752                 }
1753
1754                 // Are we still LISTENing?
1755                 if(utcp->accept) {
1756                         utcp->accept(c, c->src);
1757                 }
1758
1759                 if(c->state != ESTABLISHED) {
1760                         set_state(c, CLOSED);
1761                         c->reapable = true;
1762                         goto reset;
1763                 }
1764         }
1765
1766         if(len) {
1767                 switch(c->state) {
1768                 case SYN_SENT:
1769                 case SYN_RECEIVED:
1770                         // This should never happen.
1771 #ifdef UTCP_DEBUG
1772                         abort();
1773 #endif
1774                         return 0;
1775
1776                 case ESTABLISHED:
1777                 case FIN_WAIT_1:
1778                 case FIN_WAIT_2:
1779                         break;
1780
1781                 case CLOSE_WAIT:
1782                 case CLOSING:
1783                 case LAST_ACK:
1784                 case TIME_WAIT:
1785                         // Ehm no, We should never receive more data after a FIN.
1786                         goto reset;
1787
1788                 default:
1789 #ifdef UTCP_DEBUG
1790                         abort();
1791 #endif
1792                         return 0;
1793                 }
1794
1795                 handle_incoming_data(c, &hdr, ptr, len);
1796         }
1797
1798         // 7. Process FIN stuff
1799
1800         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1801                 switch(c->state) {
1802                 case SYN_SENT:
1803                 case SYN_RECEIVED:
1804                         // This should never happen.
1805 #ifdef UTCP_DEBUG
1806                         abort();
1807 #endif
1808                         break;
1809
1810                 case ESTABLISHED:
1811                         set_state(c, CLOSE_WAIT);
1812                         break;
1813
1814                 case FIN_WAIT_1:
1815                         set_state(c, CLOSING);
1816                         break;
1817
1818                 case FIN_WAIT_2:
1819                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1820                         c->conn_timeout.tv_sec += utcp->timeout;
1821                         set_state(c, TIME_WAIT);
1822                         break;
1823
1824                 case CLOSE_WAIT:
1825                 case CLOSING:
1826                 case LAST_ACK:
1827                 case TIME_WAIT:
1828                         // Ehm, no. We should never receive a second FIN.
1829                         goto reset;
1830
1831                 default:
1832 #ifdef UTCP_DEBUG
1833                         abort();
1834 #endif
1835                         break;
1836                 }
1837
1838                 // FIN counts as one sequence number
1839                 c->rcv.nxt++;
1840                 len++;
1841
1842                 // Inform the application that the peer closed its end of the connection.
1843                 if(c->recv) {
1844                         errno = 0;
1845                         c->recv(c, NULL, 0);
1846                 }
1847         }
1848
1849         // Now we send something back if:
1850         // - we received data, so we have to send back an ACK
1851         //   -> sendatleastone = true
1852         // - or we got an ack, so we should maybe send a bit more data
1853         //   -> sendatleastone = false
1854
1855         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1856                 ack(c, has_data);
1857         }
1858
1859         return 0;
1860
1861 reset:
1862         swap_ports(&hdr);
1863         hdr.wnd = 0;
1864         hdr.aux = 0;
1865
1866         if(hdr.ctl & ACK) {
1867                 hdr.seq = hdr.ack;
1868                 hdr.ctl = RST;
1869         } else {
1870                 hdr.ack = hdr.seq + len;
1871                 hdr.seq = 0;
1872                 hdr.ctl = RST | ACK;
1873         }
1874
1875         print_packet(c, "send", &hdr, sizeof(hdr));
1876         utcp->send(utcp, &hdr, sizeof(hdr));
1877         return 0;
1878
1879 }
1880
1881 int utcp_shutdown(struct utcp_connection *c, int dir) {
1882         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1883
1884         if(!c) {
1885                 errno = EFAULT;
1886                 return -1;
1887         }
1888
1889         if(c->reapable) {
1890                 debug(c, "shutdown() called on closed connection\n");
1891                 errno = EBADF;
1892                 return -1;
1893         }
1894
1895         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1896                 errno = EINVAL;
1897                 return -1;
1898         }
1899
1900         // TCP does not have a provision for stopping incoming packets.
1901         // The best we can do is to just ignore them.
1902         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1903                 c->recv = NULL;
1904         }
1905
1906         // The rest of the code deals with shutting down writes.
1907         if(dir == UTCP_SHUT_RD) {
1908                 return 0;
1909         }
1910
1911         // Only process shutting down writes once.
1912         if(c->shut_wr) {
1913                 return 0;
1914         }
1915
1916         c->shut_wr = true;
1917
1918         switch(c->state) {
1919         case CLOSED:
1920         case LISTEN:
1921                 errno = ENOTCONN;
1922                 return -1;
1923
1924         case SYN_SENT:
1925                 return 0;
1926
1927         case SYN_RECEIVED:
1928         case ESTABLISHED:
1929                 set_state(c, FIN_WAIT_1);
1930                 break;
1931
1932         case FIN_WAIT_1:
1933         case FIN_WAIT_2:
1934                 return 0;
1935
1936         case CLOSE_WAIT:
1937                 set_state(c, CLOSING);
1938                 break;
1939
1940         case CLOSING:
1941         case LAST_ACK:
1942         case TIME_WAIT:
1943                 return 0;
1944         }
1945
1946         c->snd.last++;
1947
1948         ack(c, false);
1949
1950         if(!timespec_isset(&c->rtrx_timeout)) {
1951                 start_retransmit_timer(c);
1952         }
1953
1954         return 0;
1955 }
1956
1957 static bool reset_connection(struct utcp_connection *c) {
1958         if(!c) {
1959                 errno = EFAULT;
1960                 return false;
1961         }
1962
1963         if(c->reapable) {
1964                 debug(c, "abort() called on closed connection\n");
1965                 errno = EBADF;
1966                 return false;
1967         }
1968
1969         c->recv = NULL;
1970         c->poll = NULL;
1971
1972         switch(c->state) {
1973         case CLOSED:
1974                 return true;
1975
1976         case LISTEN:
1977         case SYN_SENT:
1978         case CLOSING:
1979         case LAST_ACK:
1980         case TIME_WAIT:
1981                 set_state(c, CLOSED);
1982                 return true;
1983
1984         case SYN_RECEIVED:
1985         case ESTABLISHED:
1986         case FIN_WAIT_1:
1987         case FIN_WAIT_2:
1988         case CLOSE_WAIT:
1989                 set_state(c, CLOSED);
1990                 break;
1991         }
1992
1993         // Send RST
1994
1995         struct hdr hdr;
1996
1997         hdr.src = c->src;
1998         hdr.dst = c->dst;
1999         hdr.seq = c->snd.nxt;
2000         hdr.ack = 0;
2001         hdr.wnd = 0;
2002         hdr.ctl = RST;
2003
2004         print_packet(c, "send", &hdr, sizeof(hdr));
2005         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2006         return true;
2007 }
2008
2009 // Closes all the opened connections
2010 void utcp_abort_all_connections(struct utcp *utcp) {
2011         if(!utcp) {
2012                 errno = EINVAL;
2013                 return;
2014         }
2015
2016         for(int i = 0; i < utcp->nconnections; i++) {
2017                 struct utcp_connection *c = utcp->connections[i];
2018
2019                 if(c->reapable || c->state == CLOSED) {
2020                         continue;
2021                 }
2022
2023                 utcp_recv_t old_recv = c->recv;
2024                 utcp_poll_t old_poll = c->poll;
2025
2026                 reset_connection(c);
2027
2028                 if(old_recv) {
2029                         errno = 0;
2030                         old_recv(c, NULL, 0);
2031                 }
2032
2033                 if(old_poll && !c->reapable) {
2034                         errno = 0;
2035                         old_poll(c, 0);
2036                 }
2037         }
2038
2039         return;
2040 }
2041
2042 int utcp_close(struct utcp_connection *c) {
2043         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2044                 return -1;
2045         }
2046
2047         c->recv = NULL;
2048         c->poll = NULL;
2049         c->reapable = true;
2050         return 0;
2051 }
2052
2053 int utcp_abort(struct utcp_connection *c) {
2054         if(!reset_connection(c)) {
2055                 return -1;
2056         }
2057
2058         c->reapable = true;
2059         return 0;
2060 }
2061
2062 /* Handle timeouts.
2063  * One call to this function will loop through all connections,
2064  * checking if something needs to be resent or not.
2065  * The return value is the time to the next timeout in milliseconds,
2066  * or maybe a negative value if the timeout is infinite.
2067  */
2068 struct timespec utcp_timeout(struct utcp *utcp) {
2069         struct timespec now;
2070         clock_gettime(UTCP_CLOCK, &now);
2071         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2072
2073         for(int i = 0; i < utcp->nconnections; i++) {
2074                 struct utcp_connection *c = utcp->connections[i];
2075
2076                 if(!c) {
2077                         continue;
2078                 }
2079
2080                 // delete connections that have been utcp_close()d.
2081                 if(c->state == CLOSED) {
2082                         if(c->reapable) {
2083                                 debug(c, "reaping\n");
2084                                 free_connection(c);
2085                                 i--;
2086                         }
2087
2088                         continue;
2089                 }
2090
2091                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2092                         errno = ETIMEDOUT;
2093                         c->state = CLOSED;
2094
2095                         if(c->recv) {
2096                                 c->recv(c, NULL, 0);
2097                         }
2098
2099                         if(c->poll && !c->reapable) {
2100                                 c->poll(c, 0);
2101                         }
2102
2103                         continue;
2104                 }
2105
2106                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2107                         debug(c, "retransmitting after timeout\n");
2108                         retransmit(c);
2109                 }
2110
2111                 if(c->poll) {
2112                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2113                                 c->do_poll = false;
2114                                 uint32_t len = buffer_free(&c->sndbuf);
2115
2116                                 if(len) {
2117                                         c->poll(c, len);
2118                                 }
2119                         } else if(c->state == CLOSED) {
2120                                 c->poll(c, 0);
2121                         }
2122                 }
2123
2124                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2125                         next = c->conn_timeout;
2126                 }
2127
2128                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2129                         next = c->rtrx_timeout;
2130                 }
2131         }
2132
2133         struct timespec diff;
2134
2135         timespec_sub(&next, &now, &diff);
2136
2137         return diff;
2138 }
2139
2140 bool utcp_is_active(struct utcp *utcp) {
2141         if(!utcp) {
2142                 return false;
2143         }
2144
2145         for(int i = 0; i < utcp->nconnections; i++)
2146                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2147                         return true;
2148                 }
2149
2150         return false;
2151 }
2152
2153 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2154         if(!send) {
2155                 errno = EFAULT;
2156                 return NULL;
2157         }
2158
2159         struct utcp *utcp = calloc(1, sizeof(*utcp));
2160
2161         if(!utcp) {
2162                 return NULL;
2163         }
2164
2165         if(!CLOCK_GRANULARITY) {
2166                 struct timespec res;
2167                 clock_getres(UTCP_CLOCK, &res);
2168                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2169         }
2170
2171         utcp->accept = accept;
2172         utcp->pre_accept = pre_accept;
2173         utcp->send = send;
2174         utcp->priv = priv;
2175         utcp_set_mtu(utcp, DEFAULT_MTU);
2176         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2177         utcp->rto = START_RTO; // usec
2178
2179         return utcp;
2180 }
2181
2182 void utcp_exit(struct utcp *utcp) {
2183         if(!utcp) {
2184                 return;
2185         }
2186
2187         for(int i = 0; i < utcp->nconnections; i++) {
2188                 struct utcp_connection *c = utcp->connections[i];
2189
2190                 if(!c->reapable) {
2191                         if(c->recv) {
2192                                 c->recv(c, NULL, 0);
2193                         }
2194
2195                         if(c->poll && !c->reapable) {
2196                                 c->poll(c, 0);
2197                         }
2198                 }
2199
2200                 buffer_exit(&c->rcvbuf);
2201                 buffer_exit(&c->sndbuf);
2202                 free(c);
2203         }
2204
2205         free(utcp->connections);
2206         free(utcp->pkt);
2207         free(utcp);
2208 }
2209
2210 uint16_t utcp_get_mtu(struct utcp *utcp) {
2211         return utcp ? utcp->mtu : 0;
2212 }
2213
2214 uint16_t utcp_get_mss(struct utcp *utcp) {
2215         return utcp ? utcp->mss : 0;
2216 }
2217
2218 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2219         if(!utcp) {
2220                 return;
2221         }
2222
2223         if(mtu <= sizeof(struct hdr)) {
2224                 return;
2225         }
2226
2227         if(mtu > utcp->mtu) {
2228                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2229
2230                 if(!new) {
2231                         return;
2232                 }
2233
2234                 utcp->pkt = new;
2235         }
2236
2237         utcp->mtu = mtu;
2238         utcp->mss = mtu - sizeof(struct hdr);
2239 }
2240
2241 void utcp_reset_timers(struct utcp *utcp) {
2242         if(!utcp) {
2243                 return;
2244         }
2245
2246         struct timespec now, then;
2247
2248         clock_gettime(UTCP_CLOCK, &now);
2249
2250         then = now;
2251
2252         then.tv_sec += utcp->timeout;
2253
2254         for(int i = 0; i < utcp->nconnections; i++) {
2255                 struct utcp_connection *c = utcp->connections[i];
2256
2257                 if(c->reapable) {
2258                         continue;
2259                 }
2260
2261                 if(timespec_isset(&c->rtrx_timeout)) {
2262                         c->rtrx_timeout = now;
2263                 }
2264
2265                 if(timespec_isset(&c->conn_timeout)) {
2266                         c->conn_timeout = then;
2267                 }
2268
2269                 c->rtt_start.tv_sec = 0;
2270         }
2271
2272         if(utcp->rto > START_RTO) {
2273                 utcp->rto = START_RTO;
2274         }
2275 }
2276
2277 int utcp_get_user_timeout(struct utcp *u) {
2278         return u ? u->timeout : 0;
2279 }
2280
2281 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2282         if(u) {
2283                 u->timeout = timeout;
2284         }
2285 }
2286
2287 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2288         return c ? c->sndbuf.maxsize : 0;
2289 }
2290
2291 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2292         if(!c) {
2293                 return 0;
2294         }
2295
2296         switch(c->state) {
2297         case SYN_SENT:
2298         case SYN_RECEIVED:
2299         case ESTABLISHED:
2300         case CLOSE_WAIT:
2301                 return buffer_free(&c->sndbuf);
2302
2303         default:
2304                 return 0;
2305         }
2306 }
2307
2308 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2309         if(!c) {
2310                 return;
2311         }
2312
2313         c->sndbuf.maxsize = size;
2314
2315         if(c->sndbuf.maxsize != size) {
2316                 c->sndbuf.maxsize = -1;
2317         }
2318
2319         c->do_poll = buffer_free(&c->sndbuf);
2320 }
2321
2322 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2323         return c ? c->rcvbuf.maxsize : 0;
2324 }
2325
2326 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2327         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2328                 return buffer_free(&c->rcvbuf);
2329         } else {
2330                 return 0;
2331         }
2332 }
2333
2334 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2335         if(!c) {
2336                 return;
2337         }
2338
2339         c->rcvbuf.maxsize = size;
2340
2341         if(c->rcvbuf.maxsize != size) {
2342                 c->rcvbuf.maxsize = -1;
2343         }
2344 }
2345
2346 size_t utcp_get_sendq(struct utcp_connection *c) {
2347         return c->sndbuf.used;
2348 }
2349
2350 size_t utcp_get_recvq(struct utcp_connection *c) {
2351         return c->rcvbuf.used;
2352 }
2353
2354 bool utcp_get_nodelay(struct utcp_connection *c) {
2355         return c ? c->nodelay : false;
2356 }
2357
2358 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2359         if(c) {
2360                 c->nodelay = nodelay;
2361         }
2362 }
2363
2364 bool utcp_get_keepalive(struct utcp_connection *c) {
2365         return c ? c->keepalive : false;
2366 }
2367
2368 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2369         if(c) {
2370                 c->keepalive = keepalive;
2371         }
2372 }
2373
2374 size_t utcp_get_outq(struct utcp_connection *c) {
2375         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2376 }
2377
2378 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2379         if(c) {
2380                 c->recv = recv;
2381         }
2382 }
2383
2384 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2385         if(c) {
2386                 c->poll = poll;
2387                 c->do_poll = buffer_free(&c->sndbuf);
2388         }
2389 }
2390
2391 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2392         if(utcp) {
2393                 utcp->accept = accept;
2394                 utcp->pre_accept = pre_accept;
2395         }
2396 }
2397
2398 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2399         if(!c || c->reapable) {
2400                 return;
2401         }
2402
2403         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2404                 return;
2405         }
2406
2407         if(expect) {
2408                 // If we expect data, start the connection timer.
2409                 if(!timespec_isset(&c->conn_timeout)) {
2410                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2411                         c->conn_timeout.tv_sec += c->utcp->timeout;
2412                 }
2413         } else {
2414                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2415                 if(c->snd.una == c->snd.last) {
2416                         timespec_clear(&c->conn_timeout);
2417                 }
2418         }
2419 }
2420
2421 void utcp_offline(struct utcp *utcp, bool offline) {
2422         struct timespec now;
2423         clock_gettime(UTCP_CLOCK, &now);
2424
2425         for(int i = 0; i < utcp->nconnections; i++) {
2426                 struct utcp_connection *c = utcp->connections[i];
2427
2428                 if(c->reapable) {
2429                         continue;
2430                 }
2431
2432                 utcp_expect_data(c, offline);
2433
2434                 if(!offline) {
2435                         if(timespec_isset(&c->rtrx_timeout)) {
2436                                 c->rtrx_timeout = now;
2437                         }
2438
2439                         utcp->connections[i]->rtt_start.tv_sec = 0;
2440                 }
2441         }
2442
2443         if(!offline && utcp->rto > START_RTO) {
2444                 utcp->rto = START_RTO;
2445         }
2446 }
2447
2448 void utcp_set_clock_granularity(long granularity) {
2449         CLOCK_GRANULARITY = granularity;
2450 }