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

   1 /*
   2     utcp.c -- Userspace TCP
   3     Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
   4
   5     This program is free software; you can redistribute it and/or modify
   6     it under the terms of the GNU General Public License as published by
   7     the Free Software Foundation; either version 2 of the License, or
   8     (at your option) any later version.
   9
  10     This program is distributed in the hope that it will be useful,
  11     but WITHOUT ANY WARRANTY; without even the implied warranty of
  12     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13     GNU General Public License for more details.
  14
  15     You should have received a copy of the GNU General Public License along
  16     with this program; if not, write to the Free Software Foundation, Inc.,
  17     51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18 */
  19
  20 #define _GNU_SOURCE
  21
  22 #include <assert.h>
  23 #include <errno.h>
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <stdint.h>
  27 #include <stdbool.h>
  28 #include <string.h>
  29 #include <unistd.h>
  30 #include <time.h>
  31
  32 #include "utcp_priv.h"
  33
  34 #ifndef EBADMSG
  35 #define EBADMSG         104
  36 #endif
  37
  38 #ifndef SHUT_RDWR
  39 #define SHUT_RDWR 2
  40 #endif
  41
  42 #ifdef poll
  43 #undef poll
  44 #endif
  45
  46 #ifndef UTCP_CLOCK
  47 #if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
  48 #define UTCP_CLOCK CLOCK_MONOTONIC_RAW
  49 #else
  50 #define UTCP_CLOCK CLOCK_MONOTONIC
  51 #endif
  52 #endif
  53
  54 static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
  55         r->tv_sec = a->tv_sec - b->tv_sec;
  56         r->tv_nsec = a->tv_nsec - b->tv_nsec;
  57
  58         if(r->tv_nsec < 0) {
  59                 r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
  60         }
  61 }
  62
  63 static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
  64         return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
  65 }
  66
  67 static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
  68         if(a->tv_sec == b->tv_sec) {
  69                 return a->tv_nsec < b->tv_nsec;
  70         } else {
  71                 return a->tv_sec < b->tv_sec;
  72         }
  73 }
  74
  75 static void timespec_clear(struct timespec *a) {
  76         a->tv_sec = 0;
  77         a->tv_nsec = 0;
  78 }
  79
  80 static bool timespec_isset(const struct timespec *a) {
  81         return a->tv_sec;
  82 }
  83
  84 static long CLOCK_GRANULARITY; // usec
  85
  86 static inline size_t min(size_t a, size_t b) {
  87         return a < b ? a : b;
  88 }
  89
  90 static inline size_t max(size_t a, size_t b) {
  91         return a > b ? a : b;
  92 }
  93
  94 #ifdef UTCP_DEBUG
  95 #include <stdarg.h>
  96
  97 #ifndef UTCP_DEBUG_DATALEN
  98 #define UTCP_DEBUG_DATALEN 20
  99 #endif
 100
 101 static void debug(struct utcp_connection *c, const char *format, ...) {
 102         struct timespec tv;
 103         char buf[1024];
 104         int len;
 105
 106         clock_gettime(CLOCK_REALTIME, &tv);
 107         len = snprintf(buf, sizeof(buf), "%ld.%06lu %u:%u ", (long)tv.tv_sec, tv.tv_nsec / 1000, c ? c->src : 0, c ? c->dst : 0);
 108         va_list ap;
 109         va_start(ap, format);
 110         len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
 111         va_end(ap);
 112
 113         if(len > 0 && (size_t)len < sizeof(buf)) {
 114                 fwrite(buf, len, 1, stderr);
 115         }
 116 }
 117
 118 static void print_packet(struct utcp_connection *c, const char *dir, const void *pkt, size_t len) {
 119         struct hdr hdr;
 120
 121         if(len < sizeof(hdr)) {
 122                 debug(c, "%s: short packet (%lu bytes)\n", dir, (unsigned long)len);
 123                 return;
 124         }
 125
 126         memcpy(&hdr, pkt, sizeof(hdr));
 127
 128         uint32_t datalen;
 129
 130         if(len > sizeof(hdr)) {
 131                 datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
 132         } else {
 133                 datalen = 0;
 134         }
 135
 136
 137         const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 138         char str[datalen * 2 + 1];
 139         char *p = str;
 140
 141         for(uint32_t i = 0; i < datalen; i++) {
 142                 *p++ = "0123456789ABCDEF"[data[i] >> 4];
 143                 *p++ = "0123456789ABCDEF"[data[i] & 15];
 144         }
 145
 146         *p = 0;
 147
 148         debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s%s data %s\n",
 149               dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
 150               hdr.ctl & SYN ? "SYN" : "",
 151               hdr.ctl & RST ? "RST" : "",
 152               hdr.ctl & FIN ? "FIN" : "",
 153               hdr.ctl & ACK ? "ACK" : "",
 154               hdr.ctl & MF ? "MF" : "",
 155               str
 156              );
 157 }
 158
 159 static void debug_cwnd(struct utcp_connection *c) {
 160         debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
 161 }
 162 #else
 163 #define debug(...) do {} while(0)
 164 #define print_packet(...) do {} while(0)
 165 #define debug_cwnd(...) do {} while(0)
 166 #endif
 167
 168 static void set_state(struct utcp_connection *c, enum state state) {
 169         c->state = state;
 170
 171         if(state == ESTABLISHED) {
 172                 timespec_clear(&c->conn_timeout);
 173         }
 174
 175         debug(c, "state %s\n", strstate[state]);
 176 }
 177
 178 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 179         if(seq != c->snd.last) {
 180                 return false;
 181         }
 182
 183         switch(c->state) {
 184         case FIN_WAIT_1:
 185         case CLOSING:
 186         case LAST_ACK:
 187                 return true;
 188
 189         default:
 190                 return false;
 191         }
 192 }
 193
 194 static bool is_reliable(struct utcp_connection *c) {
 195         return c->flags & UTCP_RELIABLE;
 196 }
 197
 198 static int32_t seqdiff(uint32_t a, uint32_t b) {
 199         return a - b;
 200 }
 201
 202 // Buffer functions
 203 static bool buffer_wraps(struct buffer *buf) {
 204         return buf->size - buf->offset < buf->used;
 205 }
 206
 207 static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
 208         char *newdata = realloc(buf->data, newsize);
 209
 210         if(!newdata) {
 211                 return false;
 212         }
 213
 214         buf->data = newdata;
 215
 216         if(buffer_wraps(buf)) {
 217                 // Shift the right part of the buffer until it hits the end of the new buffer.
 218                 // Old situation:
 219                 // [345......012]
 220                 // New situation:
 221                 // [345.........|........012]
 222                 uint32_t tailsize = buf->size - buf->offset;
 223                 uint32_t newoffset = newsize - tailsize;
 224                 memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
 225                 buf->offset = newoffset;
 226         }
 227
 228         buf->size = newsize;
 229         return true;
 230 }
 231
 232 // Store data into the buffer
 233 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 234         debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 235
 236         // Ensure we don't store more than maxsize bytes in total
 237         size_t required = offset + len;
 238
 239         if(required > buf->maxsize) {
 240                 if(offset >= buf->maxsize) {
 241                         return 0;
 242                 }
 243
 244                 len = buf->maxsize - offset;
 245                 required = buf->maxsize;
 246         }
 247
 248         // Check if we need to resize the buffer
 249         if(required > buf->size) {
 250                 size_t newsize = buf->size;
 251
 252                 if(!newsize) {
 253                         newsize = 4096;
 254                 }
 255
 256                 do {
 257                         newsize *= 2;
 258                 } while(newsize < required);
 259
 260                 if(newsize > buf->maxsize) {
 261                         newsize = buf->maxsize;
 262                 }
 263
 264                 if(!buffer_resize(buf, newsize)) {
 265                         return -1;
 266                 }
 267         }
 268
 269         uint32_t realoffset = buf->offset + offset;
 270
 271         if(buf->size - buf->offset < offset) {
 272                 // The offset wrapped
 273                 realoffset -= buf->size;
 274         }
 275
 276         if(buf->size - realoffset < len) {
 277                 // The new chunk of data must be wrapped
 278                 memcpy(buf->data + realoffset, data, buf->size - realoffset);
 279                 memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
 280         } else {
 281                 memcpy(buf->data + realoffset, data, len);
 282         }
 283
 284         if(required > buf->used) {
 285                 buf->used = required;
 286         }
 287
 288         return len;
 289 }
 290
 291 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 292         return buffer_put_at(buf, buf->used, data, len);
 293 }
 294
 295 // Copy data from the buffer without removing it.
 296 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 297         // Ensure we don't copy more than is actually stored in the buffer
 298         if(offset >= buf->used) {
 299                 return 0;
 300         }
 301
 302         if(buf->used - offset < len) {
 303                 len = buf->used - offset;
 304         }
 305
 306         uint32_t realoffset = buf->offset + offset;
 307
 308         if(buf->size - buf->offset < offset) {
 309                 // The offset wrapped
 310                 realoffset -= buf->size;
 311         }
 312
 313         if(buf->size - realoffset < len) {
 314                 // The data is wrapped
 315                 memcpy(data, buf->data + realoffset, buf->size - realoffset);
 316                 memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
 317         } else {
 318                 memcpy(data, buf->data + realoffset, len);
 319         }
 320
 321         return len;
 322 }
 323
 324 // Copy data from the buffer without removing it.
 325 static ssize_t buffer_call(struct buffer *buf, utcp_recv_t cb, void *arg, size_t offset, size_t len) {
 326         // Ensure we don't copy more than is actually stored in the buffer
 327         if(offset >= buf->used) {
 328                 return 0;
 329         }
 330
 331         if(buf->used - offset < len) {
 332                 len = buf->used - offset;
 333         }
 334
 335         uint32_t realoffset = buf->offset + offset;
 336
 337         if(buf->size - buf->offset < offset) {
 338                 // The offset wrapped
 339                 realoffset -= buf->size;
 340         }
 341
 342         if(buf->size - realoffset < len) {
 343                 // The data is wrapped
 344                 ssize_t rx1 = cb(arg, buf->data + realoffset, buf->size - realoffset);
 345
 346                 if(rx1 < buf->size - realoffset) {
 347                         return rx1;
 348                 }
 349
 350                 ssize_t rx2 = cb(arg, buf->data, len - (buf->size - realoffset));
 351
 352                 if(rx2 < 0) {
 353                         return rx2;
 354                 } else {
 355                         return rx1 + rx2;
 356                 }
 357         } else {
 358                 return cb(arg, buf->data + realoffset, len);
 359         }
 360 }
 361
 362 // Discard data from the buffer.
 363 static ssize_t buffer_discard(struct buffer *buf, size_t len) {
 364         if(buf->used < len) {
 365                 len = buf->used;
 366         }
 367
 368         if(buf->size - buf->offset < len) {
 369                 buf->offset -= buf->size;
 370         }
 371
 372         if(buf->used == len) {
 373                 buf->offset = 0;
 374         } else {
 375                 buf->offset += len;
 376         }
 377
 378         buf->used -= len;
 379
 380         return len;
 381 }
 382
 383 static void buffer_clear(struct buffer *buf) {
 384         buf->used = 0;
 385         buf->offset = 0;
 386 }
 387
 388 static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
 389         if(maxsize < minsize) {
 390                 maxsize = minsize;
 391         }
 392
 393         buf->maxsize = maxsize;
 394
 395         return buf->size >= minsize || buffer_resize(buf, minsize);
 396 }
 397
 398 static void buffer_exit(struct buffer *buf) {
 399         free(buf->data);
 400         memset(buf, 0, sizeof(*buf));
 401 }
 402
 403 static uint32_t buffer_free(const struct buffer *buf) {
 404         return buf->maxsize > buf->used ? buf->maxsize - buf->used : 0;
 405 }
 406
 407 // Connections are stored in a sorted list.
 408 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 409
 410 static int compare(const void *va, const void *vb) {
 411         assert(va && vb);
 412
 413         const struct utcp_connection *a = *(struct utcp_connection **)va;
 414         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 415
 416         assert(a && b);
 417         assert(a->src && b->src);
 418
 419         int c = (int)a->src - (int)b->src;
 420
 421         if(c) {
 422                 return c;
 423         }
 424
 425         c = (int)a->dst - (int)b->dst;
 426         return c;
 427 }
 428
 429 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 430         if(!utcp->nconnections) {
 431                 return NULL;
 432         }
 433
 434         struct utcp_connection key = {
 435                 .src = src,
 436                 .dst = dst,
 437         }, *keyp = &key;
 438         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 439         return match ? *match : NULL;
 440 }
 441
 442 static void free_connection(struct utcp_connection *c) {
 443         struct utcp *utcp = c->utcp;
 444         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 445
 446         assert(cp);
 447
 448         int i = cp - utcp->connections;
 449         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 450         utcp->nconnections--;
 451
 452         buffer_exit(&c->rcvbuf);
 453         buffer_exit(&c->sndbuf);
 454         free(c);
 455 }
 456
 457 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 458         // Check whether this combination of src and dst is free
 459
 460         if(src) {
 461                 if(find_connection(utcp, src, dst)) {
 462                         errno = EADDRINUSE;
 463                         return NULL;
 464                 }
 465         } else { // If src == 0, generate a random port number with the high bit set
 466                 if(utcp->nconnections >= 32767) {
 467                         errno = ENOMEM;
 468                         return NULL;
 469                 }
 470
 471                 src = rand() | 0x8000;
 472
 473                 while(find_connection(utcp, src, dst)) {
 474                         src++;
 475                 }
 476         }
 477
 478         // Allocate memory for the new connection
 479
 480         if(utcp->nconnections >= utcp->nallocated) {
 481                 if(!utcp->nallocated) {
 482                         utcp->nallocated = 4;
 483                 } else {
 484                         utcp->nallocated *= 2;
 485                 }
 486
 487                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 488
 489                 if(!new_array) {
 490                         return NULL;
 491                 }
 492
 493                 utcp->connections = new_array;
 494         }
 495
 496         struct utcp_connection *c = calloc(1, sizeof(*c));
 497
 498         if(!c) {
 499                 return NULL;
 500         }
 501
 502         if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 503                 free(c);
 504                 return NULL;
 505         }
 506
 507         if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 508                 buffer_exit(&c->sndbuf);
 509                 free(c);
 510                 return NULL;
 511         }
 512
 513         // Fill in the details
 514
 515         c->src = src;
 516         c->dst = dst;
 517 #ifdef UTCP_DEBUG
 518         c->snd.iss = 0;
 519 #else
 520         c->snd.iss = rand();
 521 #endif
 522         c->snd.una = c->snd.iss;
 523         c->snd.nxt = c->snd.iss + 1;
 524         c->snd.last = c->snd.nxt;
 525         c->snd.cwnd = (utcp->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
 526         c->snd.ssthresh = ~0;
 527         debug_cwnd(c);
 528         c->srtt = 0;
 529         c->rttvar = 0;
 530         c->rto = START_RTO;
 531         c->utcp = utcp;
 532
 533         // Add it to the sorted list of connections
 534
 535         utcp->connections[utcp->nconnections++] = c;
 536         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 537
 538         return c;
 539 }
 540
 541 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 542         if(a > b) {
 543                 return a - b;
 544         } else {
 545                 return b - a;
 546         }
 547 }
 548
 549 // Update RTT variables. See RFC 6298.
 550 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 551         if(!rtt) {
 552                 debug(c, "invalid rtt\n");
 553                 return;
 554         }
 555
 556         if(!c->srtt) {
 557                 c->srtt = rtt;
 558                 c->rttvar = rtt / 2;
 559         } else {
 560                 c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
 561                 c->srtt = (c->srtt * 7 + rtt) / 8;
 562         }
 563
 564         c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
 565
 566         if(c->rto > MAX_RTO) {
 567                 c->rto = MAX_RTO;
 568         }
 569
 570         debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
 571 }
 572
 573 static void start_retransmit_timer(struct utcp_connection *c) {
 574         clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
 575
 576         uint32_t rto = c->rto;
 577
 578         while(rto > USEC_PER_SEC) {
 579                 c->rtrx_timeout.tv_sec++;
 580                 rto -= USEC_PER_SEC;
 581         }
 582
 583         c->rtrx_timeout.tv_nsec += rto * 1000;
 584
 585         if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
 586                 c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
 587                 c->rtrx_timeout.tv_sec++;
 588         }
 589
 590         debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
 591 }
 592
 593 static void stop_retransmit_timer(struct utcp_connection *c) {
 594         timespec_clear(&c->rtrx_timeout);
 595         debug(c, "rtrx_timeout cleared\n");
 596 }
 597
 598 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 599         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 600
 601         if(!c) {
 602                 return NULL;
 603         }
 604
 605         assert((flags & ~0x1f) == 0);
 606
 607         c->flags = flags;
 608         c->recv = recv;
 609         c->priv = priv;
 610
 611         struct {
 612                 struct hdr hdr;
 613                 uint8_t init[4];
 614         } pkt;
 615
 616         pkt.hdr.src = c->src;
 617         pkt.hdr.dst = c->dst;
 618         pkt.hdr.seq = c->snd.iss;
 619         pkt.hdr.ack = 0;
 620         pkt.hdr.wnd = c->rcvbuf.maxsize;
 621         pkt.hdr.ctl = SYN;
 622         pkt.hdr.aux = 0x0101;
 623         pkt.init[0] = 1;
 624         pkt.init[1] = 0;
 625         pkt.init[2] = 0;
 626         pkt.init[3] = flags & 0x7;
 627
 628         set_state(c, SYN_SENT);
 629
 630         print_packet(c, "send", &pkt, sizeof(pkt));
 631         utcp->send(utcp, &pkt, sizeof(pkt));
 632
 633         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 634         c->conn_timeout.tv_sec += utcp->timeout;
 635
 636         start_retransmit_timer(c);
 637
 638         return c;
 639 }
 640
 641 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 642         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 643 }
 644
 645 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 646         if(c->reapable || c->state != SYN_RECEIVED) {
 647                 debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
 648                 return;
 649         }
 650
 651         debug(c, "accepted %p %p\n", c, recv, priv);
 652         c->recv = recv;
 653         c->priv = priv;
 654         set_state(c, ESTABLISHED);
 655 }
 656
 657 static void ack(struct utcp_connection *c, bool sendatleastone) {
 658         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 659         int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
 660
 661         assert(left >= 0);
 662
 663         if(cwndleft <= 0) {
 664                 left = 0;
 665         } else if(cwndleft < left) {
 666                 left = cwndleft;
 667
 668                 if(!sendatleastone || cwndleft > c->utcp->mss) {
 669                         left -= left % c->utcp->mss;
 670                 }
 671         }
 672
 673         debug(c, "cwndleft %d left %d\n", cwndleft, left);
 674
 675         if(!left && !sendatleastone) {
 676                 return;
 677         }
 678
 679         struct {
 680                 struct hdr hdr;
 681                 uint8_t data[];
 682         } *pkt = c->utcp->pkt;
 683
 684         pkt->hdr.src = c->src;
 685         pkt->hdr.dst = c->dst;
 686         pkt->hdr.ack = c->rcv.nxt;
 687         pkt->hdr.wnd = is_reliable(c) ? c->rcvbuf.maxsize : 0;
 688         pkt->hdr.ctl = ACK;
 689         pkt->hdr.aux = 0;
 690
 691         do {
 692                 uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
 693                 pkt->hdr.seq = c->snd.nxt;
 694
 695                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 696
 697                 c->snd.nxt += seglen;
 698                 left -= seglen;
 699
 700                 if(!is_reliable(c)) {
 701                         if(left) {
 702                                 pkt->hdr.ctl |= MF;
 703                         } else {
 704                                 pkt->hdr.ctl &= ~MF;
 705                         }
 706                 }
 707
 708                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 709                         seglen--;
 710                         pkt->hdr.ctl |= FIN;
 711                 }
 712
 713                 if(!c->rtt_start.tv_sec) {
 714                         // Start RTT measurement
 715                         clock_gettime(UTCP_CLOCK, &c->rtt_start);
 716                         c->rtt_seq = pkt->hdr.seq + seglen;
 717                         debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 718                 }
 719
 720                 print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
 721                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 722
 723                 if(left && !is_reliable(c)) {
 724                         pkt->hdr.wnd += seglen;
 725                 }
 726         } while(left);
 727 }
 728
 729 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 730         if(c->reapable) {
 731                 debug(c, "send() called on closed connection\n");
 732                 errno = EBADF;
 733                 return -1;
 734         }
 735
 736         switch(c->state) {
 737         case CLOSED:
 738         case LISTEN:
 739                 debug(c, "send() called on unconnected connection\n");
 740                 errno = ENOTCONN;
 741                 return -1;
 742
 743         case SYN_SENT:
 744         case SYN_RECEIVED:
 745         case ESTABLISHED:
 746         case CLOSE_WAIT:
 747                 break;
 748
 749         case FIN_WAIT_1:
 750         case FIN_WAIT_2:
 751         case CLOSING:
 752         case LAST_ACK:
 753         case TIME_WAIT:
 754                 debug(c, "send() called on closed connection\n");
 755                 errno = EPIPE;
 756                 return -1;
 757         }
 758
 759         // Exit early if we have nothing to send.
 760
 761         if(!len) {
 762                 return 0;
 763         }
 764
 765         if(!data) {
 766                 errno = EFAULT;
 767                 return -1;
 768         }
 769
 770         // Check if we need to be able to buffer all data
 771
 772         if(c->flags & UTCP_NO_PARTIAL) {
 773                 if(len > buffer_free(&c->sndbuf)) {
 774                         if(len > c->sndbuf.maxsize) {
 775                                 errno = EMSGSIZE;
 776                                 return -1;
 777                         } else {
 778                                 errno = EWOULDBLOCK;
 779                                 return 0;
 780                         }
 781                 }
 782         }
 783
 784         // Add data to send buffer.
 785
 786         if(is_reliable(c)) {
 787                 len = buffer_put(&c->sndbuf, data, len);
 788         } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
 789                 if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
 790                         errno = EMSGSIZE;
 791                         return -1;
 792                 }
 793         } else {
 794                 return 0;
 795         }
 796
 797         if(len <= 0) {
 798                 if(is_reliable(c)) {
 799                         errno = EWOULDBLOCK;
 800                         return 0;
 801                 } else {
 802                         return len;
 803                 }
 804         }
 805
 806         c->snd.last += len;
 807
 808         // Don't send anything yet if the connection has not fully established yet
 809
 810         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 811                 return len;
 812         }
 813
 814         ack(c, false);
 815
 816         if(!is_reliable(c)) {
 817                 c->snd.una = c->snd.nxt = c->snd.last;
 818                 buffer_discard(&c->sndbuf, c->sndbuf.used);
 819         }
 820
 821         if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
 822                 start_retransmit_timer(c);
 823         }
 824
 825         if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
 826                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
 827                 c->conn_timeout.tv_sec += c->utcp->timeout;
 828         }
 829
 830         return len;
 831 }
 832
 833 static void swap_ports(struct hdr *hdr) {
 834         uint16_t tmp = hdr->src;
 835         hdr->src = hdr->dst;
 836         hdr->dst = tmp;
 837 }
 838
 839 static void fast_retransmit(struct utcp_connection *c) {
 840         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 841                 debug(c, "fast_retransmit() called but nothing to retransmit!\n");
 842                 return;
 843         }
 844
 845         struct utcp *utcp = c->utcp;
 846
 847         struct {
 848                 struct hdr hdr;
 849                 uint8_t data[];
 850         } *pkt = c->utcp->pkt;
 851
 852         pkt->hdr.src = c->src;
 853         pkt->hdr.dst = c->dst;
 854         pkt->hdr.wnd = c->rcvbuf.maxsize;
 855         pkt->hdr.aux = 0;
 856
 857         switch(c->state) {
 858         case ESTABLISHED:
 859         case FIN_WAIT_1:
 860         case CLOSE_WAIT:
 861         case CLOSING:
 862         case LAST_ACK:
 863                 // Send unacked data again.
 864                 pkt->hdr.seq = c->snd.una;
 865                 pkt->hdr.ack = c->rcv.nxt;
 866                 pkt->hdr.ctl = ACK;
 867                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 868
 869                 if(fin_wanted(c, c->snd.una + len)) {
 870                         len--;
 871                         pkt->hdr.ctl |= FIN;
 872                 }
 873
 874                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 875                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 876                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 877                 break;
 878
 879         default:
 880                 break;
 881         }
 882 }
 883
 884 static void retransmit(struct utcp_connection *c) {
 885         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 886                 debug(c, "retransmit() called but nothing to retransmit!\n");
 887                 stop_retransmit_timer(c);
 888                 return;
 889         }
 890
 891         struct utcp *utcp = c->utcp;
 892
 893         if (utcp->retransmit) {
 894                 utcp->retransmit(c);
 895         }
 896
 897         struct {
 898                 struct hdr hdr;
 899                 uint8_t data[];
 900         } *pkt = c->utcp->pkt;
 901
 902         pkt->hdr.src = c->src;
 903         pkt->hdr.dst = c->dst;
 904         pkt->hdr.wnd = c->rcvbuf.maxsize;
 905         pkt->hdr.aux = 0;
 906
 907         switch(c->state) {
 908         case SYN_SENT:
 909                 // Send our SYN again
 910                 pkt->hdr.seq = c->snd.iss;
 911                 pkt->hdr.ack = 0;
 912                 pkt->hdr.ctl = SYN;
 913                 pkt->hdr.aux = 0x0101;
 914                 pkt->data[0] = 1;
 915                 pkt->data[1] = 0;
 916                 pkt->data[2] = 0;
 917                 pkt->data[3] = c->flags & 0x7;
 918                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 919                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 920                 break;
 921
 922         case SYN_RECEIVED:
 923                 // Send SYNACK again
 924                 pkt->hdr.seq = c->snd.nxt;
 925                 pkt->hdr.ack = c->rcv.nxt;
 926                 pkt->hdr.ctl = SYN | ACK;
 927                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
 928                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 929                 break;
 930
 931         case ESTABLISHED:
 932         case FIN_WAIT_1:
 933         case CLOSE_WAIT:
 934         case CLOSING:
 935         case LAST_ACK:
 936                 // Send unacked data again.
 937                 pkt->hdr.seq = c->snd.una;
 938                 pkt->hdr.ack = c->rcv.nxt;
 939                 pkt->hdr.ctl = ACK;
 940                 uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mss);
 941
 942                 if(fin_wanted(c, c->snd.una + len)) {
 943                         len--;
 944                         pkt->hdr.ctl |= FIN;
 945                 }
 946
 947                 // RFC 5681 slow start after timeout
 948                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
 949                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
 950                 c->snd.cwnd = utcp->mss;
 951                 debug_cwnd(c);
 952
 953                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 954                 print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
 955                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 956
 957                 c->snd.nxt = c->snd.una + len;
 958                 break;
 959
 960         case CLOSED:
 961         case LISTEN:
 962         case TIME_WAIT:
 963         case FIN_WAIT_2:
 964                 // We shouldn't need to retransmit anything in this state.
 965 #ifdef UTCP_DEBUG
 966                 abort();
 967 #endif
 968                 stop_retransmit_timer(c);
 969                 goto cleanup;
 970         }
 971
 972         start_retransmit_timer(c);
 973         c->rto *= 2;
 974
 975         if(c->rto > MAX_RTO) {
 976                 c->rto = MAX_RTO;
 977         }
 978
 979         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 980         c->dupack = 0; // cancel any ongoing fast recovery
 981
 982 cleanup:
 983         return;
 984 }
 985
 986 /* Update receive buffer and SACK entries after consuming data.
 987  *
 988  * Situation:
 989  *
 990  * |.....0000..1111111111.....22222......3333|
 991  * |---------------^
 992  *
 993  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 994  * to remove data from the receive buffer. The idea is to substract "len"
 995  * from the offset of all the SACK entries, and then remove/cut down entries
 996  * that are shifted to before the start of the receive buffer.
 997  *
 998  * There are three cases:
 999  * - the SACK entry is after ^, in that case just change the offset.
1000  * - the SACK entry starts before and ends after ^, so we have to
1001  *   change both its offset and size.
1002  * - the SACK entry is completely before ^, in that case delete it.
1003  */
1004 static void sack_consume(struct utcp_connection *c, size_t len) {
1005         debug(c, "sack_consume %lu\n", (unsigned long)len);
1006
1007         if(len > c->rcvbuf.used) {
1008                 debug(c, "all SACK entries consumed\n");
1009                 c->sacks[0].len = 0;
1010                 return;
1011         }
1012
1013         buffer_discard(&c->rcvbuf, len);
1014
1015         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
1016                 if(len < c->sacks[i].offset) {
1017                         c->sacks[i].offset -= len;
1018                         i++;
1019                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
1020                         c->sacks[i].len -= len - c->sacks[i].offset;
1021                         c->sacks[i].offset = 0;
1022                         i++;
1023                 } else {
1024                         if(i < NSACKS - 1) {
1025                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
1026                                 c->sacks[NSACKS - 1].len = 0;
1027                         } else {
1028                                 c->sacks[i].len = 0;
1029                                 break;
1030                         }
1031                 }
1032         }
1033
1034         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1035                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1036         }
1037 }
1038
1039 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
1040         debug(c, "out of order packet, offset %u\n", offset);
1041         // Packet loss or reordering occured. Store the data in the buffer.
1042         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
1043
1044         if(rxd <= 0) {
1045                 debug(c, "packet outside receive buffer, dropping\n");
1046                 return;
1047         }
1048
1049         if((size_t)rxd < len) {
1050                 debug(c, "packet partially outside receive buffer\n");
1051                 len = rxd;
1052         }
1053
1054         // Make note of where we put it.
1055         for(int i = 0; i < NSACKS; i++) {
1056                 if(!c->sacks[i].len) { // nothing to merge, add new entry
1057                         debug(c, "new SACK entry %d\n", i);
1058                         c->sacks[i].offset = offset;
1059                         c->sacks[i].len = rxd;
1060                         break;
1061                 } else if(offset < c->sacks[i].offset) {
1062                         if(offset + rxd < c->sacks[i].offset) { // insert before
1063                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
1064                                         debug(c, "insert SACK entry at %d\n", i);
1065                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
1066                                         c->sacks[i].offset = offset;
1067                                         c->sacks[i].len = rxd;
1068                                 } else {
1069                                         debug(c, "SACK entries full, dropping packet\n");
1070                                 }
1071
1072                                 break;
1073                         } else { // merge
1074                                 debug(c, "merge with start of SACK entry at %d\n", i);
1075                                 c->sacks[i].offset = offset;
1076                                 break;
1077                         }
1078                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
1079                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
1080                                 debug(c, "merge with end of SACK entry at %d\n", i);
1081                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
1082                                 // TODO: handle potential merge with next entry
1083                         }
1084
1085                         break;
1086                 }
1087         }
1088
1089         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
1090                 debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
1091         }
1092 }
1093
1094 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
1095         if(c->recv) {
1096                 ssize_t rxd = c->recv(c, data, len);
1097
1098                 if(rxd != (ssize_t)len) {
1099                         // TODO: handle the application not accepting all data.
1100                         abort();
1101                 }
1102         }
1103
1104         // Check if we can process out-of-order data now.
1105         if(c->sacks[0].len && len >= c->sacks[0].offset) {
1106                 debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
1107
1108                 if(len < c->sacks[0].offset + c->sacks[0].len) {
1109                         size_t offset = len;
1110                         len = c->sacks[0].offset + c->sacks[0].len;
1111                         size_t remainder = len - offset;
1112
1113                         if(c->recv) {
1114                                 ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
1115
1116                                 if(rxd != (ssize_t)remainder) {
1117                                         // TODO: handle the application not accepting all data.
1118                                         abort();
1119                                 }
1120                         }
1121                 }
1122         }
1123
1124         if(c->rcvbuf.used) {
1125                 sack_consume(c, len);
1126         }
1127
1128         c->rcv.nxt += len;
1129 }
1130
1131 static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1132         // Fast path for unfragmented packets
1133         if(!hdr->wnd && !(hdr->ctl & MF)) {
1134                 if(c->recv) {
1135                         c->recv(c, data, len);
1136                 }
1137
1138                 c->rcv.nxt = hdr->seq + len;
1139                 return;
1140         }
1141
1142         // Ensure reassembled packet are not larger than 64 kiB
1143         if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
1144                 return;
1145         }
1146
1147         // Don't accept out of order fragments
1148         if(hdr->wnd && hdr->seq != c->rcv.nxt) {
1149                 return;
1150         }
1151
1152         // Reset the receive buffer for the first fragment
1153         if(!hdr->wnd) {
1154                 buffer_clear(&c->rcvbuf);
1155         }
1156
1157         ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
1158
1159         if(rxd != (ssize_t)len) {
1160                 return;
1161         }
1162
1163         // Send the packet if it's the final fragment
1164         if(!(hdr->ctl & MF) && c->recv) {
1165                 buffer_call(&c->rcvbuf, c->recv, c, 0, hdr->wnd + len);
1166         }
1167
1168         c->rcv.nxt = hdr->seq + len;
1169 }
1170
1171 static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
1172         if(!is_reliable(c)) {
1173                 handle_unreliable(c, hdr, data, len);
1174                 return;
1175         }
1176
1177         uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
1178
1179         if(offset) {
1180                 handle_out_of_order(c, offset, data, len);
1181         } else {
1182                 handle_in_order(c, data, len);
1183         }
1184 }
1185
1186
1187 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
1188         const uint8_t *ptr = data;
1189
1190         if(!utcp) {
1191                 errno = EFAULT;
1192                 return -1;
1193         }
1194
1195         if(!len) {
1196                 return 0;
1197         }
1198
1199         if(!data) {
1200                 errno = EFAULT;
1201                 return -1;
1202         }
1203
1204         // Drop packets smaller than the header
1205
1206         struct hdr hdr;
1207
1208         if(len < sizeof(hdr)) {
1209                 print_packet(NULL, "recv", data, len);
1210                 errno = EBADMSG;
1211                 return -1;
1212         }
1213
1214         // Make a copy from the potentially unaligned data to a struct hdr
1215
1216         memcpy(&hdr, ptr, sizeof(hdr));
1217
1218         // Try to match the packet to an existing connection
1219
1220         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1221         print_packet(c, "recv", data, len);
1222
1223         // Process the header
1224
1225         ptr += sizeof(hdr);
1226         len -= sizeof(hdr);
1227
1228         // Drop packets with an unknown CTL flag
1229
1230         if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
1231                 print_packet(NULL, "recv", data, len);
1232                 errno = EBADMSG;
1233                 return -1;
1234         }
1235
1236         // Check for auxiliary headers
1237
1238         const uint8_t *init = NULL;
1239
1240         uint16_t aux = hdr.aux;
1241
1242         while(aux) {
1243                 size_t auxlen = 4 * (aux >> 8) & 0xf;
1244                 uint8_t auxtype = aux & 0xff;
1245
1246                 if(len < auxlen) {
1247                         errno = EBADMSG;
1248                         return -1;
1249                 }
1250
1251                 switch(auxtype) {
1252                 case AUX_INIT:
1253                         if(!(hdr.ctl & SYN) || auxlen != 4) {
1254                                 errno = EBADMSG;
1255                                 return -1;
1256                         }
1257
1258                         init = ptr;
1259                         break;
1260
1261                 default:
1262                         errno = EBADMSG;
1263                         return -1;
1264                 }
1265
1266                 len -= auxlen;
1267                 ptr += auxlen;
1268
1269                 if(!(aux & 0x800)) {
1270                         break;
1271                 }
1272
1273                 if(len < 2) {
1274                         errno = EBADMSG;
1275                         return -1;
1276                 }
1277
1278                 memcpy(&aux, ptr, 2);
1279                 len -= 2;
1280                 ptr += 2;
1281         }
1282
1283         bool has_data = len || (hdr.ctl & (SYN | FIN));
1284
1285         // Is it for a new connection?
1286
1287         if(!c) {
1288                 // Ignore RST packets
1289
1290                 if(hdr.ctl & RST) {
1291                         return 0;
1292                 }
1293
1294                 // Is it a SYN packet and are we LISTENing?
1295
1296                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1297                         // If we don't want to accept it, send a RST back
1298                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1299                                 len = 1;
1300                                 goto reset;
1301                         }
1302
1303                         // Try to allocate memory, otherwise send a RST back
1304                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1305
1306                         if(!c) {
1307                                 len = 1;
1308                                 goto reset;
1309                         }
1310
1311                         // Parse auxilliary information
1312                         if(init) {
1313                                 if(init[0] < 1) {
1314                                         len = 1;
1315                                         goto reset;
1316                                 }
1317
1318                                 c->flags = init[3] & 0x7;
1319                         } else {
1320                                 c->flags = UTCP_TCP;
1321                         }
1322
1323 synack:
1324                         // Return SYN+ACK, go to SYN_RECEIVED state
1325                         c->snd.wnd = hdr.wnd;
1326                         c->rcv.irs = hdr.seq;
1327                         c->rcv.nxt = c->rcv.irs + 1;
1328                         set_state(c, SYN_RECEIVED);
1329
1330                         struct {
1331                                 struct hdr hdr;
1332                                 uint8_t data[4];
1333                         } pkt;
1334
1335                         pkt.hdr.src = c->src;
1336                         pkt.hdr.dst = c->dst;
1337                         pkt.hdr.ack = c->rcv.irs + 1;
1338                         pkt.hdr.seq = c->snd.iss;
1339                         pkt.hdr.wnd = c->rcvbuf.maxsize;
1340                         pkt.hdr.ctl = SYN | ACK;
1341
1342                         if(init) {
1343                                 pkt.hdr.aux = 0x0101;
1344                                 pkt.data[0] = 1;
1345                                 pkt.data[1] = 0;
1346                                 pkt.data[2] = 0;
1347                                 pkt.data[3] = c->flags & 0x7;
1348                                 print_packet(c, "send", &pkt, sizeof(hdr) + 4);
1349                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1350                         } else {
1351                                 pkt.hdr.aux = 0;
1352                                 print_packet(c, "send", &pkt, sizeof(hdr));
1353                                 utcp->send(utcp, &pkt, sizeof(hdr));
1354                         }
1355                 } else {
1356                         // No, we don't want your packets, send a RST back
1357                         len = 1;
1358                         goto reset;
1359                 }
1360
1361                 return 0;
1362         }
1363
1364         debug(c, "state %s\n", strstate[c->state]);
1365
1366         // In case this is for a CLOSED connection, ignore the packet.
1367         // TODO: make it so incoming packets can never match a CLOSED connection.
1368
1369         if(c->state == CLOSED) {
1370                 debug(c, "got packet for closed connection\n");
1371                 return 0;
1372         }
1373
1374         // It is for an existing connection.
1375
1376         // 1. Drop invalid packets.
1377
1378         // 1a. Drop packets that should not happen in our current state.
1379
1380         switch(c->state) {
1381         case SYN_SENT:
1382         case SYN_RECEIVED:
1383         case ESTABLISHED:
1384         case FIN_WAIT_1:
1385         case FIN_WAIT_2:
1386         case CLOSE_WAIT:
1387         case CLOSING:
1388         case LAST_ACK:
1389         case TIME_WAIT:
1390                 break;
1391
1392         default:
1393 #ifdef UTCP_DEBUG
1394                 abort();
1395 #endif
1396                 break;
1397         }
1398
1399         // 1b. Discard data that is not in our receive window.
1400
1401         if(is_reliable(c)) {
1402                 bool acceptable;
1403
1404                 if(c->state == SYN_SENT) {
1405                         acceptable = true;
1406                 } else if(len == 0) {
1407                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1408                 } else {
1409                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1410
1411                         // cut already accepted front overlapping
1412                         if(rcv_offset < 0) {
1413                                 acceptable = len > (size_t) - rcv_offset;
1414
1415                                 if(acceptable) {
1416                                         ptr -= rcv_offset;
1417                                         len += rcv_offset;
1418                                         hdr.seq -= rcv_offset;
1419                                 }
1420                         } else {
1421                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1422                         }
1423                 }
1424
1425                 if(!acceptable) {
1426                         debug(c, "packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1427
1428                         // Ignore unacceptable RST packets.
1429                         if(hdr.ctl & RST) {
1430                                 return 0;
1431                         }
1432
1433                         // Otherwise, continue processing.
1434                         len = 0;
1435                 }
1436         } else {
1437 #if UTCP_DEBUG
1438                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1439
1440                 if(rcv_offset) {
1441                         debug(c, "packet out of order, offset %u bytes", rcv_offset);
1442                 }
1443
1444 #endif
1445         }
1446
1447         c->snd.wnd = hdr.wnd; // TODO: move below
1448
1449         // 1c. Drop packets with an invalid ACK.
1450         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1451         // (= snd.una + c->sndbuf.used).
1452
1453         if(!is_reliable(c)) {
1454                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1455                         hdr.ack = c->snd.una;
1456                 }
1457         }
1458
1459         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1460                 debug(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1461
1462                 // Ignore unacceptable RST packets.
1463                 if(hdr.ctl & RST) {
1464                         return 0;
1465                 }
1466
1467                 goto reset;
1468         }
1469
1470         // 2. Handle RST packets
1471
1472         if(hdr.ctl & RST) {
1473                 switch(c->state) {
1474                 case SYN_SENT:
1475                         if(!(hdr.ctl & ACK)) {
1476                                 return 0;
1477                         }
1478
1479                         // The peer has refused our connection.
1480                         set_state(c, CLOSED);
1481                         errno = ECONNREFUSED;
1482
1483                         if(c->recv) {
1484                                 c->recv(c, NULL, 0);
1485                         }
1486
1487                         if(c->poll && !c->reapable) {
1488                                 c->poll(c, 0);
1489                         }
1490
1491                         return 0;
1492
1493                 case SYN_RECEIVED:
1494                         if(hdr.ctl & ACK) {
1495                                 return 0;
1496                         }
1497
1498                         // We haven't told the application about this connection yet. Silently delete.
1499                         free_connection(c);
1500                         return 0;
1501
1502                 case ESTABLISHED:
1503                 case FIN_WAIT_1:
1504                 case FIN_WAIT_2:
1505                 case CLOSE_WAIT:
1506                         if(hdr.ctl & ACK) {
1507                                 return 0;
1508                         }
1509
1510                         // The peer has aborted our connection.
1511                         set_state(c, CLOSED);
1512                         errno = ECONNRESET;
1513
1514                         if(c->recv) {
1515                                 c->recv(c, NULL, 0);
1516                         }
1517
1518                         if(c->poll && !c->reapable) {
1519                                 c->poll(c, 0);
1520                         }
1521
1522                         return 0;
1523
1524                 case CLOSING:
1525                 case LAST_ACK:
1526                 case TIME_WAIT:
1527                         if(hdr.ctl & ACK) {
1528                                 return 0;
1529                         }
1530
1531                         // As far as the application is concerned, the connection has already been closed.
1532                         // If it has called utcp_close() already, we can immediately free this connection.
1533                         if(c->reapable) {
1534                                 free_connection(c);
1535                                 return 0;
1536                         }
1537
1538                         // Otherwise, immediately move to the CLOSED state.
1539                         set_state(c, CLOSED);
1540                         return 0;
1541
1542                 default:
1543 #ifdef UTCP_DEBUG
1544                         abort();
1545 #endif
1546                         break;
1547                 }
1548         }
1549
1550         uint32_t advanced;
1551
1552         if(!(hdr.ctl & ACK)) {
1553                 advanced = 0;
1554                 goto skip_ack;
1555         }
1556
1557         // 3. Advance snd.una
1558
1559         advanced = seqdiff(hdr.ack, c->snd.una);
1560
1561         if(advanced) {
1562                 // RTT measurement
1563                 if(c->rtt_start.tv_sec) {
1564                         if(c->rtt_seq == hdr.ack) {
1565                                 struct timespec now;
1566                                 clock_gettime(UTCP_CLOCK, &now);
1567                                 int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
1568                                 update_rtt(c, diff);
1569                                 c->rtt_start.tv_sec = 0;
1570                         } else if(c->rtt_seq < hdr.ack) {
1571                                 debug(c, "cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1572                                 c->rtt_start.tv_sec = 0;
1573                         }
1574                 }
1575
1576                 int32_t data_acked = advanced;
1577
1578                 switch(c->state) {
1579                 case SYN_SENT:
1580                 case SYN_RECEIVED:
1581                         data_acked--;
1582                         break;
1583
1584                 // TODO: handle FIN as well.
1585                 default:
1586                         break;
1587                 }
1588
1589                 assert(data_acked >= 0);
1590
1591 #ifndef NDEBUG
1592                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1593                 assert(data_acked <= bufused);
1594 #endif
1595
1596                 if(data_acked) {
1597                         buffer_discard(&c->sndbuf, data_acked);
1598
1599                         if(is_reliable(c)) {
1600                                 c->do_poll = true;
1601                         }
1602                 }
1603
1604                 // Also advance snd.nxt if possible
1605                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1606                         c->snd.nxt = hdr.ack;
1607                 }
1608
1609                 c->snd.una = hdr.ack;
1610
1611                 if(c->dupack) {
1612                         if(c->dupack >= 3) {
1613                                 debug(c, "fast recovery ended\n");
1614                                 c->snd.cwnd = c->snd.ssthresh;
1615                         }
1616
1617                         c->dupack = 0;
1618                 }
1619
1620                 // Increase the congestion window according to RFC 5681
1621                 if(c->snd.cwnd < c->snd.ssthresh) {
1622                         c->snd.cwnd += min(advanced, utcp->mss); // eq. 2
1623                 } else {
1624                         c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / c->snd.cwnd); // eq. 3
1625                 }
1626
1627                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1628                         c->snd.cwnd = c->sndbuf.maxsize;
1629                 }
1630
1631                 debug_cwnd(c);
1632
1633                 // Check if we have sent a FIN that is now ACKed.
1634                 switch(c->state) {
1635                 case FIN_WAIT_1:
1636                         if(c->snd.una == c->snd.last) {
1637                                 set_state(c, FIN_WAIT_2);
1638                         }
1639
1640                         break;
1641
1642                 case CLOSING:
1643                         if(c->snd.una == c->snd.last) {
1644                                 clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1645                                 c->conn_timeout.tv_sec += utcp->timeout;
1646                                 set_state(c, TIME_WAIT);
1647                         }
1648
1649                         break;
1650
1651                 default:
1652                         break;
1653                 }
1654         } else {
1655                 if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
1656                         c->dupack++;
1657                         debug(c, "duplicate ACK %d\n", c->dupack);
1658
1659                         if(c->dupack == 3) {
1660                                 // RFC 5681 fast recovery
1661                                 debug(c, "fast recovery started\n", c->dupack);
1662                                 uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
1663                                 c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
1664                                 c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, c->sndbuf.maxsize);
1665
1666                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1667                                         c->snd.cwnd = c->sndbuf.maxsize;
1668                                 }
1669
1670                                 debug_cwnd(c);
1671
1672                                 fast_retransmit(c);
1673                         } else if(c->dupack > 3) {
1674                                 c->snd.cwnd += utcp->mss;
1675
1676                                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1677                                         c->snd.cwnd = c->sndbuf.maxsize;
1678                                 }
1679
1680                                 debug_cwnd(c);
1681                         }
1682
1683                         // We got an ACK which indicates the other side did get one of our packets.
1684                         // Reset the retransmission timer to avoid going to slow start,
1685                         // but don't touch the connection timeout.
1686                         start_retransmit_timer(c);
1687                 }
1688         }
1689
1690         // 4. Update timers
1691
1692         if(advanced) {
1693                 if(c->snd.una == c->snd.last) {
1694                         stop_retransmit_timer(c);
1695                         timespec_clear(&c->conn_timeout);
1696                 } else if(is_reliable(c)) {
1697                         start_retransmit_timer(c);
1698                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1699                         c->conn_timeout.tv_sec += utcp->timeout;
1700                 }
1701         }
1702
1703 skip_ack:
1704         // 5. Process SYN stuff
1705
1706         if(hdr.ctl & SYN) {
1707                 switch(c->state) {
1708                 case SYN_SENT:
1709
1710                         // This is a SYNACK. It should always have ACKed the SYN.
1711                         if(!advanced) {
1712                                 goto reset;
1713                         }
1714
1715                         c->rcv.irs = hdr.seq;
1716                         c->rcv.nxt = hdr.seq;
1717
1718                         if(c->shut_wr) {
1719                                 c->snd.last++;
1720                                 set_state(c, FIN_WAIT_1);
1721                         } else {
1722                                 set_state(c, ESTABLISHED);
1723                         }
1724
1725                         // TODO: notify application of this somehow.
1726                         break;
1727
1728                 case SYN_RECEIVED:
1729                         // This is a retransmit of a SYN, send back the SYNACK.
1730                         goto synack;
1731
1732                 case ESTABLISHED:
1733                 case FIN_WAIT_1:
1734                 case FIN_WAIT_2:
1735                 case CLOSE_WAIT:
1736                 case CLOSING:
1737                 case LAST_ACK:
1738                 case TIME_WAIT:
1739                         // Ehm, no. We should never receive a second SYN.
1740                         return 0;
1741
1742                 default:
1743 #ifdef UTCP_DEBUG
1744                         abort();
1745 #endif
1746                         return 0;
1747                 }
1748
1749                 // SYN counts as one sequence number
1750                 c->rcv.nxt++;
1751         }
1752
1753         // 6. Process new data
1754
1755         if(c->state == SYN_RECEIVED) {
1756                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1757                 if(!advanced) {
1758                         goto reset;
1759                 }
1760
1761                 // Are we still LISTENing?
1762                 if(utcp->accept) {
1763                         utcp->accept(c, c->src);
1764                 }
1765
1766                 if(c->state != ESTABLISHED) {
1767                         set_state(c, CLOSED);
1768                         c->reapable = true;
1769                         goto reset;
1770                 }
1771         }
1772
1773         if(len) {
1774                 switch(c->state) {
1775                 case SYN_SENT:
1776                 case SYN_RECEIVED:
1777                         // This should never happen.
1778 #ifdef UTCP_DEBUG
1779                         abort();
1780 #endif
1781                         return 0;
1782
1783                 case ESTABLISHED:
1784                 case FIN_WAIT_1:
1785                 case FIN_WAIT_2:
1786                         break;
1787
1788                 case CLOSE_WAIT:
1789                 case CLOSING:
1790                 case LAST_ACK:
1791                 case TIME_WAIT:
1792                         // Ehm no, We should never receive more data after a FIN.
1793                         goto reset;
1794
1795                 default:
1796 #ifdef UTCP_DEBUG
1797                         abort();
1798 #endif
1799                         return 0;
1800                 }
1801
1802                 handle_incoming_data(c, &hdr, ptr, len);
1803         }
1804
1805         // 7. Process FIN stuff
1806
1807         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1808                 switch(c->state) {
1809                 case SYN_SENT:
1810                 case SYN_RECEIVED:
1811                         // This should never happen.
1812 #ifdef UTCP_DEBUG
1813                         abort();
1814 #endif
1815                         break;
1816
1817                 case ESTABLISHED:
1818                         set_state(c, CLOSE_WAIT);
1819                         break;
1820
1821                 case FIN_WAIT_1:
1822                         set_state(c, CLOSING);
1823                         break;
1824
1825                 case FIN_WAIT_2:
1826                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
1827                         c->conn_timeout.tv_sec += utcp->timeout;
1828                         set_state(c, TIME_WAIT);
1829                         break;
1830
1831                 case CLOSE_WAIT:
1832                 case CLOSING:
1833                 case LAST_ACK:
1834                 case TIME_WAIT:
1835                         // Ehm, no. We should never receive a second FIN.
1836                         goto reset;
1837
1838                 default:
1839 #ifdef UTCP_DEBUG
1840                         abort();
1841 #endif
1842                         break;
1843                 }
1844
1845                 // FIN counts as one sequence number
1846                 c->rcv.nxt++;
1847                 len++;
1848
1849                 // Inform the application that the peer closed its end of the connection.
1850                 if(c->recv) {
1851                         errno = 0;
1852                         c->recv(c, NULL, 0);
1853                 }
1854         }
1855
1856         // Now we send something back if:
1857         // - we received data, so we have to send back an ACK
1858         //   -> sendatleastone = true
1859         // - or we got an ack, so we should maybe send a bit more data
1860         //   -> sendatleastone = false
1861
1862         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1863                 ack(c, has_data);
1864         }
1865
1866         return 0;
1867
1868 reset:
1869         swap_ports(&hdr);
1870         hdr.wnd = 0;
1871         hdr.aux = 0;
1872
1873         if(hdr.ctl & ACK) {
1874                 hdr.seq = hdr.ack;
1875                 hdr.ctl = RST;
1876         } else {
1877                 hdr.ack = hdr.seq + len;
1878                 hdr.seq = 0;
1879                 hdr.ctl = RST | ACK;
1880         }
1881
1882         print_packet(c, "send", &hdr, sizeof(hdr));
1883         utcp->send(utcp, &hdr, sizeof(hdr));
1884         return 0;
1885
1886 }
1887
1888 int utcp_shutdown(struct utcp_connection *c, int dir) {
1889         debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
1890
1891         if(!c) {
1892                 errno = EFAULT;
1893                 return -1;
1894         }
1895
1896         if(c->reapable) {
1897                 debug(c, "shutdown() called on closed connection\n");
1898                 errno = EBADF;
1899                 return -1;
1900         }
1901
1902         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1903                 errno = EINVAL;
1904                 return -1;
1905         }
1906
1907         // TCP does not have a provision for stopping incoming packets.
1908         // The best we can do is to just ignore them.
1909         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1910                 c->recv = NULL;
1911         }
1912
1913         // The rest of the code deals with shutting down writes.
1914         if(dir == UTCP_SHUT_RD) {
1915                 return 0;
1916         }
1917
1918         // Only process shutting down writes once.
1919         if(c->shut_wr) {
1920                 return 0;
1921         }
1922
1923         c->shut_wr = true;
1924
1925         switch(c->state) {
1926         case CLOSED:
1927         case LISTEN:
1928                 errno = ENOTCONN;
1929                 return -1;
1930
1931         case SYN_SENT:
1932                 return 0;
1933
1934         case SYN_RECEIVED:
1935         case ESTABLISHED:
1936                 set_state(c, FIN_WAIT_1);
1937                 break;
1938
1939         case FIN_WAIT_1:
1940         case FIN_WAIT_2:
1941                 return 0;
1942
1943         case CLOSE_WAIT:
1944                 set_state(c, CLOSING);
1945                 break;
1946
1947         case CLOSING:
1948         case LAST_ACK:
1949         case TIME_WAIT:
1950                 return 0;
1951         }
1952
1953         c->snd.last++;
1954
1955         ack(c, false);
1956
1957         if(!timespec_isset(&c->rtrx_timeout)) {
1958                 start_retransmit_timer(c);
1959         }
1960
1961         return 0;
1962 }
1963
1964 static bool reset_connection(struct utcp_connection *c) {
1965         if(!c) {
1966                 errno = EFAULT;
1967                 return false;
1968         }
1969
1970         if(c->reapable) {
1971                 debug(c, "abort() called on closed connection\n");
1972                 errno = EBADF;
1973                 return false;
1974         }
1975
1976         c->recv = NULL;
1977         c->poll = NULL;
1978
1979         switch(c->state) {
1980         case CLOSED:
1981                 return true;
1982
1983         case LISTEN:
1984         case SYN_SENT:
1985         case CLOSING:
1986         case LAST_ACK:
1987         case TIME_WAIT:
1988                 set_state(c, CLOSED);
1989                 return true;
1990
1991         case SYN_RECEIVED:
1992         case ESTABLISHED:
1993         case FIN_WAIT_1:
1994         case FIN_WAIT_2:
1995         case CLOSE_WAIT:
1996                 set_state(c, CLOSED);
1997                 break;
1998         }
1999
2000         // Send RST
2001
2002         struct hdr hdr;
2003
2004         hdr.src = c->src;
2005         hdr.dst = c->dst;
2006         hdr.seq = c->snd.nxt;
2007         hdr.ack = 0;
2008         hdr.wnd = 0;
2009         hdr.ctl = RST;
2010
2011         print_packet(c, "send", &hdr, sizeof(hdr));
2012         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
2013         return true;
2014 }
2015
2016 // Closes all the opened connections
2017 void utcp_abort_all_connections(struct utcp *utcp) {
2018         if(!utcp) {
2019                 errno = EINVAL;
2020                 return;
2021         }
2022
2023         for(int i = 0; i < utcp->nconnections; i++) {
2024                 struct utcp_connection *c = utcp->connections[i];
2025
2026                 if(c->reapable || c->state == CLOSED) {
2027                         continue;
2028                 }
2029
2030                 utcp_recv_t old_recv = c->recv;
2031                 utcp_poll_t old_poll = c->poll;
2032
2033                 reset_connection(c);
2034
2035                 if(old_recv) {
2036                         errno = 0;
2037                         old_recv(c, NULL, 0);
2038                 }
2039
2040                 if(old_poll && !c->reapable) {
2041                         errno = 0;
2042                         old_poll(c, 0);
2043                 }
2044         }
2045
2046         return;
2047 }
2048
2049 int utcp_close(struct utcp_connection *c) {
2050         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
2051                 return -1;
2052         }
2053
2054         c->recv = NULL;
2055         c->poll = NULL;
2056         c->reapable = true;
2057         return 0;
2058 }
2059
2060 int utcp_abort(struct utcp_connection *c) {
2061         if(!reset_connection(c)) {
2062                 return -1;
2063         }
2064
2065         c->reapable = true;
2066         return 0;
2067 }
2068
2069 /* Handle timeouts.
2070  * One call to this function will loop through all connections,
2071  * checking if something needs to be resent or not.
2072  * The return value is the time to the next timeout in milliseconds,
2073  * or maybe a negative value if the timeout is infinite.
2074  */
2075 struct timespec utcp_timeout(struct utcp *utcp) {
2076         struct timespec now;
2077         clock_gettime(UTCP_CLOCK, &now);
2078         struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
2079
2080         for(int i = 0; i < utcp->nconnections; i++) {
2081                 struct utcp_connection *c = utcp->connections[i];
2082
2083                 if(!c) {
2084                         continue;
2085                 }
2086
2087                 // delete connections that have been utcp_close()d.
2088                 if(c->state == CLOSED) {
2089                         if(c->reapable) {
2090                                 debug(c, "reaping\n");
2091                                 free_connection(c);
2092                                 i--;
2093                         }
2094
2095                         continue;
2096                 }
2097
2098                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
2099                         errno = ETIMEDOUT;
2100                         c->state = CLOSED;
2101
2102                         if(c->recv) {
2103                                 c->recv(c, NULL, 0);
2104                         }
2105
2106                         if(c->poll && !c->reapable) {
2107                                 c->poll(c, 0);
2108                         }
2109
2110                         continue;
2111                 }
2112
2113                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
2114                         debug(c, "retransmitting after timeout\n");
2115                         retransmit(c);
2116                 }
2117
2118                 if(c->poll) {
2119                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
2120                                 c->do_poll = false;
2121                                 uint32_t len = buffer_free(&c->sndbuf);
2122
2123                                 if(len) {
2124                                         c->poll(c, len);
2125                                 }
2126                         } else if(c->state == CLOSED) {
2127                                 c->poll(c, 0);
2128                         }
2129                 }
2130
2131                 if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
2132                         next = c->conn_timeout;
2133                 }
2134
2135                 if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
2136                         next = c->rtrx_timeout;
2137                 }
2138         }
2139
2140         struct timespec diff;
2141
2142         timespec_sub(&next, &now, &diff);
2143
2144         return diff;
2145 }
2146
2147 bool utcp_is_active(struct utcp *utcp) {
2148         if(!utcp) {
2149                 return false;
2150         }
2151
2152         for(int i = 0; i < utcp->nconnections; i++)
2153                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
2154                         return true;
2155                 }
2156
2157         return false;
2158 }
2159
2160 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
2161         if(!send) {
2162                 errno = EFAULT;
2163                 return NULL;
2164         }
2165
2166         struct utcp *utcp = calloc(1, sizeof(*utcp));
2167
2168         if(!utcp) {
2169                 return NULL;
2170         }
2171
2172         utcp_set_mtu(utcp, DEFAULT_MTU);
2173
2174         if(!utcp->pkt) {
2175                 free(utcp);
2176                 return NULL;
2177         }
2178
2179         if(!CLOCK_GRANULARITY) {
2180                 struct timespec res;
2181                 clock_getres(UTCP_CLOCK, &res);
2182                 CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
2183         }
2184
2185         utcp->accept = accept;
2186         utcp->pre_accept = pre_accept;
2187         utcp->send = send;
2188         utcp->priv = priv;
2189         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
2190
2191         return utcp;
2192 }
2193
2194 void utcp_exit(struct utcp *utcp) {
2195         if(!utcp) {
2196                 return;
2197         }
2198
2199         for(int i = 0; i < utcp->nconnections; i++) {
2200                 struct utcp_connection *c = utcp->connections[i];
2201
2202                 if(!c->reapable) {
2203                         if(c->recv) {
2204                                 c->recv(c, NULL, 0);
2205                         }
2206
2207                         if(c->poll && !c->reapable) {
2208                                 c->poll(c, 0);
2209                         }
2210                 }
2211
2212                 buffer_exit(&c->rcvbuf);
2213                 buffer_exit(&c->sndbuf);
2214                 free(c);
2215         }
2216
2217         free(utcp->connections);
2218         free(utcp->pkt);
2219         free(utcp);
2220 }
2221
2222 uint16_t utcp_get_mtu(struct utcp *utcp) {
2223         return utcp ? utcp->mtu : 0;
2224 }
2225
2226 uint16_t utcp_get_mss(struct utcp *utcp) {
2227         return utcp ? utcp->mss : 0;
2228 }
2229
2230 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
2231         if(!utcp) {
2232                 return;
2233         }
2234
2235         if(mtu <= sizeof(struct hdr)) {
2236                 return;
2237         }
2238
2239         if(mtu > utcp->mtu) {
2240                 char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
2241
2242                 if(!new) {
2243                         return;
2244                 }
2245
2246                 utcp->pkt = new;
2247         }
2248
2249         utcp->mtu = mtu;
2250         utcp->mss = mtu - sizeof(struct hdr);
2251 }
2252
2253 void utcp_reset_timers(struct utcp *utcp) {
2254         if(!utcp) {
2255                 return;
2256         }
2257
2258         struct timespec now, then;
2259
2260         clock_gettime(UTCP_CLOCK, &now);
2261
2262         then = now;
2263
2264         then.tv_sec += utcp->timeout;
2265
2266         for(int i = 0; i < utcp->nconnections; i++) {
2267                 struct utcp_connection *c = utcp->connections[i];
2268
2269                 if(c->reapable) {
2270                         continue;
2271                 }
2272
2273                 if(timespec_isset(&c->rtrx_timeout)) {
2274                         c->rtrx_timeout = now;
2275                 }
2276
2277                 if(timespec_isset(&c->conn_timeout)) {
2278                         c->conn_timeout = then;
2279                 }
2280
2281                 c->rtt_start.tv_sec = 0;
2282
2283                 if(c->rto > START_RTO) {
2284                         c->rto = START_RTO;
2285                 }
2286         }
2287 }
2288
2289 int utcp_get_user_timeout(struct utcp *u) {
2290         return u ? u->timeout : 0;
2291 }
2292
2293 void utcp_set_user_timeout(struct utcp *u, int timeout) {
2294         if(u) {
2295                 u->timeout = timeout;
2296         }
2297 }
2298
2299 size_t utcp_get_sndbuf(struct utcp_connection *c) {
2300         return c ? c->sndbuf.maxsize : 0;
2301 }
2302
2303 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
2304         if(!c) {
2305                 return 0;
2306         }
2307
2308         switch(c->state) {
2309         case SYN_SENT:
2310         case SYN_RECEIVED:
2311         case ESTABLISHED:
2312         case CLOSE_WAIT:
2313                 return buffer_free(&c->sndbuf);
2314
2315         default:
2316                 return 0;
2317         }
2318 }
2319
2320 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
2321         if(!c) {
2322                 return;
2323         }
2324
2325         c->sndbuf.maxsize = size;
2326
2327         if(c->sndbuf.maxsize != size) {
2328                 c->sndbuf.maxsize = -1;
2329         }
2330
2331         c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2332 }
2333
2334 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
2335         return c ? c->rcvbuf.maxsize : 0;
2336 }
2337
2338 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
2339         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
2340                 return buffer_free(&c->rcvbuf);
2341         } else {
2342                 return 0;
2343         }
2344 }
2345
2346 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
2347         if(!c) {
2348                 return;
2349         }
2350
2351         c->rcvbuf.maxsize = size;
2352
2353         if(c->rcvbuf.maxsize != size) {
2354                 c->rcvbuf.maxsize = -1;
2355         }
2356 }
2357
2358 size_t utcp_get_sendq(struct utcp_connection *c) {
2359         return c->sndbuf.used;
2360 }
2361
2362 size_t utcp_get_recvq(struct utcp_connection *c) {
2363         return c->rcvbuf.used;
2364 }
2365
2366 bool utcp_get_nodelay(struct utcp_connection *c) {
2367         return c ? c->nodelay : false;
2368 }
2369
2370 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2371         if(c) {
2372                 c->nodelay = nodelay;
2373         }
2374 }
2375
2376 bool utcp_get_keepalive(struct utcp_connection *c) {
2377         return c ? c->keepalive : false;
2378 }
2379
2380 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2381         if(c) {
2382                 c->keepalive = keepalive;
2383         }
2384 }
2385
2386 size_t utcp_get_outq(struct utcp_connection *c) {
2387         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2388 }
2389
2390 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2391         if(c) {
2392                 c->recv = recv;
2393         }
2394 }
2395
2396 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2397         if(c) {
2398                 c->poll = poll;
2399                 c->do_poll = is_reliable(c) && buffer_free(&c->sndbuf);
2400         }
2401 }
2402
2403 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2404         if(utcp) {
2405                 utcp->accept = accept;
2406                 utcp->pre_accept = pre_accept;
2407         }
2408 }
2409
2410 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2411         if(!c || c->reapable) {
2412                 return;
2413         }
2414
2415         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2416                 return;
2417         }
2418
2419         if(expect) {
2420                 // If we expect data, start the connection timer.
2421                 if(!timespec_isset(&c->conn_timeout)) {
2422                         clock_gettime(UTCP_CLOCK, &c->conn_timeout);
2423                         c->conn_timeout.tv_sec += c->utcp->timeout;
2424                 }
2425         } else {
2426                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2427                 if(c->snd.una == c->snd.last) {
2428                         timespec_clear(&c->conn_timeout);
2429                 }
2430         }
2431 }
2432
2433 void utcp_offline(struct utcp *utcp, bool offline) {
2434         struct timespec now;
2435         clock_gettime(UTCP_CLOCK, &now);
2436
2437         for(int i = 0; i < utcp->nconnections; i++) {
2438                 struct utcp_connection *c = utcp->connections[i];
2439
2440                 if(c->reapable) {
2441                         continue;
2442                 }
2443
2444                 utcp_expect_data(c, offline);
2445
2446                 if(!offline) {
2447                         if(timespec_isset(&c->rtrx_timeout)) {
2448                                 c->rtrx_timeout = now;
2449                         }
2450
2451                         utcp->connections[i]->rtt_start.tv_sec = 0;
2452
2453                         if(c->rto > START_RTO) {
2454                                 c->rto = START_RTO;
2455                         }
2456                 }
2457         }
2458 }
2459
2460 void utcp_set_retransmit_cb(struct utcp *utcp, utcp_retransmit_t retransmit) {
2461         utcp->retransmit = retransmit;
2462 }
2463
2464 void utcp_set_clock_granularity(long granularity) {
2465         CLOCK_GRANULARITY = granularity;
2466 }