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

   1 /*
   2     utcp.c -- Userspace TCP
   3     Copyright (C) 2014 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) do {\
  49         (r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
  50         (r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
  51         if((r)->tv_usec < 0)\
  52                 (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
  53 } while (0)
  54 #endif
  55
  56 static inline size_t max(size_t a, size_t b) {
  57         return a > b ? a : b;
  58 }
  59
  60 #ifdef UTCP_DEBUG
  61 #include <stdarg.h>
  62
  63 static void debug(const char *format, ...) {
  64         va_list ap;
  65         va_start(ap, format);
  66         vfprintf(stderr, format, ap);
  67         va_end(ap);
  68 }
  69
  70 static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
  71         struct hdr hdr;
  72         if(len < sizeof hdr) {
  73                 debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
  74                 return;
  75         }
  76
  77         memcpy(&hdr, pkt, sizeof hdr);
  78         debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
  79         if(hdr.ctl & SYN)
  80                 debug("SYN");
  81         if(hdr.ctl & RST)
  82                 debug("RST");
  83         if(hdr.ctl & FIN)
  84                 debug("FIN");
  85         if(hdr.ctl & ACK)
  86                 debug("ACK");
  87
  88         if(len > sizeof hdr) {
  89                 uint32_t datalen = len - sizeof hdr;
  90                 const uint8_t *data = (uint8_t *)pkt + sizeof hdr;
  91                 char str[datalen * 2 + 1];
  92                 char *p = str;
  93
  94                 for(uint32_t i = 0; i < datalen; i++) {
  95                         *p++ = "0123456789ABCDEF"[data[i] >> 4];
  96                         *p++ = "0123456789ABCDEF"[data[i] & 15];
  97                 }
  98                 *p = 0;
  99
 100                 debug(" data=%s", str);
 101         }
 102
 103         debug("\n");
 104 }
 105 #else
 106 #define debug(...)
 107 #define print_packet(...)
 108 #endif
 109
 110 static void set_state(struct utcp_connection *c, enum state state) {
 111         c->state = state;
 112         if(state == ESTABLISHED)
 113                 timerclear(&c->conn_timeout);
 114         debug("%p new state: %s\n", c->utcp, strstate[state]);
 115 }
 116
 117 static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
 118         if(seq != c->snd.last)
 119                 return false;
 120         switch(c->state) {
 121         case FIN_WAIT_1:
 122         case CLOSING:
 123         case LAST_ACK:
 124                 return true;
 125         default:
 126                 return false;
 127         }
 128 }
 129
 130 static bool is_reliable(struct utcp_connection *c) {
 131         return c->flags & UTCP_RELIABLE;
 132 }
 133
 134 static inline void list_connections(struct utcp *utcp) {
 135         debug("%p has %d connections:\n", utcp, utcp->nconnections);
 136         for(int i = 0; i < utcp->nconnections; i++)
 137                 debug("  %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
 138 }
 139
 140 static int32_t seqdiff(uint32_t a, uint32_t b) {
 141         return a - b;
 142 }
 143
 144 // Buffer functions
 145 // TODO: convert to ringbuffers to avoid memmove() operations.
 146
 147 // Store data into the buffer
 148 static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
 149         debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
 150
 151         size_t required = offset + len;
 152         if(required > buf->maxsize) {
 153                 if(offset >= buf->maxsize)
 154                         return 0;
 155                 len = buf->maxsize - offset;
 156                 required = buf->maxsize;
 157         }
 158
 159         if(required > buf->size) {
 160                 size_t newsize = buf->size;
 161                 if(!newsize) {
 162                         newsize = required;
 163                 } else {
 164                         do {
 165                                 newsize *= 2;
 166                         } while(newsize < required);
 167                 }
 168                 if(newsize > buf->maxsize)
 169                         newsize = buf->maxsize;
 170                 char *newdata = realloc(buf->data, newsize);
 171                 if(!newdata)
 172                         return -1;
 173                 buf->data = newdata;
 174                 buf->size = newsize;
 175         }
 176
 177         memcpy(buf->data + offset, data, len);
 178         if(required > buf->used)
 179                 buf->used = required;
 180         return len;
 181 }
 182
 183 static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
 184         return buffer_put_at(buf, buf->used, data, len);
 185 }
 186
 187 // Get data from the buffer. data can be NULL.
 188 static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
 189         if(len > buf->used)
 190                 len = buf->used;
 191         if(data)
 192                 memcpy(data, buf->data, len);
 193         if(len < buf->used)
 194                 memmove(buf->data, buf->data + len, buf->used - len);
 195         buf->used -= len;
 196         return len;
 197 }
 198
 199 // Copy data from the buffer without removing it.
 200 static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
 201         if(offset >= buf->used)
 202                 return 0;
 203         if(offset + len > buf->used)
 204                 len = buf->used - offset;
 205         memcpy(data, buf->data + offset, len);
 206         return len;
 207 }
 208
 209 static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
 210         memset(buf, 0, sizeof *buf);
 211         if(len) {
 212                 buf->data = malloc(len);
 213                 if(!buf->data)
 214                         return false;
 215         }
 216         buf->size = len;
 217         buf->maxsize = maxlen;
 218         return true;
 219 }
 220
 221 static void buffer_exit(struct buffer *buf) {
 222         free(buf->data);
 223         memset(buf, 0, sizeof *buf);
 224 }
 225
 226 static uint32_t buffer_free(const struct buffer *buf) {
 227         return buf->maxsize - buf->used;
 228 }
 229
 230 // Connections are stored in a sorted list.
 231 // This gives O(log(N)) lookup time, O(N log(N)) insertion time and O(N) deletion time.
 232
 233 static int compare(const void *va, const void *vb) {
 234         assert(va && vb);
 235
 236         const struct utcp_connection *a = *(struct utcp_connection **)va;
 237         const struct utcp_connection *b = *(struct utcp_connection **)vb;
 238
 239         assert(a && b);
 240         assert(a->src && b->src);
 241
 242         int c = (int)a->src - (int)b->src;
 243         if(c)
 244                 return c;
 245         c = (int)a->dst - (int)b->dst;
 246         return c;
 247 }
 248
 249 static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
 250         if(!utcp->nconnections)
 251                 return NULL;
 252         struct utcp_connection key = {
 253                 .src = src,
 254                 .dst = dst,
 255         }, *keyp = &key;
 256         struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 257         return match ? *match : NULL;
 258 }
 259
 260 static void free_connection(struct utcp_connection *c) {
 261         struct utcp *utcp = c->utcp;
 262         struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 263
 264         assert(cp);
 265
 266         int i = cp - utcp->connections;
 267         memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
 268         utcp->nconnections--;
 269
 270         buffer_exit(&c->rcvbuf);
 271         buffer_exit(&c->sndbuf);
 272         free(c);
 273 }
 274
 275 static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t src, uint16_t dst) {
 276         // Check whether this combination of src and dst is free
 277
 278         if(src) {
 279                 if(find_connection(utcp, src, dst)) {
 280                         errno = EADDRINUSE;
 281                         return NULL;
 282                 }
 283         } else { // If src == 0, generate a random port number with the high bit set
 284                 if(utcp->nconnections >= 32767) {
 285                         errno = ENOMEM;
 286                         return NULL;
 287                 }
 288                 src = rand() | 0x8000;
 289                 while(find_connection(utcp, src, dst))
 290                         src++;
 291         }
 292
 293         // Allocate memory for the new connection
 294
 295         if(utcp->nconnections >= utcp->nallocated) {
 296                 if(!utcp->nallocated)
 297                         utcp->nallocated = 4;
 298                 else
 299                         utcp->nallocated *= 2;
 300                 struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
 301                 if(!new_array)
 302                         return NULL;
 303                 utcp->connections = new_array;
 304         }
 305
 306         struct utcp_connection *c = calloc(1, sizeof *c);
 307         if(!c)
 308                 return NULL;
 309
 310         if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
 311                 free(c);
 312                 return NULL;
 313         }
 314
 315         if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
 316                 buffer_exit(&c->sndbuf);
 317                 free(c);
 318                 return NULL;
 319         }
 320
 321         // Fill in the details
 322
 323         c->src = src;
 324         c->dst = dst;
 325 #ifdef UTCP_DEBUG
 326         c->snd.iss = 0;
 327 #else
 328         c->snd.iss = rand();
 329 #endif
 330         c->snd.una = c->snd.iss;
 331         c->snd.nxt = c->snd.iss + 1;
 332         c->rcv.wnd = utcp->mtu;
 333         c->snd.last = c->snd.nxt;
 334         c->snd.cwnd = utcp->mtu;
 335         c->utcp = utcp;
 336
 337         // Add it to the sorted list of connections
 338
 339         utcp->connections[utcp->nconnections++] = c;
 340         qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
 341
 342         return c;
 343 }
 344
 345 // Update RTT variables. See RFC 6298.
 346 static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
 347         if(!rtt) {
 348                 debug("invalid rtt\n");
 349                 return;
 350         }
 351
 352         struct utcp *utcp = c->utcp;
 353
 354         if(!utcp->srtt) {
 355                 utcp->srtt = rtt;
 356                 utcp->rttvar = rtt / 2;
 357                 utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
 358         } else {
 359                 utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
 360                 utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
 361                 utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
 362         }
 363
 364         if(utcp->rto > MAX_RTO)
 365                 utcp->rto = MAX_RTO;
 366
 367         debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
 368 }
 369
 370 static void start_retransmit_timer(struct utcp_connection *c) {
 371         gettimeofday(&c->rtrx_timeout, NULL);
 372         c->rtrx_timeout.tv_usec += c->utcp->rto;
 373         while(c->rtrx_timeout.tv_usec >= 1000000) {
 374                 c->rtrx_timeout.tv_usec -= 1000000;
 375                 c->rtrx_timeout.tv_sec++;
 376         }
 377         debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
 378 }
 379
 380 static void stop_retransmit_timer(struct utcp_connection *c) {
 381         timerclear(&c->rtrx_timeout);
 382         debug("timeout cleared\n");
 383 }
 384
 385 struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
 386         struct utcp_connection *c = allocate_connection(utcp, 0, dst);
 387         if(!c)
 388                 return NULL;
 389
 390         assert((flags & ~0xf) == 0);
 391
 392         c->flags = flags;
 393         c->recv = recv;
 394         c->priv = priv;
 395
 396         struct {
 397                 struct hdr hdr;
 398                 uint8_t init[4];
 399         } pkt;
 400
 401         pkt.hdr.src = c->src;
 402         pkt.hdr.dst = c->dst;
 403         pkt.hdr.seq = c->snd.iss;
 404         pkt.hdr.ack = 0;
 405         pkt.hdr.wnd = c->rcv.wnd;
 406         pkt.hdr.ctl = SYN;
 407         pkt.hdr.aux = 0x0101;
 408         pkt.init[0] = 1;
 409         pkt.init[1] = 0;
 410         pkt.init[2] = 0;
 411         pkt.init[3] = flags & 0x7;
 412
 413         set_state(c, SYN_SENT);
 414
 415         print_packet(utcp, "send", &pkt, sizeof pkt);
 416         utcp->send(utcp, &pkt, sizeof pkt);
 417
 418         gettimeofday(&c->conn_timeout, NULL);
 419         c->conn_timeout.tv_sec += utcp->timeout;
 420
 421         start_retransmit_timer(c);
 422
 423         return c;
 424 }
 425
 426 struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
 427         return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
 428 }
 429
 430 void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
 431         if(c->reapable || c->state != SYN_RECEIVED) {
 432                 debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
 433                 return;
 434         }
 435
 436         debug("%p accepted, %p %p\n", c, recv, priv);
 437         c->recv = recv;
 438         c->priv = priv;
 439         set_state(c, ESTABLISHED);
 440 }
 441
 442 static void ack(struct utcp_connection *c, bool sendatleastone) {
 443         int32_t left = seqdiff(c->snd.last, c->snd.nxt);
 444         int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
 445         debug("cwndleft = %d\n", cwndleft);
 446
 447         assert(left >= 0);
 448
 449         if(cwndleft <= 0)
 450                 cwndleft = 0;
 451
 452         if(cwndleft < left)
 453                 left = cwndleft;
 454
 455         if(!left && !sendatleastone)
 456                 return;
 457
 458         struct {
 459                 struct hdr hdr;
 460                 char data[];
 461         } *pkt;
 462
 463         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 464         if(!pkt)
 465                 return;
 466
 467         pkt->hdr.src = c->src;
 468         pkt->hdr.dst = c->dst;
 469         pkt->hdr.ack = c->rcv.nxt;
 470         pkt->hdr.wnd = c->snd.wnd;
 471         pkt->hdr.ctl = ACK;
 472         pkt->hdr.aux = 0;
 473
 474         do {
 475                 uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
 476                 pkt->hdr.seq = c->snd.nxt;
 477
 478                 buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
 479
 480                 c->snd.nxt += seglen;
 481                 left -= seglen;
 482
 483                 if(seglen && fin_wanted(c, c->snd.nxt)) {
 484                         seglen--;
 485                         pkt->hdr.ctl |= FIN;
 486                 }
 487
 488                 if(!c->rtt_start.tv_sec) {
 489                         // Start RTT measurement
 490                         gettimeofday(&c->rtt_start, NULL);
 491                         c->rtt_seq = pkt->hdr.seq + seglen;
 492                         debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
 493                 }
 494
 495                 print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
 496                 c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
 497         } while(left);
 498
 499         free(pkt);
 500 }
 501
 502 ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
 503         if(c->reapable) {
 504                 debug("Error: send() called on closed connection %p\n", c);
 505                 errno = EBADF;
 506                 return -1;
 507         }
 508
 509         switch(c->state) {
 510         case CLOSED:
 511         case LISTEN:
 512         case SYN_SENT:
 513         case SYN_RECEIVED:
 514                 debug("Error: send() called on unconnected connection %p\n", c);
 515                 errno = ENOTCONN;
 516                 return -1;
 517         case ESTABLISHED:
 518         case CLOSE_WAIT:
 519                 break;
 520         case FIN_WAIT_1:
 521         case FIN_WAIT_2:
 522         case CLOSING:
 523         case LAST_ACK:
 524         case TIME_WAIT:
 525                 debug("Error: send() called on closing connection %p\n", c);
 526                 errno = EPIPE;
 527                 return -1;
 528         }
 529
 530         // Exit early if we have nothing to send.
 531
 532         if(!len)
 533                 return 0;
 534
 535         if(!data) {
 536                 errno = EFAULT;
 537                 return -1;
 538         }
 539
 540         // Add data to send buffer.
 541
 542         len = buffer_put(&c->sndbuf, data, len);
 543         if(len <= 0) {
 544                 errno = EWOULDBLOCK;
 545                 return 0;
 546         }
 547
 548         c->snd.last += len;
 549         ack(c, false);
 550         if(!is_reliable(c)) {
 551                 c->snd.una = c->snd.nxt = c->snd.last;
 552                 buffer_get(&c->sndbuf, NULL, c->sndbuf.used);
 553         }
 554         if(is_reliable(c) && !timerisset(&c->rtrx_timeout))
 555                 start_retransmit_timer(c);
 556         return len;
 557 }
 558
 559 static void swap_ports(struct hdr *hdr) {
 560         uint16_t tmp = hdr->src;
 561         hdr->src = hdr->dst;
 562         hdr->dst = tmp;
 563 }
 564
 565 static void retransmit(struct utcp_connection *c) {
 566         if(c->state == CLOSED || c->snd.last == c->snd.una) {
 567                 debug("Retransmit() called but nothing to retransmit!\n");
 568                 stop_retransmit_timer(c);
 569                 return;
 570         }
 571
 572         struct utcp *utcp = c->utcp;
 573
 574         struct {
 575                 struct hdr hdr;
 576                 char data[];
 577         } *pkt;
 578
 579         pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
 580         if(!pkt)
 581                 return;
 582
 583         pkt->hdr.src = c->src;
 584         pkt->hdr.dst = c->dst;
 585         pkt->hdr.wnd = c->rcv.wnd;
 586         pkt->hdr.aux = 0;
 587
 588         switch(c->state) {
 589                 case SYN_SENT:
 590                         fprintf(stderr, "Retransmitting SYN\n");
 591                         // Send our SYN again
 592                         pkt->hdr.seq = c->snd.iss;
 593                         pkt->hdr.ack = 0;
 594                         pkt->hdr.ctl = SYN;
 595                         pkt->hdr.aux = 0x0101;
 596                         pkt->data[0] = 1;
 597                         pkt->data[1] = 0;
 598                         pkt->data[2] = 0;
 599                         pkt->data[3] = c->flags & 0x7;
 600                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + 4);
 601                         utcp->send(utcp, pkt, sizeof pkt->hdr + 4);
 602                         break;
 603
 604                 case SYN_RECEIVED:
 605                         // Send SYNACK again
 606                         pkt->hdr.seq = c->snd.nxt;
 607                         pkt->hdr.ack = c->rcv.nxt;
 608                         pkt->hdr.ctl = SYN | ACK;
 609                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
 610                         utcp->send(utcp, pkt, sizeof pkt->hdr);
 611                         break;
 612
 613                 case ESTABLISHED:
 614                 case FIN_WAIT_1:
 615                 case CLOSE_WAIT:
 616                 case CLOSING:
 617                 case LAST_ACK:
 618                         // Send unacked data again.
 619                         pkt->hdr.seq = c->snd.una;
 620                         pkt->hdr.ack = c->rcv.nxt;
 621                         pkt->hdr.ctl = ACK;
 622                         uint32_t len = seqdiff(c->snd.last, c->snd.una);
 623                         if(len > utcp->mtu)
 624                                 len = utcp->mtu;
 625                         if(fin_wanted(c, c->snd.una + len)) {
 626                                 len--;
 627                                 pkt->hdr.ctl |= FIN;
 628                         }
 629                         c->snd.nxt = c->snd.una + len;
 630                         c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
 631                         buffer_copy(&c->sndbuf, pkt->data, 0, len);
 632                         print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr + len);
 633                         utcp->send(utcp, pkt, sizeof pkt->hdr + len);
 634                         break;
 635
 636                 case CLOSED:
 637                 case LISTEN:
 638                 case TIME_WAIT:
 639                 case FIN_WAIT_2:
 640                         // We shouldn't need to retransmit anything in this state.
 641 #ifdef UTCP_DEBUG
 642                         abort();
 643 #endif
 644                         stop_retransmit_timer(c);
 645                         goto cleanup;
 646         }
 647
 648         start_retransmit_timer(c);
 649         utcp->rto *= 2;
 650         if(utcp->rto > MAX_RTO)
 651                 utcp->rto = MAX_RTO;
 652         c->rtt_start.tv_sec = 0; // invalidate RTT timer
 653
 654 cleanup:
 655         free(pkt);
 656 }
 657
 658 /* Update receive buffer and SACK entries after consuming data.
 659  *
 660  * Situation:
 661  *
 662  * |.....0000..1111111111.....22222......3333|
 663  * |---------------^
 664  *
 665  * 0..3 represent the SACK entries. The ^ indicates up to which point we want
 666  * to remove data from the receive buffer. The idea is to substract "len"
 667  * from the offset of all the SACK entries, and then remove/cut down entries
 668  * that are shifted to before the start of the receive buffer.
 669  *
 670  * There are three cases:
 671  * - the SACK entry is after ^, in that case just change the offset.
 672  * - the SACK entry starts before and ends after ^, so we have to
 673  *   change both its offset and size.
 674  * - the SACK entry is completely before ^, in that case delete it.
 675  */
 676 static void sack_consume(struct utcp_connection *c, size_t len) {
 677         debug("sack_consume %lu\n", (unsigned long)len);
 678         if(len > c->rcvbuf.used) {
 679                 debug("All SACK entries consumed");
 680                 c->sacks[0].len = 0;
 681                 return;
 682         }
 683
 684         buffer_get(&c->rcvbuf, NULL, len);
 685
 686         for(int i = 0; i < NSACKS && c->sacks[i].len; ) {
 687                 if(len < c->sacks[i].offset) {
 688                         c->sacks[i].offset -= len;
 689                         i++;
 690                 } else if(len < c->sacks[i].offset + c->sacks[i].len) {
 691                         c->sacks[i].len -= len - c->sacks[i].offset;
 692                         c->sacks[i].offset = 0;
 693                         i++;
 694                 } else {
 695                         if(i < NSACKS - 1) {
 696                                 memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
 697                                 c->sacks[NSACKS - 1].len = 0;
 698                         } else {
 699                                 c->sacks[i].len = 0;
 700                                 break;
 701                         }
 702                 }
 703         }
 704
 705         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 706                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 707 }
 708
 709 static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
 710         debug("out of order packet, offset %u\n", offset);
 711         // Packet loss or reordering occured. Store the data in the buffer.
 712         ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
 713         if(rxd < len)
 714                 abort();
 715
 716         // Make note of where we put it.
 717         for(int i = 0; i < NSACKS; i++) {
 718                 if(!c->sacks[i].len) { // nothing to merge, add new entry
 719                         debug("New SACK entry %d\n", i);
 720                         c->sacks[i].offset = offset;
 721                         c->sacks[i].len = rxd;
 722                         break;
 723                 } else if(offset < c->sacks[i].offset) {
 724                         if(offset + rxd < c->sacks[i].offset) { // insert before
 725                                 if(!c->sacks[NSACKS - 1].len) { // only if room left
 726                                         debug("Insert SACK entry at %d\n", i);
 727                                         memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
 728                                         c->sacks[i].offset = offset;
 729                                         c->sacks[i].len = rxd;
 730                                 } else {
 731                                         debug("SACK entries full, dropping packet\n");
 732                                 }
 733                                 break;
 734                         } else { // merge
 735                                 debug("Merge with start of SACK entry at %d\n", i);
 736                                 c->sacks[i].offset = offset;
 737                                 break;
 738                         }
 739                 } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
 740                         if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
 741                                 debug("Merge with end of SACK entry at %d\n", i);
 742                                 c->sacks[i].len = offset + rxd - c->sacks[i].offset;
 743                                 // TODO: handle potential merge with next entry
 744                         }
 745                         break;
 746                 }
 747         }
 748
 749         for(int i = 0; i < NSACKS && c->sacks[i].len; i++)
 750                 debug("SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
 751 }
 752
 753 static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
 754         // Check if we can process out-of-order data now.
 755         if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
 756                 debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
 757                 buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
 758                 len = max(len, c->sacks[0].offset + c->sacks[0].len);
 759                 data = c->rcvbuf.data;
 760         }
 761
 762         if(c->recv) {
 763                 ssize_t rxd = c->recv(c, data, len);
 764                 if(rxd != len) {
 765                         // TODO: handle the application not accepting all data.
 766                         abort();
 767                 }
 768         }
 769
 770         if(c->rcvbuf.used)
 771                 sack_consume(c, len);
 772
 773         c->rcv.nxt += len;
 774 }
 775
 776
 777 static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) {
 778         if(!is_reliable(c)) {
 779                 c->recv(c, data, len);
 780                 c->rcv.nxt = seq + len;
 781                 return;
 782         }
 783
 784         uint32_t offset = seqdiff(seq, c->rcv.nxt);
 785         if(offset + len > c->rcvbuf.maxsize)
 786                 abort();
 787
 788         if(offset)
 789                 handle_out_of_order(c, offset, data, len);
 790         else
 791                 handle_in_order(c, data, len);
 792 }
 793
 794
 795 ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
 796         if(!utcp) {
 797                 errno = EFAULT;
 798                 return -1;
 799         }
 800
 801         if(!len)
 802                 return 0;
 803
 804         if(!data) {
 805                 errno = EFAULT;
 806                 return -1;
 807         }
 808
 809         print_packet(utcp, "recv", data, len);
 810
 811         // Drop packets smaller than the header
 812
 813         struct hdr hdr;
 814         if(len < sizeof hdr) {
 815                 errno = EBADMSG;
 816                 return -1;
 817         }
 818
 819         // Make a copy from the potentially unaligned data to a struct hdr
 820
 821         memcpy(&hdr, data, sizeof hdr);
 822         data += sizeof hdr;
 823         len -= sizeof hdr;
 824
 825         // Drop packets with an unknown CTL flag
 826
 827         if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
 828                 errno = EBADMSG;
 829                 return -1;
 830         }
 831
 832         // Try to match the packet to an existing connection
 833
 834         struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
 835
 836         // Is it for a new connection?
 837
 838         if(!c) {
 839                 // Ignore RST packets
 840
 841                 if(hdr.ctl & RST)
 842                         return 0;
 843
 844                 // Is it a SYN packet and are we LISTENing?
 845
 846                 if(hdr.ctl & SYN && !(hdr.ctl & ACK) && utcp->accept) {
 847                         // If we don't want to accept it, send a RST back
 848                         if((utcp->pre_accept && !utcp->pre_accept(utcp, hdr.dst))) {
 849                                 len = 1;
 850                                 goto reset;
 851                         }
 852
 853                         // Try to allocate memory, otherwise send a RST back
 854                         c = allocate_connection(utcp, hdr.dst, hdr.src);
 855                         if(!c) {
 856                                 len = 1;
 857                                 goto reset;
 858                         }
 859
 860                         // Parse auxilliary information
 861                         if(hdr.aux) {
 862                                 if(hdr.aux != 0x0101 || len < 4 || ((uint8_t *)data)[0] != 1) {
 863                                         len = 1;
 864                                         goto reset;
 865                                 }
 866                                 c->flags = ((uint8_t *)data)[3] & 0x7;
 867                                 data += 4;
 868                                 len -= 4;
 869                         } else {
 870                                 c->flags = UTCP_TCP;
 871                         }
 872
 873                         // Return SYN+ACK, go to SYN_RECEIVED state
 874                         c->snd.wnd = hdr.wnd;
 875                         c->rcv.irs = hdr.seq;
 876                         c->rcv.nxt = c->rcv.irs + 1;
 877                         set_state(c, SYN_RECEIVED);
 878
 879                         hdr.dst = c->dst;
 880                         hdr.src = c->src;
 881                         hdr.ack = c->rcv.irs + 1;
 882                         hdr.seq = c->snd.iss;
 883                         hdr.ctl = SYN | ACK;
 884                         print_packet(c->utcp, "send", &hdr, sizeof hdr);
 885                         utcp->send(utcp, &hdr, sizeof hdr);
 886                 } else {
 887                         // No, we don't want your packets, send a RST back
 888                         len = 1;
 889                         goto reset;
 890                 }
 891
 892                 return 0;
 893         }
 894
 895         debug("%p state %s\n", c->utcp, strstate[c->state]);
 896
 897         // In case this is for a CLOSED connection, ignore the packet.
 898         // TODO: make it so incoming packets can never match a CLOSED connection.
 899
 900         if(c->state == CLOSED) {
 901                 debug("Got packet for closed connection\n");
 902                 return 0;
 903         }
 904
 905         // It is for an existing connection.
 906
 907         uint32_t prevrcvnxt = c->rcv.nxt;
 908
 909         // 1. Drop invalid packets.
 910
 911         // 1a. Drop packets that should not happen in our current state.
 912
 913         switch(c->state) {
 914         case SYN_SENT:
 915         case SYN_RECEIVED:
 916         case ESTABLISHED:
 917         case FIN_WAIT_1:
 918         case FIN_WAIT_2:
 919         case CLOSE_WAIT:
 920         case CLOSING:
 921         case LAST_ACK:
 922         case TIME_WAIT:
 923                 break;
 924         default:
 925 #ifdef UTCP_DEBUG
 926                 abort();
 927 #endif
 928                 break;
 929         }
 930
 931         // 1b. Drop packets with a sequence number not in our receive window.
 932
 933         bool acceptable;
 934
 935         if(c->state == SYN_SENT)
 936                 acceptable = true;
 937         else if(len == 0)
 938                 acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
 939         else {
 940                 int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
 941
 942                 // cut already accepted front overlapping
 943                 if(rcv_offset < 0) {
 944                         acceptable = len > -rcv_offset;
 945                         if(acceptable) {
 946                                 data -= rcv_offset;
 947                                 len += rcv_offset;
 948                                 hdr.seq -= rcv_offset;
 949                         }
 950                 } else {
 951                         acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
 952                 }
 953         }
 954
 955         if(!acceptable) {
 956                 debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
 957                 // Ignore unacceptable RST packets.
 958                 if(hdr.ctl & RST)
 959                         return 0;
 960                 // Otherwise, continue processing.
 961                 len = 0;
 962         }
 963
 964         c->snd.wnd = hdr.wnd; // TODO: move below
 965
 966         // 1c. Drop packets with an invalid ACK.
 967         // ackno should not roll back, and it should also not be bigger than what we ever could have sent
 968         // (= snd.una + c->sndbuf.used).
 969
 970         if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
 971                 debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
 972                 // Ignore unacceptable RST packets.
 973                 if(hdr.ctl & RST)
 974                         return 0;
 975                 goto reset;
 976         }
 977
 978         // 2. Handle RST packets
 979
 980         if(hdr.ctl & RST) {
 981                 switch(c->state) {
 982                 case SYN_SENT:
 983                         if(!(hdr.ctl & ACK))
 984                                 return 0;
 985                         // The peer has refused our connection.
 986                         set_state(c, CLOSED);
 987                         errno = ECONNREFUSED;
 988                         if(c->recv)
 989                                 c->recv(c, NULL, 0);
 990                         return 0;
 991                 case SYN_RECEIVED:
 992                         if(hdr.ctl & ACK)
 993                                 return 0;
 994                         // We haven't told the application about this connection yet. Silently delete.
 995                         free_connection(c);
 996                         return 0;
 997                 case ESTABLISHED:
 998                 case FIN_WAIT_1:
 999                 case FIN_WAIT_2:
1000                 case CLOSE_WAIT:
1001                         if(hdr.ctl & ACK)
1002                                 return 0;
1003                         // The peer has aborted our connection.
1004                         set_state(c, CLOSED);
1005                         errno = ECONNRESET;
1006                         if(c->recv)
1007                                 c->recv(c, NULL, 0);
1008                         return 0;
1009                 case CLOSING:
1010                 case LAST_ACK:
1011                 case TIME_WAIT:
1012                         if(hdr.ctl & ACK)
1013                                 return 0;
1014                         // As far as the application is concerned, the connection has already been closed.
1015                         // If it has called utcp_close() already, we can immediately free this connection.
1016                         if(c->reapable) {
1017                                 free_connection(c);
1018                                 return 0;
1019                         }
1020                         // Otherwise, immediately move to the CLOSED state.
1021                         set_state(c, CLOSED);
1022                         return 0;
1023                 default:
1024 #ifdef UTCP_DEBUG
1025                         abort();
1026 #endif
1027                         break;
1028                 }
1029         }
1030
1031         // 3. Advance snd.una
1032
1033         uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
1034         prevrcvnxt = c->rcv.nxt;
1035
1036         if(advanced) {
1037                 // RTT measurement
1038                 if(c->rtt_start.tv_sec) {
1039                         if(c->rtt_seq == hdr.ack) {
1040                                 struct timeval now, diff;
1041                                 gettimeofday(&now, NULL);
1042                                 timersub(&now, &c->rtt_start, &diff);
1043                                 update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
1044                                 c->rtt_start.tv_sec = 0;
1045                         } else if(c->rtt_seq < hdr.ack) {
1046                                 debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
1047                                 c->rtt_start.tv_sec = 0;
1048                         }
1049                 }
1050
1051                 int32_t data_acked = advanced;
1052
1053                 switch(c->state) {
1054                         case SYN_SENT:
1055                         case SYN_RECEIVED:
1056                                 data_acked--;
1057                                 break;
1058                         // TODO: handle FIN as well.
1059                         default:
1060                                 break;
1061                 }
1062
1063                 assert(data_acked >= 0);
1064
1065                 int32_t bufused = seqdiff(c->snd.last, c->snd.una);
1066                 assert(data_acked <= bufused);
1067
1068                 if(data_acked)
1069                         buffer_get(&c->sndbuf, NULL, data_acked);
1070
1071                 // Also advance snd.nxt if possible
1072                 if(seqdiff(c->snd.nxt, hdr.ack) < 0)
1073                         c->snd.nxt = hdr.ack;
1074
1075                 c->snd.una = hdr.ack;
1076
1077                 c->dupack = 0;
1078                 c->snd.cwnd += utcp->mtu;
1079                 if(c->snd.cwnd > c->sndbuf.maxsize)
1080                         c->snd.cwnd = c->sndbuf.maxsize;
1081
1082                 // Check if we have sent a FIN that is now ACKed.
1083                 switch(c->state) {
1084                 case FIN_WAIT_1:
1085                         if(c->snd.una == c->snd.last)
1086                                 set_state(c, FIN_WAIT_2);
1087                         break;
1088                 case CLOSING:
1089                         if(c->snd.una == c->snd.last) {
1090                                 gettimeofday(&c->conn_timeout, NULL);
1091                                 c->conn_timeout.tv_sec += 60;
1092                                 set_state(c, TIME_WAIT);
1093                         }
1094                         break;
1095                 default:
1096                         break;
1097                 }
1098         } else {
1099                 if(!len && is_reliable(c)) {
1100                         c->dupack++;
1101                         if(c->dupack == 3) {
1102                                 debug("Triplicate ACK\n");
1103                                 //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
1104                                 //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
1105                                 //Reset the congestion window so we wait for ACKs.
1106                                 c->snd.nxt = c->snd.una;
1107                                 c->snd.cwnd = utcp->mtu;
1108                                 start_retransmit_timer(c);
1109                         }
1110                 }
1111         }
1112
1113         // 4. Update timers
1114
1115         if(advanced) {
1116                 timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
1117                 if(c->snd.una == c->snd.last)
1118                         stop_retransmit_timer(c);
1119                 else if(is_reliable(c))
1120                         start_retransmit_timer(c);
1121         }
1122
1123         // 5. Process SYN stuff
1124
1125         if(hdr.ctl & SYN) {
1126                 switch(c->state) {
1127                 case SYN_SENT:
1128                         // This is a SYNACK. It should always have ACKed the SYN.
1129                         if(!advanced)
1130                                 goto reset;
1131                         c->rcv.irs = hdr.seq;
1132                         c->rcv.nxt = hdr.seq;
1133                         set_state(c, ESTABLISHED);
1134                         // TODO: notify application of this somehow.
1135                         break;
1136                 case SYN_RECEIVED:
1137                 case ESTABLISHED:
1138                 case FIN_WAIT_1:
1139                 case FIN_WAIT_2:
1140                 case CLOSE_WAIT:
1141                 case CLOSING:
1142                 case LAST_ACK:
1143                 case TIME_WAIT:
1144                         // Ehm, no. We should never receive a second SYN.
1145                         goto reset;
1146                 default:
1147 #ifdef UTCP_DEBUG
1148                         abort();
1149 #endif
1150                         return 0;
1151                 }
1152
1153                 // SYN counts as one sequence number
1154                 c->rcv.nxt++;
1155         }
1156
1157         // 6. Process new data
1158
1159         if(c->state == SYN_RECEIVED) {
1160                 // This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
1161                 if(!advanced)
1162                         goto reset;
1163
1164                 // Are we still LISTENing?
1165                 if(utcp->accept)
1166                         utcp->accept(c, c->src);
1167
1168                 if(c->state != ESTABLISHED) {
1169                         set_state(c, CLOSED);
1170                         c->reapable = true;
1171                         goto reset;
1172                 }
1173         }
1174
1175         if(len) {
1176                 switch(c->state) {
1177                 case SYN_SENT:
1178                 case SYN_RECEIVED:
1179                         // This should never happen.
1180 #ifdef UTCP_DEBUG
1181                         abort();
1182 #endif
1183                         return 0;
1184                 case ESTABLISHED:
1185                 case FIN_WAIT_1:
1186                 case FIN_WAIT_2:
1187                         break;
1188                 case CLOSE_WAIT:
1189                 case CLOSING:
1190                 case LAST_ACK:
1191                 case TIME_WAIT:
1192                         // Ehm no, We should never receive more data after a FIN.
1193                         goto reset;
1194                 default:
1195 #ifdef UTCP_DEBUG
1196                         abort();
1197 #endif
1198                         return 0;
1199                 }
1200
1201                 handle_incoming_data(c, hdr.seq, data, len);
1202         }
1203
1204         // 7. Process FIN stuff
1205
1206         if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
1207                 switch(c->state) {
1208                 case SYN_SENT:
1209                 case SYN_RECEIVED:
1210                         // This should never happen.
1211 #ifdef UTCP_DEBUG
1212                         abort();
1213 #endif
1214                         break;
1215                 case ESTABLISHED:
1216                         set_state(c, CLOSE_WAIT);
1217                         break;
1218                 case FIN_WAIT_1:
1219                         set_state(c, CLOSING);
1220                         break;
1221                 case FIN_WAIT_2:
1222                         gettimeofday(&c->conn_timeout, NULL);
1223                         c->conn_timeout.tv_sec += 60;
1224                         set_state(c, TIME_WAIT);
1225                         break;
1226                 case CLOSE_WAIT:
1227                 case CLOSING:
1228                 case LAST_ACK:
1229                 case TIME_WAIT:
1230                         // Ehm, no. We should never receive a second FIN.
1231                         goto reset;
1232                 default:
1233 #ifdef UTCP_DEBUG
1234                         abort();
1235 #endif
1236                         break;
1237                 }
1238
1239                 // FIN counts as one sequence number
1240                 c->rcv.nxt++;
1241                 len++;
1242
1243                 // Inform the application that the peer closed the connection.
1244                 if(c->recv) {
1245                         errno = 0;
1246                         c->recv(c, NULL, 0);
1247                 }
1248         }
1249
1250         // Now we send something back if:
1251         // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
1252         //   -> sendatleastone = true
1253         // - or we got an ack, so we should maybe send a bit more data
1254         //   -> sendatleastone = false
1255
1256         ack(c, len || prevrcvnxt != c->rcv.nxt);
1257         return 0;
1258
1259 reset:
1260         swap_ports(&hdr);
1261         hdr.wnd = 0;
1262         if(hdr.ctl & ACK) {
1263                 hdr.seq = hdr.ack;
1264                 hdr.ctl = RST;
1265         } else {
1266                 hdr.ack = hdr.seq + len;
1267                 hdr.seq = 0;
1268                 hdr.ctl = RST | ACK;
1269         }
1270         print_packet(utcp, "send", &hdr, sizeof hdr);
1271         utcp->send(utcp, &hdr, sizeof hdr);
1272         return 0;
1273
1274 }
1275
1276 int utcp_shutdown(struct utcp_connection *c, int dir) {
1277         debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c ? c->snd.last : 0);
1278         if(!c) {
1279                 errno = EFAULT;
1280                 return -1;
1281         }
1282
1283         if(c->reapable) {
1284                 debug("Error: shutdown() called on closed connection %p\n", c);
1285                 errno = EBADF;
1286                 return -1;
1287         }
1288
1289         if(!(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_WR || dir == UTCP_SHUT_RDWR)) {
1290                 errno = EINVAL;
1291                 return -1;
1292         }
1293
1294         // TCP does not have a provision for stopping incoming packets.
1295         // The best we can do is to just ignore them.
1296         if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
1297                 c->recv = NULL;
1298
1299         // The rest of the code deals with shutting down writes.
1300         if(dir == UTCP_SHUT_RD)
1301                 return 0;
1302
1303         switch(c->state) {
1304         case CLOSED:
1305         case LISTEN:
1306                 errno = ENOTCONN;
1307                 return -1;
1308
1309         case SYN_SENT:
1310                 set_state(c, CLOSED);
1311                 return 0;
1312
1313         case SYN_RECEIVED:
1314         case ESTABLISHED:
1315                 set_state(c, FIN_WAIT_1);
1316                 break;
1317         case FIN_WAIT_1:
1318         case FIN_WAIT_2:
1319                 return 0;
1320         case CLOSE_WAIT:
1321                 set_state(c, CLOSING);
1322                 break;
1323
1324         case CLOSING:
1325         case LAST_ACK:
1326         case TIME_WAIT:
1327                 return 0;
1328         }
1329
1330         c->snd.last++;
1331
1332         ack(c, false);
1333         if(!timerisset(&c->rtrx_timeout))
1334                 start_retransmit_timer(c);
1335         return 0;
1336 }
1337
1338 int utcp_close(struct utcp_connection *c) {
1339         if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
1340                 return -1;
1341         c->recv = NULL;
1342         c->poll = NULL;
1343         c->reapable = true;
1344         return 0;
1345 }
1346
1347 int utcp_abort(struct utcp_connection *c) {
1348         if(!c) {
1349                 errno = EFAULT;
1350                 return -1;
1351         }
1352
1353         if(c->reapable) {
1354                 debug("Error: abort() called on closed connection %p\n", c);
1355                 errno = EBADF;
1356                 return -1;
1357         }
1358
1359         c->recv = NULL;
1360         c->poll = NULL;
1361         c->reapable = true;
1362
1363         switch(c->state) {
1364         case CLOSED:
1365                 return 0;
1366         case LISTEN:
1367         case SYN_SENT:
1368         case CLOSING:
1369         case LAST_ACK:
1370         case TIME_WAIT:
1371                 set_state(c, CLOSED);
1372                 return 0;
1373
1374         case SYN_RECEIVED:
1375         case ESTABLISHED:
1376         case FIN_WAIT_1:
1377         case FIN_WAIT_2:
1378         case CLOSE_WAIT:
1379                 set_state(c, CLOSED);
1380                 break;
1381         }
1382
1383         // Send RST
1384
1385         struct hdr hdr;
1386
1387         hdr.src = c->src;
1388         hdr.dst = c->dst;
1389         hdr.seq = c->snd.nxt;
1390         hdr.ack = 0;
1391         hdr.wnd = 0;
1392         hdr.ctl = RST;
1393
1394         print_packet(c->utcp, "send", &hdr, sizeof hdr);
1395         c->utcp->send(c->utcp, &hdr, sizeof hdr);
1396         return 0;
1397 }
1398
1399 /* Handle timeouts.
1400  * One call to this function will loop through all connections,
1401  * checking if something needs to be resent or not.
1402  * The return value is the time to the next timeout in milliseconds,
1403  * or maybe a negative value if the timeout is infinite.
1404  */
1405 struct timeval utcp_timeout(struct utcp *utcp) {
1406         struct timeval now;
1407         gettimeofday(&now, NULL);
1408         struct timeval next = {now.tv_sec + 3600, now.tv_usec};
1409
1410         for(int i = 0; i < utcp->nconnections; i++) {
1411                 struct utcp_connection *c = utcp->connections[i];
1412                 if(!c)
1413                         continue;
1414
1415                 // delete connections that have been utcp_close()d.
1416                 if(c->state == CLOSED) {
1417                         if(c->reapable) {
1418                                 debug("Reaping %p\n", c);
1419                                 free_connection(c);
1420                                 i--;
1421                         }
1422                         continue;
1423                 }
1424
1425                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
1426                         errno = ETIMEDOUT;
1427                         c->state = CLOSED;
1428                         if(c->recv)
1429                                 c->recv(c, NULL, 0);
1430                         continue;
1431                 }
1432
1433                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
1434                         debug("retransmit()\n");
1435                         retransmit(c);
1436                 }
1437
1438                 if(c->poll) {
1439                         if((c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
1440                                 uint32_t len =  buffer_free(&c->sndbuf);
1441                                 if(len)
1442                                         c->poll(c, len);
1443                         } else if(c->state == CLOSED) {
1444                                 c->poll(c, 0);
1445                         }
1446                 }
1447
1448                 if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
1449                         next = c->conn_timeout;
1450
1451                 if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
1452                         next = c->rtrx_timeout;
1453         }
1454
1455         struct timeval diff;
1456         timersub(&next, &now, &diff);
1457         return diff;
1458 }
1459
1460 bool utcp_is_active(struct utcp *utcp) {
1461         if(!utcp)
1462                 return false;
1463
1464         for(int i = 0; i < utcp->nconnections; i++)
1465                 if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
1466                         return true;
1467
1468         return false;
1469 }
1470
1471 struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
1472         if(!send) {
1473                 errno = EFAULT;
1474                 return NULL;
1475         }
1476
1477         struct utcp *utcp = calloc(1, sizeof *utcp);
1478         if(!utcp)
1479                 return NULL;
1480
1481         utcp->accept = accept;
1482         utcp->pre_accept = pre_accept;
1483         utcp->send = send;
1484         utcp->priv = priv;
1485         utcp->mtu = DEFAULT_MTU;
1486         utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
1487         utcp->rto = START_RTO; // usec
1488
1489         return utcp;
1490 }
1491
1492 void utcp_exit(struct utcp *utcp) {
1493         if(!utcp)
1494                 return;
1495         for(int i = 0; i < utcp->nconnections; i++) {
1496                 if(!utcp->connections[i]->reapable)
1497                         debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
1498                 buffer_exit(&utcp->connections[i]->rcvbuf);
1499                 buffer_exit(&utcp->connections[i]->sndbuf);
1500                 free(utcp->connections[i]);
1501         }
1502         free(utcp->connections);
1503         free(utcp);
1504 }
1505
1506 uint16_t utcp_get_mtu(struct utcp *utcp) {
1507         return utcp ? utcp->mtu : 0;
1508 }
1509
1510 void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
1511         // TODO: handle overhead of the header
1512         if(utcp)
1513                 utcp->mtu = mtu;
1514 }
1515
1516 int utcp_get_user_timeout(struct utcp *u) {
1517         return u ? u->timeout : 0;
1518 }
1519
1520 void utcp_set_user_timeout(struct utcp *u, int timeout) {
1521         if(u)
1522                 u->timeout = timeout;
1523 }
1524
1525 size_t utcp_get_sndbuf(struct utcp_connection *c) {
1526         return c ? c->sndbuf.maxsize : 0;
1527 }
1528
1529 size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
1530         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1531                 return buffer_free(&c->sndbuf);
1532         else
1533                 return 0;
1534 }
1535
1536 void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
1537         if(!c)
1538                 return;
1539         c->sndbuf.maxsize = size;
1540         if(c->sndbuf.maxsize != size)
1541                 c->sndbuf.maxsize = -1;
1542 }
1543
1544 size_t utcp_get_rcvbuf(struct utcp_connection *c) {
1545         return c ? c->rcvbuf.maxsize : 0;
1546 }
1547
1548 size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
1549         if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
1550                 return buffer_free(&c->rcvbuf);
1551         else
1552                 return 0;
1553 }
1554
1555 void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
1556         if(!c)
1557                 return;
1558         c->rcvbuf.maxsize = size;
1559         if(c->rcvbuf.maxsize != size)
1560                 c->rcvbuf.maxsize = -1;
1561 }
1562
1563 bool utcp_get_nodelay(struct utcp_connection *c) {
1564         return c ? c->nodelay : false;
1565 }
1566
1567 void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
1568         if(c)
1569                 c->nodelay = nodelay;
1570 }
1571
1572 bool utcp_get_keepalive(struct utcp_connection *c) {
1573         return c ? c->keepalive : false;
1574 }
1575
1576 void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
1577         if(c)
1578                 c->keepalive = keepalive;
1579 }
1580
1581 size_t utcp_get_outq(struct utcp_connection *c) {
1582         return c ? seqdiff(c->snd.nxt, c->snd.una) : 0;
1583 }
1584
1585 void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
1586         if(c)
1587                 c->recv = recv;
1588 }
1589
1590 void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
1591         if(c)
1592                 c->poll = poll;
1593 }
1594
1595 void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
1596         if(utcp) {
1597                 utcp->accept = accept;
1598                 utcp->pre_accept = pre_accept;
1599         }
1600 }