contrib/unbound/util/netevent.c

   1 /*
   2  * util/netevent.c - event notification
   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 LIMITED
  25  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  26  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE
  27  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  30  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  31  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  32  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  * POSSIBILITY OF SUCH DAMAGE.
  34  */
  35
  36 /**
  37  * \file
  38  *
  39  * This file contains event notification functions.
  40  */
  41 #include "config.h"
  42 #include <ldns/wire2host.h>
  43 #include "util/netevent.h"
  44 #include "util/log.h"
  45 #include "util/net_help.h"
  46 #include "util/fptr_wlist.h"
  47 #ifdef HAVE_OPENSSL_SSL_H
  48 #include <openssl/ssl.h>
  49 #endif
  50 #ifdef HAVE_OPENSSL_ERR_H
  51 #include <openssl/err.h>
  52 #endif
  53
  54 /* -------- Start of local definitions -------- */
  55 /** if CMSG_ALIGN is not defined on this platform, a workaround */
  56 #ifndef CMSG_ALIGN
  57 #  ifdef _CMSG_DATA_ALIGN
  58 #    define CMSG_ALIGN _CMSG_DATA_ALIGN
  59 #  else
  60 #    define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1))
  61 #  endif
  62 #endif
  63
  64 /** if CMSG_LEN is not defined on this platform, a workaround */
  65 #ifndef CMSG_LEN
  66 #  define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len))
  67 #endif
  68
  69 /** if CMSG_SPACE is not defined on this platform, a workaround */
  70 #ifndef CMSG_SPACE
  71 #  ifdef _CMSG_HDR_ALIGN
  72 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr)))
  73 #  else
  74 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr)))
  75 #  endif
  76 #endif
  77
  78 /** The TCP reading or writing query timeout in seconds */
  79 #define TCP_QUERY_TIMEOUT 120
  80
  81 #ifndef NONBLOCKING_IS_BROKEN
  82 /** number of UDP reads to perform per read indication from select */
  83 #define NUM_UDP_PER_SELECT 100
  84 #else
  85 #define NUM_UDP_PER_SELECT 1
  86 #endif
  87
  88 /* We define libevent structures here to hide the libevent stuff. */
  89
  90 #ifdef USE_MINI_EVENT
  91 #  ifdef USE_WINSOCK
  92 #    include "util/winsock_event.h"
  93 #  else
  94 #    include "util/mini_event.h"
  95 #  endif /* USE_WINSOCK */
  96 #else /* USE_MINI_EVENT */
  97    /* we use libevent */
  98 #  ifdef HAVE_EVENT_H
  99 #    include <event.h>
 100 #  else
 101 #    include "event2/event.h"
 102 #    include "event2/event_struct.h"
 103 #    include "event2/event_compat.h"
 104 #  endif
 105 #endif /* USE_MINI_EVENT */
 106
 107 /**
 108  * The internal event structure for keeping libevent info for the event.
 109  * Possibly other structures (list, tree) this is part of.
 110  */
 111 struct internal_event {
 112         /** the comm base */
 113         struct comm_base* base;
 114         /** libevent event type, alloced here */
 115         struct event ev;
 116 };
 117
 118 /**
 119  * Internal base structure, so that every thread has its own events.
 120  */
 121 struct internal_base {
 122         /** libevent event_base type. */
 123         struct event_base* base;
 124         /** seconds time pointer points here */
 125         uint32_t secs;
 126         /** timeval with current time */
 127         struct timeval now;
 128         /** the event used for slow_accept timeouts */
 129         struct event slow_accept;
 130         /** true if slow_accept is enabled */
 131         int slow_accept_enabled;
 132 };
 133
 134 /**
 135  * Internal timer structure, to store timer event in.
 136  */
 137 struct internal_timer {
 138         /** the comm base */
 139         struct comm_base* base;
 140         /** libevent event type, alloced here */
 141         struct event ev;
 142         /** is timer enabled */
 143         uint8_t enabled;
 144 };
 145
 146 /**
 147  * Internal signal structure, to store signal event in.
 148  */
 149 struct internal_signal {
 150         /** libevent event type, alloced here */
 151         struct event ev;
 152         /** next in signal list */
 153         struct internal_signal* next;
 154 };
 155
 156 /** create a tcp handler with a parent */
 157 static struct comm_point* comm_point_create_tcp_handler(
 158         struct comm_base *base, struct comm_point* parent, size_t bufsize,
 159         comm_point_callback_t* callback, void* callback_arg);
 160
 161 /* -------- End of local definitions -------- */
 162
 163 #ifdef USE_MINI_EVENT
 164 /** minievent updates the time when it blocks. */
 165 #define comm_base_now(x) /* nothing to do */
 166 #else /* !USE_MINI_EVENT */
 167 /** fillup the time values in the event base */
 168 static void
 169 comm_base_now(struct comm_base* b)
 170 {
 171         if(gettimeofday(&b->eb->now, NULL) < 0) {
 172                 log_err("gettimeofday: %s", strerror(errno));
 173         }
 174         b->eb->secs = (uint32_t)b->eb->now.tv_sec;
 175 }
 176 #endif /* USE_MINI_EVENT */
 177
 178 struct comm_base*
 179 comm_base_create(int sigs)
 180 {
 181         struct comm_base* b = (struct comm_base*)calloc(1,
 182                 sizeof(struct comm_base));
 183         if(!b)
 184                 return NULL;
 185         b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
 186         if(!b->eb) {
 187                 free(b);
 188                 return NULL;
 189         }
 190 #ifdef USE_MINI_EVENT
 191         (void)sigs;
 192         /* use mini event time-sharing feature */
 193         b->eb->base = event_init(&b->eb->secs, &b->eb->now);
 194 #else
 195 #  if defined(HAVE_EV_LOOP) || defined(HAVE_EV_DEFAULT_LOOP)
 196         /* libev */
 197         if(sigs)
 198                 b->eb->base=(struct event_base *)ev_default_loop(EVFLAG_AUTO);
 199         else
 200                 b->eb->base=(struct event_base *)ev_loop_new(EVFLAG_AUTO);
 201 #  else
 202         (void)sigs;
 203 #    ifdef HAVE_EVENT_BASE_NEW
 204         b->eb->base = event_base_new();
 205 #    else
 206         b->eb->base = event_init();
 207 #    endif
 208 #  endif
 209 #endif
 210         if(!b->eb->base) {
 211                 free(b->eb);
 212                 free(b);
 213                 return NULL;
 214         }
 215         comm_base_now(b);
 216         /* avoid event_get_method call which causes crashes even when
 217          * not printing, because its result is passed */
 218         verbose(VERB_ALGO,
 219 #if defined(HAVE_EV_LOOP) || defined(HAVE_EV_DEFAULT_LOOP)
 220                 "libev"
 221 #elif defined(USE_MINI_EVENT)
 222                 "event "
 223 #else
 224                 "libevent "
 225 #endif
 226                 "%s uses %s method.",
 227                 event_get_version(),
 228 #ifdef HAVE_EVENT_BASE_GET_METHOD
 229                 event_base_get_method(b->eb->base)
 230 #else
 231                 "not_obtainable"
 232 #endif
 233         );
 234         return b;
 235 }
 236
 237 void
 238 comm_base_delete(struct comm_base* b)
 239 {
 240         if(!b)
 241                 return;
 242         if(b->eb->slow_accept_enabled) {
 243                 if(event_del(&b->eb->slow_accept) != 0) {
 244                         log_err("could not event_del slow_accept");
 245                 }
 246         }
 247 #ifdef USE_MINI_EVENT
 248         event_base_free(b->eb->base);
 249 #elif defined(HAVE_EVENT_BASE_FREE) && defined(HAVE_EVENT_BASE_ONCE)
 250         /* only libevent 1.2+ has it, but in 1.2 it is broken -
 251            assertion fails on signal handling ev that is not deleted
 252            in libevent 1.3c (event_base_once appears) this is fixed. */
 253         event_base_free(b->eb->base);
 254 #endif /* HAVE_EVENT_BASE_FREE and HAVE_EVENT_BASE_ONCE */
 255         b->eb->base = NULL;
 256         free(b->eb);
 257         free(b);
 258 }
 259
 260 void
 261 comm_base_timept(struct comm_base* b, uint32_t** tt, struct timeval** tv)
 262 {
 263         *tt = &b->eb->secs;
 264         *tv = &b->eb->now;
 265 }
 266
 267 void
 268 comm_base_dispatch(struct comm_base* b)
 269 {
 270         int retval;
 271         retval = event_base_dispatch(b->eb->base);
 272         if(retval != 0) {
 273                 fatal_exit("event_dispatch returned error %d, "
 274                         "errno is %s", retval, strerror(errno));
 275         }
 276 }
 277
 278 void comm_base_exit(struct comm_base* b)
 279 {
 280         if(event_base_loopexit(b->eb->base, NULL) != 0) {
 281                 log_err("Could not loopexit");
 282         }
 283 }
 284
 285 void comm_base_set_slow_accept_handlers(struct comm_base* b,
 286         void (*stop_acc)(void*), void (*start_acc)(void*), void* arg)
 287 {
 288         b->stop_accept = stop_acc;
 289         b->start_accept = start_acc;
 290         b->cb_arg = arg;
 291 }
 292
 293 struct event_base* comm_base_internal(struct comm_base* b)
 294 {
 295         return b->eb->base;
 296 }
 297
 298 /** see if errno for udp has to be logged or not uses globals */
 299 static int
 300 udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
 301 {
 302         /* do not log transient errors (unless high verbosity) */
 303 #if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN)
 304         switch(errno) {
 305 #  ifdef ENETUNREACH
 306                 case ENETUNREACH:
 307 #  endif
 308 #  ifdef EHOSTDOWN
 309                 case EHOSTDOWN:
 310 #  endif
 311 #  ifdef EHOSTUNREACH
 312                 case EHOSTUNREACH:
 313 #  endif
 314 #  ifdef ENETDOWN
 315                 case ENETDOWN:
 316 #  endif
 317                         if(verbosity < VERB_ALGO)
 318                                 return 0;
 319                 default:
 320                         break;
 321         }
 322 #endif
 323         /* squelch errors where people deploy AAAA ::ffff:bla for
 324          * authority servers, which we try for intranets. */
 325         if(errno == EINVAL && addr_is_ip4mapped(
 326                 (struct sockaddr_storage*)addr, addrlen) &&
 327                 verbosity < VERB_DETAIL)
 328                 return 0;
 329         /* SO_BROADCAST sockopt can give access to 255.255.255.255,
 330          * but a dns cache does not need it. */
 331         if(errno == EACCES && addr_is_broadcast(
 332                 (struct sockaddr_storage*)addr, addrlen) &&
 333                 verbosity < VERB_DETAIL)
 334                 return 0;
 335         return 1;
 336 }
 337
 338 int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
 339 {
 340         return udp_send_errno_needs_log(addr, addrlen);
 341 }
 342
 343 /* send a UDP reply */
 344 int
 345 comm_point_send_udp_msg(struct comm_point *c, ldns_buffer* packet,
 346         struct sockaddr* addr, socklen_t addrlen)
 347 {
 348         ssize_t sent;
 349         log_assert(c->fd != -1);
 350 #ifdef UNBOUND_DEBUG
 351         if(ldns_buffer_remaining(packet) == 0)
 352                 log_err("error: send empty UDP packet");
 353 #endif
 354         log_assert(addr && addrlen > 0);
 355         sent = sendto(c->fd, (void*)ldns_buffer_begin(packet),
 356                 ldns_buffer_remaining(packet), 0,
 357                 addr, addrlen);
 358         if(sent == -1) {
 359                 if(!udp_send_errno_needs_log(addr, addrlen))
 360                         return 0;
 361 #ifndef USE_WINSOCK
 362                 verbose(VERB_OPS, "sendto failed: %s", strerror(errno));
 363 #else
 364                 verbose(VERB_OPS, "sendto failed: %s",
 365                         wsa_strerror(WSAGetLastError()));
 366 #endif
 367                 log_addr(VERB_OPS, "remote address is",
 368                         (struct sockaddr_storage*)addr, addrlen);
 369                 return 0;
 370         } else if((size_t)sent != ldns_buffer_remaining(packet)) {
 371                 log_err("sent %d in place of %d bytes",
 372                         (int)sent, (int)ldns_buffer_remaining(packet));
 373                 return 0;
 374         }
 375         return 1;
 376 }
 377
 378 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG))
 379 /** print debug ancillary info */
 380 static void p_ancil(const char* str, struct comm_reply* r)
 381 {
 382         if(r->srctype != 4 && r->srctype != 6) {
 383                 log_info("%s: unknown srctype %d", str, r->srctype);
 384                 return;
 385         }
 386         if(r->srctype == 6) {
 387                 char buf[1024];
 388                 if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr,
 389                         buf, (socklen_t)sizeof(buf)) == 0) {
 390                         strncpy(buf, "(inet_ntop error)", sizeof(buf));
 391                 }
 392                 buf[sizeof(buf)-1]=0;
 393                 log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex);
 394         } else if(r->srctype == 4) {
 395 #ifdef IP_PKTINFO
 396                 char buf1[1024], buf2[1024];
 397                 if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr,
 398                         buf1, (socklen_t)sizeof(buf1)) == 0) {
 399                         strncpy(buf1, "(inet_ntop error)", sizeof(buf1));
 400                 }
 401                 buf1[sizeof(buf1)-1]=0;
 402 #ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
 403                 if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst,
 404                         buf2, (socklen_t)sizeof(buf2)) == 0) {
 405                         strncpy(buf2, "(inet_ntop error)", sizeof(buf2));
 406                 }
 407                 buf2[sizeof(buf2)-1]=0;
 408 #else
 409                 buf2[0]=0;
 410 #endif
 411                 log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex,
 412                         buf1, buf2);
 413 #elif defined(IP_RECVDSTADDR)
 414                 char buf1[1024];
 415                 if(inet_ntop(AF_INET, &r->pktinfo.v4addr,
 416                         buf1, (socklen_t)sizeof(buf1)) == 0) {
 417                         strncpy(buf1, "(inet_ntop error)", sizeof(buf1));
 418                 }
 419                 buf1[sizeof(buf1)-1]=0;
 420                 log_info("%s: %s", str, buf1);
 421 #endif /* IP_PKTINFO or PI_RECVDSTDADDR */
 422         }
 423 }
 424 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */
 425
 426 /** send a UDP reply over specified interface*/
 427 static int
 428 comm_point_send_udp_msg_if(struct comm_point *c, ldns_buffer* packet,
 429         struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r)
 430 {
 431 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG)
 432         ssize_t sent;
 433         struct msghdr msg;
 434         struct iovec iov[1];
 435         char control[256];
 436 #ifndef S_SPLINT_S
 437         struct cmsghdr *cmsg;
 438 #endif /* S_SPLINT_S */
 439
 440         log_assert(c->fd != -1);
 441 #ifdef UNBOUND_DEBUG
 442         if(ldns_buffer_remaining(packet) == 0)
 443                 log_err("error: send empty UDP packet");
 444 #endif
 445         log_assert(addr && addrlen > 0);
 446
 447         msg.msg_name = addr;
 448         msg.msg_namelen = addrlen;
 449         iov[0].iov_base = ldns_buffer_begin(packet);
 450         iov[0].iov_len = ldns_buffer_remaining(packet);
 451         msg.msg_iov = iov;
 452         msg.msg_iovlen = 1;
 453         msg.msg_control = control;
 454 #ifndef S_SPLINT_S
 455         msg.msg_controllen = sizeof(control);
 456 #endif /* S_SPLINT_S */
 457         msg.msg_flags = 0;
 458
 459 #ifndef S_SPLINT_S
 460         cmsg = CMSG_FIRSTHDR(&msg);
 461         if(r->srctype == 4) {
 462 #ifdef IP_PKTINFO
 463                 msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
 464                 log_assert(msg.msg_controllen <= sizeof(control));
 465                 cmsg->cmsg_level = IPPROTO_IP;
 466                 cmsg->cmsg_type = IP_PKTINFO;
 467                 memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info,
 468                         sizeof(struct in_pktinfo));
 469                 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
 470 #elif defined(IP_SENDSRCADDR)
 471                 msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
 472                 log_assert(msg.msg_controllen <= sizeof(control));
 473                 cmsg->cmsg_level = IPPROTO_IP;
 474                 cmsg->cmsg_type = IP_SENDSRCADDR;
 475                 memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr,
 476                         sizeof(struct in_addr));
 477                 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
 478 #else
 479                 verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR");
 480                 msg.msg_control = NULL;
 481 #endif /* IP_PKTINFO or IP_SENDSRCADDR */
 482         } else if(r->srctype == 6) {
 483                 msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
 484                 log_assert(msg.msg_controllen <= sizeof(control));
 485                 cmsg->cmsg_level = IPPROTO_IPV6;
 486                 cmsg->cmsg_type = IPV6_PKTINFO;
 487                 memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info,
 488                         sizeof(struct in6_pktinfo));
 489                 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
 490         } else {
 491                 /* try to pass all 0 to use default route */
 492                 msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
 493                 log_assert(msg.msg_controllen <= sizeof(control));
 494                 cmsg->cmsg_level = IPPROTO_IPV6;
 495                 cmsg->cmsg_type = IPV6_PKTINFO;
 496                 memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo));
 497                 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
 498         }
 499 #endif /* S_SPLINT_S */
 500         if(verbosity >= VERB_ALGO)
 501                 p_ancil("send_udp over interface", r);
 502         sent = sendmsg(c->fd, &msg, 0);
 503         if(sent == -1) {
 504                 if(!udp_send_errno_needs_log(addr, addrlen))
 505                         return 0;
 506                 verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno));
 507                 log_addr(VERB_OPS, "remote address is",
 508                         (struct sockaddr_storage*)addr, addrlen);
 509                 return 0;
 510         } else if((size_t)sent != ldns_buffer_remaining(packet)) {
 511                 log_err("sent %d in place of %d bytes",
 512                         (int)sent, (int)ldns_buffer_remaining(packet));
 513                 return 0;
 514         }
 515         return 1;
 516 #else
 517         (void)c;
 518         (void)packet;
 519         (void)addr;
 520         (void)addrlen;
 521         (void)r;
 522         log_err("sendmsg: IPV6_PKTINFO not supported");
 523         return 0;
 524 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */
 525 }
 526
 527 void
 528 comm_point_udp_ancil_callback(int fd, short event, void* arg)
 529 {
 530 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
 531         struct comm_reply rep;
 532         struct msghdr msg;
 533         struct iovec iov[1];
 534         ssize_t rcv;
 535         char ancil[256];
 536         int i;
 537 #ifndef S_SPLINT_S
 538         struct cmsghdr* cmsg;
 539 #endif /* S_SPLINT_S */
 540
 541         rep.c = (struct comm_point*)arg;
 542         log_assert(rep.c->type == comm_udp);
 543
 544         if(!(event&EV_READ))
 545                 return;
 546         log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
 547         comm_base_now(rep.c->ev->base);
 548         for(i=0; i<NUM_UDP_PER_SELECT; i++) {
 549                 ldns_buffer_clear(rep.c->buffer);
 550                 rep.addrlen = (socklen_t)sizeof(rep.addr);
 551                 log_assert(fd != -1);
 552                 log_assert(ldns_buffer_remaining(rep.c->buffer) > 0);
 553                 msg.msg_name = &rep.addr;
 554                 msg.msg_namelen = (socklen_t)sizeof(rep.addr);
 555                 iov[0].iov_base = ldns_buffer_begin(rep.c->buffer);
 556                 iov[0].iov_len = ldns_buffer_remaining(rep.c->buffer);
 557                 msg.msg_iov = iov;
 558                 msg.msg_iovlen = 1;
 559                 msg.msg_control = ancil;
 560 #ifndef S_SPLINT_S
 561                 msg.msg_controllen = sizeof(ancil);
 562 #endif /* S_SPLINT_S */
 563                 msg.msg_flags = 0;
 564                 rcv = recvmsg(fd, &msg, 0);
 565                 if(rcv == -1) {
 566                         if(errno != EAGAIN && errno != EINTR) {
 567                                 log_err("recvmsg failed: %s", strerror(errno));
 568                         }
 569                         return;
 570                 }
 571                 rep.addrlen = msg.msg_namelen;
 572                 ldns_buffer_skip(rep.c->buffer, rcv);
 573                 ldns_buffer_flip(rep.c->buffer);
 574                 rep.srctype = 0;
 575 #ifndef S_SPLINT_S
 576                 for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL;
 577                         cmsg = CMSG_NXTHDR(&msg, cmsg)) {
 578                         if( cmsg->cmsg_level == IPPROTO_IPV6 &&
 579                                 cmsg->cmsg_type == IPV6_PKTINFO) {
 580                                 rep.srctype = 6;
 581                                 memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg),
 582                                         sizeof(struct in6_pktinfo));
 583                                 break;
 584 #ifdef IP_PKTINFO
 585                         } else if( cmsg->cmsg_level == IPPROTO_IP &&
 586                                 cmsg->cmsg_type == IP_PKTINFO) {
 587                                 rep.srctype = 4;
 588                                 memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg),
 589                                         sizeof(struct in_pktinfo));
 590                                 break;
 591 #elif defined(IP_RECVDSTADDR)
 592                         } else if( cmsg->cmsg_level == IPPROTO_IP &&
 593                                 cmsg->cmsg_type == IP_RECVDSTADDR) {
 594                                 rep.srctype = 4;
 595                                 memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg),
 596                                         sizeof(struct in_addr));
 597                                 break;
 598 #endif /* IP_PKTINFO or IP_RECVDSTADDR */
 599                         }
 600                 }
 601                 if(verbosity >= VERB_ALGO)
 602                         p_ancil("receive_udp on interface", &rep);
 603 #endif /* S_SPLINT_S */
 604                 fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
 605                 if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
 606                         /* send back immediate reply */
 607                         (void)comm_point_send_udp_msg_if(rep.c, rep.c->buffer,
 608                                 (struct sockaddr*)&rep.addr, rep.addrlen, &rep);
 609                 }
 610                 if(rep.c->fd == -1) /* commpoint closed */
 611                         break;
 612         }
 613 #else
 614         (void)fd;
 615         (void)event;
 616         (void)arg;
 617         fatal_exit("recvmsg: No support for IPV6_PKTINFO. "
 618                 "Please disable interface-automatic");
 619 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */
 620 }
 621
 622 void
 623 comm_point_udp_callback(int fd, short event, void* arg)
 624 {
 625         struct comm_reply rep;
 626         ssize_t rcv;
 627         int i;
 628
 629         rep.c = (struct comm_point*)arg;
 630         log_assert(rep.c->type == comm_udp);
 631
 632         if(!(event&EV_READ))
 633                 return;
 634         log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
 635         comm_base_now(rep.c->ev->base);
 636         for(i=0; i<NUM_UDP_PER_SELECT; i++) {
 637                 ldns_buffer_clear(rep.c->buffer);
 638                 rep.addrlen = (socklen_t)sizeof(rep.addr);
 639                 log_assert(fd != -1);
 640                 log_assert(ldns_buffer_remaining(rep.c->buffer) > 0);
 641                 rcv = recvfrom(fd, (void*)ldns_buffer_begin(rep.c->buffer),
 642                         ldns_buffer_remaining(rep.c->buffer), 0,
 643                         (struct sockaddr*)&rep.addr, &rep.addrlen);
 644                 if(rcv == -1) {
 645 #ifndef USE_WINSOCK
 646                         if(errno != EAGAIN && errno != EINTR)
 647                                 log_err("recvfrom %d failed: %s",
 648                                         fd, strerror(errno));
 649 #else
 650                         if(WSAGetLastError() != WSAEINPROGRESS &&
 651                                 WSAGetLastError() != WSAECONNRESET &&
 652                                 WSAGetLastError()!= WSAEWOULDBLOCK)
 653                                 log_err("recvfrom failed: %s",
 654                                         wsa_strerror(WSAGetLastError()));
 655 #endif
 656                         return;
 657                 }
 658                 ldns_buffer_skip(rep.c->buffer, rcv);
 659                 ldns_buffer_flip(rep.c->buffer);
 660                 rep.srctype = 0;
 661                 fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
 662                 if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
 663                         /* send back immediate reply */
 664                         (void)comm_point_send_udp_msg(rep.c, rep.c->buffer,
 665                                 (struct sockaddr*)&rep.addr, rep.addrlen);
 666                 }
 667                 if(rep.c->fd != fd) /* commpoint closed to -1 or reused for
 668                 another UDP port. Note rep.c cannot be reused with TCP fd. */
 669                         break;
 670         }
 671 }
 672
 673 /** Use a new tcp handler for new query fd, set to read query */
 674 static void
 675 setup_tcp_handler(struct comm_point* c, int fd)
 676 {
 677         log_assert(c->type == comm_tcp);
 678         log_assert(c->fd == -1);
 679         ldns_buffer_clear(c->buffer);
 680         c->tcp_is_reading = 1;
 681         c->tcp_byte_count = 0;
 682         comm_point_start_listening(c, fd, TCP_QUERY_TIMEOUT);
 683 }
 684
 685 void comm_base_handle_slow_accept(int ATTR_UNUSED(fd),
 686         short ATTR_UNUSED(event), void* arg)
 687 {
 688         struct comm_base* b = (struct comm_base*)arg;
 689         /* timeout for the slow accept, re-enable accepts again */
 690         if(b->start_accept) {
 691                 verbose(VERB_ALGO, "wait is over, slow accept disabled");
 692                 fptr_ok(fptr_whitelist_start_accept(b->start_accept));
 693                 (*b->start_accept)(b->cb_arg);
 694                 b->eb->slow_accept_enabled = 0;
 695         }
 696 }
 697
 698 int comm_point_perform_accept(struct comm_point* c,
 699         struct sockaddr_storage* addr, socklen_t* addrlen)
 700 {
 701         int new_fd;
 702         *addrlen = (socklen_t)sizeof(*addr);
 703         new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen);
 704         if(new_fd == -1) {
 705 #ifndef USE_WINSOCK
 706                 /* EINTR is signal interrupt. others are closed connection. */
 707                 if(     errno == EINTR || errno == EAGAIN
 708 #ifdef EWOULDBLOCK
 709                         || errno == EWOULDBLOCK
 710 #endif
 711 #ifdef ECONNABORTED
 712                         || errno == ECONNABORTED
 713 #endif
 714 #ifdef EPROTO
 715                         || errno == EPROTO
 716 #endif /* EPROTO */
 717                         )
 718                         return -1;
 719 #if defined(ENFILE) && defined(EMFILE)
 720                 if(errno == ENFILE || errno == EMFILE) {
 721                         /* out of file descriptors, likely outside of our
 722                          * control. stop accept() calls for some time */
 723                         if(c->ev->base->stop_accept) {
 724                                 struct comm_base* b = c->ev->base;
 725                                 struct timeval tv;
 726                                 verbose(VERB_ALGO, "out of file descriptors: "
 727                                         "slow accept");
 728                                 b->eb->slow_accept_enabled = 1;
 729                                 fptr_ok(fptr_whitelist_stop_accept(
 730                                         b->stop_accept));
 731                                 (*b->stop_accept)(b->cb_arg);
 732                                 /* set timeout, no mallocs */
 733                                 tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000;
 734                                 tv.tv_usec = NETEVENT_SLOW_ACCEPT_TIME%1000;
 735                                 event_set(&b->eb->slow_accept, -1, EV_TIMEOUT,
 736                                         comm_base_handle_slow_accept, b);
 737                                 if(event_base_set(b->eb->base,
 738                                         &b->eb->slow_accept) != 0) {
 739                                         /* we do not want to log here, because
 740                                          * that would spam the logfiles.
 741                                          * error: "event_base_set failed." */
 742                                 }
 743                                 if(event_add(&b->eb->slow_accept, &tv) != 0) {
 744                                         /* we do not want to log here,
 745                                          * error: "event_add failed." */
 746                                 }
 747                         }
 748                         return -1;
 749                 }
 750 #endif
 751                 log_err("accept failed: %s", strerror(errno));
 752 #else /* USE_WINSOCK */
 753                 if(WSAGetLastError() == WSAEINPROGRESS ||
 754                         WSAGetLastError() == WSAECONNRESET)
 755                         return -1;
 756                 if(WSAGetLastError() == WSAEWOULDBLOCK) {
 757                         winsock_tcp_wouldblock(&c->ev->ev, EV_READ);
 758                         return -1;
 759                 }
 760                 log_err("accept failed: %s", wsa_strerror(WSAGetLastError()));
 761 #endif
 762                 log_addr(0, "remote address is", addr, *addrlen);
 763                 return -1;
 764         }
 765         fd_set_nonblock(new_fd);
 766         return new_fd;
 767 }
 768
 769 #ifdef USE_WINSOCK
 770 static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp),
 771         int ATTR_UNUSED(argi), long argl, long retvalue)
 772 {
 773         verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper,
 774                 (oper&BIO_CB_RETURN)?"return":"before",
 775                 (oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"),
 776                 WSAGetLastError()==WSAEWOULDBLOCK?"wsawb":"");
 777         /* on windows, check if previous operation caused EWOULDBLOCK */
 778         if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) ||
 779                 (oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) {
 780                 if(WSAGetLastError() == WSAEWOULDBLOCK)
 781                         winsock_tcp_wouldblock((struct event*)
 782                                 BIO_get_callback_arg(b), EV_READ);
 783         }
 784         if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) ||
 785                 (oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) {
 786                 if(WSAGetLastError() == WSAEWOULDBLOCK)
 787                         winsock_tcp_wouldblock((struct event*)
 788                                 BIO_get_callback_arg(b), EV_WRITE);
 789         }
 790         /* return original return value */
 791         return retvalue;
 792 }
 793
 794 /** set win bio callbacks for nonblocking operations */
 795 void
 796 comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl)
 797 {
 798         SSL* ssl = (SSL*)thessl;
 799         /* set them both just in case, but usually they are the same BIO */
 800         BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb);
 801         BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)&c->ev->ev);
 802         BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb);
 803         BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)&c->ev->ev);
 804 }
 805 #endif
 806
 807 void
 808 comm_point_tcp_accept_callback(int fd, short event, void* arg)
 809 {
 810         struct comm_point* c = (struct comm_point*)arg, *c_hdl;
 811         int new_fd;
 812         log_assert(c->type == comm_tcp_accept);
 813         if(!(event & EV_READ)) {
 814                 log_info("ignoring tcp accept event %d", (int)event);
 815                 return;
 816         }
 817         comm_base_now(c->ev->base);
 818         /* find free tcp handler. */
 819         if(!c->tcp_free) {
 820                 log_warn("accepted too many tcp, connections full");
 821                 return;
 822         }
 823         /* accept incoming connection. */
 824         c_hdl = c->tcp_free;
 825         log_assert(fd != -1);
 826         new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.addr,
 827                 &c_hdl->repinfo.addrlen);
 828         if(new_fd == -1)
 829                 return;
 830         if(c->ssl) {
 831                 c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd);
 832                 if(!c_hdl->ssl) {
 833                         c_hdl->fd = new_fd;
 834                         comm_point_close(c_hdl);
 835                         return;
 836                 }
 837                 c_hdl->ssl_shake_state = comm_ssl_shake_read;
 838 #ifdef USE_WINSOCK
 839                 comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl);
 840 #endif
 841         }
 842
 843         /* grab the tcp handler buffers */
 844         c->tcp_free = c_hdl->tcp_free;
 845         if(!c->tcp_free) {
 846                 /* stop accepting incoming queries for now. */
 847                 comm_point_stop_listening(c);
 848         }
 849         /* addr is dropped. Not needed for tcp reply. */
 850         setup_tcp_handler(c_hdl, new_fd);
 851 }
 852
 853 /** Make tcp handler free for next assignment */
 854 static void
 855 reclaim_tcp_handler(struct comm_point* c)
 856 {
 857         log_assert(c->type == comm_tcp);
 858         if(c->ssl) {
 859 #ifdef HAVE_SSL
 860                 SSL_shutdown(c->ssl);
 861                 SSL_free(c->ssl);
 862                 c->ssl = NULL;
 863 #endif
 864         }
 865         comm_point_close(c);
 866         if(c->tcp_parent) {
 867                 c->tcp_free = c->tcp_parent->tcp_free;
 868                 c->tcp_parent->tcp_free = c;
 869                 if(!c->tcp_free) {
 870                         /* re-enable listening on accept socket */
 871                         comm_point_start_listening(c->tcp_parent, -1, -1);
 872                 }
 873         }
 874 }
 875
 876 /** do the callback when writing is done */
 877 static void
 878 tcp_callback_writer(struct comm_point* c)
 879 {
 880         log_assert(c->type == comm_tcp);
 881         ldns_buffer_clear(c->buffer);
 882         if(c->tcp_do_toggle_rw)
 883                 c->tcp_is_reading = 1;
 884         c->tcp_byte_count = 0;
 885         /* switch from listening(write) to listening(read) */
 886         comm_point_stop_listening(c);
 887         comm_point_start_listening(c, -1, -1);
 888 }
 889
 890 /** do the callback when reading is done */
 891 static void
 892 tcp_callback_reader(struct comm_point* c)
 893 {
 894         log_assert(c->type == comm_tcp || c->type == comm_local);
 895         ldns_buffer_flip(c->buffer);
 896         if(c->tcp_do_toggle_rw)
 897                 c->tcp_is_reading = 0;
 898         c->tcp_byte_count = 0;
 899         if(c->type == comm_tcp)
 900                 comm_point_stop_listening(c);
 901         fptr_ok(fptr_whitelist_comm_point(c->callback));
 902         if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
 903                 comm_point_start_listening(c, -1, TCP_QUERY_TIMEOUT);
 904         }
 905 }
 906
 907 /** continue ssl handshake */
 908 #ifdef HAVE_SSL
 909 static int
 910 ssl_handshake(struct comm_point* c)
 911 {
 912         int r;
 913         if(c->ssl_shake_state == comm_ssl_shake_hs_read) {
 914                 /* read condition satisfied back to writing */
 915                 comm_point_listen_for_rw(c, 1, 1);
 916                 c->ssl_shake_state = comm_ssl_shake_none;
 917                 return 1;
 918         }
 919         if(c->ssl_shake_state == comm_ssl_shake_hs_write) {
 920                 /* write condition satisfied, back to reading */
 921                 comm_point_listen_for_rw(c, 1, 0);
 922                 c->ssl_shake_state = comm_ssl_shake_none;
 923                 return 1;
 924         }
 925
 926         ERR_clear_error();
 927         r = SSL_do_handshake(c->ssl);
 928         if(r != 1) {
 929                 int want = SSL_get_error(c->ssl, r);
 930                 if(want == SSL_ERROR_WANT_READ) {
 931                         if(c->ssl_shake_state == comm_ssl_shake_read)
 932                                 return 1;
 933                         c->ssl_shake_state = comm_ssl_shake_read;
 934                         comm_point_listen_for_rw(c, 1, 0);
 935                         return 1;
 936                 } else if(want == SSL_ERROR_WANT_WRITE) {
 937                         if(c->ssl_shake_state == comm_ssl_shake_write)
 938                                 return 1;
 939                         c->ssl_shake_state = comm_ssl_shake_write;
 940                         comm_point_listen_for_rw(c, 0, 1);
 941                         return 1;
 942                 } else if(r == 0) {
 943                         return 0; /* closed */
 944                 } else if(want == SSL_ERROR_SYSCALL) {
 945                         /* SYSCALL and errno==0 means closed uncleanly */
 946                         if(errno != 0)
 947                                 log_err("SSL_handshake syscall: %s",
 948                                         strerror(errno));
 949                         return 0;
 950                 } else {
 951                         log_crypto_err("ssl handshake failed");
 952                         log_addr(1, "ssl handshake failed", &c->repinfo.addr,
 953                                 c->repinfo.addrlen);
 954                         return 0;
 955                 }
 956         }
 957         /* this is where peer verification could take place */
 958         log_addr(VERB_ALGO, "SSL DNS connection", &c->repinfo.addr,
 959                 c->repinfo.addrlen);
 960
 961         /* setup listen rw correctly */
 962         if(c->tcp_is_reading) {
 963                 if(c->ssl_shake_state != comm_ssl_shake_read)
 964                         comm_point_listen_for_rw(c, 1, 0);
 965         } else {
 966                 comm_point_listen_for_rw(c, 1, 1);
 967         }
 968         c->ssl_shake_state = comm_ssl_shake_none;
 969         return 1;
 970 }
 971 #endif /* HAVE_SSL */
 972
 973 /** ssl read callback on TCP */
 974 static int
 975 ssl_handle_read(struct comm_point* c)
 976 {
 977 #ifdef HAVE_SSL
 978         int r;
 979         if(c->ssl_shake_state != comm_ssl_shake_none) {
 980                 if(!ssl_handshake(c))
 981                         return 0;
 982                 if(c->ssl_shake_state != comm_ssl_shake_none)
 983                         return 1;
 984         }
 985         if(c->tcp_byte_count < sizeof(uint16_t)) {
 986                 /* read length bytes */
 987                 ERR_clear_error();
 988                 if((r=SSL_read(c->ssl, (void*)ldns_buffer_at(c->buffer,
 989                         c->tcp_byte_count), (int)(sizeof(uint16_t) -
 990                         c->tcp_byte_count))) <= 0) {
 991                         int want = SSL_get_error(c->ssl, r);
 992                         if(want == SSL_ERROR_ZERO_RETURN) {
 993                                 return 0; /* shutdown, closed */
 994                         } else if(want == SSL_ERROR_WANT_READ) {
 995                                 return 1; /* read more later */
 996                         } else if(want == SSL_ERROR_WANT_WRITE) {
 997                                 c->ssl_shake_state = comm_ssl_shake_hs_write;
 998                                 comm_point_listen_for_rw(c, 0, 1);
 999                                 return 1;
1000                         } else if(want == SSL_ERROR_SYSCALL) {
1001                                 if(errno != 0)
1002                                         log_err("SSL_read syscall: %s",
1003                                                 strerror(errno));
1004                                 return 0;
1005                         }
1006                         log_crypto_err("could not SSL_read");
1007                         return 0;
1008                 }
1009                 c->tcp_byte_count += r;
1010                 if(c->tcp_byte_count != sizeof(uint16_t))
1011                         return 1;
1012                 if(ldns_buffer_read_u16_at(c->buffer, 0) >
1013                         ldns_buffer_capacity(c->buffer)) {
1014                         verbose(VERB_QUERY, "ssl: dropped larger than buffer");
1015                         return 0;
1016                 }
1017                 ldns_buffer_set_limit(c->buffer,
1018                         ldns_buffer_read_u16_at(c->buffer, 0));
1019                 if(ldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1020                         verbose(VERB_QUERY, "ssl: dropped bogus too short.");
1021                         return 0;
1022                 }
1023                 verbose(VERB_ALGO, "Reading ssl tcp query of length %d",
1024                         (int)ldns_buffer_limit(c->buffer));
1025         }
1026         log_assert(ldns_buffer_remaining(c->buffer) > 0);
1027         ERR_clear_error();
1028         r = SSL_read(c->ssl, (void*)ldns_buffer_current(c->buffer),
1029                 (int)ldns_buffer_remaining(c->buffer));
1030         if(r <= 0) {
1031                 int want = SSL_get_error(c->ssl, r);
1032                 if(want == SSL_ERROR_ZERO_RETURN) {
1033                         return 0; /* shutdown, closed */
1034                 } else if(want == SSL_ERROR_WANT_READ) {
1035                         return 1; /* read more later */
1036                 } else if(want == SSL_ERROR_WANT_WRITE) {
1037                         c->ssl_shake_state = comm_ssl_shake_hs_write;
1038                         comm_point_listen_for_rw(c, 0, 1);
1039                         return 1;
1040                 } else if(want == SSL_ERROR_SYSCALL) {
1041                         if(errno != 0)
1042                                 log_err("SSL_read syscall: %s",
1043                                         strerror(errno));
1044                         return 0;
1045                 }
1046                 log_crypto_err("could not SSL_read");
1047                 return 0;
1048         }
1049         ldns_buffer_skip(c->buffer, (ssize_t)r);
1050         if(ldns_buffer_remaining(c->buffer) <= 0) {
1051                 tcp_callback_reader(c);
1052         }
1053         return 1;
1054 #else
1055         (void)c;
1056         return 0;
1057 #endif /* HAVE_SSL */
1058 }
1059
1060 /** ssl write callback on TCP */
1061 static int
1062 ssl_handle_write(struct comm_point* c)
1063 {
1064 #ifdef HAVE_SSL
1065         int r;
1066         if(c->ssl_shake_state != comm_ssl_shake_none) {
1067                 if(!ssl_handshake(c))
1068                         return 0;
1069                 if(c->ssl_shake_state != comm_ssl_shake_none)
1070                         return 1;
1071         }
1072         /* ignore return, if fails we may simply block */
1073         (void)SSL_set_mode(c->ssl, SSL_MODE_ENABLE_PARTIAL_WRITE);
1074         if(c->tcp_byte_count < sizeof(uint16_t)) {
1075                 uint16_t len = htons(ldns_buffer_limit(c->buffer));
1076                 ERR_clear_error();
1077                 r = SSL_write(c->ssl,
1078                         (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1079                         (int)(sizeof(uint16_t)-c->tcp_byte_count));
1080                 if(r <= 0) {
1081                         int want = SSL_get_error(c->ssl, r);
1082                         if(want == SSL_ERROR_ZERO_RETURN) {
1083                                 return 0; /* closed */
1084                         } else if(want == SSL_ERROR_WANT_READ) {
1085                                 c->ssl_shake_state = comm_ssl_shake_read;
1086                                 comm_point_listen_for_rw(c, 1, 0);
1087                                 return 1; /* wait for read condition */
1088                         } else if(want == SSL_ERROR_WANT_WRITE) {
1089                                 return 1; /* write more later */
1090                         } else if(want == SSL_ERROR_SYSCALL) {
1091                                 if(errno != 0)
1092                                         log_err("SSL_write syscall: %s",
1093                                                 strerror(errno));
1094                                 return 0;
1095                         }
1096                         log_crypto_err("could not SSL_write");
1097                         return 0;
1098                 }
1099                 c->tcp_byte_count += r;
1100                 if(c->tcp_byte_count < sizeof(uint16_t))
1101                         return 1;
1102                 ldns_buffer_set_position(c->buffer, c->tcp_byte_count -
1103                         sizeof(uint16_t));
1104                 if(ldns_buffer_remaining(c->buffer) == 0) {
1105                         tcp_callback_writer(c);
1106                         return 1;
1107                 }
1108         }
1109         log_assert(ldns_buffer_remaining(c->buffer) > 0);
1110         ERR_clear_error();
1111         r = SSL_write(c->ssl, (void*)ldns_buffer_current(c->buffer),
1112                 (int)ldns_buffer_remaining(c->buffer));
1113         if(r <= 0) {
1114                 int want = SSL_get_error(c->ssl, r);
1115                 if(want == SSL_ERROR_ZERO_RETURN) {
1116                         return 0; /* closed */
1117                 } else if(want == SSL_ERROR_WANT_READ) {
1118                         c->ssl_shake_state = comm_ssl_shake_read;
1119                         comm_point_listen_for_rw(c, 1, 0);
1120                         return 1; /* wait for read condition */
1121                 } else if(want == SSL_ERROR_WANT_WRITE) {
1122                         return 1; /* write more later */
1123                 } else if(want == SSL_ERROR_SYSCALL) {
1124                         if(errno != 0)
1125                                 log_err("SSL_write syscall: %s",
1126                                         strerror(errno));
1127                         return 0;
1128                 }
1129                 log_crypto_err("could not SSL_write");
1130                 return 0;
1131         }
1132         ldns_buffer_skip(c->buffer, (ssize_t)r);
1133
1134         if(ldns_buffer_remaining(c->buffer) == 0) {
1135                 tcp_callback_writer(c);
1136         }
1137         return 1;
1138 #else
1139         (void)c;
1140         return 0;
1141 #endif /* HAVE_SSL */
1142 }
1143
1144 /** handle ssl tcp connection with dns contents */
1145 static int
1146 ssl_handle_it(struct comm_point* c)
1147 {
1148         if(c->tcp_is_reading)
1149                 return ssl_handle_read(c);
1150         return ssl_handle_write(c);
1151 }
1152
1153 /** Handle tcp reading callback.
1154  * @param fd: file descriptor of socket.
1155  * @param c: comm point to read from into buffer.
1156  * @param short_ok: if true, very short packets are OK (for comm_local).
1157  * @return: 0 on error
1158  */
1159 static int
1160 comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok)
1161 {
1162         ssize_t r;
1163         log_assert(c->type == comm_tcp || c->type == comm_local);
1164         if(c->ssl)
1165                 return ssl_handle_it(c);
1166         if(!c->tcp_is_reading)
1167                 return 0;
1168
1169         log_assert(fd != -1);
1170         if(c->tcp_byte_count < sizeof(uint16_t)) {
1171                 /* read length bytes */
1172                 r = recv(fd,(void*)ldns_buffer_at(c->buffer,c->tcp_byte_count),
1173                         sizeof(uint16_t)-c->tcp_byte_count, 0);
1174                 if(r == 0)
1175                         return 0;
1176                 else if(r == -1) {
1177 #ifndef USE_WINSOCK
1178                         if(errno == EINTR || errno == EAGAIN)
1179                                 return 1;
1180 #ifdef ECONNRESET
1181                         if(errno == ECONNRESET && verbosity < 2)
1182                                 return 0; /* silence reset by peer */
1183 #endif
1184                         log_err("read (in tcp s): %s", strerror(errno));
1185 #else /* USE_WINSOCK */
1186                         if(WSAGetLastError() == WSAECONNRESET)
1187                                 return 0;
1188                         if(WSAGetLastError() == WSAEINPROGRESS)
1189                                 return 1;
1190                         if(WSAGetLastError() == WSAEWOULDBLOCK) {
1191                                 winsock_tcp_wouldblock(&c->ev->ev, EV_READ);
1192                                 return 1;
1193                         }
1194                         log_err("read (in tcp s): %s",
1195                                 wsa_strerror(WSAGetLastError()));
1196 #endif
1197                         log_addr(0, "remote address is", &c->repinfo.addr,
1198                                 c->repinfo.addrlen);
1199                         return 0;
1200                 }
1201                 c->tcp_byte_count += r;
1202                 if(c->tcp_byte_count != sizeof(uint16_t))
1203                         return 1;
1204                 if(ldns_buffer_read_u16_at(c->buffer, 0) >
1205                         ldns_buffer_capacity(c->buffer)) {
1206                         verbose(VERB_QUERY, "tcp: dropped larger than buffer");
1207                         return 0;
1208                 }
1209                 ldns_buffer_set_limit(c->buffer,
1210                         ldns_buffer_read_u16_at(c->buffer, 0));
1211                 if(!short_ok &&
1212                         ldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1213                         verbose(VERB_QUERY, "tcp: dropped bogus too short.");
1214                         return 0;
1215                 }
1216                 verbose(VERB_ALGO, "Reading tcp query of length %d",
1217                         (int)ldns_buffer_limit(c->buffer));
1218         }
1219
1220         log_assert(ldns_buffer_remaining(c->buffer) > 0);
1221         r = recv(fd, (void*)ldns_buffer_current(c->buffer),
1222                 ldns_buffer_remaining(c->buffer), 0);
1223         if(r == 0) {
1224                 return 0;
1225         } else if(r == -1) {
1226 #ifndef USE_WINSOCK
1227                 if(errno == EINTR || errno == EAGAIN)
1228                         return 1;
1229                 log_err("read (in tcp r): %s", strerror(errno));
1230 #else /* USE_WINSOCK */
1231                 if(WSAGetLastError() == WSAECONNRESET)
1232                         return 0;
1233                 if(WSAGetLastError() == WSAEINPROGRESS)
1234                         return 1;
1235                 if(WSAGetLastError() == WSAEWOULDBLOCK) {
1236                         winsock_tcp_wouldblock(&c->ev->ev, EV_READ);
1237                         return 1;
1238                 }
1239                 log_err("read (in tcp r): %s",
1240                         wsa_strerror(WSAGetLastError()));
1241 #endif
1242                 log_addr(0, "remote address is", &c->repinfo.addr,
1243                         c->repinfo.addrlen);
1244                 return 0;
1245         }
1246         ldns_buffer_skip(c->buffer, r);
1247         if(ldns_buffer_remaining(c->buffer) <= 0) {
1248                 tcp_callback_reader(c);
1249         }
1250         return 1;
1251 }
1252
1253 /**
1254  * Handle tcp writing callback.
1255  * @param fd: file descriptor of socket.
1256  * @param c: comm point to write buffer out of.
1257  * @return: 0 on error
1258  */
1259 static int
1260 comm_point_tcp_handle_write(int fd, struct comm_point* c)
1261 {
1262         ssize_t r;
1263         log_assert(c->type == comm_tcp);
1264         if(c->tcp_is_reading && !c->ssl)
1265                 return 0;
1266         log_assert(fd != -1);
1267         if(c->tcp_byte_count == 0 && c->tcp_check_nb_connect) {
1268                 /* check for pending error from nonblocking connect */
1269                 /* from Stevens, unix network programming, vol1, 3rd ed, p450*/
1270                 int error = 0;
1271                 socklen_t len = (socklen_t)sizeof(error);
1272                 if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
1273                         &len) < 0){
1274 #ifndef USE_WINSOCK
1275                         error = errno; /* on solaris errno is error */
1276 #else /* USE_WINSOCK */
1277                         error = WSAGetLastError();
1278 #endif
1279                 }
1280 #ifndef USE_WINSOCK
1281 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1282                 if(error == EINPROGRESS || error == EWOULDBLOCK)
1283                         return 1; /* try again later */
1284                 else
1285 #endif
1286                 if(error != 0 && verbosity < 2)
1287                         return 0; /* silence lots of chatter in the logs */
1288                 else if(error != 0) {
1289                         log_err("tcp connect: %s", strerror(error));
1290 #else /* USE_WINSOCK */
1291                 /* examine error */
1292                 if(error == WSAEINPROGRESS)
1293                         return 1;
1294                 else if(error == WSAEWOULDBLOCK) {
1295                         winsock_tcp_wouldblock(&c->ev->ev, EV_WRITE);
1296                         return 1;
1297                 } else if(error != 0 && verbosity < 2)
1298                         return 0;
1299                 else if(error != 0) {
1300                         log_err("tcp connect: %s", wsa_strerror(error));
1301 #endif /* USE_WINSOCK */
1302                         log_addr(0, "remote address is", &c->repinfo.addr,
1303                                 c->repinfo.addrlen);
1304                         return 0;
1305                 }
1306         }
1307         if(c->ssl)
1308                 return ssl_handle_it(c);
1309
1310         if(c->tcp_byte_count < sizeof(uint16_t)) {
1311                 uint16_t len = htons(ldns_buffer_limit(c->buffer));
1312 #ifdef HAVE_WRITEV
1313                 struct iovec iov[2];
1314                 iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1315                 iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1316                 iov[1].iov_base = ldns_buffer_begin(c->buffer);
1317                 iov[1].iov_len = ldns_buffer_limit(c->buffer);
1318                 log_assert(iov[0].iov_len > 0);
1319                 log_assert(iov[1].iov_len > 0);
1320                 r = writev(fd, iov, 2);
1321 #else /* HAVE_WRITEV */
1322                 r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1323                         sizeof(uint16_t)-c->tcp_byte_count, 0);
1324 #endif /* HAVE_WRITEV */
1325                 if(r == -1) {
1326 #ifndef USE_WINSOCK
1327 #ifdef EPIPE
1328                         if(errno == EPIPE && verbosity < 2)
1329                                 return 0; /* silence 'broken pipe' */
1330 #endif
1331                         if(errno == EINTR || errno == EAGAIN)
1332                                 return 1;
1333                         log_err("tcp writev: %s", strerror(errno));
1334 #else
1335                         if(WSAGetLastError() == WSAENOTCONN)
1336                                 return 1;
1337                         if(WSAGetLastError() == WSAEINPROGRESS)
1338                                 return 1;
1339                         if(WSAGetLastError() == WSAEWOULDBLOCK) {
1340                                 winsock_tcp_wouldblock(&c->ev->ev, EV_WRITE);
1341                                 return 1;
1342                         }
1343                         log_err("tcp send s: %s",
1344                                 wsa_strerror(WSAGetLastError()));
1345 #endif
1346                         log_addr(0, "remote address is", &c->repinfo.addr,
1347                                 c->repinfo.addrlen);
1348                         return 0;
1349                 }
1350                 c->tcp_byte_count += r;
1351                 if(c->tcp_byte_count < sizeof(uint16_t))
1352                         return 1;
1353                 ldns_buffer_set_position(c->buffer, c->tcp_byte_count -
1354                         sizeof(uint16_t));
1355                 if(ldns_buffer_remaining(c->buffer) == 0) {
1356                         tcp_callback_writer(c);
1357                         return 1;
1358                 }
1359         }
1360         log_assert(ldns_buffer_remaining(c->buffer) > 0);
1361         r = send(fd, (void*)ldns_buffer_current(c->buffer),
1362                 ldns_buffer_remaining(c->buffer), 0);
1363         if(r == -1) {
1364 #ifndef USE_WINSOCK
1365                 if(errno == EINTR || errno == EAGAIN)
1366                         return 1;
1367                 log_err("tcp send r: %s", strerror(errno));
1368 #else
1369                 if(WSAGetLastError() == WSAEINPROGRESS)
1370                         return 1;
1371                 if(WSAGetLastError() == WSAEWOULDBLOCK) {
1372                         winsock_tcp_wouldblock(&c->ev->ev, EV_WRITE);
1373                         return 1;
1374                 }
1375                 log_err("tcp send r: %s",
1376                         wsa_strerror(WSAGetLastError()));
1377 #endif
1378                 log_addr(0, "remote address is", &c->repinfo.addr,
1379                         c->repinfo.addrlen);
1380                 return 0;
1381         }
1382         ldns_buffer_skip(c->buffer, r);
1383
1384         if(ldns_buffer_remaining(c->buffer) == 0) {
1385                 tcp_callback_writer(c);
1386         }
1387
1388         return 1;
1389 }
1390
1391 void
1392 comm_point_tcp_handle_callback(int fd, short event, void* arg)
1393 {
1394         struct comm_point* c = (struct comm_point*)arg;
1395         log_assert(c->type == comm_tcp);
1396         comm_base_now(c->ev->base);
1397
1398         if(event&EV_READ) {
1399                 if(!comm_point_tcp_handle_read(fd, c, 0)) {
1400                         reclaim_tcp_handler(c);
1401                         if(!c->tcp_do_close) {
1402                                 fptr_ok(fptr_whitelist_comm_point(
1403                                         c->callback));
1404                                 (void)(*c->callback)(c, c->cb_arg,
1405                                         NETEVENT_CLOSED, NULL);
1406                         }
1407                 }
1408                 return;
1409         }
1410         if(event&EV_WRITE) {
1411                 if(!comm_point_tcp_handle_write(fd, c)) {
1412                         reclaim_tcp_handler(c);
1413                         if(!c->tcp_do_close) {
1414                                 fptr_ok(fptr_whitelist_comm_point(
1415                                         c->callback));
1416                                 (void)(*c->callback)(c, c->cb_arg,
1417                                         NETEVENT_CLOSED, NULL);
1418                         }
1419                 }
1420                 return;
1421         }
1422         if(event&EV_TIMEOUT) {
1423                 verbose(VERB_QUERY, "tcp took too long, dropped");
1424                 reclaim_tcp_handler(c);
1425                 if(!c->tcp_do_close) {
1426                         fptr_ok(fptr_whitelist_comm_point(c->callback));
1427                         (void)(*c->callback)(c, c->cb_arg,
1428                                 NETEVENT_TIMEOUT, NULL);
1429                 }
1430                 return;
1431         }
1432         log_err("Ignored event %d for tcphdl.", event);
1433 }
1434
1435 void comm_point_local_handle_callback(int fd, short event, void* arg)
1436 {
1437         struct comm_point* c = (struct comm_point*)arg;
1438         log_assert(c->type == comm_local);
1439         comm_base_now(c->ev->base);
1440
1441         if(event&EV_READ) {
1442                 if(!comm_point_tcp_handle_read(fd, c, 1)) {
1443                         fptr_ok(fptr_whitelist_comm_point(c->callback));
1444                         (void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED,
1445                                 NULL);
1446                 }
1447                 return;
1448         }
1449         log_err("Ignored event %d for localhdl.", event);
1450 }
1451
1452 void comm_point_raw_handle_callback(int ATTR_UNUSED(fd),
1453         short event, void* arg)
1454 {
1455         struct comm_point* c = (struct comm_point*)arg;
1456         int err = NETEVENT_NOERROR;
1457         log_assert(c->type == comm_raw);
1458         comm_base_now(c->ev->base);
1459
1460         if(event&EV_TIMEOUT)
1461                 err = NETEVENT_TIMEOUT;
1462         fptr_ok(fptr_whitelist_comm_point_raw(c->callback));
1463         (void)(*c->callback)(c, c->cb_arg, err, NULL);
1464 }
1465
1466 struct comm_point*
1467 comm_point_create_udp(struct comm_base *base, int fd, ldns_buffer* buffer,
1468         comm_point_callback_t* callback, void* callback_arg)
1469 {
1470         struct comm_point* c = (struct comm_point*)calloc(1,
1471                 sizeof(struct comm_point));
1472         short evbits;
1473         if(!c)
1474                 return NULL;
1475         c->ev = (struct internal_event*)calloc(1,
1476                 sizeof(struct internal_event));
1477         if(!c->ev) {
1478                 free(c);
1479                 return NULL;
1480         }
1481         c->ev->base = base;
1482         c->fd = fd;
1483         c->buffer = buffer;
1484         c->timeout = NULL;
1485         c->tcp_is_reading = 0;
1486         c->tcp_byte_count = 0;
1487         c->tcp_parent = NULL;
1488         c->max_tcp_count = 0;
1489         c->tcp_handlers = NULL;
1490         c->tcp_free = NULL;
1491         c->type = comm_udp;
1492         c->tcp_do_close = 0;
1493         c->do_not_close = 0;
1494         c->tcp_do_toggle_rw = 0;
1495         c->tcp_check_nb_connect = 0;
1496         c->inuse = 0;
1497         c->callback = callback;
1498         c->cb_arg = callback_arg;
1499         evbits = EV_READ | EV_PERSIST;
1500         /* libevent stuff */
1501         event_set(&c->ev->ev, c->fd, evbits, comm_point_udp_callback, c);
1502         if(event_base_set(base->eb->base, &c->ev->ev) != 0) {
1503                 log_err("could not baseset udp event");
1504                 comm_point_delete(c);
1505                 return NULL;
1506         }
1507         if(fd!=-1 && event_add(&c->ev->ev, c->timeout) != 0 ) {
1508                 log_err("could not add udp event");
1509                 comm_point_delete(c);
1510                 return NULL;
1511         }
1512         return c;
1513 }
1514
1515 struct comm_point*
1516 comm_point_create_udp_ancil(struct comm_base *base, int fd,
1517         ldns_buffer* buffer,
1518         comm_point_callback_t* callback, void* callback_arg)
1519 {
1520         struct comm_point* c = (struct comm_point*)calloc(1,
1521                 sizeof(struct comm_point));
1522         short evbits;
1523         if(!c)
1524                 return NULL;
1525         c->ev = (struct internal_event*)calloc(1,
1526                 sizeof(struct internal_event));
1527         if(!c->ev) {
1528                 free(c);
1529                 return NULL;
1530         }
1531         c->ev->base = base;
1532         c->fd = fd;
1533         c->buffer = buffer;
1534         c->timeout = NULL;
1535         c->tcp_is_reading = 0;
1536         c->tcp_byte_count = 0;
1537         c->tcp_parent = NULL;
1538         c->max_tcp_count = 0;
1539         c->tcp_handlers = NULL;
1540         c->tcp_free = NULL;
1541         c->type = comm_udp;
1542         c->tcp_do_close = 0;
1543         c->do_not_close = 0;
1544         c->inuse = 0;
1545         c->tcp_do_toggle_rw = 0;
1546         c->tcp_check_nb_connect = 0;
1547         c->callback = callback;
1548         c->cb_arg = callback_arg;
1549         evbits = EV_READ | EV_PERSIST;
1550         /* libevent stuff */
1551         event_set(&c->ev->ev, c->fd, evbits, comm_point_udp_ancil_callback, c);
1552         if(event_base_set(base->eb->base, &c->ev->ev) != 0) {
1553                 log_err("could not baseset udp event");
1554                 comm_point_delete(c);
1555                 return NULL;
1556         }
1557         if(fd!=-1 && event_add(&c->ev->ev, c->timeout) != 0 ) {
1558                 log_err("could not add udp event");
1559                 comm_point_delete(c);
1560                 return NULL;
1561         }
1562         return c;
1563 }
1564
1565 static struct comm_point*
1566 comm_point_create_tcp_handler(struct comm_base *base,
1567         struct comm_point* parent, size_t bufsize,
1568         comm_point_callback_t* callback, void* callback_arg)
1569 {
1570         struct comm_point* c = (struct comm_point*)calloc(1,
1571                 sizeof(struct comm_point));
1572         short evbits;
1573         if(!c)
1574                 return NULL;
1575         c->ev = (struct internal_event*)calloc(1,
1576                 sizeof(struct internal_event));
1577         if(!c->ev) {
1578                 free(c);
1579                 return NULL;
1580         }
1581         c->ev->base = base;
1582         c->fd = -1;
1583         c->buffer = ldns_buffer_new(bufsize);
1584         if(!c->buffer) {
1585                 free(c->ev);
1586                 free(c);
1587                 return NULL;
1588         }
1589         c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
1590         if(!c->timeout) {
1591                 ldns_buffer_free(c->buffer);
1592                 free(c->ev);
1593                 free(c);
1594                 return NULL;
1595         }
1596         c->tcp_is_reading = 0;
1597         c->tcp_byte_count = 0;
1598         c->tcp_parent = parent;
1599         c->max_tcp_count = 0;
1600         c->tcp_handlers = NULL;
1601         c->tcp_free = NULL;
1602         c->type = comm_tcp;
1603         c->tcp_do_close = 0;
1604         c->do_not_close = 0;
1605         c->tcp_do_toggle_rw = 1;
1606         c->tcp_check_nb_connect = 0;
1607         c->repinfo.c = c;
1608         c->callback = callback;
1609         c->cb_arg = callback_arg;
1610         /* add to parent free list */
1611         c->tcp_free = parent->tcp_free;
1612         parent->tcp_free = c;
1613         /* libevent stuff */
1614         evbits = EV_PERSIST | EV_READ | EV_TIMEOUT;
1615         event_set(&c->ev->ev, c->fd, evbits, comm_point_tcp_handle_callback, c);
1616         if(event_base_set(base->eb->base, &c->ev->ev) != 0)
1617         {
1618                 log_err("could not basetset tcphdl event");
1619                 parent->tcp_free = c->tcp_free;
1620                 free(c->ev);
1621                 free(c);
1622                 return NULL;
1623         }
1624         return c;
1625 }
1626
1627 struct comm_point*
1628 comm_point_create_tcp(struct comm_base *base, int fd, int num, size_t bufsize,
1629         comm_point_callback_t* callback, void* callback_arg)
1630 {
1631         struct comm_point* c = (struct comm_point*)calloc(1,
1632                 sizeof(struct comm_point));
1633         short evbits;
1634         int i;
1635         /* first allocate the TCP accept listener */
1636         if(!c)
1637                 return NULL;
1638         c->ev = (struct internal_event*)calloc(1,
1639                 sizeof(struct internal_event));
1640         if(!c->ev) {
1641                 free(c);
1642                 return NULL;
1643         }
1644         c->ev->base = base;
1645         c->fd = fd;
1646         c->buffer = NULL;
1647         c->timeout = NULL;
1648         c->tcp_is_reading = 0;
1649         c->tcp_byte_count = 0;
1650         c->tcp_parent = NULL;
1651         c->max_tcp_count = num;
1652         c->tcp_handlers = (struct comm_point**)calloc((size_t)num,
1653                 sizeof(struct comm_point*));
1654         if(!c->tcp_handlers) {
1655                 free(c->ev);
1656                 free(c);
1657                 return NULL;
1658         }
1659         c->tcp_free = NULL;
1660         c->type = comm_tcp_accept;
1661         c->tcp_do_close = 0;
1662         c->do_not_close = 0;
1663         c->tcp_do_toggle_rw = 0;
1664         c->tcp_check_nb_connect = 0;
1665         c->callback = NULL;
1666         c->cb_arg = NULL;
1667         evbits = EV_READ | EV_PERSIST;
1668         /* libevent stuff */
1669         event_set(&c->ev->ev, c->fd, evbits, comm_point_tcp_accept_callback, c);
1670         if(event_base_set(base->eb->base, &c->ev->ev) != 0 ||
1671                 event_add(&c->ev->ev, c->timeout) != 0 )
1672         {
1673                 log_err("could not add tcpacc event");
1674                 comm_point_delete(c);
1675                 return NULL;
1676         }
1677
1678         /* now prealloc the tcp handlers */
1679         for(i=0; i<num; i++) {
1680                 c->tcp_handlers[i] = comm_point_create_tcp_handler(base,
1681                         c, bufsize, callback, callback_arg);
1682                 if(!c->tcp_handlers[i]) {
1683                         comm_point_delete(c);
1684                         return NULL;
1685                 }
1686         }
1687
1688         return c;
1689 }
1690
1691 struct comm_point*
1692 comm_point_create_tcp_out(struct comm_base *base, size_t bufsize,
1693         comm_point_callback_t* callback, void* callback_arg)
1694 {
1695         struct comm_point* c = (struct comm_point*)calloc(1,
1696                 sizeof(struct comm_point));
1697         short evbits;
1698         if(!c)
1699                 return NULL;
1700         c->ev = (struct internal_event*)calloc(1,
1701                 sizeof(struct internal_event));
1702         if(!c->ev) {
1703                 free(c);
1704                 return NULL;
1705         }
1706         c->ev->base = base;
1707         c->fd = -1;
1708         c->buffer = ldns_buffer_new(bufsize);
1709         if(!c->buffer) {
1710                 free(c->ev);
1711                 free(c);
1712                 return NULL;
1713         }
1714         c->timeout = NULL;
1715         c->tcp_is_reading = 0;
1716         c->tcp_byte_count = 0;
1717         c->tcp_parent = NULL;
1718         c->max_tcp_count = 0;
1719         c->tcp_handlers = NULL;
1720         c->tcp_free = NULL;
1721         c->type = comm_tcp;
1722         c->tcp_do_close = 0;
1723         c->do_not_close = 0;
1724         c->tcp_do_toggle_rw = 1;
1725         c->tcp_check_nb_connect = 1;
1726         c->repinfo.c = c;
1727         c->callback = callback;
1728         c->cb_arg = callback_arg;
1729         evbits = EV_PERSIST | EV_WRITE;
1730         event_set(&c->ev->ev, c->fd, evbits, comm_point_tcp_handle_callback, c);
1731         if(event_base_set(base->eb->base, &c->ev->ev) != 0)
1732         {
1733                 log_err("could not basetset tcpout event");
1734                 ldns_buffer_free(c->buffer);
1735                 free(c->ev);
1736                 free(c);
1737                 return NULL;
1738         }
1739
1740         return c;
1741 }
1742
1743 struct comm_point*
1744 comm_point_create_local(struct comm_base *base, int fd, size_t bufsize,
1745         comm_point_callback_t* callback, void* callback_arg)
1746 {
1747         struct comm_point* c = (struct comm_point*)calloc(1,
1748                 sizeof(struct comm_point));
1749         short evbits;
1750         if(!c)
1751                 return NULL;
1752         c->ev = (struct internal_event*)calloc(1,
1753                 sizeof(struct internal_event));
1754         if(!c->ev) {
1755                 free(c);
1756                 return NULL;
1757         }
1758         c->ev->base = base;
1759         c->fd = fd;
1760         c->buffer = ldns_buffer_new(bufsize);
1761         if(!c->buffer) {
1762                 free(c->ev);
1763                 free(c);
1764                 return NULL;
1765         }
1766         c->timeout = NULL;
1767         c->tcp_is_reading = 1;
1768         c->tcp_byte_count = 0;
1769         c->tcp_parent = NULL;
1770         c->max_tcp_count = 0;
1771         c->tcp_handlers = NULL;
1772         c->tcp_free = NULL;
1773         c->type = comm_local;
1774         c->tcp_do_close = 0;
1775         c->do_not_close = 1;
1776         c->tcp_do_toggle_rw = 0;
1777         c->tcp_check_nb_connect = 0;
1778         c->callback = callback;
1779         c->cb_arg = callback_arg;
1780         /* libevent stuff */
1781         evbits = EV_PERSIST | EV_READ;
1782         event_set(&c->ev->ev, c->fd, evbits, comm_point_local_handle_callback,
1783                 c);
1784         if(event_base_set(base->eb->base, &c->ev->ev) != 0 ||
1785                 event_add(&c->ev->ev, c->timeout) != 0 )
1786         {
1787                 log_err("could not add localhdl event");
1788                 free(c->ev);
1789                 free(c);
1790                 return NULL;
1791         }
1792         return c;
1793 }
1794
1795 struct comm_point*
1796 comm_point_create_raw(struct comm_base* base, int fd, int writing,
1797         comm_point_callback_t* callback, void* callback_arg)
1798 {
1799         struct comm_point* c = (struct comm_point*)calloc(1,
1800                 sizeof(struct comm_point));
1801         short evbits;
1802         if(!c)
1803                 return NULL;
1804         c->ev = (struct internal_event*)calloc(1,
1805                 sizeof(struct internal_event));
1806         if(!c->ev) {
1807                 free(c);
1808                 return NULL;
1809         }
1810         c->ev->base = base;
1811         c->fd = fd;
1812         c->buffer = NULL;
1813         c->timeout = NULL;
1814         c->tcp_is_reading = 0;
1815         c->tcp_byte_count = 0;
1816         c->tcp_parent = NULL;
1817         c->max_tcp_count = 0;
1818         c->tcp_handlers = NULL;
1819         c->tcp_free = NULL;
1820         c->type = comm_raw;
1821         c->tcp_do_close = 0;
1822         c->do_not_close = 1;
1823         c->tcp_do_toggle_rw = 0;
1824         c->tcp_check_nb_connect = 0;
1825         c->callback = callback;
1826         c->cb_arg = callback_arg;
1827         /* libevent stuff */
1828         if(writing)
1829                 evbits = EV_PERSIST | EV_WRITE;
1830         else    evbits = EV_PERSIST | EV_READ;
1831         event_set(&c->ev->ev, c->fd, evbits, comm_point_raw_handle_callback,
1832                 c);
1833         if(event_base_set(base->eb->base, &c->ev->ev) != 0 ||
1834                 event_add(&c->ev->ev, c->timeout) != 0 )
1835         {
1836                 log_err("could not add rawhdl event");
1837                 free(c->ev);
1838                 free(c);
1839                 return NULL;
1840         }
1841         return c;
1842 }
1843
1844 void
1845 comm_point_close(struct comm_point* c)
1846 {
1847         if(!c)
1848                 return;
1849         if(c->fd != -1)
1850                 if(event_del(&c->ev->ev) != 0) {
1851                         log_err("could not event_del on close");
1852                 }
1853         /* close fd after removing from event lists, or epoll.. is messed up */
1854         if(c->fd != -1 && !c->do_not_close) {
1855                 verbose(VERB_ALGO, "close fd %d", c->fd);
1856 #ifndef USE_WINSOCK
1857                 close(c->fd);
1858 #else
1859                 closesocket(c->fd);
1860 #endif
1861         }
1862         c->fd = -1;
1863 }
1864
1865 void
1866 comm_point_delete(struct comm_point* c)
1867 {
1868         if(!c)
1869                 return;
1870         if(c->type == comm_tcp && c->ssl) {
1871 #ifdef HAVE_SSL
1872                 SSL_shutdown(c->ssl);
1873                 SSL_free(c->ssl);
1874 #endif
1875         }
1876         comm_point_close(c);
1877         if(c->tcp_handlers) {
1878                 int i;
1879                 for(i=0; i<c->max_tcp_count; i++)
1880                         comm_point_delete(c->tcp_handlers[i]);
1881                 free(c->tcp_handlers);
1882         }
1883         free(c->timeout);
1884         if(c->type == comm_tcp || c->type == comm_local)
1885                 ldns_buffer_free(c->buffer);
1886         free(c->ev);
1887         free(c);
1888 }
1889
1890 void
1891 comm_point_send_reply(struct comm_reply *repinfo)
1892 {
1893         log_assert(repinfo && repinfo->c);
1894         if(repinfo->c->type == comm_udp) {
1895                 if(repinfo->srctype)
1896                         comm_point_send_udp_msg_if(repinfo->c,
1897                         repinfo->c->buffer, (struct sockaddr*)&repinfo->addr,
1898                         repinfo->addrlen, repinfo);
1899                 else
1900                         comm_point_send_udp_msg(repinfo->c, repinfo->c->buffer,
1901                         (struct sockaddr*)&repinfo->addr, repinfo->addrlen);
1902         } else {
1903                 comm_point_start_listening(repinfo->c, -1, TCP_QUERY_TIMEOUT);
1904         }
1905 }
1906
1907 void
1908 comm_point_drop_reply(struct comm_reply* repinfo)
1909 {
1910         if(!repinfo)
1911                 return;
1912         log_assert(repinfo && repinfo->c);
1913         log_assert(repinfo->c->type != comm_tcp_accept);
1914         if(repinfo->c->type == comm_udp)
1915                 return;
1916         reclaim_tcp_handler(repinfo->c);
1917 }
1918
1919 void
1920 comm_point_stop_listening(struct comm_point* c)
1921 {
1922         verbose(VERB_ALGO, "comm point stop listening %d", c->fd);
1923         if(event_del(&c->ev->ev) != 0) {
1924                 log_err("event_del error to stoplisten");
1925         }
1926 }
1927
1928 void
1929 comm_point_start_listening(struct comm_point* c, int newfd, int sec)
1930 {
1931         verbose(VERB_ALGO, "comm point start listening %d",
1932                 c->fd==-1?newfd:c->fd);
1933         if(c->type == comm_tcp_accept && !c->tcp_free) {
1934                 /* no use to start listening no free slots. */
1935                 return;
1936         }
1937         if(sec != -1 && sec != 0) {
1938                 if(!c->timeout) {
1939                         c->timeout = (struct timeval*)malloc(sizeof(
1940                                 struct timeval));
1941                         if(!c->timeout) {
1942                                 log_err("cpsl: malloc failed. No net read.");
1943                                 return;
1944                         }
1945                 }
1946                 c->ev->ev.ev_events |= EV_TIMEOUT;
1947 #ifndef S_SPLINT_S /* splint fails on struct timeval. */
1948                 c->timeout->tv_sec = sec;
1949                 c->timeout->tv_usec = 0;
1950 #endif /* S_SPLINT_S */
1951         }
1952         if(c->type == comm_tcp) {
1953                 c->ev->ev.ev_events &= ~(EV_READ|EV_WRITE);
1954                 if(c->tcp_is_reading)
1955                         c->ev->ev.ev_events |= EV_READ;
1956                 else    c->ev->ev.ev_events |= EV_WRITE;
1957         }
1958         if(newfd != -1) {
1959                 if(c->fd != -1) {
1960 #ifndef USE_WINSOCK
1961                         close(c->fd);
1962 #else
1963                         closesocket(c->fd);
1964 #endif
1965                 }
1966                 c->fd = newfd;
1967                 c->ev->ev.ev_fd = c->fd;
1968         }
1969         if(event_add(&c->ev->ev, sec==0?NULL:c->timeout) != 0) {
1970                 log_err("event_add failed. in cpsl.");
1971         }
1972 }
1973
1974 void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr)
1975 {
1976         verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr);
1977         if(event_del(&c->ev->ev) != 0) {
1978                 log_err("event_del error to cplf");
1979         }
1980         c->ev->ev.ev_events &= ~(EV_READ|EV_WRITE);
1981         if(rd) c->ev->ev.ev_events |= EV_READ;
1982         if(wr) c->ev->ev.ev_events |= EV_WRITE;
1983         if(event_add(&c->ev->ev, c->timeout) != 0) {
1984                 log_err("event_add failed. in cplf.");
1985         }
1986 }
1987
1988 size_t comm_point_get_mem(struct comm_point* c)
1989 {
1990         size_t s;
1991         if(!c)
1992                 return 0;
1993         s = sizeof(*c) + sizeof(*c->ev);
1994         if(c->timeout)
1995                 s += sizeof(*c->timeout);
1996         if(c->type == comm_tcp || c->type == comm_local)
1997                 s += sizeof(*c->buffer) + ldns_buffer_capacity(c->buffer);
1998         if(c->type == comm_tcp_accept) {
1999                 int i;
2000                 for(i=0; i<c->max_tcp_count; i++)
2001                         s += comm_point_get_mem(c->tcp_handlers[i]);
2002         }
2003         return s;
2004 }
2005
2006 struct comm_timer*
2007 comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg)
2008 {
2009         struct comm_timer *tm = (struct comm_timer*)calloc(1,
2010                 sizeof(struct comm_timer));
2011         if(!tm)
2012                 return NULL;
2013         tm->ev_timer = (struct internal_timer*)calloc(1,
2014                 sizeof(struct internal_timer));
2015         if(!tm->ev_timer) {
2016                 log_err("malloc failed");
2017                 free(tm);
2018                 return NULL;
2019         }
2020         tm->ev_timer->base = base;
2021         tm->callback = cb;
2022         tm->cb_arg = cb_arg;
2023         event_set(&tm->ev_timer->ev, -1, EV_TIMEOUT,
2024                 comm_timer_callback, tm);
2025         if(event_base_set(base->eb->base, &tm->ev_timer->ev) != 0) {
2026                 log_err("timer_create: event_base_set failed.");
2027                 free(tm->ev_timer);
2028                 free(tm);
2029                 return NULL;
2030         }
2031         return tm;
2032 }
2033
2034 void
2035 comm_timer_disable(struct comm_timer* timer)
2036 {
2037         if(!timer)
2038                 return;
2039         evtimer_del(&timer->ev_timer->ev);
2040         timer->ev_timer->enabled = 0;
2041 }
2042
2043 void
2044 comm_timer_set(struct comm_timer* timer, struct timeval* tv)
2045 {
2046         log_assert(tv);
2047         if(timer->ev_timer->enabled)
2048                 comm_timer_disable(timer);
2049         event_set(&timer->ev_timer->ev, -1, EV_TIMEOUT,
2050                 comm_timer_callback, timer);
2051         if(event_base_set(timer->ev_timer->base->eb->base,
2052                 &timer->ev_timer->ev) != 0)
2053                 log_err("comm_timer_set: set_base failed.");
2054         if(evtimer_add(&timer->ev_timer->ev, tv) != 0)
2055                 log_err("comm_timer_set: evtimer_add failed.");
2056         timer->ev_timer->enabled = 1;
2057 }
2058
2059 void
2060 comm_timer_delete(struct comm_timer* timer)
2061 {
2062         if(!timer)
2063                 return;
2064         comm_timer_disable(timer);
2065         free(timer->ev_timer);
2066         free(timer);
2067 }
2068
2069 void
2070 comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg)
2071 {
2072         struct comm_timer* tm = (struct comm_timer*)arg;
2073         if(!(event&EV_TIMEOUT))
2074                 return;
2075         comm_base_now(tm->ev_timer->base);
2076         tm->ev_timer->enabled = 0;
2077         fptr_ok(fptr_whitelist_comm_timer(tm->callback));
2078         (*tm->callback)(tm->cb_arg);
2079 }
2080
2081 int
2082 comm_timer_is_set(struct comm_timer* timer)
2083 {
2084         return (int)timer->ev_timer->enabled;
2085 }
2086
2087 size_t
2088 comm_timer_get_mem(struct comm_timer* timer)
2089 {
2090         return sizeof(*timer) + sizeof(struct internal_timer);
2091 }
2092
2093 struct comm_signal*
2094 comm_signal_create(struct comm_base* base,
2095         void (*callback)(int, void*), void* cb_arg)
2096 {
2097         struct comm_signal* com = (struct comm_signal*)malloc(
2098                 sizeof(struct comm_signal));
2099         if(!com) {
2100                 log_err("malloc failed");
2101                 return NULL;
2102         }
2103         com->base = base;
2104         com->callback = callback;
2105         com->cb_arg = cb_arg;
2106         com->ev_signal = NULL;
2107         return com;
2108 }
2109
2110 void
2111 comm_signal_callback(int sig, short event, void* arg)
2112 {
2113         struct comm_signal* comsig = (struct comm_signal*)arg;
2114         if(!(event & EV_SIGNAL))
2115                 return;
2116         comm_base_now(comsig->base);
2117         fptr_ok(fptr_whitelist_comm_signal(comsig->callback));
2118         (*comsig->callback)(sig, comsig->cb_arg);
2119 }
2120
2121 int
2122 comm_signal_bind(struct comm_signal* comsig, int sig)
2123 {
2124         struct internal_signal* entry = (struct internal_signal*)calloc(1,
2125                 sizeof(struct internal_signal));
2126         if(!entry) {
2127                 log_err("malloc failed");
2128                 return 0;
2129         }
2130         log_assert(comsig);
2131         /* add signal event */
2132         signal_set(&entry->ev, sig, comm_signal_callback, comsig);
2133         if(event_base_set(comsig->base->eb->base, &entry->ev) != 0) {
2134                 log_err("Could not set signal base");
2135                 free(entry);
2136                 return 0;
2137         }
2138         if(signal_add(&entry->ev, NULL) != 0) {
2139                 log_err("Could not add signal handler");
2140                 free(entry);
2141                 return 0;
2142         }
2143         /* link into list */
2144         entry->next = comsig->ev_signal;
2145         comsig->ev_signal = entry;
2146         return 1;
2147 }
2148
2149 void
2150 comm_signal_delete(struct comm_signal* comsig)
2151 {
2152         struct internal_signal* p, *np;
2153         if(!comsig)
2154                 return;
2155         p=comsig->ev_signal;
2156         while(p) {
2157                 np = p->next;
2158                 signal_del(&p->ev);
2159                 free(p);
2160                 p = np;
2161         }
2162         free(comsig);
2163 }