services/listen_dnsport.c

   1 /*
   2  * services/listen_dnsport.c - listen on port 53 for incoming DNS queries.
   3  *
   4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
   5  *
   6  * This software is open source.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * Redistributions of source code must retain the above copyright notice,
  13  * this list of conditions and the following disclaimer.
  14  *
  15  * Redistributions in binary form must reproduce the above copyright notice,
  16  * this list of conditions and the following disclaimer in the documentation
  17  * and/or other materials provided with the distribution.
  18  *
  19  * Neither the name of the NLNET LABS nor the names of its contributors may
  20  * be used to endorse or promote products derived from this software without
  21  * specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
  29  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
  30  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  31  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  32  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  33  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  34  */
  35
  36 /**
  37  * \file
  38  *
  39  * This file has functions to get queries from clients.
  40  */
  41 #include "config.h"
  42 #ifdef HAVE_SYS_TYPES_H
  43 #  include <sys/types.h>
  44 #endif
  45 #include <sys/time.h>
  46 #ifdef USE_TCP_FASTOPEN
  47 #include <netinet/tcp.h>
  48 #endif
  49 #include "services/listen_dnsport.h"
  50 #include "services/outside_network.h"
  51 #include "util/netevent.h"
  52 #include "util/log.h"
  53 #include "util/config_file.h"
  54 #include "util/net_help.h"
  55 #include "sldns/sbuffer.h"
  56 #include "services/mesh.h"
  57 #include "util/fptr_wlist.h"
  58 #include "util/locks.h"
  59
  60 #ifdef HAVE_NETDB_H
  61 #include <netdb.h>
  62 #endif
  63 #include <fcntl.h>
  64
  65 #ifdef HAVE_SYS_UN_H
  66 #include <sys/un.h>
  67 #endif
  68
  69 #ifdef HAVE_SYSTEMD
  70 #include <systemd/sd-daemon.h>
  71 #endif
  72
  73 /** number of queued TCP connections for listen() */
  74 #define TCP_BACKLOG 256
  75
  76 /** number of simultaneous requests a client can have */
  77 #define TCP_MAX_REQ_SIMULTANEOUS 32
  78
  79 #ifndef THREADS_DISABLED
  80 /** lock on the counter of stream buffer memory */
  81 static lock_basic_type stream_wait_count_lock;
  82 #endif
  83 /** size (in bytes) of stream wait buffers */
  84 static size_t stream_wait_count = 0;
  85 /** is the lock initialised for stream wait buffers */
  86 static int stream_wait_lock_inited = 0;
  87
  88 /**
  89  * Debug print of the getaddrinfo returned address.
  90  * @param addr: the address returned.
  91  */
  92 static void
  93 verbose_print_addr(struct addrinfo *addr)
  94 {
  95         if(verbosity >= VERB_ALGO) {
  96                 char buf[100];
  97                 void* sinaddr = &((struct sockaddr_in*)addr->ai_addr)->sin_addr;
  98 #ifdef INET6
  99                 if(addr->ai_family == AF_INET6)
 100                         sinaddr = &((struct sockaddr_in6*)addr->ai_addr)->
 101                                 sin6_addr;
 102 #endif /* INET6 */
 103                 if(inet_ntop(addr->ai_family, sinaddr, buf,
 104                         (socklen_t)sizeof(buf)) == 0) {
 105                         (void)strlcpy(buf, "(null)", sizeof(buf));
 106                 }
 107                 buf[sizeof(buf)-1] = 0;
 108                 verbose(VERB_ALGO, "creating %s%s socket %s %d",
 109                         addr->ai_socktype==SOCK_DGRAM?"udp":
 110                         addr->ai_socktype==SOCK_STREAM?"tcp":"otherproto",
 111                         addr->ai_family==AF_INET?"4":
 112                         addr->ai_family==AF_INET6?"6":
 113                         "_otherfam", buf,
 114                         ntohs(((struct sockaddr_in*)addr->ai_addr)->sin_port));
 115         }
 116 }
 117
 118 #ifdef HAVE_SYSTEMD
 119 static int
 120 systemd_get_activated(int family, int socktype, int listen,
 121                       struct sockaddr *addr, socklen_t addrlen,
 122                       const char *path)
 123 {
 124         int i = 0;
 125         int r = 0;
 126         int s = -1;
 127         const char* listen_pid, *listen_fds;
 128
 129         /* We should use "listen" option only for stream protocols. For UDP it should be -1 */
 130
 131         if((r = sd_booted()) < 1) {
 132                 if(r == 0)
 133                         log_warn("systemd is not running");
 134                 else
 135                         log_err("systemd sd_booted(): %s", strerror(-r));
 136                 return -1;
 137         }
 138
 139         listen_pid = getenv("LISTEN_PID");
 140         listen_fds = getenv("LISTEN_FDS");
 141
 142         if (!listen_pid) {
 143                 log_warn("Systemd mandatory ENV variable is not defined: LISTEN_PID");
 144                 return -1;
 145         }
 146
 147         if (!listen_fds) {
 148                 log_warn("Systemd mandatory ENV variable is not defined: LISTEN_FDS");
 149                 return -1;
 150         }
 151
 152         if((r = sd_listen_fds(0)) < 1) {
 153                 if(r == 0)
 154                         log_warn("systemd: did not return socket, check unit configuration");
 155                 else
 156                         log_err("systemd sd_listen_fds(): %s", strerror(-r));
 157                 return -1;
 158         }
 159
 160         for(i = 0; i < r; i++) {
 161                 if(sd_is_socket(SD_LISTEN_FDS_START + i, family, socktype, listen)) {
 162                         s = SD_LISTEN_FDS_START + i;
 163                         break;
 164                 }
 165         }
 166         if (s == -1) {
 167                 if (addr)
 168                         log_err_addr("systemd sd_listen_fds()",
 169                                      "no such socket",
 170                                      (struct sockaddr_storage *)addr, addrlen);
 171                 else
 172                         log_err("systemd sd_listen_fds(): %s", path);
 173         }
 174         return s;
 175 }
 176 #endif
 177
 178 int
 179 create_udp_sock(int family, int socktype, struct sockaddr* addr,
 180         socklen_t addrlen, int v6only, int* inuse, int* noproto,
 181         int rcv, int snd, int listen, int* reuseport, int transparent,
 182         int freebind, int use_systemd)
 183 {
 184         int s;
 185 #if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_USE_MIN_MTU)  || defined(IP_TRANSPARENT) || defined(IP_BINDANY) || defined(IP_FREEBIND) || defined (SO_BINDANY)
 186         int on=1;
 187 #endif
 188 #ifdef IPV6_MTU
 189         int mtu = IPV6_MIN_MTU;
 190 #endif
 191 #if !defined(SO_RCVBUFFORCE) && !defined(SO_RCVBUF)
 192         (void)rcv;
 193 #endif
 194 #if !defined(SO_SNDBUFFORCE) && !defined(SO_SNDBUF)
 195         (void)snd;
 196 #endif
 197 #ifndef IPV6_V6ONLY
 198         (void)v6only;
 199 #endif
 200 #if !defined(IP_TRANSPARENT) && !defined(IP_BINDANY) && !defined(SO_BINDANY)
 201         (void)transparent;
 202 #endif
 203 #if !defined(IP_FREEBIND)
 204         (void)freebind;
 205 #endif
 206 #ifdef HAVE_SYSTEMD
 207         int got_fd_from_systemd = 0;
 208
 209         if (!use_systemd
 210             || (use_systemd
 211                 && (s = systemd_get_activated(family, socktype, -1, addr,
 212                                               addrlen, NULL)) == -1)) {
 213 #else
 214         (void)use_systemd;
 215 #endif
 216         if((s = socket(family, socktype, 0)) == -1) {
 217                 *inuse = 0;
 218 #ifndef USE_WINSOCK
 219                 if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
 220                         *noproto = 1;
 221                         return -1;
 222                 }
 223                 log_err("can't create socket: %s", strerror(errno));
 224 #else
 225                 if(WSAGetLastError() == WSAEAFNOSUPPORT ||
 226                         WSAGetLastError() == WSAEPROTONOSUPPORT) {
 227                         *noproto = 1;
 228                         return -1;
 229                 }
 230                 log_err("can't create socket: %s",
 231                         wsa_strerror(WSAGetLastError()));
 232 #endif
 233                 *noproto = 0;
 234                 return -1;
 235         }
 236 #ifdef HAVE_SYSTEMD
 237         } else {
 238                 got_fd_from_systemd = 1;
 239         }
 240 #endif
 241         if(listen) {
 242 #ifdef SO_REUSEADDR
 243                 if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
 244                         (socklen_t)sizeof(on)) < 0) {
 245 #ifndef USE_WINSOCK
 246                         log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
 247                                 strerror(errno));
 248                         if(errno != ENOSYS) {
 249                                 close(s);
 250                                 *noproto = 0;
 251                                 *inuse = 0;
 252                                 return -1;
 253                         }
 254 #else
 255                         log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
 256                                 wsa_strerror(WSAGetLastError()));
 257                         closesocket(s);
 258                         *noproto = 0;
 259                         *inuse = 0;
 260                         return -1;
 261 #endif
 262                 }
 263 #endif /* SO_REUSEADDR */
 264 #ifdef SO_REUSEPORT
 265 #  ifdef SO_REUSEPORT_LB
 266                 /* on FreeBSD 12 we have SO_REUSEPORT_LB that does loadbalance
 267                  * like SO_REUSEPORT on Linux.  This is what the users want
 268                  * with the config option in unbound.conf; if we actually
 269                  * need local address and port reuse they'll also need to
 270                  * have SO_REUSEPORT set for them, assume it was _LB they want.
 271                  */
 272                 if (reuseport && *reuseport &&
 273                     setsockopt(s, SOL_SOCKET, SO_REUSEPORT_LB, (void*)&on,
 274                         (socklen_t)sizeof(on)) < 0) {
 275 #ifdef ENOPROTOOPT
 276                         if(errno != ENOPROTOOPT || verbosity >= 3)
 277                                 log_warn("setsockopt(.. SO_REUSEPORT_LB ..) failed: %s",
 278                                         strerror(errno));
 279 #endif
 280                         /* this option is not essential, we can continue */
 281                         *reuseport = 0;
 282                 }
 283 #  else /* no SO_REUSEPORT_LB */
 284
 285                 /* try to set SO_REUSEPORT so that incoming
 286                  * queries are distributed evenly among the receiving threads.
 287                  * Each thread must have its own socket bound to the same port,
 288                  * with SO_REUSEPORT set on each socket.
 289                  */
 290                 if (reuseport && *reuseport &&
 291                     setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
 292                         (socklen_t)sizeof(on)) < 0) {
 293 #ifdef ENOPROTOOPT
 294                         if(errno != ENOPROTOOPT || verbosity >= 3)
 295                                 log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
 296                                         strerror(errno));
 297 #endif
 298                         /* this option is not essential, we can continue */
 299                         *reuseport = 0;
 300                 }
 301 #  endif /* SO_REUSEPORT_LB */
 302 #else
 303                 (void)reuseport;
 304 #endif /* defined(SO_REUSEPORT) */
 305 #ifdef IP_TRANSPARENT
 306                 if (transparent &&
 307                     setsockopt(s, IPPROTO_IP, IP_TRANSPARENT, (void*)&on,
 308                     (socklen_t)sizeof(on)) < 0) {
 309                         log_warn("setsockopt(.. IP_TRANSPARENT ..) failed: %s",
 310                         strerror(errno));
 311                 }
 312 #elif defined(IP_BINDANY)
 313                 if (transparent &&
 314                     setsockopt(s, (family==AF_INET6? IPPROTO_IPV6:IPPROTO_IP),
 315                     (family == AF_INET6? IPV6_BINDANY:IP_BINDANY),
 316                     (void*)&on, (socklen_t)sizeof(on)) < 0) {
 317                         log_warn("setsockopt(.. IP%s_BINDANY ..) failed: %s",
 318                         (family==AF_INET6?"V6":""), strerror(errno));
 319                 }
 320 #elif defined(SO_BINDANY)
 321                 if (transparent &&
 322                     setsockopt(s, SOL_SOCKET, SO_BINDANY, (void*)&on,
 323                     (socklen_t)sizeof(on)) < 0) {
 324                         log_warn("setsockopt(.. SO_BINDANY ..) failed: %s",
 325                         strerror(errno));
 326                 }
 327 #endif /* IP_TRANSPARENT || IP_BINDANY || SO_BINDANY */
 328         }
 329 #ifdef IP_FREEBIND
 330         if(freebind &&
 331             setsockopt(s, IPPROTO_IP, IP_FREEBIND, (void*)&on,
 332             (socklen_t)sizeof(on)) < 0) {
 333                 log_warn("setsockopt(.. IP_FREEBIND ..) failed: %s",
 334                 strerror(errno));
 335         }
 336 #endif /* IP_FREEBIND */
 337         if(rcv) {
 338 #ifdef SO_RCVBUF
 339                 int got;
 340                 socklen_t slen = (socklen_t)sizeof(got);
 341 #  ifdef SO_RCVBUFFORCE
 342                 /* Linux specific: try to use root permission to override
 343                  * system limits on rcvbuf. The limit is stored in
 344                  * /proc/sys/net/core/rmem_max or sysctl net.core.rmem_max */
 345                 if(setsockopt(s, SOL_SOCKET, SO_RCVBUFFORCE, (void*)&rcv,
 346                         (socklen_t)sizeof(rcv)) < 0) {
 347                         if(errno != EPERM) {
 348 #    ifndef USE_WINSOCK
 349                                 log_err("setsockopt(..., SO_RCVBUFFORCE, "
 350                                         "...) failed: %s", strerror(errno));
 351                                 close(s);
 352 #    else
 353                                 log_err("setsockopt(..., SO_RCVBUFFORCE, "
 354                                         "...) failed: %s",
 355                                         wsa_strerror(WSAGetLastError()));
 356                                 closesocket(s);
 357 #    endif
 358                                 *noproto = 0;
 359                                 *inuse = 0;
 360                                 return -1;
 361                         }
 362 #  endif /* SO_RCVBUFFORCE */
 363                         if(setsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&rcv,
 364                                 (socklen_t)sizeof(rcv)) < 0) {
 365 #  ifndef USE_WINSOCK
 366                                 log_err("setsockopt(..., SO_RCVBUF, "
 367                                         "...) failed: %s", strerror(errno));
 368                                 close(s);
 369 #  else
 370                                 log_err("setsockopt(..., SO_RCVBUF, "
 371                                         "...) failed: %s",
 372                                         wsa_strerror(WSAGetLastError()));
 373                                 closesocket(s);
 374 #  endif
 375                                 *noproto = 0;
 376                                 *inuse = 0;
 377                                 return -1;
 378                         }
 379                         /* check if we got the right thing or if system
 380                          * reduced to some system max.  Warn if so */
 381                         if(getsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&got,
 382                                 &slen) >= 0 && got < rcv/2) {
 383                                 log_warn("so-rcvbuf %u was not granted. "
 384                                         "Got %u. To fix: start with "
 385                                         "root permissions(linux) or sysctl "
 386                                         "bigger net.core.rmem_max(linux) or "
 387                                         "kern.ipc.maxsockbuf(bsd) values.",
 388                                         (unsigned)rcv, (unsigned)got);
 389                         }
 390 #  ifdef SO_RCVBUFFORCE
 391                 }
 392 #  endif
 393 #endif /* SO_RCVBUF */
 394         }
 395         /* first do RCVBUF as the receive buffer is more important */
 396         if(snd) {
 397 #ifdef SO_SNDBUF
 398                 int got;
 399                 socklen_t slen = (socklen_t)sizeof(got);
 400 #  ifdef SO_SNDBUFFORCE
 401                 /* Linux specific: try to use root permission to override
 402                  * system limits on sndbuf. The limit is stored in
 403                  * /proc/sys/net/core/wmem_max or sysctl net.core.wmem_max */
 404                 if(setsockopt(s, SOL_SOCKET, SO_SNDBUFFORCE, (void*)&snd,
 405                         (socklen_t)sizeof(snd)) < 0) {
 406                         if(errno != EPERM) {
 407 #    ifndef USE_WINSOCK
 408                                 log_err("setsockopt(..., SO_SNDBUFFORCE, "
 409                                         "...) failed: %s", strerror(errno));
 410                                 close(s);
 411 #    else
 412                                 log_err("setsockopt(..., SO_SNDBUFFORCE, "
 413                                         "...) failed: %s",
 414                                         wsa_strerror(WSAGetLastError()));
 415                                 closesocket(s);
 416 #    endif
 417                                 *noproto = 0;
 418                                 *inuse = 0;
 419                                 return -1;
 420                         }
 421 #  endif /* SO_SNDBUFFORCE */
 422                         if(setsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&snd,
 423                                 (socklen_t)sizeof(snd)) < 0) {
 424 #  ifndef USE_WINSOCK
 425                                 log_err("setsockopt(..., SO_SNDBUF, "
 426                                         "...) failed: %s", strerror(errno));
 427                                 close(s);
 428 #  else
 429                                 log_err("setsockopt(..., SO_SNDBUF, "
 430                                         "...) failed: %s",
 431                                         wsa_strerror(WSAGetLastError()));
 432                                 closesocket(s);
 433 #  endif
 434                                 *noproto = 0;
 435                                 *inuse = 0;
 436                                 return -1;
 437                         }
 438                         /* check if we got the right thing or if system
 439                          * reduced to some system max.  Warn if so */
 440                         if(getsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&got,
 441                                 &slen) >= 0 && got < snd/2) {
 442                                 log_warn("so-sndbuf %u was not granted. "
 443                                         "Got %u. To fix: start with "
 444                                         "root permissions(linux) or sysctl "
 445                                         "bigger net.core.wmem_max(linux) or "
 446                                         "kern.ipc.maxsockbuf(bsd) values.",
 447                                         (unsigned)snd, (unsigned)got);
 448                         }
 449 #  ifdef SO_SNDBUFFORCE
 450                 }
 451 #  endif
 452 #endif /* SO_SNDBUF */
 453         }
 454         if(family == AF_INET6) {
 455 # if defined(IPV6_V6ONLY)
 456                 if(v6only) {
 457                         int val=(v6only==2)?0:1;
 458                         if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
 459                                 (void*)&val, (socklen_t)sizeof(val)) < 0) {
 460 #ifndef USE_WINSOCK
 461                                 log_err("setsockopt(..., IPV6_V6ONLY"
 462                                         ", ...) failed: %s", strerror(errno));
 463                                 close(s);
 464 #else
 465                                 log_err("setsockopt(..., IPV6_V6ONLY"
 466                                         ", ...) failed: %s",
 467                                         wsa_strerror(WSAGetLastError()));
 468                                 closesocket(s);
 469 #endif
 470                                 *noproto = 0;
 471                                 *inuse = 0;
 472                                 return -1;
 473                         }
 474                 }
 475 # endif
 476 # if defined(IPV6_USE_MIN_MTU)
 477                 /*
 478                  * There is no fragmentation of IPv6 datagrams
 479                  * during forwarding in the network. Therefore
 480                  * we do not send UDP datagrams larger than
 481                  * the minimum IPv6 MTU of 1280 octets. The
 482                  * EDNS0 message length can be larger if the
 483                  * network stack supports IPV6_USE_MIN_MTU.
 484                  */
 485                 if (setsockopt(s, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
 486                         (void*)&on, (socklen_t)sizeof(on)) < 0) {
 487 #  ifndef USE_WINSOCK
 488                         log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
 489                                 "...) failed: %s", strerror(errno));
 490                         close(s);
 491 #  else
 492                         log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
 493                                 "...) failed: %s",
 494                                 wsa_strerror(WSAGetLastError()));
 495                         closesocket(s);
 496 #  endif
 497                         *noproto = 0;
 498                         *inuse = 0;
 499                         return -1;
 500                 }
 501 # elif defined(IPV6_MTU)
 502                 /*
 503                  * On Linux, to send no larger than 1280, the PMTUD is
 504                  * disabled by default for datagrams anyway, so we set
 505                  * the MTU to use.
 506                  */
 507                 if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU,
 508                         (void*)&mtu, (socklen_t)sizeof(mtu)) < 0) {
 509 #  ifndef USE_WINSOCK
 510                         log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
 511                                 strerror(errno));
 512                         close(s);
 513 #  else
 514                         log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
 515                                 wsa_strerror(WSAGetLastError()));
 516                         closesocket(s);
 517 #  endif
 518                         *noproto = 0;
 519                         *inuse = 0;
 520                         return -1;
 521                 }
 522 # endif /* IPv6 MTU */
 523         } else if(family == AF_INET) {
 524 #  if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
 525 /* linux 3.15 has IP_PMTUDISC_OMIT, Hannes Frederic Sowa made it so that
 526  * PMTU information is not accepted, but fragmentation is allowed
 527  * if and only if the packet size exceeds the outgoing interface MTU
 528  * (and also uses the interface mtu to determine the size of the packets).
 529  * So there won't be any EMSGSIZE error.  Against DNS fragmentation attacks.
 530  * FreeBSD already has same semantics without setting the option. */
 531                 int omit_set = 0;
 532                 int action;
 533 #   if defined(IP_PMTUDISC_OMIT)
 534                 action = IP_PMTUDISC_OMIT;
 535                 if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
 536                         &action, (socklen_t)sizeof(action)) < 0) {
 537
 538                         if (errno != EINVAL) {
 539                                 log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_OMIT...) failed: %s",
 540                                         strerror(errno));
 541
 542 #    ifndef USE_WINSOCK
 543                                 close(s);
 544 #    else
 545                                 closesocket(s);
 546 #    endif
 547                                 *noproto = 0;
 548                                 *inuse = 0;
 549                                 return -1;
 550                         }
 551                 }
 552                 else
 553                 {
 554                     omit_set = 1;
 555                 }
 556 #   endif
 557                 if (omit_set == 0) {
 558                         action = IP_PMTUDISC_DONT;
 559                         if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
 560                                 &action, (socklen_t)sizeof(action)) < 0) {
 561                                 log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_DONT...) failed: %s",
 562                                         strerror(errno));
 563 #    ifndef USE_WINSOCK
 564                                 close(s);
 565 #    else
 566                                 closesocket(s);
 567 #    endif
 568                                 *noproto = 0;
 569                                 *inuse = 0;
 570                                 return -1;
 571                         }
 572                 }
 573 #  elif defined(IP_DONTFRAG)
 574                 int off = 0;
 575                 if (setsockopt(s, IPPROTO_IP, IP_DONTFRAG,
 576                         &off, (socklen_t)sizeof(off)) < 0) {
 577                         log_err("setsockopt(..., IP_DONTFRAG, ...) failed: %s",
 578                                 strerror(errno));
 579 #    ifndef USE_WINSOCK
 580                         close(s);
 581 #    else
 582                         closesocket(s);
 583 #    endif
 584                         *noproto = 0;
 585                         *inuse = 0;
 586                         return -1;
 587                 }
 588 #  endif /* IPv4 MTU */
 589         }
 590         if(
 591 #ifdef HAVE_SYSTEMD
 592                 !got_fd_from_systemd &&
 593 #endif
 594                 bind(s, (struct sockaddr*)addr, addrlen) != 0) {
 595                 *noproto = 0;
 596                 *inuse = 0;
 597 #ifndef USE_WINSOCK
 598 #ifdef EADDRINUSE
 599                 *inuse = (errno == EADDRINUSE);
 600                 /* detect freebsd jail with no ipv6 permission */
 601                 if(family==AF_INET6 && errno==EINVAL)
 602                         *noproto = 1;
 603                 else if(errno != EADDRINUSE &&
 604                         !(errno == EACCES && verbosity < 4 && !listen)
 605 #ifdef EADDRNOTAVAIL
 606                         && !(errno == EADDRNOTAVAIL && verbosity < 4 && !listen)
 607 #endif
 608                         ) {
 609                         log_err_addr("can't bind socket", strerror(errno),
 610                                 (struct sockaddr_storage*)addr, addrlen);
 611                 }
 612 #endif /* EADDRINUSE */
 613                 close(s);
 614 #else /* USE_WINSOCK */
 615                 if(WSAGetLastError() != WSAEADDRINUSE &&
 616                         WSAGetLastError() != WSAEADDRNOTAVAIL &&
 617                         !(WSAGetLastError() == WSAEACCES && verbosity < 4 && !listen)) {
 618                         log_err_addr("can't bind socket",
 619                                 wsa_strerror(WSAGetLastError()),
 620                                 (struct sockaddr_storage*)addr, addrlen);
 621                 }
 622                 closesocket(s);
 623 #endif /* USE_WINSOCK */
 624                 return -1;
 625         }
 626         if(!fd_set_nonblock(s)) {
 627                 *noproto = 0;
 628                 *inuse = 0;
 629 #ifndef USE_WINSOCK
 630                 close(s);
 631 #else
 632                 closesocket(s);
 633 #endif
 634                 return -1;
 635         }
 636         return s;
 637 }
 638
 639 int
 640 create_tcp_accept_sock(struct addrinfo *addr, int v6only, int* noproto,
 641         int* reuseport, int transparent, int mss, int freebind, int use_systemd)
 642 {
 643         int s;
 644 #if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_V6ONLY) || defined(IP_TRANSPARENT) || defined(IP_BINDANY) || defined(IP_FREEBIND) || defined(SO_BINDANY)
 645         int on = 1;
 646 #endif
 647 #ifdef HAVE_SYSTEMD
 648         int got_fd_from_systemd = 0;
 649 #endif
 650 #ifdef USE_TCP_FASTOPEN
 651         int qlen;
 652 #endif
 653 #if !defined(IP_TRANSPARENT) && !defined(IP_BINDANY) && !defined(SO_BINDANY)
 654         (void)transparent;
 655 #endif
 656 #if !defined(IP_FREEBIND)
 657         (void)freebind;
 658 #endif
 659         verbose_print_addr(addr);
 660         *noproto = 0;
 661 #ifdef HAVE_SYSTEMD
 662         if (!use_systemd ||
 663             (use_systemd
 664              && (s = systemd_get_activated(addr->ai_family, addr->ai_socktype, 1,
 665                                            addr->ai_addr, addr->ai_addrlen,
 666                                            NULL)) == -1)) {
 667 #else
 668         (void)use_systemd;
 669 #endif
 670         if((s = socket(addr->ai_family, addr->ai_socktype, 0)) == -1) {
 671 #ifndef USE_WINSOCK
 672                 if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
 673                         *noproto = 1;
 674                         return -1;
 675                 }
 676                 log_err("can't create socket: %s", strerror(errno));
 677 #else
 678                 if(WSAGetLastError() == WSAEAFNOSUPPORT ||
 679                         WSAGetLastError() == WSAEPROTONOSUPPORT) {
 680                         *noproto = 1;
 681                         return -1;
 682                 }
 683                 log_err("can't create socket: %s",
 684                         wsa_strerror(WSAGetLastError()));
 685 #endif
 686                 return -1;
 687         }
 688         if (mss > 0) {
 689 #if defined(IPPROTO_TCP) && defined(TCP_MAXSEG)
 690                 if(setsockopt(s, IPPROTO_TCP, TCP_MAXSEG, (void*)&mss,
 691                         (socklen_t)sizeof(mss)) < 0) {
 692                         #ifndef USE_WINSOCK
 693                         log_err(" setsockopt(.. TCP_MAXSEG ..) failed: %s",
 694                                 strerror(errno));
 695                         #else
 696                         log_err(" setsockopt(.. TCP_MAXSEG ..) failed: %s",
 697                                 wsa_strerror(WSAGetLastError()));
 698                         #endif
 699                 } else {
 700                         verbose(VERB_ALGO,
 701                                 " tcp socket mss set to %d", mss);
 702                 }
 703 #else
 704                 log_warn(" setsockopt(TCP_MAXSEG) unsupported");
 705 #endif /* defined(IPPROTO_TCP) && defined(TCP_MAXSEG) */
 706         }
 707 #ifdef HAVE_SYSTEMD
 708         } else {
 709                 got_fd_from_systemd = 1;
 710     }
 711 #endif
 712 #ifdef SO_REUSEADDR
 713         if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
 714                 (socklen_t)sizeof(on)) < 0) {
 715 #ifndef USE_WINSOCK
 716                 log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
 717                         strerror(errno));
 718                 close(s);
 719 #else
 720                 log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
 721                         wsa_strerror(WSAGetLastError()));
 722                 closesocket(s);
 723 #endif
 724                 return -1;
 725         }
 726 #endif /* SO_REUSEADDR */
 727 #ifdef IP_FREEBIND
 728         if (freebind && setsockopt(s, IPPROTO_IP, IP_FREEBIND, (void*)&on,
 729             (socklen_t)sizeof(on)) < 0) {
 730                 log_warn("setsockopt(.. IP_FREEBIND ..) failed: %s",
 731                 strerror(errno));
 732         }
 733 #endif /* IP_FREEBIND */
 734 #ifdef SO_REUSEPORT
 735         /* try to set SO_REUSEPORT so that incoming
 736          * connections are distributed evenly among the receiving threads.
 737          * Each thread must have its own socket bound to the same port,
 738          * with SO_REUSEPORT set on each socket.
 739          */
 740         if (reuseport && *reuseport &&
 741                 setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
 742                 (socklen_t)sizeof(on)) < 0) {
 743 #ifdef ENOPROTOOPT
 744                 if(errno != ENOPROTOOPT || verbosity >= 3)
 745                         log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
 746                                 strerror(errno));
 747 #endif
 748                 /* this option is not essential, we can continue */
 749                 *reuseport = 0;
 750         }
 751 #else
 752         (void)reuseport;
 753 #endif /* defined(SO_REUSEPORT) */
 754 #if defined(IPV6_V6ONLY)
 755         if(addr->ai_family == AF_INET6 && v6only) {
 756                 if(setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
 757                         (void*)&on, (socklen_t)sizeof(on)) < 0) {
 758 #ifndef USE_WINSOCK
 759                         log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
 760                                 strerror(errno));
 761                         close(s);
 762 #else
 763                         log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
 764                                 wsa_strerror(WSAGetLastError()));
 765                         closesocket(s);
 766 #endif
 767                         return -1;
 768                 }
 769         }
 770 #else
 771         (void)v6only;
 772 #endif /* IPV6_V6ONLY */
 773 #ifdef IP_TRANSPARENT
 774         if (transparent &&
 775             setsockopt(s, IPPROTO_IP, IP_TRANSPARENT, (void*)&on,
 776             (socklen_t)sizeof(on)) < 0) {
 777                 log_warn("setsockopt(.. IP_TRANSPARENT ..) failed: %s",
 778                         strerror(errno));
 779         }
 780 #elif defined(IP_BINDANY)
 781         if (transparent &&
 782             setsockopt(s, (addr->ai_family==AF_INET6? IPPROTO_IPV6:IPPROTO_IP),
 783             (addr->ai_family == AF_INET6? IPV6_BINDANY:IP_BINDANY),
 784             (void*)&on, (socklen_t)sizeof(on)) < 0) {
 785                 log_warn("setsockopt(.. IP%s_BINDANY ..) failed: %s",
 786                 (addr->ai_family==AF_INET6?"V6":""), strerror(errno));
 787         }
 788 #elif defined(SO_BINDANY)
 789         if (transparent &&
 790             setsockopt(s, SOL_SOCKET, SO_BINDANY, (void*)&on, (socklen_t)
 791             sizeof(on)) < 0) {
 792                 log_warn("setsockopt(.. SO_BINDANY ..) failed: %s",
 793                 strerror(errno));
 794         }
 795 #endif /* IP_TRANSPARENT || IP_BINDANY || SO_BINDANY */
 796         if(
 797 #ifdef HAVE_SYSTEMD
 798                 !got_fd_from_systemd &&
 799 #endif
 800         bind(s, addr->ai_addr, addr->ai_addrlen) != 0) {
 801 #ifndef USE_WINSOCK
 802                 /* detect freebsd jail with no ipv6 permission */
 803                 if(addr->ai_family==AF_INET6 && errno==EINVAL)
 804                         *noproto = 1;
 805                 else {
 806                         log_err_addr("can't bind socket", strerror(errno),
 807                                 (struct sockaddr_storage*)addr->ai_addr,
 808                                 addr->ai_addrlen);
 809                 }
 810                 close(s);
 811 #else
 812                 log_err_addr("can't bind socket",
 813                         wsa_strerror(WSAGetLastError()),
 814                         (struct sockaddr_storage*)addr->ai_addr,
 815                         addr->ai_addrlen);
 816                 closesocket(s);
 817 #endif
 818                 return -1;
 819         }
 820         if(!fd_set_nonblock(s)) {
 821 #ifndef USE_WINSOCK
 822                 close(s);
 823 #else
 824                 closesocket(s);
 825 #endif
 826                 return -1;
 827         }
 828         if(listen(s, TCP_BACKLOG) == -1) {
 829 #ifndef USE_WINSOCK
 830                 log_err("can't listen: %s", strerror(errno));
 831                 close(s);
 832 #else
 833                 log_err("can't listen: %s", wsa_strerror(WSAGetLastError()));
 834                 closesocket(s);
 835 #endif
 836                 return -1;
 837         }
 838 #ifdef USE_TCP_FASTOPEN
 839         /* qlen specifies how many outstanding TFO requests to allow. Limit is a defense
 840            against IP spoofing attacks as suggested in RFC7413 */
 841 #ifdef __APPLE__
 842         /* OS X implementation only supports qlen of 1 via this call. Actual
 843            value is configured by the net.inet.tcp.fastopen_backlog kernel parm. */
 844         qlen = 1;
 845 #else
 846         /* 5 is recommended on linux */
 847         qlen = 5;
 848 #endif
 849         if ((setsockopt(s, IPPROTO_TCP, TCP_FASTOPEN, &qlen,
 850                   sizeof(qlen))) == -1 ) {
 851 #ifdef ENOPROTOOPT
 852                 /* squelch ENOPROTOOPT: freebsd server mode with kernel support
 853                    disabled, except when verbosity enabled for debugging */
 854                 if(errno != ENOPROTOOPT || verbosity >= 3)
 855 #endif
 856                   if(errno == EPERM) {
 857                         log_warn("Setting TCP Fast Open as server failed: %s ; this could likely be because sysctl net.inet.tcp.fastopen.enabled, net.inet.tcp.fastopen.server_enable, or net.ipv4.tcp_fastopen is disabled", strerror(errno));
 858                   } else {
 859                         log_err("Setting TCP Fast Open as server failed: %s", strerror(errno));
 860                   }
 861         }
 862 #endif
 863         return s;
 864 }
 865
 866 int
 867 create_local_accept_sock(const char *path, int* noproto, int use_systemd)
 868 {
 869 #ifdef HAVE_SYSTEMD
 870         int ret;
 871
 872         if (use_systemd && (ret = systemd_get_activated(AF_LOCAL, SOCK_STREAM, 1, NULL, 0, path)) != -1)
 873                 return ret;
 874         else {
 875 #endif
 876 #ifdef HAVE_SYS_UN_H
 877         int s;
 878         struct sockaddr_un usock;
 879 #ifndef HAVE_SYSTEMD
 880         (void)use_systemd;
 881 #endif
 882
 883         verbose(VERB_ALGO, "creating unix socket %s", path);
 884 #ifdef HAVE_STRUCT_SOCKADDR_UN_SUN_LEN
 885         /* this member exists on BSDs, not Linux */
 886         usock.sun_len = (unsigned)sizeof(usock);
 887 #endif
 888         usock.sun_family = AF_LOCAL;
 889         /* length is 92-108, 104 on FreeBSD */
 890         (void)strlcpy(usock.sun_path, path, sizeof(usock.sun_path));
 891
 892         if ((s = socket(AF_LOCAL, SOCK_STREAM, 0)) == -1) {
 893                 log_err("Cannot create local socket %s (%s)",
 894                         path, strerror(errno));
 895                 return -1;
 896         }
 897
 898         if (unlink(path) && errno != ENOENT) {
 899                 /* The socket already exists and cannot be removed */
 900                 log_err("Cannot remove old local socket %s (%s)",
 901                         path, strerror(errno));
 902                 goto err;
 903         }
 904
 905         if (bind(s, (struct sockaddr *)&usock,
 906                 (socklen_t)sizeof(struct sockaddr_un)) == -1) {
 907                 log_err("Cannot bind local socket %s (%s)",
 908                         path, strerror(errno));
 909                 goto err;
 910         }
 911
 912         if (!fd_set_nonblock(s)) {
 913                 log_err("Cannot set non-blocking mode");
 914                 goto err;
 915         }
 916
 917         if (listen(s, TCP_BACKLOG) == -1) {
 918                 log_err("can't listen: %s", strerror(errno));
 919                 goto err;
 920         }
 921
 922         (void)noproto; /*unused*/
 923         return s;
 924
 925 err:
 926 #ifndef USE_WINSOCK
 927         close(s);
 928 #else
 929         closesocket(s);
 930 #endif
 931         return -1;
 932
 933 #ifdef HAVE_SYSTEMD
 934         }
 935 #endif
 936 #else
 937         (void)use_systemd;
 938         (void)path;
 939         log_err("Local sockets are not supported");
 940         *noproto = 1;
 941         return -1;
 942 #endif
 943 }
 944
 945
 946 /**
 947  * Create socket from getaddrinfo results
 948  */
 949 static int
 950 make_sock(int stype, const char* ifname, const char* port,
 951         struct addrinfo *hints, int v6only, int* noip6, size_t rcv, size_t snd,
 952         int* reuseport, int transparent, int tcp_mss, int freebind, int use_systemd)
 953 {
 954         struct addrinfo *res = NULL;
 955         int r, s, inuse, noproto;
 956         hints->ai_socktype = stype;
 957         *noip6 = 0;
 958         if((r=getaddrinfo(ifname, port, hints, &res)) != 0 || !res) {
 959 #ifdef USE_WINSOCK
 960                 if(r == EAI_NONAME && hints->ai_family == AF_INET6){
 961                         *noip6 = 1; /* 'Host not found' for IP6 on winXP */
 962                         return -1;
 963                 }
 964 #endif
 965                 log_err("node %s:%s getaddrinfo: %s %s",
 966                         ifname?ifname:"default", port, gai_strerror(r),
 967 #ifdef EAI_SYSTEM
 968                         r==EAI_SYSTEM?(char*)strerror(errno):""
 969 #else
 970                         ""
 971 #endif
 972                 );
 973                 return -1;
 974         }
 975         if(stype == SOCK_DGRAM) {
 976                 verbose_print_addr(res);
 977                 s = create_udp_sock(res->ai_family, res->ai_socktype,
 978                         (struct sockaddr*)res->ai_addr, res->ai_addrlen,
 979                         v6only, &inuse, &noproto, (int)rcv, (int)snd, 1,
 980                         reuseport, transparent, freebind, use_systemd);
 981                 if(s == -1 && inuse) {
 982                         log_err("bind: address already in use");
 983                 } else if(s == -1 && noproto && hints->ai_family == AF_INET6){
 984                         *noip6 = 1;
 985                 }
 986         } else  {
 987                 s = create_tcp_accept_sock(res, v6only, &noproto, reuseport,
 988                         transparent, tcp_mss, freebind, use_systemd);
 989                 if(s == -1 && noproto && hints->ai_family == AF_INET6){
 990                         *noip6 = 1;
 991                 }
 992         }
 993         freeaddrinfo(res);
 994         return s;
 995 }
 996
 997 /** make socket and first see if ifname contains port override info */
 998 static int
 999 make_sock_port(int stype, const char* ifname, const char* port,
1000         struct addrinfo *hints, int v6only, int* noip6, size_t rcv, size_t snd,
1001         int* reuseport, int transparent, int tcp_mss, int freebind, int use_systemd)
1002 {
1003         char* s = strchr(ifname, '@');
1004         if(s) {
1005                 /* override port with ifspec@port */
1006                 char p[16];
1007                 char newif[128];
1008                 if((size_t)(s-ifname) >= sizeof(newif)) {
1009                         log_err("ifname too long: %s", ifname);
1010                         *noip6 = 0;
1011                         return -1;
1012                 }
1013                 if(strlen(s+1) >= sizeof(p)) {
1014                         log_err("portnumber too long: %s", ifname);
1015                         *noip6 = 0;
1016                         return -1;
1017                 }
1018                 (void)strlcpy(newif, ifname, sizeof(newif));
1019                 newif[s-ifname] = 0;
1020                 (void)strlcpy(p, s+1, sizeof(p));
1021                 p[strlen(s+1)]=0;
1022                 return make_sock(stype, newif, p, hints, v6only, noip6,
1023                         rcv, snd, reuseport, transparent, tcp_mss, freebind, use_systemd);
1024         }
1025         return make_sock(stype, ifname, port, hints, v6only, noip6, rcv, snd,
1026                 reuseport, transparent, tcp_mss, freebind, use_systemd);
1027 }
1028
1029 /**
1030  * Add port to open ports list.
1031  * @param list: list head. changed.
1032  * @param s: fd.
1033  * @param ftype: if fd is UDP.
1034  * @return false on failure. list in unchanged then.
1035  */
1036 static int
1037 port_insert(struct listen_port** list, int s, enum listen_type ftype)
1038 {
1039         struct listen_port* item = (struct listen_port*)malloc(
1040                 sizeof(struct listen_port));
1041         if(!item)
1042                 return 0;
1043         item->next = *list;
1044         item->fd = s;
1045         item->ftype = ftype;
1046         *list = item;
1047         return 1;
1048 }
1049
1050 /** set fd to receive source address packet info */
1051 static int
1052 set_recvpktinfo(int s, int family)
1053 {
1054 #if defined(IPV6_RECVPKTINFO) || defined(IPV6_PKTINFO) || (defined(IP_RECVDSTADDR) && defined(IP_SENDSRCADDR)) || defined(IP_PKTINFO)
1055         int on = 1;
1056 #else
1057         (void)s;
1058 #endif
1059         if(family == AF_INET6) {
1060 #           ifdef IPV6_RECVPKTINFO
1061                 if(setsockopt(s, IPPROTO_IPV6, IPV6_RECVPKTINFO,
1062                         (void*)&on, (socklen_t)sizeof(on)) < 0) {
1063                         log_err("setsockopt(..., IPV6_RECVPKTINFO, ...) failed: %s",
1064                                 strerror(errno));
1065                         return 0;
1066                 }
1067 #           elif defined(IPV6_PKTINFO)
1068                 if(setsockopt(s, IPPROTO_IPV6, IPV6_PKTINFO,
1069                         (void*)&on, (socklen_t)sizeof(on)) < 0) {
1070                         log_err("setsockopt(..., IPV6_PKTINFO, ...) failed: %s",
1071                                 strerror(errno));
1072                         return 0;
1073                 }
1074 #           else
1075                 log_err("no IPV6_RECVPKTINFO and no IPV6_PKTINFO option, please "
1076                         "disable interface-automatic or do-ip6 in config");
1077                 return 0;
1078 #           endif /* defined IPV6_RECVPKTINFO */
1079
1080         } else if(family == AF_INET) {
1081 #           ifdef IP_PKTINFO
1082                 if(setsockopt(s, IPPROTO_IP, IP_PKTINFO,
1083                         (void*)&on, (socklen_t)sizeof(on)) < 0) {
1084                         log_err("setsockopt(..., IP_PKTINFO, ...) failed: %s",
1085                                 strerror(errno));
1086                         return 0;
1087                 }
1088 #           elif defined(IP_RECVDSTADDR) && defined(IP_SENDSRCADDR)
1089                 if(setsockopt(s, IPPROTO_IP, IP_RECVDSTADDR,
1090                         (void*)&on, (socklen_t)sizeof(on)) < 0) {
1091                         log_err("setsockopt(..., IP_RECVDSTADDR, ...) failed: %s",
1092                                 strerror(errno));
1093                         return 0;
1094                 }
1095 #           else
1096                 log_err("no IP_SENDSRCADDR or IP_PKTINFO option, please disable "
1097                         "interface-automatic or do-ip4 in config");
1098                 return 0;
1099 #           endif /* IP_PKTINFO */
1100
1101         }
1102         return 1;
1103 }
1104
1105 /** see if interface is ssl, its port number == the ssl port number */
1106 static int
1107 if_is_ssl(const char* ifname, const char* port, int ssl_port,
1108         struct config_strlist* tls_additional_port)
1109 {
1110         struct config_strlist* s;
1111         char* p = strchr(ifname, '@');
1112         if(!p && atoi(port) == ssl_port)
1113                 return 1;
1114         if(p && atoi(p+1) == ssl_port)
1115                 return 1;
1116         for(s = tls_additional_port; s; s = s->next) {
1117                 if(p && atoi(p+1) == atoi(s->str))
1118                         return 1;
1119                 if(!p && atoi(port) == atoi(s->str))
1120                         return 1;
1121         }
1122         return 0;
1123 }
1124
1125 /**
1126  * Helper for ports_open. Creates one interface (or NULL for default).
1127  * @param ifname: The interface ip address.
1128  * @param do_auto: use automatic interface detection.
1129  *      If enabled, then ifname must be the wildcard name.
1130  * @param do_udp: if udp should be used.
1131  * @param do_tcp: if udp should be used.
1132  * @param hints: for getaddrinfo. family and flags have to be set by caller.
1133  * @param port: Port number to use (as string).
1134  * @param list: list of open ports, appended to, changed to point to list head.
1135  * @param rcv: receive buffer size for UDP
1136  * @param snd: send buffer size for UDP
1137  * @param ssl_port: ssl service port number
1138  * @param tls_additional_port: list of additional ssl service port numbers.
1139  * @param reuseport: try to set SO_REUSEPORT if nonNULL and true.
1140  *      set to false on exit if reuseport failed due to no kernel support.
1141  * @param transparent: set IP_TRANSPARENT socket option.
1142  * @param tcp_mss: maximum segment size of tcp socket. default if zero.
1143  * @param freebind: set IP_FREEBIND socket option.
1144  * @param use_systemd: if true, fetch sockets from systemd.
1145  * @param dnscrypt_port: dnscrypt service port number
1146  * @return: returns false on error.
1147  */
1148 static int
1149 ports_create_if(const char* ifname, int do_auto, int do_udp, int do_tcp,
1150         struct addrinfo *hints, const char* port, struct listen_port** list,
1151         size_t rcv, size_t snd, int ssl_port,
1152         struct config_strlist* tls_additional_port, int* reuseport,
1153         int transparent, int tcp_mss, int freebind, int use_systemd,
1154         int dnscrypt_port)
1155 {
1156         int s, noip6=0;
1157 #ifdef USE_DNSCRYPT
1158         int is_dnscrypt = ((strchr(ifname, '@') &&
1159                         atoi(strchr(ifname, '@')+1) == dnscrypt_port) ||
1160                         (!strchr(ifname, '@') && atoi(port) == dnscrypt_port));
1161 #else
1162         int is_dnscrypt = 0;
1163         (void)dnscrypt_port;
1164 #endif
1165
1166         if(!do_udp && !do_tcp)
1167                 return 0;
1168         if(do_auto) {
1169                 if((s = make_sock_port(SOCK_DGRAM, ifname, port, hints, 1,
1170                         &noip6, rcv, snd, reuseport, transparent,
1171                         tcp_mss, freebind, use_systemd)) == -1) {
1172                         if(noip6) {
1173                                 log_warn("IPv6 protocol not available");
1174                                 return 1;
1175                         }
1176                         return 0;
1177                 }
1178                 /* getting source addr packet info is highly non-portable */
1179                 if(!set_recvpktinfo(s, hints->ai_family)) {
1180 #ifndef USE_WINSOCK
1181                         close(s);
1182 #else
1183                         closesocket(s);
1184 #endif
1185                         return 0;
1186                 }
1187                 if(!port_insert(list, s,
1188                    is_dnscrypt?listen_type_udpancil_dnscrypt:listen_type_udpancil)) {
1189 #ifndef USE_WINSOCK
1190                         close(s);
1191 #else
1192                         closesocket(s);
1193 #endif
1194                         return 0;
1195                 }
1196         } else if(do_udp) {
1197                 /* regular udp socket */
1198                 if((s = make_sock_port(SOCK_DGRAM, ifname, port, hints, 1,
1199                         &noip6, rcv, snd, reuseport, transparent,
1200                         tcp_mss, freebind, use_systemd)) == -1) {
1201                         if(noip6) {
1202                                 log_warn("IPv6 protocol not available");
1203                                 return 1;
1204                         }
1205                         return 0;
1206                 }
1207                 if(!port_insert(list, s,
1208                    is_dnscrypt?listen_type_udp_dnscrypt:listen_type_udp)) {
1209 #ifndef USE_WINSOCK
1210                         close(s);
1211 #else
1212                         closesocket(s);
1213 #endif
1214                         return 0;
1215                 }
1216         }
1217         if(do_tcp) {
1218                 int is_ssl = if_is_ssl(ifname, port, ssl_port,
1219                         tls_additional_port);
1220                 if((s = make_sock_port(SOCK_STREAM, ifname, port, hints, 1,
1221                         &noip6, 0, 0, reuseport, transparent, tcp_mss,
1222                         freebind, use_systemd)) == -1) {
1223                         if(noip6) {
1224                                 /*log_warn("IPv6 protocol not available");*/
1225                                 return 1;
1226                         }
1227                         return 0;
1228                 }
1229                 if(is_ssl)
1230                         verbose(VERB_ALGO, "setup TCP for SSL service");
1231                 if(!port_insert(list, s, is_ssl?listen_type_ssl:
1232                         (is_dnscrypt?listen_type_tcp_dnscrypt:listen_type_tcp))) {
1233 #ifndef USE_WINSOCK
1234                         close(s);
1235 #else
1236                         closesocket(s);
1237 #endif
1238                         return 0;
1239                 }
1240         }
1241         return 1;
1242 }
1243
1244 /**
1245  * Add items to commpoint list in front.
1246  * @param c: commpoint to add.
1247  * @param front: listen struct.
1248  * @return: false on failure.
1249  */
1250 static int
1251 listen_cp_insert(struct comm_point* c, struct listen_dnsport* front)
1252 {
1253         struct listen_list* item = (struct listen_list*)malloc(
1254                 sizeof(struct listen_list));
1255         if(!item)
1256                 return 0;
1257         item->com = c;
1258         item->next = front->cps;
1259         front->cps = item;
1260         return 1;
1261 }
1262
1263 struct listen_dnsport*
1264 listen_create(struct comm_base* base, struct listen_port* ports,
1265         size_t bufsize, int tcp_accept_count, int tcp_idle_timeout,
1266         struct tcl_list* tcp_conn_limit, void* sslctx,
1267         struct dt_env* dtenv, comm_point_callback_type* cb, void *cb_arg)
1268 {
1269         struct listen_dnsport* front = (struct listen_dnsport*)
1270                 malloc(sizeof(struct listen_dnsport));
1271         if(!front)
1272                 return NULL;
1273         front->cps = NULL;
1274         front->udp_buff = sldns_buffer_new(bufsize);
1275 #ifdef USE_DNSCRYPT
1276         front->dnscrypt_udp_buff = NULL;
1277 #endif
1278         if(!front->udp_buff) {
1279                 free(front);
1280                 return NULL;
1281         }
1282         if(!stream_wait_lock_inited) {
1283                 lock_basic_init(&stream_wait_count_lock);
1284                 stream_wait_lock_inited = 1;
1285         }
1286
1287         /* create comm points as needed */
1288         while(ports) {
1289                 struct comm_point* cp = NULL;
1290                 if(ports->ftype == listen_type_udp ||
1291                    ports->ftype == listen_type_udp_dnscrypt)
1292                         cp = comm_point_create_udp(base, ports->fd,
1293                                 front->udp_buff, cb, cb_arg);
1294                 else if(ports->ftype == listen_type_tcp ||
1295                                 ports->ftype == listen_type_tcp_dnscrypt)
1296                         cp = comm_point_create_tcp(base, ports->fd,
1297                                 tcp_accept_count, tcp_idle_timeout,
1298                                 tcp_conn_limit, bufsize, front->udp_buff,
1299                                 cb, cb_arg);
1300                 else if(ports->ftype == listen_type_ssl) {
1301                         cp = comm_point_create_tcp(base, ports->fd,
1302                                 tcp_accept_count, tcp_idle_timeout,
1303                                 tcp_conn_limit, bufsize, front->udp_buff,
1304                                 cb, cb_arg);
1305                         cp->ssl = sslctx;
1306                 } else if(ports->ftype == listen_type_udpancil ||
1307                                   ports->ftype == listen_type_udpancil_dnscrypt)
1308                         cp = comm_point_create_udp_ancil(base, ports->fd,
1309                                 front->udp_buff, cb, cb_arg);
1310                 if(!cp) {
1311                         log_err("can't create commpoint");
1312                         listen_delete(front);
1313                         return NULL;
1314                 }
1315                 cp->dtenv = dtenv;
1316                 cp->do_not_close = 1;
1317 #ifdef USE_DNSCRYPT
1318                 if (ports->ftype == listen_type_udp_dnscrypt ||
1319                         ports->ftype == listen_type_tcp_dnscrypt ||
1320                         ports->ftype == listen_type_udpancil_dnscrypt) {
1321                         cp->dnscrypt = 1;
1322                         cp->dnscrypt_buffer = sldns_buffer_new(bufsize);
1323                         if(!cp->dnscrypt_buffer) {
1324                                 log_err("can't alloc dnscrypt_buffer");
1325                                 comm_point_delete(cp);
1326                                 listen_delete(front);
1327                                 return NULL;
1328                         }
1329                         front->dnscrypt_udp_buff = cp->dnscrypt_buffer;
1330                 }
1331 #endif
1332                 if(!listen_cp_insert(cp, front)) {
1333                         log_err("malloc failed");
1334                         comm_point_delete(cp);
1335                         listen_delete(front);
1336                         return NULL;
1337                 }
1338                 ports = ports->next;
1339         }
1340         if(!front->cps) {
1341                 log_err("Could not open sockets to accept queries.");
1342                 listen_delete(front);
1343                 return NULL;
1344         }
1345
1346         return front;
1347 }
1348
1349 void
1350 listen_list_delete(struct listen_list* list)
1351 {
1352         struct listen_list *p = list, *pn;
1353         while(p) {
1354                 pn = p->next;
1355                 comm_point_delete(p->com);
1356                 free(p);
1357                 p = pn;
1358         }
1359 }
1360
1361 void
1362 listen_delete(struct listen_dnsport* front)
1363 {
1364         if(!front)
1365                 return;
1366         listen_list_delete(front->cps);
1367 #ifdef USE_DNSCRYPT
1368         if(front->dnscrypt_udp_buff &&
1369                 front->udp_buff != front->dnscrypt_udp_buff) {
1370                 sldns_buffer_free(front->dnscrypt_udp_buff);
1371         }
1372 #endif
1373         sldns_buffer_free(front->udp_buff);
1374         free(front);
1375         if(stream_wait_lock_inited) {
1376                 stream_wait_lock_inited = 0;
1377                 lock_basic_destroy(&stream_wait_count_lock);
1378         }
1379 }
1380
1381 struct listen_port*
1382 listening_ports_open(struct config_file* cfg, int* reuseport)
1383 {
1384         struct listen_port* list = NULL;
1385         struct addrinfo hints;
1386         int i, do_ip4, do_ip6;
1387         int do_tcp, do_auto;
1388         char portbuf[32];
1389         snprintf(portbuf, sizeof(portbuf), "%d", cfg->port);
1390         do_ip4 = cfg->do_ip4;
1391         do_ip6 = cfg->do_ip6;
1392         do_tcp = cfg->do_tcp;
1393         do_auto = cfg->if_automatic && cfg->do_udp;
1394         if(cfg->incoming_num_tcp == 0)
1395                 do_tcp = 0;
1396
1397         /* getaddrinfo */
1398         memset(&hints, 0, sizeof(hints));
1399         hints.ai_flags = AI_PASSIVE;
1400         /* no name lookups on our listening ports */
1401         if(cfg->num_ifs > 0)
1402                 hints.ai_flags |= AI_NUMERICHOST;
1403         hints.ai_family = AF_UNSPEC;
1404 #ifndef INET6
1405         do_ip6 = 0;
1406 #endif
1407         if(!do_ip4 && !do_ip6) {
1408                 return NULL;
1409         }
1410         /* create ip4 and ip6 ports so that return addresses are nice. */
1411         if(do_auto || cfg->num_ifs == 0) {
1412                 if(do_ip6) {
1413                         hints.ai_family = AF_INET6;
1414                         if(!ports_create_if(do_auto?"::0":"::1",
1415                                 do_auto, cfg->do_udp, do_tcp,
1416                                 &hints, portbuf, &list,
1417                                 cfg->so_rcvbuf, cfg->so_sndbuf,
1418                                 cfg->ssl_port, cfg->tls_additional_port,
1419                                 reuseport, cfg->ip_transparent,
1420                                 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1421                                 cfg->dnscrypt_port)) {
1422                                 listening_ports_free(list);
1423                                 return NULL;
1424                         }
1425                 }
1426                 if(do_ip4) {
1427                         hints.ai_family = AF_INET;
1428                         if(!ports_create_if(do_auto?"0.0.0.0":"127.0.0.1",
1429                                 do_auto, cfg->do_udp, do_tcp,
1430                                 &hints, portbuf, &list,
1431                                 cfg->so_rcvbuf, cfg->so_sndbuf,
1432                                 cfg->ssl_port, cfg->tls_additional_port,
1433                                 reuseport, cfg->ip_transparent,
1434                                 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1435                                 cfg->dnscrypt_port)) {
1436                                 listening_ports_free(list);
1437                                 return NULL;
1438                         }
1439                 }
1440         } else for(i = 0; i<cfg->num_ifs; i++) {
1441                 if(str_is_ip6(cfg->ifs[i])) {
1442                         if(!do_ip6)
1443                                 continue;
1444                         hints.ai_family = AF_INET6;
1445                         if(!ports_create_if(cfg->ifs[i], 0, cfg->do_udp,
1446                                 do_tcp, &hints, portbuf, &list,
1447                                 cfg->so_rcvbuf, cfg->so_sndbuf,
1448                                 cfg->ssl_port, cfg->tls_additional_port,
1449                                 reuseport, cfg->ip_transparent,
1450                                 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1451                                 cfg->dnscrypt_port)) {
1452                                 listening_ports_free(list);
1453                                 return NULL;
1454                         }
1455                 } else {
1456                         if(!do_ip4)
1457                                 continue;
1458                         hints.ai_family = AF_INET;
1459                         if(!ports_create_if(cfg->ifs[i], 0, cfg->do_udp,
1460                                 do_tcp, &hints, portbuf, &list,
1461                                 cfg->so_rcvbuf, cfg->so_sndbuf,
1462                                 cfg->ssl_port, cfg->tls_additional_port,
1463                                 reuseport, cfg->ip_transparent,
1464                                 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1465                                 cfg->dnscrypt_port)) {
1466                                 listening_ports_free(list);
1467                                 return NULL;
1468                         }
1469                 }
1470         }
1471         return list;
1472 }
1473
1474 void listening_ports_free(struct listen_port* list)
1475 {
1476         struct listen_port* nx;
1477         while(list) {
1478                 nx = list->next;
1479                 if(list->fd != -1) {
1480 #ifndef USE_WINSOCK
1481                         close(list->fd);
1482 #else
1483                         closesocket(list->fd);
1484 #endif
1485                 }
1486                 free(list);
1487                 list = nx;
1488         }
1489 }
1490
1491 size_t listen_get_mem(struct listen_dnsport* listen)
1492 {
1493         struct listen_list* p;
1494         size_t s = sizeof(*listen) + sizeof(*listen->base) +
1495                 sizeof(*listen->udp_buff) +
1496                 sldns_buffer_capacity(listen->udp_buff);
1497 #ifdef USE_DNSCRYPT
1498         s += sizeof(*listen->dnscrypt_udp_buff);
1499         if(listen->udp_buff != listen->dnscrypt_udp_buff){
1500                 s += sldns_buffer_capacity(listen->dnscrypt_udp_buff);
1501         }
1502 #endif
1503         for(p = listen->cps; p; p = p->next) {
1504                 s += sizeof(*p);
1505                 s += comm_point_get_mem(p->com);
1506         }
1507         return s;
1508 }
1509
1510 void listen_stop_accept(struct listen_dnsport* listen)
1511 {
1512         /* do not stop the ones that have no tcp_free list
1513          * (they have already stopped listening) */
1514         struct listen_list* p;
1515         for(p=listen->cps; p; p=p->next) {
1516                 if(p->com->type == comm_tcp_accept &&
1517                         p->com->tcp_free != NULL) {
1518                         comm_point_stop_listening(p->com);
1519                 }
1520         }
1521 }
1522
1523 void listen_start_accept(struct listen_dnsport* listen)
1524 {
1525         /* do not start the ones that have no tcp_free list, it is no
1526          * use to listen to them because they have no free tcp handlers */
1527         struct listen_list* p;
1528         for(p=listen->cps; p; p=p->next) {
1529                 if(p->com->type == comm_tcp_accept &&
1530                         p->com->tcp_free != NULL) {
1531                         comm_point_start_listening(p->com, -1, -1);
1532                 }
1533         }
1534 }
1535
1536 struct tcp_req_info*
1537 tcp_req_info_create(struct sldns_buffer* spoolbuf)
1538 {
1539         struct tcp_req_info* req = (struct tcp_req_info*)malloc(sizeof(*req));
1540         if(!req) {
1541                 log_err("malloc failure for new stream outoforder processing structure");
1542                 return NULL;
1543         }
1544         memset(req, 0, sizeof(*req));
1545         req->spool_buffer = spoolbuf;
1546         return req;
1547 }
1548
1549 void
1550 tcp_req_info_delete(struct tcp_req_info* req)
1551 {
1552         if(!req) return;
1553         tcp_req_info_clear(req);
1554         /* cp is pointer back to commpoint that owns this struct and
1555          * called delete on us */
1556         /* spool_buffer is shared udp buffer, not deleted here */
1557         free(req);
1558 }
1559
1560 void tcp_req_info_clear(struct tcp_req_info* req)
1561 {
1562         struct tcp_req_open_item* open, *nopen;
1563         struct tcp_req_done_item* item, *nitem;
1564         if(!req) return;
1565
1566         /* free outstanding request mesh reply entries */
1567         open = req->open_req_list;
1568         while(open) {
1569                 nopen = open->next;
1570                 mesh_state_remove_reply(open->mesh, open->mesh_state, req->cp);
1571                 free(open);
1572                 open = nopen;
1573         }
1574         req->open_req_list = NULL;
1575         req->num_open_req = 0;
1576
1577         /* free pending writable result packets */
1578         item = req->done_req_list;
1579         while(item) {
1580                 nitem = item->next;
1581                 lock_basic_lock(&stream_wait_count_lock);
1582                 stream_wait_count -= (sizeof(struct tcp_req_done_item)
1583                         +item->len);
1584                 lock_basic_unlock(&stream_wait_count_lock);
1585                 free(item->buf);
1586                 free(item);
1587                 item = nitem;
1588         }
1589         req->done_req_list = NULL;
1590         req->num_done_req = 0;
1591         req->read_is_closed = 0;
1592 }
1593
1594 void
1595 tcp_req_info_remove_mesh_state(struct tcp_req_info* req, struct mesh_state* m)
1596 {
1597         struct tcp_req_open_item* open, *prev = NULL;
1598         if(!req || !m) return;
1599         open = req->open_req_list;
1600         while(open) {
1601                 if(open->mesh_state == m) {
1602                         struct tcp_req_open_item* next;
1603                         if(prev) prev->next = open->next;
1604                         else req->open_req_list = open->next;
1605                         /* caller has to manage the mesh state reply entry */
1606                         next = open->next;
1607                         free(open);
1608                         req->num_open_req --;
1609
1610                         /* prev = prev; */
1611                         open = next;
1612                         continue;
1613                 }
1614                 prev = open;
1615                 open = open->next;
1616         }
1617 }
1618
1619 /** setup listening for read or write */
1620 static void
1621 tcp_req_info_setup_listen(struct tcp_req_info* req)
1622 {
1623         int wr = 0;
1624         int rd = 0;
1625
1626         if(req->cp->tcp_byte_count != 0) {
1627                 /* cannot change, halfway through */
1628                 return;
1629         }
1630
1631         if(!req->cp->tcp_is_reading)
1632                 wr = 1;
1633         if(req->num_open_req + req->num_done_req < TCP_MAX_REQ_SIMULTANEOUS &&
1634                 !req->read_is_closed)
1635                 rd = 1;
1636
1637         if(wr) {
1638                 req->cp->tcp_is_reading = 0;
1639                 comm_point_start_listening(req->cp, -1,
1640                         req->cp->tcp_timeout_msec);
1641         } else if(rd) {
1642                 req->cp->tcp_is_reading = 1;
1643                 comm_point_start_listening(req->cp, -1,
1644                         req->cp->tcp_timeout_msec);
1645                 /* and also read it (from SSL stack buffers), so
1646                  * no event read event is expected since the remainder of
1647                  * the TLS frame is sitting in the buffers. */
1648                 req->read_again = 1;
1649         } else {
1650                 comm_point_start_listening(req->cp, -1,
1651                         req->cp->tcp_timeout_msec);
1652                 comm_point_listen_for_rw(req->cp, 0, 0);
1653         }
1654 }
1655
1656 /** remove first item from list of pending results */
1657 static struct tcp_req_done_item*
1658 tcp_req_info_pop_done(struct tcp_req_info* req)
1659 {
1660         struct tcp_req_done_item* item;
1661         log_assert(req->num_done_req > 0 && req->done_req_list);
1662         item = req->done_req_list;
1663         lock_basic_lock(&stream_wait_count_lock);
1664         stream_wait_count -= (sizeof(struct tcp_req_done_item)+item->len);
1665         lock_basic_unlock(&stream_wait_count_lock);
1666         req->done_req_list = req->done_req_list->next;
1667         req->num_done_req --;
1668         return item;
1669 }
1670
1671 /** Send given buffer and setup to write */
1672 static void
1673 tcp_req_info_start_write_buf(struct tcp_req_info* req, uint8_t* buf,
1674         size_t len)
1675 {
1676         sldns_buffer_clear(req->cp->buffer);
1677         sldns_buffer_write(req->cp->buffer, buf, len);
1678         sldns_buffer_flip(req->cp->buffer);
1679
1680         req->cp->tcp_is_reading = 0; /* we are now writing */
1681 }
1682
1683 /** pick up the next result and start writing it to the channel */
1684 static void
1685 tcp_req_pickup_next_result(struct tcp_req_info* req)
1686 {
1687         if(req->num_done_req > 0) {
1688                 /* unlist the done item from the list of pending results */
1689                 struct tcp_req_done_item* item = tcp_req_info_pop_done(req);
1690                 tcp_req_info_start_write_buf(req, item->buf, item->len);
1691                 free(item->buf);
1692                 free(item);
1693         }
1694 }
1695
1696 /** the read channel has closed */
1697 int
1698 tcp_req_info_handle_read_close(struct tcp_req_info* req)
1699 {
1700         verbose(VERB_ALGO, "tcp channel read side closed %d", req->cp->fd);
1701         /* reset byte count for (potential) partial read */
1702         req->cp->tcp_byte_count = 0;
1703         /* if we still have results to write, pick up next and write it */
1704         if(req->num_done_req != 0) {
1705                 tcp_req_pickup_next_result(req);
1706                 tcp_req_info_setup_listen(req);
1707                 return 1;
1708         }
1709         /* if nothing to do, this closes the connection */
1710         if(req->num_open_req == 0 && req->num_done_req == 0)
1711                 return 0;
1712         /* otherwise, we must be waiting for dns resolve, wait with timeout */
1713         req->read_is_closed = 1;
1714         tcp_req_info_setup_listen(req);
1715         return 1;
1716 }
1717
1718 void
1719 tcp_req_info_handle_writedone(struct tcp_req_info* req)
1720 {
1721         /* back to reading state, we finished this write event */
1722         sldns_buffer_clear(req->cp->buffer);
1723         if(req->num_done_req == 0 && req->read_is_closed) {
1724                 /* no more to write and nothing to read, close it */
1725                 comm_point_drop_reply(&req->cp->repinfo);
1726                 return;
1727         }
1728         req->cp->tcp_is_reading = 1;
1729         /* see if another result needs writing */
1730         tcp_req_pickup_next_result(req);
1731
1732         /* see if there is more to write, if not stop_listening for writing */
1733         /* see if new requests are allowed, if so, start_listening
1734          * for reading */
1735         tcp_req_info_setup_listen(req);
1736 }
1737
1738 void
1739 tcp_req_info_handle_readdone(struct tcp_req_info* req)
1740 {
1741         struct comm_point* c = req->cp;
1742
1743         /* we want to read up several requests, unless there are
1744          * pending answers */
1745
1746         req->is_drop = 0;
1747         req->is_reply = 0;
1748         req->in_worker_handle = 1;
1749         /* handle the current request */
1750         /* this calls the worker handle request routine that could give
1751          * a cache response, or localdata response, or drop the reply,
1752          * or schedule a mesh entry for later */
1753         fptr_ok(fptr_whitelist_comm_point(c->callback));
1754         if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
1755                 req->in_worker_handle = 0;
1756                 /* there is an answer, put it up.  It is already in the
1757                  * c->buffer, just send it. */
1758                 /* since we were just reading a query, the channel is
1759                  * clear to write to */
1760         send_it:
1761                 c->tcp_is_reading = 0;
1762                 comm_point_start_listening(c, -1, c->tcp_timeout_msec);
1763                 return;
1764         }
1765         req->in_worker_handle = 0;
1766         /* it should be waiting in the mesh for recursion.
1767          * If mesh failed to add a new entry and called commpoint_drop_reply.
1768          * Then the mesh state has been cleared. */
1769         if(req->is_drop) {
1770                 /* we can now call drop_reply without recursing into ourselves
1771                  * whilst in the callback */
1772                 /* we have to close the stream because there is no reply,
1773                  * no servfail to send, but the query needs an action, for
1774                  * a stream that is close the connection */
1775                 sldns_buffer_clear(c->buffer);
1776                 comm_point_drop_reply(&c->repinfo);
1777                 return;
1778         }
1779         /* If mesh failed(mallocfail) and called commpoint_send_reply with
1780          * something like servfail then we pick up that reply below. */
1781         if(req->is_reply) {
1782                 goto send_it;
1783         }
1784
1785         sldns_buffer_clear(c->buffer);
1786         /* if pending answers, pick up an answer and start sending it */
1787         tcp_req_pickup_next_result(req);
1788
1789         /* if answers pending, start sending answers */
1790         /* read more requests if we can have more requests */
1791         tcp_req_info_setup_listen(req);
1792 }
1793
1794 int
1795 tcp_req_info_add_meshstate(struct tcp_req_info* req,
1796         struct mesh_area* mesh, struct mesh_state* m)
1797 {
1798         struct tcp_req_open_item* item;
1799         log_assert(req && mesh && m);
1800         item = (struct tcp_req_open_item*)malloc(sizeof(*item));
1801         if(!item) return 0;
1802         item->next = req->open_req_list;
1803         item->mesh = mesh;
1804         item->mesh_state = m;
1805         req->open_req_list = item;
1806         req->num_open_req++;
1807         return 1;
1808 }
1809
1810 /** Add a result to the result list.  At the end. */
1811 static int
1812 tcp_req_info_add_result(struct tcp_req_info* req, uint8_t* buf, size_t len)
1813 {
1814         struct tcp_req_done_item* last = NULL;
1815         struct tcp_req_done_item* item;
1816         size_t space;
1817
1818         /* see if we have space */
1819         space = sizeof(struct tcp_req_done_item) + len;
1820         lock_basic_lock(&stream_wait_count_lock);
1821         if(stream_wait_count + space > stream_wait_max) {
1822                 lock_basic_unlock(&stream_wait_count_lock);
1823                 verbose(VERB_ALGO, "drop stream reply, no space left, in stream-wait-size");
1824                 return 0;
1825         }
1826         stream_wait_count += space;
1827         lock_basic_unlock(&stream_wait_count_lock);
1828
1829         /* find last element */
1830         last = req->done_req_list;
1831         while(last && last->next)
1832                 last = last->next;
1833
1834         /* create new element */
1835         item = (struct tcp_req_done_item*)malloc(sizeof(*item));
1836         if(!item) {
1837                 log_err("malloc failure, for stream result list");
1838                 return 0;
1839         }
1840         item->next = NULL;
1841         item->len = len;
1842         item->buf = memdup(buf, len);
1843         if(!item->buf) {
1844                 free(item);
1845                 log_err("malloc failure, adding reply to stream result list");
1846                 return 0;
1847         }
1848
1849         /* link in */
1850         if(last) last->next = item;
1851         else req->done_req_list = item;
1852         req->num_done_req++;
1853         return 1;
1854 }
1855
1856 void
1857 tcp_req_info_send_reply(struct tcp_req_info* req)
1858 {
1859         if(req->in_worker_handle) {
1860                 /* It is in the right buffer to answer straight away */
1861                 req->is_reply = 1;
1862                 return;
1863         }
1864         /* now that the query has been handled, that mesh_reply entry
1865          * should be removed, from the tcp_req_info list,
1866          * the mesh state cleanup removes then with region_cleanup and
1867          * replies_sent true. */
1868         /* see if we can send it straight away (we are not doing
1869          * anything else).  If so, copy to buffer and start */
1870         if(req->cp->tcp_is_reading && req->cp->tcp_byte_count == 0) {
1871                 /* buffer is free, and was ready to read new query into,
1872                  * but we are now going to use it to send this answer */
1873                 tcp_req_info_start_write_buf(req,
1874                         sldns_buffer_begin(req->spool_buffer),
1875                         sldns_buffer_limit(req->spool_buffer));
1876                 /* switch to listen to write events */
1877                 comm_point_stop_listening(req->cp);
1878                 comm_point_start_listening(req->cp, -1,
1879                         req->cp->tcp_timeout_msec);
1880                 return;
1881         }
1882         /* queue up the answer behind the others already pending */
1883         if(!tcp_req_info_add_result(req, sldns_buffer_begin(req->spool_buffer),
1884                 sldns_buffer_limit(req->spool_buffer))) {
1885                 /* drop the connection, we are out of resources */
1886                 comm_point_drop_reply(&req->cp->repinfo);
1887         }
1888 }
1889
1890 size_t tcp_req_info_get_stream_buffer_size(void)
1891 {
1892         size_t s;
1893         if(!stream_wait_lock_inited)
1894                 return stream_wait_count;
1895         lock_basic_lock(&stream_wait_count_lock);
1896         s = stream_wait_count;
1897         lock_basic_unlock(&stream_wait_count_lock);
1898         return s;
1899 }