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 <sys/time.h>
  31 #include <sys/socket.h>
  32
  33 #include "utcp_priv.h"
  34
  35 #ifndef EBADMSG
  36 #define EBADMSG         104
  37 #endif
  38
  39 #ifndef SHUT_RDWR
  40 #define SHUT_RDWR 2
  41 #endif
  42
  43 #ifdef poll
  44 #undef poll
  45 #endif
  46
  47 #ifndef timersub
  48 #define timersub(a, b, r)\
  49         do {\
  50                 (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
  51                 (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
  52                 if((r)->tv_usec < 0)\
  53                         (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
  54         } while (0)
  55 #endif
  56
  57 static inline size_t max(size_t a, size_t b) {
  58         return a > b ? a : b;
  59 }
  60
  61 #ifdef UTCP_DEBUG
  62 #include <stdarg.h>
  63
  64 static void debug(const char *format, ...) {
  65         va_list ap;
  66         va_start(ap, format);
  67         vfprintf(stderr, format, ap);
  68         va_end(ap);
  69 }
  70
  71 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
  72         struct hdr hdr;
  73
  74         if(len < sizeof(hdr)) {
  75                 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
  76                 return;
  77         }
  78
  79         memcpy(&hdr, pkt, sizeof(hdr));
  80         debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux);
  81
  82         if(hdr.ctl & SYN) {
  83                 debug("SYN");
  84         }
  85
  86         if(hdr.ctl & RST) {
  87                 debug("RST");
  88         }
  89
  90         if(hdr.ctl & FIN) {
  91                 debug("FIN");
  92         }
  93
  94         if(hdr.ctl & ACK) {
  95                 debug("ACK");
  96         }
  97
  98         if(len > sizeof(hdr)) {
  99                 uint32_t datalen = len - sizeof(hdr);
 100                 const uint8_t *data = (uint8_t *)pkt + sizeof(hdr);
 101                 char str[datalen * 2 + 1];
 102                 char *p = str;
 103
 104                 for(uint32_t i = 0; i < datalen; i++) {
 105                         *p++ = "0123456789ABCDEF"[data[i] >> 4];
 106                         *p++ = "0123456789ABCDEF"[data[i] & 15];
 107                 }
 108
 109                 *p = 0;
 110
 111                 debug(" data=%s", str);
 112         }
 113
 114         debug("\n");
 115 }
 116 #else
 117 #define debug(...) do {} while(0)
 118 #define print_packet(...) do {} while(0)
 119 #endif
 120
 121 static void set_state(struct utcp_connection *c, enum state state) {
 122         c->state = state;
 123
 124         if(state == ESTABLISHED) {
 125                 timerclear(&c->conn_timeout);
 126         }
 127
 128         debug("%p new state: %s\n", c->utcp, strstate[state]);
 129 }
 130
 131 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 132         if(seq != c->snd.last) {
 133                 return false;
 134         }
 135
 136         switch(c->state) {
 137         case FIN_WAIT_1:
 138         case CLOSING:
 139         case LAST_ACK:
 140                 return true;
 141
 142         default:
 143                 return false;
 144         }
 145 }
 146
 147 static bool is_reliable(struct utcp_connection *c) {
 148         return c->flags & UTCP_RELIABLE;
 149 }
 150
 151 static int32_t seqdiff(uint32_t a, uint32_t b) {
 152         return a - b;
 153 }
 154
 155 // Buffer functions
 156 // TODO: convert to ringbuffers to avoid memmove() operations.
 157
 158 // Store data into the buffer
 159 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 160         debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 161
 162         size_t required = offset + len;
 163
 164         if(required > buf->maxsize) {
 165                 if(offset >= buf->maxsize) {
 166                         return 0;
 167                 }
 168
 169                 len = buf->maxsize - offset;
 170                 required = buf->maxsize;
 171         }
 172
 173         if(required > buf->size) {
 174                 size_t newsize = buf->size;
 175
 176                 if(!newsize) {
 177                         newsize = required;
 178                 } else {
 179                         do {
 180                                 newsize *= 2;
 181                         } while(newsize < required);
 182                 }
 183
 184                 if(newsize > buf->maxsize) {
 185                         newsize = buf->maxsize;
 186                 }
 187
 188                 char *newdata = realloc(buf->data, newsize);
 189
 190                 if(!newdata) {
 191                         return -1;
 192                 }
 193
 194                 buf->data = newdata;
 195                 buf->size = newsize;
 196         }
 197
 198         memcpy(buf->data + offset, data, len);
 199
 200         if(required > buf->used) {
 201                 buf->used = required;
 202         }
 203
 204         return len;
 205 }
 206
 207 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 208         return buffer_put_at(buf, buf->used, data, len);
 209 }
 210
 211 // Get data from the buffer. data can be NULL.
 212 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 213         if(len > buf->used) {
 214                 len = buf->used;
 215         }
 216
 217         if(data) {
 218                 memcpy(data, buf->data, len);
 219         }
 220
 221         if(len < buf->used) {
 222                 memmove(buf->data, buf->data + len, buf->used - len);
 223         }
 224
 225         buf->used -= len;
 226         return len;
 227 }
 228
 229 // Copy data from the buffer without removing it.
 230 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 231         if(offset >= buf->used) {
 232                 return 0;
 233         }
 234
 235         if(offset + len > buf->used) {
 236                 len = buf->used - offset;
 237         }
 238
 239         memcpy(data, buf->data + offset, len);
 240         return len;
 241 }
 242
 243 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 244         memset(buf, 0, sizeof(*buf));
 245
 246         if(len) {
 247                 buf->data = malloc(len);
 248
 249                 if(!buf->data) {
 250                         return false;
 251                 }
 252         }
 253
 254         buf->size = len;
 255         buf->maxsize = maxlen;
 256         return true;
 257 }
 258
 259 static void buffer_exit(struct buffer *buf) {
 260         free(buf->data);
 261         memset(buf, 0, sizeof(*buf));
 262 }
 263
 264 static uint32_t buffer_free(const struct buffer *buf) {
 265         return buf->maxsize - buf->used;
 266 }
 267
 268 // Connections are stored in a sorted list.
 269 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 270
 271 static int compare(const void *va, const void *vb) {
 272         assert(va && vb);
 273
 274         const struct utcp_connection *a = *(struct utcp_connection **)va;
 275         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 276
 277         assert(a && b);
 278         assert(a->src && b->src);
 279
 280         int c = (int)a->src - (int)b->src;
 281
 282         if(c) {
 283                 return c;
 284         }
 285
 286         c = (int)a->dst - (int)b->dst;
 287         return c;
 288 }
 289
 290 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 291         if(!utcp->nconnections) {
 292                 return NULL;
 293         }
 294
 295         struct utcp_connection key = {
 296                 .src = src,
 297                 .dst = dst,
 298         }, *keyp = &key;
 299         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 300         return match ? *match : NULL;
 301 }
 302
 303 static void free_connection(struct utcp_connection *c) {
 304         struct utcp *utcp = c->utcp;
 305         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 306
 307         assert(cp);
 308
 309         int i = cp - utcp->connections;
 310         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
 311         utcp->nconnections--;
 312
 313         buffer_exit(&c->rcvbuf);
 314         buffer_exit(&c->sndbuf);
 315         free(c);
 316 }
 317
 318 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 319         // Check whether this combination of src and dst is free
 320
 321         if(src) {
 322                 if(find_connection(utcp, src, dst)) {
 323                         errno = EADDRINUSE;
 324                         return NULL;
 325                 }
 326         } else { // If src == 0, generate a random port number with the high bit set
 327                 if(utcp->nconnections >= 32767) {
 328                         errno = ENOMEM;
 329                         return NULL;
 330                 }
 331
 332                 src = rand() | 0x8000;
 333
 334                 while(find_connection(utcp, src, dst)) {
 335                         src++;
 336                 }
 337         }
 338
 339         // Allocate memory for the new connection
 340
 341         if(utcp->nconnections >= utcp->nallocated) {
 342                 if(!utcp->nallocated) {
 343                         utcp->nallocated = 4;
 344                 } else {
 345                         utcp->nallocated *= 2;
 346                 }
 347
 348                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
 349
 350                 if(!new_array) {
 351                         return NULL;
 352                 }
 353
 354                 utcp->connections = new_array;
 355         }
 356
 357         struct utcp_connection *c = calloc(1, sizeof(*c));
 358
 359         if(!c) {
 360                 return NULL;
 361         }
 362
 363         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 364                 free(c);
 365                 return NULL;
 366         }
 367
 368         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 369                 buffer_exit(&c->sndbuf);
 370                 free(c);
 371                 return NULL;
 372         }
 373
 374         // Fill in the details
 375
 376         c->src = src;
 377         c->dst = dst;
 378 #ifdef UTCP_DEBUG
 379         c->snd.iss = 0;
 380 #else
 381         c->snd.iss = rand();
 382 #endif
 383         c->snd.una = c->snd.iss;
 384         c->snd.nxt = c->snd.iss + 1;
 385         c->rcv.wnd = utcp->mtu;
 386         c->snd.last = c->snd.nxt;
 387         c->snd.cwnd = utcp->mtu;
 388         c->utcp = utcp;
 389
 390         // Add it to the sorted list of connections
 391
 392         utcp->connections[utcp->nconnections++] = c;
 393         qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
 394
 395         return c;
 396 }
 397
 398 static inline uint32_t absdiff(uint32_t a, uint32_t b) {
 399         if(a > b) {
 400                 return a - b;
 401         } else {
 402                 return b - a;
 403         }
 404 }
 405
 406 // Update RTT variables. See RFC 6298.
 407 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 408         if(!rtt) {
 409                 debug("invalid rtt\n");
 410                 return;
 411         }
 412
 413         struct utcp *utcp = c->utcp;
 414
 415         if(!utcp->srtt) {
 416                 utcp->srtt = rtt;
 417                 utcp->rttvar = rtt / 2;
 418         } else {
 419                 utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
 420                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 421         }
 422
 423         utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
 424
 425         if(utcp->rto > MAX_RTO) {
 426                 utcp->rto = MAX_RTO;
 427         }
 428
 429         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 430 }
 431
 432 static void start_retransmit_timer(struct utcp_connection *c) {
 433         gettimeofday(&c->rtrx_timeout, NULL);
 434         c->rtrx_timeout.tv_usec += c->utcp->rto;
 435
 436         while(c->rtrx_timeout.tv_usec >= 1000000) {
 437                 c->rtrx_timeout.tv_usec -= 1000000;
 438                 c->rtrx_timeout.tv_sec++;
 439         }
 440
 441         debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
 442 }
 443
 444 static void stop_retransmit_timer(struct utcp_connection *c) {
 445         timerclear(&c->rtrx_timeout);
 446         debug("timeout cleared\n");
 447 }
 448
 449 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 450         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 451
 452         if(!c) {
 453                 return NULL;
 454         }
 455
 456         assert((flags & ~0x1f) == 0);
 457
 458         c->flags = flags;
 459         c->recv = recv;
 460         c->priv = priv;
 461
 462         struct {
 463                 struct hdr hdr;
 464                 uint8_t init[4];
 465         } pkt;
 466
 467         pkt.hdr.src = c->src;
 468         pkt.hdr.dst = c->dst;
 469         pkt.hdr.seq = c->snd.iss;
 470         pkt.hdr.ack = 0;
 471         pkt.hdr.wnd = c->rcv.wnd;
 472         pkt.hdr.ctl = SYN;
 473         pkt.hdr.aux = 0x0101;
 474         pkt.init[0] = 1;
 475         pkt.init[1] = 0;
 476         pkt.init[2] = 0;
 477         pkt.init[3] = flags & 0x7;
 478
 479         set_state(c, SYN_SENT);
 480
 481         print_packet(utcp, "send", &pkt, sizeof(pkt));
 482         utcp->send(utcp, &pkt, sizeof(pkt));
 483
 484         gettimeofday(&c->conn_timeout, NULL);
 485         c->conn_timeout.tv_sec += utcp->timeout;
 486
 487         start_retransmit_timer(c);
 488
 489         return c;
 490 }
 491
 492 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 493         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 494 }
 495
 496 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 497         if(c->reapable || c->state != SYN_RECEIVED) {
 498                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 499                 return;
 500         }
 501
 502         debug("%p accepted, %p %p\n", c, recv, priv);
 503         c->recv = recv;
 504         c->priv = priv;
 505         set_state(c, ESTABLISHED);
 506 }
 507
 508 static void ack(struct utcp_connection *c, bool sendatleastone) {
 509         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 510         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 511         debug("cwndleft = %d\n", cwndleft);
 512
 513         assert(left >= 0);
 514
 515         if(cwndleft <= 0) {
 516                 cwndleft = 0;
 517         }
 518
 519         if(cwndleft < left) {
 520                 left = cwndleft;
 521         }
 522
 523         if(!left && !sendatleastone) {
 524                 return;
 525         }
 526
 527         struct {
 528                 struct hdr hdr;
 529                 uint8_t data[];
 530         } *pkt;
 531
 532         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 533
 534         if(!pkt) {
 535                 return;
 536         }
 537
 538         pkt->hdr.src = c->src;
 539         pkt->hdr.dst = c->dst;
 540         pkt->hdr.ack = c->rcv.nxt;
 541         pkt->hdr.wnd = c->snd.wnd;
 542         pkt->hdr.ctl = ACK;
 543         pkt->hdr.aux = 0;
 544
 545         do {
 546                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 547                 pkt->hdr.seq = c->snd.nxt;
 548
 549                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 550
 551                 c->snd.nxt += seglen;
 552                 left -= seglen;
 553
 554                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 555                         seglen--;
 556                         pkt->hdr.ctl |= FIN;
 557                 }
 558
 559                 if(!c->rtt_start.tv_sec) {
 560                         // Start RTT measurement
 561                         gettimeofday(&c->rtt_start, NULL);
 562                         c->rtt_seq = pkt->hdr.seq + seglen;
 563                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 564                 }
 565
 566                 print_packet(c->utcp, "send", pkt, sizeof(pkt->hdr) + seglen);
 567                 c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
 568         } while(left);
 569
 570         free(pkt);
 571 }
 572
 573 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 574         if(c->reapable) {
 575                 debug("Error: send() called on closed connection %p\n", c);
 576                 errno = EBADF;
 577                 return -1;
 578         }
 579
 580         switch(c->state) {
 581         case CLOSED:
 582         case LISTEN:
 583                 debug("Error: send() called on unconnected connection %p\n", c);
 584                 errno = ENOTCONN;
 585                 return -1;
 586
 587         case SYN_SENT:
 588         case SYN_RECEIVED:
 589         case ESTABLISHED:
 590         case CLOSE_WAIT:
 591                 break;
 592
 593         case FIN_WAIT_1:
 594         case FIN_WAIT_2:
 595         case CLOSING:
 596         case LAST_ACK:
 597         case TIME_WAIT:
 598                 debug("Error: send() called on closing connection %p\n", c);
 599                 errno = EPIPE;
 600                 return -1;
 601         }
 602
 603         // Exit early if we have nothing to send.
 604
 605         if(!len) {
 606                 return 0;
 607         }
 608
 609         if(!data) {
 610                 errno = EFAULT;
 611                 return -1;
 612         }
 613
 614         // Check if we need to be able to buffer all data
 615
 616         if(c->flags & UTCP_NO_PARTIAL) {
 617                 if(len > buffer_free(&c->sndbuf)) {
 618                         if(len > c->sndbuf.maxsize) {
 619                                 errno = EMSGSIZE;
 620                                 return -1;
 621                         } else {
 622                                 errno = EWOULDBLOCK;
 623                                 return 0;
 624                         }
 625                 }
 626         }
 627
 628         // Add data to send buffer.
 629
 630         if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
 631                 len = buffer_put(&c->sndbuf, data, len);
 632         }
 633
 634         if(len <= 0) {
 635                 if(is_reliable(c)) {
 636                         errno = EWOULDBLOCK;
 637                         return 0;
 638                 } else {
 639                         return len;
 640                 }
 641         }
 642
 643         c->snd.last += len;
 644
 645         // Don't send anything yet if the connection has not fully established yet
 646
 647         if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
 648                 return len;
 649         }
 650
 651         ack(c, false);
 652
 653         if(!is_reliable(c)) {
 654                 c->snd.una = c->snd.nxt = c->snd.last;
 655                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 656         }
 657
 658         if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) {
 659                 start_retransmit_timer(c);
 660         }
 661
 662         if(is_reliable(c) && !timerisset(&c->conn_timeout)) {
 663                 gettimeofday(&c->conn_timeout, NULL);
 664                 c->conn_timeout.tv_sec += c->utcp->timeout;
 665         }
 666
 667         return len;
 668 }
 669
 670 static void swap_ports(struct hdr *hdr) {
 671         uint16_t tmp = hdr->src;
 672         hdr->src = hdr->dst;
 673         hdr->dst = tmp;
 674 }
 675
 676 static void retransmit(struct utcp_connection *c) {
 677         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 678                 debug("Retransmit() called but nothing to retransmit!\n");
 679                 stop_retransmit_timer(c);
 680                 return;
 681         }
 682
 683         struct utcp *utcp = c->utcp;
 684
 685         struct {
 686                 struct hdr hdr;
 687                 uint8_t data[];
 688         } *pkt;
 689
 690         pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
 691
 692         if(!pkt) {
 693                 return;
 694         }
 695
 696         pkt->hdr.src = c->src;
 697         pkt->hdr.dst = c->dst;
 698         pkt->hdr.wnd = c->rcv.wnd;
 699         pkt->hdr.aux = 0;
 700
 701         switch(c->state) {
 702         case SYN_SENT:
 703                 // Send our SYN again
 704                 pkt->hdr.seq = c->snd.iss;
 705                 pkt->hdr.ack = 0;
 706                 pkt->hdr.ctl = SYN;
 707                 pkt->hdr.aux = 0x0101;
 708                 pkt->data[0] = 1;
 709                 pkt->data[1] = 0;
 710                 pkt->data[2] = 0;
 711                 pkt->data[3] = c->flags & 0x7;
 712                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + 4);
 713                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
 714                 break;
 715
 716         case SYN_RECEIVED:
 717                 // Send SYNACK again
 718                 pkt->hdr.seq = c->snd.nxt;
 719                 pkt->hdr.ack = c->rcv.nxt;
 720                 pkt->hdr.ctl = SYN | ACK;
 721                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr));
 722                 utcp->send(utcp, pkt, sizeof(pkt->hdr));
 723                 break;
 724
 725         case ESTABLISHED:
 726         case FIN_WAIT_1:
 727         case CLOSE_WAIT:
 728         case CLOSING:
 729         case LAST_ACK:
 730                 // Send unacked data again.
 731                 pkt->hdr.seq = c->snd.una;
 732                 pkt->hdr.ack = c->rcv.nxt;
 733                 pkt->hdr.ctl = ACK;
 734                 uint32_t len = seqdiff(c->snd.last, c->snd.una);
 735
 736                 if(len > utcp->mtu) {
 737                         len = utcp->mtu;
 738                 }
 739
 740                 if(fin_wanted(c, c->snd.una + len)) {
 741                         len--;
 742                         pkt->hdr.ctl |= FIN;
 743                 }
 744
 745                 c->snd.nxt = c->snd.una + len;
 746                 c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 747                 buffer_copy(&c->sndbuf, pkt->data, 0, len);
 748                 print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
 749                 utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
 750                 break;
 751
 752         case CLOSED:
 753         case LISTEN:
 754         case TIME_WAIT:
 755         case FIN_WAIT_2:
 756                 // We shouldn't need to retransmit anything in this state.
 757 #ifdef UTCP_DEBUG
 758                 abort();
 759 #endif
 760                 stop_retransmit_timer(c);
 761                 goto cleanup;
 762         }
 763
 764         start_retransmit_timer(c);
 765         utcp->rto *= 2;
 766
 767         if(utcp->rto > MAX_RTO) {
 768                 utcp->rto = MAX_RTO;
 769         }
 770
 771         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 772
 773 cleanup:
 774         free(pkt);
 775 }
 776
 777 /* Update receive buffer and SACK entries after consuming data.
 778  *
 779  * Situation:
 780  *
 781  * |.....0000..1111111111.....22222......3333|
 782  * |---------------^
 783  *
 784  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 785  * to remove data from the receive buffer. The idea is to substract "len"
 786  * from the offset of all the SACK entries, and then remove/cut down entries
 787  * that are shifted to before the start of the receive buffer.
 788  *
 789  * There are three cases:
 790  * - the SACK entry is after ^, in that case just change the offset.
 791  * - the SACK entry starts before and ends after ^, so we have to
 792  *   change both its offset and size.
 793  * - the SACK entry is completely before ^, in that case delete it.
 794  */
 795 static void sack_consume(struct utcp_connection *c, size_t len) {
 796         debug("sack_consume %lu\n", (unsigned long)len);
 797
 798         if(len > c->rcvbuf.used) {
 799                 debug("All SACK entries consumed");
 800                 c->sacks[0].len = 0;
 801                 return;
 802         }
 803
 804         buffer_get(&c->rcvbuf, NULL, len);
 805
 806         for(int i = 0; i < NSACKS && c->sacks[i].len;) {
 807                 if(len < c->sacks[i].offset) {
 808                         c->sacks[i].offset -= len;
 809                         i++;
 810                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 811                         c->sacks[i].len -= len - c->sacks[i].offset;
 812                         c->sacks[i].offset = 0;
 813                         i++;
 814                 } else {
 815                         if(i < NSACKS - 1) {
 816                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof(c->sacks)[i]);
 817                                 c->sacks[NSACKS - 1].len = 0;
 818                         } else {
 819                                 c->sacks[i].len = 0;
 820                                 break;
 821                         }
 822                 }
 823         }
 824
 825         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 826                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 827         }
 828 }
 829
 830 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 831         debug("out of order packet, offset %u\n", offset);
 832         // Packet loss or reordering occured. Store the data in the buffer.
 833         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 834
 835         if(rxd < 0 || (size_t)rxd < len) {
 836                 abort();
 837         }
 838
 839         // Make note of where we put it.
 840         for(int i = 0; i < NSACKS; i++) {
 841                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 842                         debug("New SACK entry %d\n", i);
 843                         c->sacks[i].offset = offset;
 844                         c->sacks[i].len = rxd;
 845                         break;
 846                 } else if(offset < c->sacks[i].offset) {
 847                         if(offset + rxd < c->sacks[i].offset) { // insert before
 848                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 849                                         debug("Insert SACK entry at %d\n", i);
 850                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof(c->sacks)[i]);
 851                                         c->sacks[i].offset = offset;
 852                                         c->sacks[i].len = rxd;
 853                                 } else {
 854                                         debug("SACK entries full, dropping packet\n");
 855                                 }
 856
 857                                 break;
 858                         } else { // merge
 859                                 debug("Merge with start of SACK entry at %d\n", i);
 860                                 c->sacks[i].offset = offset;
 861                                 break;
 862                         }
 863                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 864                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 865                                 debug("Merge with end of SACK entry at %d\n", i);
 866                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 867                                 // TODO: handle potential merge with next entry
 868                         }
 869
 870                         break;
 871                 }
 872         }
 873
 874         for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
 875                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 876         }
 877 }
 878
 879 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 880         // Check if we can process out-of-order data now.
 881         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 882                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 883                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 884                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 885                 data = c->rcvbuf.data;
 886         }
 887
 888         if(c->recv) {
 889                 ssize_t rxd = c->recv(c, data, len);
 890
 891                 if(rxd < 0 || (size_t)rxd != len) {
 892                         // TODO: handle the application not accepting all data.
 893                         abort();
 894                 }
 895         }
 896
 897         if(c->rcvbuf.used) {
 898                 sack_consume(c, len);
 899         }
 900
 901         c->rcv.nxt += len;
 902 }
 903
 904
 905 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 906         if(!is_reliable(c)) {
 907                 c->recv(c, data, len);
 908                 c->rcv.nxt = seq + len;
 909                 return;
 910         }
 911
 912         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 913
 914         if(offset + len > c->rcvbuf.maxsize) {
 915                 abort();
 916         }
 917
 918         if(offset) {
 919                 handle_out_of_order(c, offset, data, len);
 920         } else {
 921                 handle_in_order(c, data, len);
 922         }
 923 }
 924
 925
 926 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 927         const uint8_t *ptr = data;
 928
 929         if(!utcp) {
 930                 errno = EFAULT;
 931                 return -1;
 932         }
 933
 934         if(!len) {
 935                 return 0;
 936         }
 937
 938         if(!data) {
 939                 errno = EFAULT;
 940                 return -1;
 941         }
 942
 943         print_packet(utcp, "recv", data, len);
 944
 945         // Drop packets smaller than the header
 946
 947         struct hdr hdr;
 948
 949         if(len < sizeof(hdr)) {
 950                 errno = EBADMSG;
 951                 return -1;
 952         }
 953
 954         // Make a copy from the potentially unaligned data to a struct hdr
 955
 956         memcpy(&hdr, ptr, sizeof(hdr));
 957         ptr += sizeof(hdr);
 958         len -= sizeof(hdr);
 959
 960         // Drop packets with an unknown CTL flag
 961
 962         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 963                 errno = EBADMSG;
 964                 return -1;
 965         }
 966
 967         // Check for auxiliary headers
 968
 969         const uint8_t *init = NULL;
 970
 971         uint16_t aux = hdr.aux;
 972
 973         while(aux) {
 974                 size_t auxlen = 4 * (aux >> 8) & 0xf;
 975                 uint8_t auxtype = aux & 0xff;
 976
 977                 if(len < auxlen) {
 978                         errno = EBADMSG;
 979                         return -1;
 980                 }
 981
 982                 switch(auxtype) {
 983                 case AUX_INIT:
 984                         if(!(hdr.ctl & SYN) || auxlen != 4) {
 985                                 errno = EBADMSG;
 986                                 return -1;
 987                         }
 988
 989                         init = ptr;
 990                         break;
 991
 992                 default:
 993                         errno = EBADMSG;
 994                         return -1;
 995                 }
 996
 997                 len -= auxlen;
 998                 ptr += auxlen;
 999
1000                 if(!(aux & 0x800)) {
1001                         break;
1002                 }
1003
1004                 if(len < 2) {
1005                         errno = EBADMSG;
1006                         return -1;
1007                 }
1008
1009                 memcpy(&aux, ptr, 2);
1010                 len -= 2;
1011                 ptr += 2;
1012         }
1013
1014         bool has_data = len || (hdr.ctl & (SYN | FIN));
1015
1016         // Try to match the packet to an existing connection
1017
1018         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
1019
1020         // Is it for a new connection?
1021
1022         if(!c) {
1023                 // Ignore RST packets
1024
1025                 if(hdr.ctl & RST) {
1026                         return 0;
1027                 }
1028
1029                 // Is it a SYN packet and are we LISTENing?
1030
1031                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
1032                         // If we don't want to accept it, send a RST back
1033                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
1034                                 len = 1;
1035                                 goto reset;
1036                         }
1037
1038                         // Try to allocate memory, otherwise send a RST back
1039                         c = allocate_connection(utcp, hdr.dst, hdr.src);
1040
1041                         if(!c) {
1042                                 len = 1;
1043                                 goto reset;
1044                         }
1045
1046                         // Parse auxilliary information
1047                         if(init) {
1048                                 if(init[0] < 1) {
1049                                         len = 1;
1050                                         goto reset;
1051                                 }
1052
1053                                 c->flags = init[3] & 0x7;
1054                         } else {
1055                                 c->flags = UTCP_TCP;
1056                         }
1057
1058 synack:
1059                         // Return SYN+ACK, go to SYN_RECEIVED state
1060                         c->snd.wnd = hdr.wnd;
1061                         c->rcv.irs = hdr.seq;
1062                         c->rcv.nxt = c->rcv.irs + 1;
1063                         set_state(c, SYN_RECEIVED);
1064
1065                         struct {
1066                                 struct hdr hdr;
1067                                 uint8_t data[4];
1068                         } pkt;
1069
1070                         pkt.hdr.src = c->src;
1071                         pkt.hdr.dst = c->dst;
1072                         pkt.hdr.ack = c->rcv.irs + 1;
1073                         pkt.hdr.seq = c->snd.iss;
1074                         pkt.hdr.wnd = c->rcv.wnd;
1075                         pkt.hdr.ctl = SYN | ACK;
1076
1077                         if(init) {
1078                                 pkt.hdr.aux = 0x0101;
1079                                 pkt.data[0] = 1;
1080                                 pkt.data[1] = 0;
1081                                 pkt.data[2] = 0;
1082                                 pkt.data[3] = c->flags & 0x7;
1083                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr) + 4);
1084                                 utcp->send(utcp, &pkt, sizeof(hdr) + 4);
1085                         } else {
1086                                 pkt.hdr.aux = 0;
1087                                 print_packet(c->utcp, "send", &pkt, sizeof(hdr));
1088                                 utcp->send(utcp, &pkt, sizeof(hdr));
1089                         }
1090                 } else {
1091                         // No, we don't want your packets, send a RST back
1092                         len = 1;
1093                         goto reset;
1094                 }
1095
1096                 return 0;
1097         }
1098
1099         debug("%p state %s\n", c->utcp, strstate[c->state]);
1100
1101         // In case this is for a CLOSED connection, ignore the packet.
1102         // TODO: make it so incoming packets can never match a CLOSED connection.
1103
1104         if(c->state == CLOSED) {
1105                 debug("Got packet for closed connection\n");
1106                 return 0;
1107         }
1108
1109         // It is for an existing connection.
1110
1111         // 1. Drop invalid packets.
1112
1113         // 1a. Drop packets that should not happen in our current state.
1114
1115         switch(c->state) {
1116         case SYN_SENT:
1117         case SYN_RECEIVED:
1118         case ESTABLISHED:
1119         case FIN_WAIT_1:
1120         case FIN_WAIT_2:
1121         case CLOSE_WAIT:
1122         case CLOSING:
1123         case LAST_ACK:
1124         case TIME_WAIT:
1125                 break;
1126
1127         default:
1128 #ifdef UTCP_DEBUG
1129                 abort();
1130 #endif
1131                 break;
1132         }
1133
1134         // 1b. Discard data that is not in our receive window.
1135
1136         if(is_reliable(c)) {
1137                 bool acceptable;
1138
1139                 if(c->state == SYN_SENT) {
1140                         acceptable = true;
1141                 } else if(len == 0) {
1142                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
1143                 } else {
1144                         int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
1145
1146                         // cut already accepted front overlapping
1147                         if(rcv_offset < 0) {
1148                                 acceptable = len > (size_t) - rcv_offset;
1149
1150                                 if(acceptable) {
1151                                         ptr -= rcv_offset;
1152                                         len += rcv_offset;
1153                                         hdr.seq -= rcv_offset;
1154                                 }
1155                         } else {
1156                                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
1157                         }
1158                 }
1159
1160                 if(!acceptable) {
1161                         debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
1162
1163                         // Ignore unacceptable RST packets.
1164                         if(hdr.ctl & RST) {
1165                                 return 0;
1166                         }
1167
1168                         // Otherwise, continue processing.
1169                         len = 0;
1170                 }
1171         }
1172
1173         c->snd.wnd = hdr.wnd; // TODO: move below
1174
1175         // 1c. Drop packets with an invalid ACK.
1176         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
1177         // (= snd.una + c->sndbuf.used).
1178
1179         if(!is_reliable(c)) {
1180                 if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
1181                         hdr.ack = c->snd.una;
1182                 }
1183         }
1184
1185         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
1186                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
1187
1188                 // Ignore unacceptable RST packets.
1189                 if(hdr.ctl & RST) {
1190                         return 0;
1191                 }
1192
1193                 goto reset;
1194         }
1195
1196         // 2. Handle RST packets
1197
1198         if(hdr.ctl & RST) {
1199                 switch(c->state) {
1200                 case SYN_SENT:
1201                         if(!(hdr.ctl & ACK)) {
1202                                 return 0;
1203                         }
1204
1205                         // The peer has refused our connection.
1206                         set_state(c, CLOSED);
1207                         errno = ECONNREFUSED;
1208
1209                         if(c->recv) {
1210                                 c->recv(c, NULL, 0);
1211                         }
1212
1213                         if(c->poll && !c->reapable) {
1214                                 c->poll(c, 0);
1215                         }
1216
1217                         return 0;
1218
1219                 case SYN_RECEIVED:
1220                         if(hdr.ctl & ACK) {
1221                                 return 0;
1222                         }
1223
1224                         // We haven't told the application about this connection yet. Silently delete.
1225                         free_connection(c);
1226                         return 0;
1227
1228                 case ESTABLISHED:
1229                 case FIN_WAIT_1:
1230                 case FIN_WAIT_2:
1231                 case CLOSE_WAIT:
1232                         if(hdr.ctl & ACK) {
1233                                 return 0;
1234                         }
1235
1236                         // The peer has aborted our connection.
1237                         set_state(c, CLOSED);
1238                         errno = ECONNRESET;
1239
1240                         if(c->recv) {
1241                                 c->recv(c, NULL, 0);
1242                         }
1243
1244                         if(c->poll && !c->reapable) {
1245                                 c->poll(c, 0);
1246                         }
1247
1248                         return 0;
1249
1250                 case CLOSING:
1251                 case LAST_ACK:
1252                 case TIME_WAIT:
1253                         if(hdr.ctl & ACK) {
1254                                 return 0;
1255                         }
1256
1257                         // As far as the application is concerned, the connection has already been closed.
1258                         // If it has called utcp_close() already, we can immediately free this connection.
1259                         if(c->reapable) {
1260                                 free_connection(c);
1261                                 return 0;
1262                         }
1263
1264                         // Otherwise, immediately move to the CLOSED state.
1265                         set_state(c, CLOSED);
1266                         return 0;
1267
1268                 default:
1269 #ifdef UTCP_DEBUG
1270                         abort();
1271 #endif
1272                         break;
1273                 }
1274         }
1275
1276         uint32_t advanced;
1277
1278         if(!(hdr.ctl & ACK)) {
1279                 advanced = 0;
1280                 goto skip_ack;
1281         }
1282
1283         // 3. Advance snd.una
1284
1285         advanced = seqdiff(hdr.ack, c->snd.una);
1286
1287         if(advanced) {
1288                 // RTT measurement
1289                 if(c->rtt_start.tv_sec) {
1290                         if(c->rtt_seq == hdr.ack) {
1291                                 struct timeval now, diff;
1292                                 gettimeofday(&now, NULL);
1293                                 timersub(&now, &c->rtt_start, &diff);
1294                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1295                                 c->rtt_start.tv_sec = 0;
1296                         } else if(c->rtt_seq < hdr.ack) {
1297                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1298                                 c->rtt_start.tv_sec = 0;
1299                         }
1300                 }
1301
1302                 int32_t data_acked = advanced;
1303
1304                 switch(c->state) {
1305                 case SYN_SENT:
1306                 case SYN_RECEIVED:
1307                         data_acked--;
1308                         break;
1309
1310                 // TODO: handle FIN as well.
1311                 default:
1312                         break;
1313                 }
1314
1315                 assert(data_acked >= 0);
1316
1317 #ifndef NDEBUG
1318                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1319                 assert(data_acked <= bufused);
1320 #endif
1321
1322                 if(data_acked) {
1323                         buffer_get(&c->sndbuf, NULL, data_acked);
1324                 }
1325
1326                 // Also advance snd.nxt if possible
1327                 if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
1328                         c->snd.nxt = hdr.ack;
1329                 }
1330
1331                 c->snd.una = hdr.ack;
1332
1333                 c->dupack = 0;
1334                 c->snd.cwnd += utcp->mtu;
1335
1336                 if(c->snd.cwnd > c->sndbuf.maxsize) {
1337                         c->snd.cwnd = c->sndbuf.maxsize;
1338                 }
1339
1340                 // Check if we have sent a FIN that is now ACKed.
1341                 switch(c->state) {
1342                 case FIN_WAIT_1:
1343                         if(c->snd.una == c->snd.last) {
1344                                 set_state(c, FIN_WAIT_2);
1345                         }
1346
1347                         break;
1348
1349                 case CLOSING:
1350                         if(c->snd.una == c->snd.last) {
1351                                 gettimeofday(&c->conn_timeout, NULL);
1352                                 c->conn_timeout.tv_sec += utcp->timeout;
1353                                 set_state(c, TIME_WAIT);
1354                         }
1355
1356                         break;
1357
1358                 default:
1359                         break;
1360                 }
1361         } else {
1362                 if(!len && is_reliable(c)) {
1363                         c->dupack++;
1364
1365                         if(c->dupack == 3) {
1366                                 debug("Triplicate ACK\n");
1367                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1368                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1369                                 //Reset the congestion window so we wait for ACKs.
1370                                 c->snd.nxt = c->snd.una;
1371                                 c->snd.cwnd = utcp->mtu;
1372                                 start_retransmit_timer(c);
1373                         }
1374                 }
1375         }
1376
1377         // 4. Update timers
1378
1379         if(advanced) {
1380                 if(c->snd.una == c->snd.last) {
1381                         stop_retransmit_timer(c);
1382                         timerclear(&c->conn_timeout);
1383                 } else if(is_reliable(c)) {
1384                         start_retransmit_timer(c);
1385                         gettimeofday(&c->conn_timeout, NULL);
1386                         c->conn_timeout.tv_sec += utcp->timeout;
1387                 }
1388         }
1389
1390 skip_ack:
1391         // 5. Process SYN stuff
1392
1393         if(hdr.ctl & SYN) {
1394                 switch(c->state) {
1395                 case SYN_SENT:
1396
1397                         // This is a SYNACK. It should always have ACKed the SYN.
1398                         if(!advanced) {
1399                                 goto reset;
1400                         }
1401
1402                         c->rcv.irs = hdr.seq;
1403                         c->rcv.nxt = hdr.seq;
1404
1405                         if(c->shut_wr) {
1406                                 c->snd.last++;
1407                                 set_state(c, FIN_WAIT_1);
1408                         } else {
1409                                 set_state(c, ESTABLISHED);
1410                         }
1411
1412                         // TODO: notify application of this somehow.
1413                         break;
1414
1415                 case SYN_RECEIVED:
1416                         // This is a retransmit of a SYN, send back the SYNACK.
1417                         goto synack;
1418
1419                 case ESTABLISHED:
1420                 case FIN_WAIT_1:
1421                 case FIN_WAIT_2:
1422                 case CLOSE_WAIT:
1423                 case CLOSING:
1424                 case LAST_ACK:
1425                 case TIME_WAIT:
1426                         // Ehm, no. We should never receive a second SYN.
1427                         return 0;
1428
1429                 default:
1430 #ifdef UTCP_DEBUG
1431                         abort();
1432 #endif
1433                         return 0;
1434                 }
1435
1436                 // SYN counts as one sequence number
1437                 c->rcv.nxt++;
1438         }
1439
1440         // 6. Process new data
1441
1442         if(c->state == SYN_RECEIVED) {
1443                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1444                 if(!advanced) {
1445                         goto reset;
1446                 }
1447
1448                 // Are we still LISTENing?
1449                 if(utcp->accept) {
1450                         utcp->accept(c, c->src);
1451                 }
1452
1453                 if(c->state != ESTABLISHED) {
1454                         set_state(c, CLOSED);
1455                         c->reapable = true;
1456                         goto reset;
1457                 }
1458         }
1459
1460         if(len) {
1461                 switch(c->state) {
1462                 case SYN_SENT:
1463                 case SYN_RECEIVED:
1464                         // This should never happen.
1465 #ifdef UTCP_DEBUG
1466                         abort();
1467 #endif
1468                         return 0;
1469
1470                 case ESTABLISHED:
1471                 case FIN_WAIT_1:
1472                 case FIN_WAIT_2:
1473                         break;
1474
1475                 case CLOSE_WAIT:
1476                 case CLOSING:
1477                 case LAST_ACK:
1478                 case TIME_WAIT:
1479                         // Ehm no, We should never receive more data after a FIN.
1480                         goto reset;
1481
1482                 default:
1483 #ifdef UTCP_DEBUG
1484                         abort();
1485 #endif
1486                         return 0;
1487                 }
1488
1489                 handle_incoming_data(c, hdr.seq, ptr, len);
1490         }
1491
1492         // 7. Process FIN stuff
1493
1494         if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
1495                 switch(c->state) {
1496                 case SYN_SENT:
1497                 case SYN_RECEIVED:
1498                         // This should never happen.
1499 #ifdef UTCP_DEBUG
1500                         abort();
1501 #endif
1502                         break;
1503
1504                 case ESTABLISHED:
1505                         set_state(c, CLOSE_WAIT);
1506                         break;
1507
1508                 case FIN_WAIT_1:
1509                         set_state(c, CLOSING);
1510                         break;
1511
1512                 case FIN_WAIT_2:
1513                         gettimeofday(&c->conn_timeout, NULL);
1514                         c->conn_timeout.tv_sec += utcp->timeout;
1515                         set_state(c, TIME_WAIT);
1516                         break;
1517
1518                 case CLOSE_WAIT:
1519                 case CLOSING:
1520                 case LAST_ACK:
1521                 case TIME_WAIT:
1522                         // Ehm, no. We should never receive a second FIN.
1523                         goto reset;
1524
1525                 default:
1526 #ifdef UTCP_DEBUG
1527                         abort();
1528 #endif
1529                         break;
1530                 }
1531
1532                 // FIN counts as one sequence number
1533                 c->rcv.nxt++;
1534                 len++;
1535
1536                 // Inform the application that the peer closed its end of the connection.
1537                 if(c->recv) {
1538                         errno = 0;
1539                         c->recv(c, NULL, 0);
1540                 }
1541         }
1542
1543         // Now we send something back if:
1544         // - we received data, so we have to send back an ACK
1545         //   -> sendatleastone = true
1546         // - or we got an ack, so we should maybe send a bit more data
1547         //   -> sendatleastone = false
1548
1549         if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
1550                 ack(c, has_data);
1551         }
1552
1553         return 0;
1554
1555 reset:
1556         swap_ports(&hdr);
1557         hdr.wnd = 0;
1558         hdr.aux = 0;
1559
1560         if(hdr.ctl & ACK) {
1561                 hdr.seq = hdr.ack;
1562                 hdr.ctl = RST;
1563         } else {
1564                 hdr.ack = hdr.seq + len;
1565                 hdr.seq = 0;
1566                 hdr.ctl = RST | ACK;
1567         }
1568
1569         print_packet(utcp, "send", &hdr, sizeof(hdr));
1570         utcp->send(utcp, &hdr, sizeof(hdr));
1571         return 0;
1572
1573 }
1574
1575 int utcp_shutdown(struct utcp_connection *c, int dir) {
1576         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1577
1578         if(!c) {
1579                 errno = EFAULT;
1580                 return -1;
1581         }
1582
1583         if(c->reapable) {
1584                 debug("Error: shutdown() called on closed connection %p\n", c);
1585                 errno = EBADF;
1586                 return -1;
1587         }
1588
1589         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1590                 errno = EINVAL;
1591                 return -1;
1592         }
1593
1594         // TCP does not have a provision for stopping incoming packets.
1595         // The best we can do is to just ignore them.
1596         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
1597                 c->recv = NULL;
1598         }
1599
1600         // The rest of the code deals with shutting down writes.
1601         if(dir == UTCP_SHUT_RD) {
1602                 return 0;
1603         }
1604
1605         // Only process shutting down writes once.
1606         if(c->shut_wr) {
1607                 return 0;
1608         }
1609
1610         c->shut_wr = true;
1611
1612         switch(c->state) {
1613         case CLOSED:
1614         case LISTEN:
1615                 errno = ENOTCONN;
1616                 return -1;
1617
1618         case SYN_SENT:
1619                 return 0;
1620
1621         case SYN_RECEIVED:
1622         case ESTABLISHED:
1623                 set_state(c, FIN_WAIT_1);
1624                 break;
1625
1626         case FIN_WAIT_1:
1627         case FIN_WAIT_2:
1628                 return 0;
1629
1630         case CLOSE_WAIT:
1631                 set_state(c, CLOSING);
1632                 break;
1633
1634         case CLOSING:
1635         case LAST_ACK:
1636         case TIME_WAIT:
1637                 return 0;
1638         }
1639
1640         c->snd.last++;
1641
1642         ack(c, false);
1643
1644         if(!timerisset(&c->rtrx_timeout)) {
1645                 start_retransmit_timer(c);
1646         }
1647
1648         return 0;
1649 }
1650
1651 static bool reset_connection(struct utcp_connection *c) {
1652         if(!c) {
1653                 errno = EFAULT;
1654                 return false;
1655         }
1656
1657         if(c->reapable) {
1658                 debug("Error: abort() called on closed connection %p\n", c);
1659                 errno = EBADF;
1660                 return false;
1661         }
1662
1663         c->recv = NULL;
1664         c->poll = NULL;
1665
1666         switch(c->state) {
1667         case CLOSED:
1668                 return true;
1669
1670         case LISTEN:
1671         case SYN_SENT:
1672         case CLOSING:
1673         case LAST_ACK:
1674         case TIME_WAIT:
1675                 set_state(c, CLOSED);
1676                 return true;
1677
1678         case SYN_RECEIVED:
1679         case ESTABLISHED:
1680         case FIN_WAIT_1:
1681         case FIN_WAIT_2:
1682         case CLOSE_WAIT:
1683                 set_state(c, CLOSED);
1684                 break;
1685         }
1686
1687         // Send RST
1688
1689         struct hdr hdr;
1690
1691         hdr.src = c->src;
1692         hdr.dst = c->dst;
1693         hdr.seq = c->snd.nxt;
1694         hdr.ack = 0;
1695         hdr.wnd = 0;
1696         hdr.ctl = RST;
1697
1698         print_packet(c->utcp, "send", &hdr, sizeof(hdr));
1699         c->utcp->send(c->utcp, &hdr, sizeof(hdr));
1700         return true;
1701 }
1702
1703 // Closes all the opened connections
1704 void utcp_abort_all_connections(struct utcp *utcp) {
1705         if(!utcp) {
1706                 errno = EINVAL;
1707                 return;
1708         }
1709
1710         for(int i = 0; i < utcp->nconnections; i++) {
1711                 struct utcp_connection *c = utcp->connections[i];
1712
1713                 if(c->reapable || c->state == CLOSED) {
1714                         continue;
1715                 }
1716
1717                 utcp_recv_t old_recv = c->recv;
1718                 utcp_poll_t old_poll = c->poll;
1719
1720                 reset_connection(c);
1721
1722                 if(old_recv) {
1723                         errno = 0;
1724                         old_recv(c, NULL, 0);
1725                 }
1726
1727                 if(old_poll && !c->reapable) {
1728                         errno = 0;
1729                         old_poll(c, 0);
1730                 }
1731         }
1732
1733         return;
1734 }
1735
1736 int utcp_close(struct utcp_connection *c) {
1737         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
1738                 return -1;
1739         }
1740
1741         c->recv = NULL;
1742         c->poll = NULL;
1743         c->reapable = true;
1744         return 0;
1745 }
1746
1747 int utcp_abort(struct utcp_connection *c) {
1748         if(!reset_connection(c)) {
1749                 return -1;
1750         }
1751
1752         c->reapable = true;
1753         return 0;
1754 }
1755
1756 /* Handle timeouts.
1757  * One call to this function will loop through all connections,
1758  * checking if something needs to be resent or not.
1759  * The return value is the time to the next timeout in milliseconds,
1760  * or maybe a negative value if the timeout is infinite.
1761  */
1762 struct timeval utcp_timeout(struct utcp *utcp) {
1763         struct timeval now;
1764         gettimeofday(&now, NULL);
1765         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1766
1767         for(int i = 0; i < utcp->nconnections; i++) {
1768                 struct utcp_connection *c = utcp->connections[i];
1769
1770                 if(!c) {
1771                         continue;
1772                 }
1773
1774                 // delete connections that have been utcp_close()d.
1775                 if(c->state == CLOSED) {
1776                         if(c->reapable) {
1777                                 debug("Reaping %p\n", c);
1778                                 free_connection(c);
1779                                 i--;
1780                         }
1781
1782                         continue;
1783                 }
1784
1785                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1786                         errno = ETIMEDOUT;
1787                         c->state = CLOSED;
1788
1789                         if(c->recv) {
1790                                 c->recv(c, NULL, 0);
1791                         }
1792
1793                         if(c->poll && !c->reapable) {
1794                                 c->poll(c, 0);
1795                         }
1796
1797                         continue;
1798                 }
1799
1800                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1801                         debug("retransmit()\n");
1802                         retransmit(c);
1803                 }
1804
1805                 if(c->poll) {
1806                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1807                                 uint32_t len =  buffer_free(&c->sndbuf);
1808
1809                                 if(len) {
1810                                         c->poll(c, len);
1811                                 }
1812                         } else if(c->state == CLOSED) {
1813                                 c->poll(c, 0);
1814                         }
1815                 }
1816
1817                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) {
1818                         next = c->conn_timeout;
1819                 }
1820
1821                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <)) {
1822                         next = c->rtrx_timeout;
1823                 }
1824         }
1825
1826         struct timeval diff;
1827
1828         timersub(&next, &now, &diff);
1829
1830         return diff;
1831 }
1832
1833 bool utcp_is_active(struct utcp *utcp) {
1834         if(!utcp) {
1835                 return false;
1836         }
1837
1838         for(int i = 0; i < utcp->nconnections; i++)
1839                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
1840                         return true;
1841                 }
1842
1843         return false;
1844 }
1845
1846 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1847         if(!send) {
1848                 errno = EFAULT;
1849                 return NULL;
1850         }
1851
1852         struct utcp *utcp = calloc(1, sizeof(*utcp));
1853
1854         if(!utcp) {
1855                 return NULL;
1856         }
1857
1858         utcp->accept = accept;
1859         utcp->pre_accept = pre_accept;
1860         utcp->send = send;
1861         utcp->priv = priv;
1862         utcp->mtu = DEFAULT_MTU;
1863         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1864         utcp->rto = START_RTO; // usec
1865
1866         return utcp;
1867 }
1868
1869 void utcp_exit(struct utcp *utcp) {
1870         if(!utcp) {
1871                 return;
1872         }
1873
1874         for(int i = 0; i < utcp->nconnections; i++) {
1875                 struct utcp_connection *c = utcp->connections[i];
1876
1877                 if(!c->reapable) {
1878                         if(c->recv) {
1879                                 c->recv(c, NULL, 0);
1880                         }
1881
1882                         if(c->poll && !c->reapable) {
1883                                 c->poll(c, 0);
1884                         }
1885                 }
1886
1887                 buffer_exit(&c->rcvbuf);
1888                 buffer_exit(&c->sndbuf);
1889                 free(c);
1890         }
1891
1892         free(utcp->connections);
1893         free(utcp);
1894 }
1895
1896 uint16_t utcp_get_mtu(struct utcp *utcp) {
1897         return utcp ? utcp->mtu : 0;
1898 }
1899
1900 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1901         // TODO: handle overhead of the header
1902         if(utcp) {
1903                 utcp->mtu = mtu;
1904         }
1905 }
1906
1907 void utcp_reset_timers(struct utcp *utcp) {
1908         if(!utcp) {
1909                 return;
1910         }
1911
1912         struct timeval now, then;
1913
1914         gettimeofday(&now, NULL);
1915
1916         then = now;
1917
1918         then.tv_sec += utcp->timeout;
1919
1920         for(int i = 0; i < utcp->nconnections; i++) {
1921                 struct utcp_connection *c = utcp->connections[i];
1922
1923                 if(c->reapable) {
1924                         continue;
1925                 }
1926
1927                 if(timerisset(&c->rtrx_timeout)) {
1928                         c->rtrx_timeout = now;
1929                 }
1930
1931                 if(timerisset(&c->conn_timeout)) {
1932                         c->conn_timeout = then;
1933                 }
1934
1935                 c->rtt_start.tv_sec = 0;
1936         }
1937
1938         if(utcp->rto > START_RTO) {
1939                 utcp->rto = START_RTO;
1940         }
1941 }
1942
1943 int utcp_get_user_timeout(struct utcp *u) {
1944         return u ? u->timeout : 0;
1945 }
1946
1947 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1948         if(u) {
1949                 u->timeout = timeout;
1950         }
1951 }
1952
1953 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1954         return c ? c->sndbuf.maxsize : 0;
1955 }
1956
1957 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1958         if(!c) {
1959                 return 0;
1960         }
1961
1962         switch(c->state) {
1963         case SYN_SENT:
1964         case SYN_RECEIVED:
1965         case ESTABLISHED:
1966         case CLOSE_WAIT:
1967                 return buffer_free(&c->sndbuf);
1968
1969         default:
1970                 return 0;
1971         }
1972 }
1973
1974 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1975         if(!c) {
1976                 return;
1977         }
1978
1979         c->sndbuf.maxsize = size;
1980
1981         if(c->sndbuf.maxsize != size) {
1982                 c->sndbuf.maxsize = -1;
1983         }
1984 }
1985
1986 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1987         return c ? c->rcvbuf.maxsize : 0;
1988 }
1989
1990 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1991         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1992                 return buffer_free(&c->rcvbuf);
1993         } else {
1994                 return 0;
1995         }
1996 }
1997
1998 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1999         if(!c) {
2000                 return;
2001         }
2002
2003         c->rcvbuf.maxsize = size;
2004
2005         if(c->rcvbuf.maxsize != size) {
2006                 c->rcvbuf.maxsize = -1;
2007         }
2008 }
2009
2010 size_t utcp_get_sendq(struct utcp_connection *c) {
2011         return c->sndbuf.used;
2012 }
2013
2014 size_t utcp_get_recvq(struct utcp_connection *c) {
2015         return c->rcvbuf.used;
2016 }
2017
2018 bool utcp_get_nodelay(struct utcp_connection *c) {
2019         return c ? c->nodelay : false;
2020 }
2021
2022 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
2023         if(c) {
2024                 c->nodelay = nodelay;
2025         }
2026 }
2027
2028 bool utcp_get_keepalive(struct utcp_connection *c) {
2029         return c ? c->keepalive : false;
2030 }
2031
2032 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
2033         if(c) {
2034                 c->keepalive = keepalive;
2035         }
2036 }
2037
2038 size_t utcp_get_outq(struct utcp_connection *c) {
2039         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
2040 }
2041
2042 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
2043         if(c) {
2044                 c->recv = recv;
2045         }
2046 }
2047
2048 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
2049         if(c) {
2050                 c->poll = poll;
2051         }
2052 }
2053
2054 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
2055         if(utcp) {
2056                 utcp->accept = accept;
2057                 utcp->pre_accept = pre_accept;
2058         }
2059 }
2060
2061 void utcp_expect_data(struct utcp_connection *c, bool expect) {
2062         if(!c || c->reapable) {
2063                 return;
2064         }
2065
2066         if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
2067                 return;
2068         }
2069
2070         if(expect) {
2071                 // If we expect data, start the connection timer.
2072                 if(!timerisset(&c->conn_timeout)) {
2073                         gettimeofday(&c->conn_timeout, NULL);
2074                         c->conn_timeout.tv_sec += c->utcp->timeout;
2075                 }
2076         } else {
2077                 // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
2078                 if(c->snd.una == c->snd.last) {
2079                         timerclear(&c->conn_timeout);
2080                 }
2081         }
2082 }
2083
2084 void utcp_offline(struct utcp *utcp, bool offline) {
2085         struct timeval now;
2086         gettimeofday(&now, NULL);
2087
2088         for(int i = 0; i < utcp->nconnections; i++) {
2089                 struct utcp_connection *c = utcp->connections[i];
2090
2091                 if(c->reapable) {
2092                         continue;
2093                 }
2094
2095                 utcp_expect_data(c, offline);
2096
2097                 if(!offline) {
2098                         if(timerisset(&c->rtrx_timeout)) {
2099                                 c->rtrx_timeout = now;
2100                         }
2101
2102                         utcp->connections[i]->rtt_start.tv_sec = 0;
2103                 }
2104         }
2105
2106         if(!offline && utcp->rto > START_RTO) {
2107                 utcp->rto = START_RTO;
2108         }
2109 }