/src/unbound/util/netevent.c

Source
/*
 * util/netevent.c - event notification
 *
 * Copyright (c) 2007, NLnet Labs. All rights reserved.
 *
 * This software is open source.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * Redistributions of source code must retain the above copyright notice,
 * this list of conditions and the following disclaimer.
 *
 * Redistributions in binary form must reproduce the above copyright notice,
 * this list of conditions and the following disclaimer in the documentation
 * and/or other materials provided with the distribution.
 *
 * Neither the name of the NLNET LABS nor the names of its contributors may
 * be used to endorse or promote products derived from this software without
 * specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/**
 * \file
 *
 * This file contains event notification functions.
 */
#include "config.h"
#include "util/netevent.h"
#include "util/ub_event.h"
#include "util/log.h"
#include "util/net_help.h"
#include "util/tcp_conn_limit.h"
#include "util/fptr_wlist.h"
#include "util/proxy_protocol.h"
#include "util/timeval_func.h"
#include "sldns/pkthdr.h"
#include "sldns/sbuffer.h"
#include "sldns/str2wire.h"
#include "dnstap/dnstap.h"
#include "dnscrypt/dnscrypt.h"
#include "services/listen_dnsport.h"
#include "util/random.h"
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_SYS_SOCKET_H
#include <sys/socket.h>
#endif
#ifdef HAVE_NETDB_H
#include <netdb.h>
#endif
#ifdef HAVE_POLL_H
#include <poll.h>
#endif

#ifdef HAVE_OPENSSL_SSL_H
#include <openssl/ssl.h>
#endif
#ifdef HAVE_OPENSSL_ERR_H
#include <openssl/err.h>
#endif

#ifdef HAVE_NGTCP2
#include <ngtcp2/ngtcp2.h>
#include <ngtcp2/ngtcp2_crypto.h>
#endif

#ifdef HAVE_LINUX_NET_TSTAMP_H
#include <linux/net_tstamp.h>
#endif

/* -------- Start of local definitions -------- */
/** if CMSG_ALIGN is not defined on this platform, a workaround */
#ifndef CMSG_ALIGN
#  ifdef __CMSG_ALIGN
#    define CMSG_ALIGN(n) __CMSG_ALIGN(n)
#  elif defined(CMSG_DATA_ALIGN)
#    define CMSG_ALIGN _CMSG_DATA_ALIGN
#  else
#    define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1))
#  endif
#endif

/** if CMSG_LEN is not defined on this platform, a workaround */
#ifndef CMSG_LEN
#  define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len))
#endif

/** if CMSG_SPACE is not defined on this platform, a workaround */
#ifndef CMSG_SPACE
#  ifdef _CMSG_HDR_ALIGN
#    define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr)))
#  else
#    define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr)))
#  endif
#endif

/** The TCP writing query timeout in milliseconds */
#define TCP_QUERY_TIMEOUT 120000
/** The minimum actual TCP timeout to use, regardless of what we advertise,
 * in msec */
#define TCP_QUERY_TIMEOUT_MINIMUM 200

#ifndef NONBLOCKING_IS_BROKEN
/** number of UDP reads to perform per read indication from select */
#define NUM_UDP_PER_SELECT 100
#else
#define NUM_UDP_PER_SELECT 1
#endif

/** timeout in millisec to wait for write to unblock, packets dropped after.*/
#define SEND_BLOCKED_WAIT_TIMEOUT 200
/** max number of times to wait for write to unblock, packets dropped after.*/
#define SEND_BLOCKED_MAX_RETRY 5

/** Let's make timestamping code cleaner and redefine SO_TIMESTAMP* */
#ifndef SO_TIMESTAMP
#define SO_TIMESTAMP 29
#endif
#ifndef SO_TIMESTAMPNS
#define SO_TIMESTAMPNS 35
#endif
#ifndef SO_TIMESTAMPING
#define SO_TIMESTAMPING 37
#endif
/**
 * The internal event structure for keeping ub_event info for the event.
 * Possibly other structures (list, tree) this is part of.
 */
struct internal_event {
  /** the comm base */
  struct comm_base* base;
  /** ub_event event type */
  struct ub_event* ev;
};

/**
 * Internal base structure, so that every thread has its own events.
 */
struct internal_base {
  /** ub_event event_base type. */
  struct ub_event_base* base;
  /** seconds time pointer points here */
  time_t secs;
  /** timeval with current time */
  struct timeval now;
  /** the event used for slow_accept timeouts */
  struct ub_event* slow_accept;
  /** true if slow_accept is enabled */
  int slow_accept_enabled;
  /** last log time for slow logging of file descriptor errors */
  time_t last_slow_log;
  /** last log time for slow logging of write wait failures */
  time_t last_writewait_log;
};

/**
 * Internal timer structure, to store timer event in.
 */
struct internal_timer {
  /** the super struct from which derived */
  struct comm_timer super;
  /** the comm base */
  struct comm_base* base;
  /** ub_event event type */
  struct ub_event* ev;
  /** is timer enabled */
  uint8_t enabled;
};

/**
 * Internal signal structure, to store signal event in.
 */
struct internal_signal {
  /** ub_event event type */
  struct ub_event* ev;
  /** next in signal list */
  struct internal_signal* next;
};

/** create a tcp handler with a parent */
static struct comm_point* comm_point_create_tcp_handler(
  struct comm_base *base, struct comm_point* parent, size_t bufsize,
  struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
  void* callback_arg, struct unbound_socket* socket);

/* -------- End of local definitions -------- */

struct comm_base*
comm_base_create(int sigs)
{
  struct comm_base* b = (struct comm_base*)calloc(1,
    sizeof(struct comm_base));
  const char *evnm="event", *evsys="", *evmethod="";

  if(!b)
    return NULL;
  b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
  if(!b->eb) {
    free(b);
    return NULL;
  }
  b->eb->base = ub_default_event_base(sigs, &b->eb->secs, &b->eb->now);
  if(!b->eb->base) {
    free(b->eb);
    free(b);
    return NULL;
  }
  ub_comm_base_now(b);
  ub_get_event_sys(b->eb->base, &evnm, &evsys, &evmethod);
  verbose(VERB_ALGO, "%s %s uses %s method.", evnm, evsys, evmethod);
  return b;
}

struct comm_base*
comm_base_create_event(struct ub_event_base* base)
{
  struct comm_base* b = (struct comm_base*)calloc(1,
    sizeof(struct comm_base));
  if(!b)
    return NULL;
  b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
  if(!b->eb) {
    free(b);
    return NULL;
  }
  b->eb->base = base;
  ub_comm_base_now(b);
  return b;
}

void
comm_base_delete(struct comm_base* b)
{
  if(!b)
    return;
  if(b->eb->slow_accept_enabled) {
    if(ub_event_del(b->eb->slow_accept) != 0) {
      log_err("could not event_del slow_accept");
    }
    ub_event_free(b->eb->slow_accept);
  }
  ub_event_base_free(b->eb->base);
  b->eb->base = NULL;
  free(b->eb);
  free(b);
}

void
comm_base_delete_no_base(struct comm_base* b)
{
  if(!b)
    return;
  if(b->eb->slow_accept_enabled) {
    if(ub_event_del(b->eb->slow_accept) != 0) {
      log_err("could not event_del slow_accept");
    }
    ub_event_free(b->eb->slow_accept);
  }
  b->eb->base = NULL;
  free(b->eb);
  free(b);
}

void
comm_base_timept(struct comm_base* b, time_t** tt, struct timeval** tv)
{
  *tt = &b->eb->secs;
  *tv = &b->eb->now;
}

void
comm_base_dispatch(struct comm_base* b)
{
  int retval;
  retval = ub_event_base_dispatch(b->eb->base);
  if(retval < 0) {
    fatal_exit("event_dispatch returned error %d, "
      "errno is %s", retval, strerror(errno));
  }
}

void comm_base_exit(struct comm_base* b)
{
  if(ub_event_base_loopexit(b->eb->base) != 0) {
    log_err("Could not loopexit");
  }
}

void comm_base_set_slow_accept_handlers(struct comm_base* b,
  void (*stop_acc)(void*), void (*start_acc)(void*), void* arg)
{
  b->stop_accept = stop_acc;
  b->start_accept = start_acc;
  b->cb_arg = arg;
}

struct ub_event_base* comm_base_internal(struct comm_base* b)
{
  return b->eb->base;
}

struct ub_event* comm_point_internal(struct comm_point* c)
{
  return c->ev->ev;
}

/** see if errno for udp has to be logged or not uses globals */
static int
udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
{
  /* do not log transient errors (unless high verbosity) */
#if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN)
  switch(errno) {
#  ifdef ENETUNREACH
    case ENETUNREACH:
#  endif
#  ifdef EHOSTDOWN
    case EHOSTDOWN:
#  endif
#  ifdef EHOSTUNREACH
    case EHOSTUNREACH:
#  endif
#  ifdef ENETDOWN
    case ENETDOWN:
#  endif
    case EPERM:
    case EACCES:
      if(verbosity < VERB_ALGO)
        return 0;
      break;
    default:
      break;
  }
#endif
  /* permission denied is gotten for every send if the
   * network is disconnected (on some OS), squelch it */
  if( ((errno == EPERM)
#  ifdef EADDRNOTAVAIL
    /* 'Cannot assign requested address' also when disconnected */
    || (errno == EADDRNOTAVAIL)
#  endif
    ) && verbosity < VERB_ALGO)
    return 0;
#  ifdef EADDRINUSE
  /* If SO_REUSEADDR is set, we could try to connect to the same server
   * from the same source port twice. */
  if(errno == EADDRINUSE && verbosity < VERB_DETAIL)
    return 0;
#  endif
  /* squelch errors where people deploy AAAA ::ffff:bla for
   * authority servers, which we try for intranets. */
  if(errno == EINVAL && addr_is_ip4mapped(
    (struct sockaddr_storage*)addr, addrlen) &&
    verbosity < VERB_DETAIL)
    return 0;
  /* SO_BROADCAST sockopt can give access to 255.255.255.255,
   * but a dns cache does not need it. */
  if(errno == EACCES && addr_is_broadcast(
    (struct sockaddr_storage*)addr, addrlen) &&
    verbosity < VERB_DETAIL)
    return 0;
#  ifdef ENOTCONN
  /* For 0.0.0.0, ::0 targets it can return that socket is not connected.
   * This can be ignored, and the address skipped. It remains
   * possible to send there for completeness in configuration. */
  if(errno == ENOTCONN && addr_is_any(
    (struct sockaddr_storage*)addr, addrlen) &&
    verbosity < VERB_DETAIL)
    return 0;
#  endif
  return 1;
}

int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
{
  return udp_send_errno_needs_log(addr, addrlen);
}

/* send a UDP reply */
int
comm_point_send_udp_msg(struct comm_point *c, sldns_buffer* packet,
  struct sockaddr* addr, socklen_t addrlen, int is_connected)
{
  ssize_t sent;
  log_assert(c->fd != -1);
#ifdef UNBOUND_DEBUG
  if(sldns_buffer_remaining(packet) == 0)
    log_err("error: send empty UDP packet");
#endif
  log_assert(addr && addrlen > 0);
  if(!is_connected) {
    sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
      sldns_buffer_remaining(packet), 0,
      addr, addrlen);
  } else {
    sent = send(c->fd, (void*)sldns_buffer_begin(packet),
      sldns_buffer_remaining(packet), 0);
  }
  if(sent == -1) {
    /* try again and block, waiting for IO to complete,
     * we want to send the answer, and we will wait for
     * the ethernet interface buffer to have space. */
#ifndef USE_WINSOCK
    if(errno == EAGAIN || errno == EINTR ||
#  ifdef EWOULDBLOCK
      errno == EWOULDBLOCK ||
#  endif
      errno == ENOBUFS) {
#else
    if(WSAGetLastError() == WSAEINPROGRESS ||
      WSAGetLastError() == WSAEINTR ||
      WSAGetLastError() == WSAENOBUFS ||
      WSAGetLastError() == WSAEWOULDBLOCK) {
#endif
      int retries = 0;
      /* if we set the fd blocking, other threads suddenly
       * have a blocking fd that they operate on */
      while(sent == -1 && retries < SEND_BLOCKED_MAX_RETRY && (
#ifndef USE_WINSOCK
        errno == EAGAIN || errno == EINTR ||
#  ifdef EWOULDBLOCK
        errno == EWOULDBLOCK ||
#  endif
        errno == ENOBUFS
#else
        WSAGetLastError() == WSAEINPROGRESS ||
        WSAGetLastError() == WSAEINTR ||
        WSAGetLastError() == WSAENOBUFS ||
        WSAGetLastError() == WSAEWOULDBLOCK
#endif
      )) {
#if defined(HAVE_POLL) || defined(USE_WINSOCK)
        int send_nobufs = (
#ifndef USE_WINSOCK
          errno == ENOBUFS
#else
          WSAGetLastError() == WSAENOBUFS
#endif
        );
        struct pollfd p;
        int pret;
        memset(&p, 0, sizeof(p));
        p.fd = c->fd;
        p.events = POLLOUT
#ifndef USE_WINSOCK
          | POLLERR | POLLHUP
#endif
          ;
#  ifndef USE_WINSOCK
        pret = poll(&p, 1, SEND_BLOCKED_WAIT_TIMEOUT);
#  else
        pret = WSAPoll(&p, 1,
          SEND_BLOCKED_WAIT_TIMEOUT);
#  endif
        if(pret == 0) {
          /* timer expired */
          struct comm_base* b = c->ev->base;
          if(b->eb->last_writewait_log+SLOW_LOG_TIME <=
            b->eb->secs) {
            b->eb->last_writewait_log = b->eb->secs;
            verbose(VERB_OPS, "send udp blocked "
              "for long, dropping packet.");
          }
          return 0;
        } else if(pret < 0 &&
#ifndef USE_WINSOCK
          errno != EAGAIN && errno != EINTR &&
#  ifdef EWOULDBLOCK
          errno != EWOULDBLOCK &&
#  endif
          errno != ENOMEM && errno != ENOBUFS
#else
          WSAGetLastError() != WSAEINPROGRESS &&
          WSAGetLastError() != WSAEINTR &&
          WSAGetLastError() != WSAENOBUFS &&
          WSAGetLastError() != WSAEWOULDBLOCK
#endif
          ) {
          log_err("poll udp out failed: %s",
            sock_strerror(errno));
          return 0;
        } else if((pret < 0 &&
#ifndef USE_WINSOCK
          ( errno == ENOBUFS  /* Maybe some systems */
          || errno == ENOMEM  /* Linux */
          || errno == EAGAIN)  /* Macos, solaris, openbsd */
#else
          WSAGetLastError() == WSAENOBUFS
#endif
          ) || (send_nobufs && retries > 0)) {
          /* ENOBUFS/ENOMEM/EAGAIN, and poll
           * returned without
           * a timeout. Or the retried send call
           * returned ENOBUFS/ENOMEM/EAGAIN.
           * It is good to wait a bit for the
           * error to clear. */
          /* The timeout is 20*(2^(retries+1)),
           * it increases exponentially, starting
           * at 40 msec. After 5 tries, 1240 msec
           * have passed in total, when poll
           * returned the error, and 1200 msec
           * when send returned the errors. */
#ifndef USE_WINSOCK
          pret = poll(NULL, 0, (SEND_BLOCKED_WAIT_TIMEOUT/10)<<(retries+1));
#else
          Sleep((SEND_BLOCKED_WAIT_TIMEOUT/10)<<(retries+1));
          pret = 0;
#endif
          if(pret < 0
#ifndef USE_WINSOCK
            && errno != EAGAIN && errno != EINTR &&
#  ifdef EWOULDBLOCK
            errno != EWOULDBLOCK &&
#  endif
            errno != ENOMEM && errno != ENOBUFS
#else
            /* Sleep does not error */
#endif
          ) {
            log_err("poll udp out timer failed: %s",
              sock_strerror(errno));
          }
        }
#endif /* defined(HAVE_POLL) || defined(USE_WINSOCK) */
        retries++;
        if (!is_connected) {
          sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
            sldns_buffer_remaining(packet), 0,
            addr, addrlen);
        } else {
          sent = send(c->fd, (void*)sldns_buffer_begin(packet),
            sldns_buffer_remaining(packet), 0);
        }
      }
    }
  }
  if(sent == -1) {
    if(!udp_send_errno_needs_log(addr, addrlen))
      return 0;
    if (!is_connected) {
      verbose(VERB_OPS, "sendto failed: %s", sock_strerror(errno));
    } else {
      verbose(VERB_OPS, "send failed: %s", sock_strerror(errno));
    }
    if(addr)
      log_addr(VERB_OPS, "remote address is",
        (struct sockaddr_storage*)addr, addrlen);
    return 0;
  } else if((size_t)sent != sldns_buffer_remaining(packet)) {
    log_err("sent %d in place of %d bytes",
      (int)sent, (int)sldns_buffer_remaining(packet));
    return 0;
  }
  return 1;
}

#if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG))
/** print debug ancillary info */
static void p_ancil(const char* str, struct comm_reply* r)
{
  if(r->srctype != 4 && r->srctype != 6) {
    log_info("%s: unknown srctype %d", str, r->srctype);
    return;
  }

  if(r->srctype == 6) {
#ifdef IPV6_PKTINFO
    char buf[1024];
    if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr,
      buf, (socklen_t)sizeof(buf)) == 0) {
      (void)strlcpy(buf, "(inet_ntop error)", sizeof(buf));
    }
    buf[sizeof(buf)-1]=0;
    log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex);
#endif
  } else if(r->srctype == 4) {
#ifdef IP_PKTINFO
    char buf1[1024], buf2[1024];
    if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr,
      buf1, (socklen_t)sizeof(buf1)) == 0) {
      (void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
    }
    buf1[sizeof(buf1)-1]=0;
#ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
    if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst,
      buf2, (socklen_t)sizeof(buf2)) == 0) {
      (void)strlcpy(buf2, "(inet_ntop error)", sizeof(buf2));
    }
    buf2[sizeof(buf2)-1]=0;
#else
    buf2[0]=0;
#endif
    log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex,
      buf1, buf2);
#elif defined(IP_RECVDSTADDR)
    char buf1[1024];
    if(inet_ntop(AF_INET, &r->pktinfo.v4addr,
      buf1, (socklen_t)sizeof(buf1)) == 0) {
      (void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
    }
    buf1[sizeof(buf1)-1]=0;
    log_info("%s: %s", str, buf1);
#endif /* IP_PKTINFO or PI_RECVDSTDADDR */
  }
}
#endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */

/** send a UDP reply over specified interface*/
static int
comm_point_send_udp_msg_if(struct comm_point *c, sldns_buffer* packet,
  struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r)
{
#if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG)
  ssize_t sent;
  struct msghdr msg;
  struct iovec iov[1];
  union {
    struct cmsghdr hdr;
    char buf[256];
  } control;
#ifndef S_SPLINT_S
  struct cmsghdr *cmsg;
#endif /* S_SPLINT_S */

  log_assert(c->fd != -1);
#ifdef UNBOUND_DEBUG
  if(sldns_buffer_remaining(packet) == 0)
    log_err("error: send empty UDP packet");
#endif
  log_assert(addr && addrlen > 0);

  msg.msg_name = addr;
  msg.msg_namelen = addrlen;
  iov[0].iov_base = sldns_buffer_begin(packet);
  iov[0].iov_len = sldns_buffer_remaining(packet);
  msg.msg_iov = iov;
  msg.msg_iovlen = 1;
  msg.msg_control = control.buf;
#ifndef S_SPLINT_S
  msg.msg_controllen = sizeof(control.buf);
#endif /* S_SPLINT_S */
  msg.msg_flags = 0;

#ifndef S_SPLINT_S
  cmsg = CMSG_FIRSTHDR(&msg);
  if(r->srctype == 4) {
#ifdef IP_PKTINFO
    void* cmsg_data;
    msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
    log_assert(msg.msg_controllen <= sizeof(control.buf));
    cmsg->cmsg_level = IPPROTO_IP;
    cmsg->cmsg_type = IP_PKTINFO;
    memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info,
      sizeof(struct in_pktinfo));
    /* unset the ifindex to not bypass the routing tables */
    cmsg_data = CMSG_DATA(cmsg);
    ((struct in_pktinfo *) cmsg_data)->ipi_ifindex = 0;
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
    /* zero the padding bytes inserted by the CMSG_LEN */
    if(sizeof(struct in_pktinfo) < cmsg->cmsg_len)
      memset(((uint8_t*)(CMSG_DATA(cmsg))) +
        sizeof(struct in_pktinfo), 0, cmsg->cmsg_len
        - sizeof(struct in_pktinfo));
#elif defined(IP_SENDSRCADDR)
    msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
    log_assert(msg.msg_controllen <= sizeof(control.buf));
    cmsg->cmsg_level = IPPROTO_IP;
    cmsg->cmsg_type = IP_SENDSRCADDR;
    memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr,
      sizeof(struct in_addr));
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
    /* zero the padding bytes inserted by the CMSG_LEN */
    if(sizeof(struct in_addr) < cmsg->cmsg_len)
      memset(((uint8_t*)(CMSG_DATA(cmsg))) +
        sizeof(struct in_addr), 0, cmsg->cmsg_len
        - sizeof(struct in_addr));
#else
    verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR");
    msg.msg_control = NULL;
#endif /* IP_PKTINFO or IP_SENDSRCADDR */
  } else if(r->srctype == 6) {
    void* cmsg_data;
    msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
    log_assert(msg.msg_controllen <= sizeof(control.buf));
    cmsg->cmsg_level = IPPROTO_IPV6;
    cmsg->cmsg_type = IPV6_PKTINFO;
    memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info,
      sizeof(struct in6_pktinfo));
    /* unset the ifindex to not bypass the routing tables */
    cmsg_data = CMSG_DATA(cmsg);
    ((struct in6_pktinfo *) cmsg_data)->ipi6_ifindex = 0;
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
    /* zero the padding bytes inserted by the CMSG_LEN */
    if(sizeof(struct in6_pktinfo) < cmsg->cmsg_len)
      memset(((uint8_t*)(CMSG_DATA(cmsg))) +
        sizeof(struct in6_pktinfo), 0, cmsg->cmsg_len
        - sizeof(struct in6_pktinfo));
  } else {
    /* try to pass all 0 to use default route */
    msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
    log_assert(msg.msg_controllen <= sizeof(control.buf));
    cmsg->cmsg_level = IPPROTO_IPV6;
    cmsg->cmsg_type = IPV6_PKTINFO;
    memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo));
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
    /* zero the padding bytes inserted by the CMSG_LEN */
    if(sizeof(struct in6_pktinfo) < cmsg->cmsg_len)
      memset(((uint8_t*)(CMSG_DATA(cmsg))) +
        sizeof(struct in6_pktinfo), 0, cmsg->cmsg_len
        - sizeof(struct in6_pktinfo));
  }
#endif /* S_SPLINT_S */
  if(verbosity >= VERB_ALGO && r->srctype != 0)
    p_ancil("send_udp over interface", r);
  sent = sendmsg(c->fd, &msg, 0);
  if(sent == -1) {
    /* try again and block, waiting for IO to complete,
     * we want to send the answer, and we will wait for
     * the ethernet interface buffer to have space. */
#ifndef USE_WINSOCK
    if(errno == EAGAIN || errno == EINTR ||
#  ifdef EWOULDBLOCK
      errno == EWOULDBLOCK ||
#  endif
      errno == ENOBUFS) {
#else
    if(WSAGetLastError() == WSAEINPROGRESS ||
      WSAGetLastError() == WSAEINTR ||
      WSAGetLastError() == WSAENOBUFS ||
      WSAGetLastError() == WSAEWOULDBLOCK) {
#endif
      int retries = 0;
      while(sent == -1 && retries < SEND_BLOCKED_MAX_RETRY && (
#ifndef USE_WINSOCK
        errno == EAGAIN || errno == EINTR ||
#  ifdef EWOULDBLOCK
        errno == EWOULDBLOCK ||
#  endif
        errno == ENOBUFS
#else
        WSAGetLastError() == WSAEINPROGRESS ||
        WSAGetLastError() == WSAEINTR ||
        WSAGetLastError() == WSAENOBUFS ||
        WSAGetLastError() == WSAEWOULDBLOCK
#endif
      )) {
#if defined(HAVE_POLL) || defined(USE_WINSOCK)
        int send_nobufs = (
#ifndef USE_WINSOCK
          errno == ENOBUFS
#else
          WSAGetLastError() == WSAENOBUFS
#endif
        );
        struct pollfd p;
        int pret;
        memset(&p, 0, sizeof(p));
        p.fd = c->fd;
        p.events = POLLOUT
#ifndef USE_WINSOCK
          | POLLERR | POLLHUP
#endif
          ;
#  ifndef USE_WINSOCK
        pret = poll(&p, 1, SEND_BLOCKED_WAIT_TIMEOUT);
#  else
        pret = WSAPoll(&p, 1,
          SEND_BLOCKED_WAIT_TIMEOUT);
#  endif
        if(pret == 0) {
          /* timer expired */
          struct comm_base* b = c->ev->base;
          if(b->eb->last_writewait_log+SLOW_LOG_TIME <=
            b->eb->secs) {
            b->eb->last_writewait_log = b->eb->secs;
            verbose(VERB_OPS, "send udp blocked "
              "for long, dropping packet.");
          }
          return 0;
        } else if(pret < 0 &&
#ifndef USE_WINSOCK
          errno != EAGAIN && errno != EINTR &&
#  ifdef EWOULDBLOCK
          errno != EWOULDBLOCK &&
#  endif
          errno != ENOMEM && errno != ENOBUFS
#else
          WSAGetLastError() != WSAEINPROGRESS &&
          WSAGetLastError() != WSAEINTR &&
          WSAGetLastError() != WSAENOBUFS &&
          WSAGetLastError() != WSAEWOULDBLOCK
#endif
          ) {
          log_err("poll udp out failed: %s",
            sock_strerror(errno));
          return 0;
        } else if((pret < 0 &&
#ifndef USE_WINSOCK
          ( errno == ENOBUFS  /* Maybe some systems */
          || errno == ENOMEM  /* Linux */
          || errno == EAGAIN)  /* Macos, solaris, openbsd */
#else
          WSAGetLastError() == WSAENOBUFS
#endif
          ) || (send_nobufs && retries > 0)) {
          /* ENOBUFS/ENOMEM/EAGAIN, and poll
           * returned without
           * a timeout. Or the retried send call
           * returned ENOBUFS/ENOMEM/EAGAIN.
           * It is good to wait a bit for the
           * error to clear. */
          /* The timeout is 20*(2^(retries+1)),
           * it increases exponentially, starting
           * at 40 msec. After 5 tries, 1240 msec
           * have passed in total, when poll
           * returned the error, and 1200 msec
           * when send returned the errors. */
#ifndef USE_WINSOCK
          pret = poll(NULL, 0, (SEND_BLOCKED_WAIT_TIMEOUT/10)<<(retries+1));
#else
          Sleep((SEND_BLOCKED_WAIT_TIMEOUT/10)<<(retries+1));
          pret = 0;
#endif
          if(pret < 0
#ifndef USE_WINSOCK
            && errno != EAGAIN && errno != EINTR &&
#  ifdef EWOULDBLOCK
            errno != EWOULDBLOCK &&
#  endif
            errno != ENOMEM && errno != ENOBUFS
#else  /* USE_WINSOCK */
            /* Sleep does not error */
#endif
          ) {
            log_err("poll udp out timer failed: %s",
              sock_strerror(errno));
          }
        }
#endif /* defined(HAVE_POLL) || defined(USE_WINSOCK) */
        retries++;
        sent = sendmsg(c->fd, &msg, 0);
      }
    }
  }
  if(sent == -1) {
    if(!udp_send_errno_needs_log(addr, addrlen))
      return 0;
    verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno));
    log_addr(VERB_OPS, "remote address is",
      (struct sockaddr_storage*)addr, addrlen);
#ifdef __NetBSD__
    /* netbsd 7 has IP_PKTINFO for recv but not send */
    if(errno == EINVAL && r->srctype == 4)
      log_err("sendmsg: No support for sendmsg(IP_PKTINFO). "
        "Please disable interface-automatic");
#endif
    return 0;
  } else if((size_t)sent != sldns_buffer_remaining(packet)) {
    log_err("sent %d in place of %d bytes",
      (int)sent, (int)sldns_buffer_remaining(packet));
    return 0;
  }
  return 1;
#else
  (void)c;
  (void)packet;
  (void)addr;
  (void)addrlen;
  (void)r;
  log_err("sendmsg: IPV6_PKTINFO not supported");
  return 0;
#endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */
}

/** return true is UDP receive error needs to be logged */
static int udp_recv_needs_log(int err)
{
  switch(err) {
  case EACCES: /* some hosts send ICMP 'Permission Denied' */
#ifndef USE_WINSOCK
  case ECONNREFUSED:
#  ifdef ENETUNREACH
  case ENETUNREACH:
#  endif
#  ifdef EHOSTDOWN
  case EHOSTDOWN:
#  endif
#  ifdef EHOSTUNREACH
  case EHOSTUNREACH:
#  endif
#  ifdef ENETDOWN
  case ENETDOWN:
#  endif
#else /* USE_WINSOCK */
  case WSAECONNREFUSED:
  case WSAENETUNREACH:
  case WSAEHOSTDOWN:
  case WSAEHOSTUNREACH:
  case WSAENETDOWN:
#endif
    if(verbosity >= VERB_ALGO)
      return 1;
    return 0;
  default:
    break;
  }
  return 1;
}

/** Parses the PROXYv2 header from buf and updates the comm_reply struct.
 *  Returns 1 on success, 0 on failure. */
static int consume_pp2_header(struct sldns_buffer* buf, struct comm_reply* rep,
  int stream) {
  size_t size;
  struct pp2_header *header;
  int err = pp2_read_header(sldns_buffer_begin(buf),
    sldns_buffer_remaining(buf));
  if(err) return 0;
  header = (struct pp2_header*)sldns_buffer_begin(buf);
  size = PP2_HEADER_SIZE + ntohs(header->len);
  if((header->ver_cmd & 0xF) == PP2_CMD_LOCAL) {
    /* A connection from the proxy itself.
     * No need to do anything with addresses. */
    goto done;
  }
  if(header->fam_prot == PP2_UNSPEC_UNSPEC) {
    /* Unspecified family and protocol. This could be used for
     * health checks by proxies.
     * No need to do anything with addresses. */
    goto done;
  }
  /* Read the proxied address */
  switch(header->fam_prot) {
    case PP2_INET_STREAM:
    case PP2_INET_DGRAM:
      {
      struct sockaddr_in* addr =
        (struct sockaddr_in*)&rep->client_addr;
      addr->sin_family = AF_INET;
      addr->sin_addr.s_addr = header->addr.addr4.src_addr;
      addr->sin_port = header->addr.addr4.src_port;
      rep->client_addrlen = (socklen_t)sizeof(struct sockaddr_in);
      }
      /* Ignore the destination address; it should be us. */
      break;
    case PP2_INET6_STREAM:
    case PP2_INET6_DGRAM:
      {
      struct sockaddr_in6* addr =
        (struct sockaddr_in6*)&rep->client_addr;
      memset(addr, 0, sizeof(*addr));
      addr->sin6_family = AF_INET6;
      memcpy(&addr->sin6_addr,
        header->addr.addr6.src_addr, 16);
      addr->sin6_port = header->addr.addr6.src_port;
      rep->client_addrlen = (socklen_t)sizeof(struct sockaddr_in6);
      }
      /* Ignore the destination address; it should be us. */
      break;
    default:
      log_err("proxy_protocol: unsupported family and "
        "protocol 0x%x", (int)header->fam_prot);
      return 0;
  }
  rep->is_proxied = 1;
done:
  if(!stream) {
    /* We are reading a whole packet;
     * Move the rest of the data to overwrite the PROXYv2 header */
    /* XXX can we do better to avoid memmove? */
    memmove(header, ((char*)header)+size,
      sldns_buffer_limit(buf)-size);
    sldns_buffer_set_limit(buf, sldns_buffer_limit(buf)-size);
  }
  return 1;
}

#if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
void
comm_point_udp_ancil_callback(int fd, short event, void* arg)
{
  struct comm_reply rep;
  struct msghdr msg;
  struct iovec iov[1];
  ssize_t rcv;
  union {
    struct cmsghdr hdr;
    char buf[256];
  } ancil;
  int i;
#ifndef S_SPLINT_S
  struct cmsghdr* cmsg;
#endif /* S_SPLINT_S */
#ifdef HAVE_LINUX_NET_TSTAMP_H
  struct timespec *ts;
#endif /* HAVE_LINUX_NET_TSTAMP_H */

  rep.c = (struct comm_point*)arg;
  log_assert(rep.c->type == comm_udp);

  if(!(event&UB_EV_READ))
    return;
  log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
  ub_comm_base_now(rep.c->ev->base);
  for(i=0; i<NUM_UDP_PER_SELECT; i++) {
    sldns_buffer_clear(rep.c->buffer);
    timeval_clear(&rep.c->recv_tv);
    rep.remote_addrlen = (socklen_t)sizeof(rep.remote_addr);
    log_assert(fd != -1);
    log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
    msg.msg_name = &rep.remote_addr;
    msg.msg_namelen = (socklen_t)sizeof(rep.remote_addr);
    iov[0].iov_base = sldns_buffer_begin(rep.c->buffer);
    iov[0].iov_len = sldns_buffer_remaining(rep.c->buffer);
    msg.msg_iov = iov;
    msg.msg_iovlen = 1;
    msg.msg_control = ancil.buf;
#ifndef S_SPLINT_S
    msg.msg_controllen = sizeof(ancil.buf);
#endif /* S_SPLINT_S */
    msg.msg_flags = 0;
    rcv = recvmsg(fd, &msg, MSG_DONTWAIT);
    if(rcv == -1) {
      if(errno != EAGAIN && errno != EINTR
        && udp_recv_needs_log(errno)) {
        log_err("recvmsg failed: %s", strerror(errno));
      }
      return;
    }
    rep.remote_addrlen = msg.msg_namelen;
    sldns_buffer_skip(rep.c->buffer, rcv);
    sldns_buffer_flip(rep.c->buffer);
    rep.srctype = 0;
    rep.is_proxied = 0;
#ifndef S_SPLINT_S
    for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL;
      cmsg = CMSG_NXTHDR(&msg, cmsg)) {
      if( cmsg->cmsg_level == IPPROTO_IPV6 &&
        cmsg->cmsg_type == IPV6_PKTINFO) {
        rep.srctype = 6;
        memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg),
          sizeof(struct in6_pktinfo));
        break;
#ifdef IP_PKTINFO
      } else if( cmsg->cmsg_level == IPPROTO_IP &&
        cmsg->cmsg_type == IP_PKTINFO) {
        rep.srctype = 4;
        memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg),
          sizeof(struct in_pktinfo));
        break;
#elif defined(IP_RECVDSTADDR)
      } else if( cmsg->cmsg_level == IPPROTO_IP &&
        cmsg->cmsg_type == IP_RECVDSTADDR) {
        rep.srctype = 4;
        memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg),
          sizeof(struct in_addr));
        break;
#endif /* IP_PKTINFO or IP_RECVDSTADDR */
#ifdef HAVE_LINUX_NET_TSTAMP_H
      } else if( cmsg->cmsg_level == SOL_SOCKET &&
        cmsg->cmsg_type == SO_TIMESTAMPNS) {
        ts = (struct timespec *)CMSG_DATA(cmsg);
        TIMESPEC_TO_TIMEVAL(&rep.c->recv_tv, ts);
      } else if( cmsg->cmsg_level == SOL_SOCKET &&
        cmsg->cmsg_type == SO_TIMESTAMPING) {
        ts = (struct timespec *)CMSG_DATA(cmsg);
        TIMESPEC_TO_TIMEVAL(&rep.c->recv_tv, ts);
      } else if( cmsg->cmsg_level == SOL_SOCKET &&
        cmsg->cmsg_type == SO_TIMESTAMP) {
        memmove(&rep.c->recv_tv, CMSG_DATA(cmsg), sizeof(struct timeval));
#elif defined(SO_TIMESTAMP) && defined(SCM_TIMESTAMP)
      } else if( cmsg->cmsg_level == SOL_SOCKET &&
        cmsg->cmsg_type == SCM_TIMESTAMP) {
        /* FreeBSD and also Linux. */
        memmove(&rep.c->recv_tv, CMSG_DATA(cmsg), sizeof(struct timeval));
#endif /* HAVE_LINUX_NET_TSTAMP_H */
      }
    }

    if(verbosity >= VERB_ALGO && rep.srctype != 0)
      p_ancil("receive_udp on interface", &rep);
#endif /* S_SPLINT_S */

    if(rep.c->pp2_enabled && !consume_pp2_header(rep.c->buffer,
      &rep, 0)) {
      log_err("proxy_protocol: could not consume PROXYv2 header");
      return;
    }
    if(!rep.is_proxied) {
      rep.client_addrlen = rep.remote_addrlen;
      memmove(&rep.client_addr, &rep.remote_addr,
        rep.remote_addrlen);
    }

    fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
    if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
      /* send back immediate reply */
      struct sldns_buffer *buffer;
#ifdef USE_DNSCRYPT
      buffer = rep.c->dnscrypt_buffer;
#else
      buffer = rep.c->buffer;
#endif
      (void)comm_point_send_udp_msg_if(rep.c, buffer,
        (struct sockaddr*)&rep.remote_addr,
        rep.remote_addrlen, &rep);
    }
    if(!rep.c || rep.c->fd == -1) /* commpoint closed */
      break;
  }
}
#endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */

void
comm_point_udp_callback(int fd, short event, void* arg)
{
  struct comm_reply rep;
  ssize_t rcv;
  int i;
  struct sldns_buffer *buffer;

  rep.c = (struct comm_point*)arg;
  log_assert(rep.c->type == comm_udp);

  if(!(event&UB_EV_READ))
    return;
  log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
  ub_comm_base_now(rep.c->ev->base);
  for(i=0; i<NUM_UDP_PER_SELECT; i++) {
    sldns_buffer_clear(rep.c->buffer);
    rep.remote_addrlen = (socklen_t)sizeof(rep.remote_addr);
    log_assert(fd != -1);
    log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
    rcv = recvfrom(fd, (void*)sldns_buffer_begin(rep.c->buffer),
      sldns_buffer_remaining(rep.c->buffer), MSG_DONTWAIT,
      (struct sockaddr*)&rep.remote_addr, &rep.remote_addrlen);
    if(rcv == -1) {
#ifndef USE_WINSOCK
      if(errno != EAGAIN && errno != EINTR
        && udp_recv_needs_log(errno))
        log_err("recvfrom %d failed: %s",
          fd, strerror(errno));
#else
      if(WSAGetLastError() != WSAEINPROGRESS &&
        WSAGetLastError() != WSAECONNRESET &&
        WSAGetLastError()!= WSAEWOULDBLOCK &&
        udp_recv_needs_log(WSAGetLastError()))
        log_err("recvfrom failed: %s",
          wsa_strerror(WSAGetLastError()));
#endif
      return;
    }
    sldns_buffer_skip(rep.c->buffer, rcv);
    sldns_buffer_flip(rep.c->buffer);
    rep.srctype = 0;
    rep.is_proxied = 0;

    if(rep.c->pp2_enabled && !consume_pp2_header(rep.c->buffer,
      &rep, 0)) {
      log_err("proxy_protocol: could not consume PROXYv2 header");
      return;
    }
    if(!rep.is_proxied) {
      rep.client_addrlen = rep.remote_addrlen;
      memmove(&rep.client_addr, &rep.remote_addr,
        rep.remote_addrlen);
    }

    fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
    if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
      /* send back immediate reply */
#ifdef USE_DNSCRYPT
      buffer = rep.c->dnscrypt_buffer;
#else
      buffer = rep.c->buffer;
#endif
      (void)comm_point_send_udp_msg(rep.c, buffer,
        (struct sockaddr*)&rep.remote_addr,
        rep.remote_addrlen, 0);
    }
    if(!rep.c || rep.c->fd != fd) /* commpoint closed to -1 or reused for
    another UDP port. Note rep.c cannot be reused with TCP fd. */
      break;
  }
}

#ifdef HAVE_NGTCP2
void
doq_pkt_addr_init(struct doq_pkt_addr* paddr)
{
  paddr->addrlen = (socklen_t)sizeof(paddr->addr);
  paddr->localaddrlen = (socklen_t)sizeof(paddr->localaddr);
  paddr->ifindex = 0;
}

/** set the ecn on the transmission */
static void
doq_set_ecn(int fd, int family, uint32_t ecn)
{
  unsigned int val = ecn;
  if(family == AF_INET6) {
    if(setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val,
      (socklen_t)sizeof(val)) == -1) {
      log_err("setsockopt(.. IPV6_TCLASS ..): %s",
        strerror(errno));
    }
    return;
  }
  if(setsockopt(fd, IPPROTO_IP, IP_TOS, &val,
    (socklen_t)sizeof(val)) == -1) {
    log_err("setsockopt(.. IP_TOS ..): %s",
      strerror(errno));
  }
}

/** set the local address in the control ancillary data */
static void
doq_set_localaddr_cmsg(struct msghdr* msg, size_t control_size,
  struct doq_addr_storage* localaddr, socklen_t localaddrlen,
  int ifindex)
{
#ifndef S_SPLINT_S
  struct cmsghdr* cmsg;
#endif /* S_SPLINT_S */
#ifndef S_SPLINT_S
  cmsg = CMSG_FIRSTHDR(msg);
  if(localaddr->sockaddr.in.sin_family == AF_INET) {
#ifdef IP_PKTINFO
    struct sockaddr_in* sa = (struct sockaddr_in*)localaddr;
    struct in_pktinfo v4info;
    log_assert(localaddrlen >= sizeof(struct sockaddr_in));
    msg->msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
    memset(msg->msg_control, 0, msg->msg_controllen);
    log_assert(msg->msg_controllen <= control_size);
    cmsg->cmsg_level = IPPROTO_IP;
    cmsg->cmsg_type = IP_PKTINFO;
    memset(&v4info, 0, sizeof(v4info));
#  ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
    memmove(&v4info.ipi_spec_dst, &sa->sin_addr,
      sizeof(struct in_addr));
#  else
    memmove(&v4info.ipi_addr, &sa->sin_addr,
      sizeof(struct in_addr));
#  endif
    v4info.ipi_ifindex = ifindex;
    memmove(CMSG_DATA(cmsg), &v4info, sizeof(struct in_pktinfo));
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
#elif defined(IP_SENDSRCADDR)
    struct sockaddr_in* sa= (struct sockaddr_in*)localaddr;
    log_assert(localaddrlen >= sizeof(struct sockaddr_in));
    msg->msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
    memset(msg->msg_control, 0, msg->msg_controllen);
    log_assert(msg->msg_controllen <= control_size);
    cmsg->cmsg_level = IPPROTO_IP;
    cmsg->cmsg_type = IP_SENDSRCADDR;
    memmove(CMSG_DATA(cmsg),  &sa->sin_addr,
      sizeof(struct in_addr));
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
#endif
  } else {
    struct sockaddr_in6* sa6 = (struct sockaddr_in6*)localaddr;
    struct in6_pktinfo v6info;
    log_assert(localaddrlen >= sizeof(struct sockaddr_in6));
    msg->msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
    memset(msg->msg_control, 0, msg->msg_controllen);
    log_assert(msg->msg_controllen <= control_size);
    cmsg->cmsg_level = IPPROTO_IPV6;
    cmsg->cmsg_type = IPV6_PKTINFO;
    memset(&v6info, 0, sizeof(v6info));
    memmove(&v6info.ipi6_addr, &sa6->sin6_addr,
      sizeof(struct in6_addr));
    v6info.ipi6_ifindex = ifindex;
    memmove(CMSG_DATA(cmsg), &v6info, sizeof(struct in6_pktinfo));
    cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
  }
#endif /* S_SPLINT_S */
  /* Ignore unused variables, if no assertions are compiled. */
  (void)localaddrlen;
  (void)control_size;
}

/** write address and port into strings */
static int
doq_print_addr_port(struct doq_addr_storage* addr, socklen_t addrlen,
  char* host, size_t hostlen, char* port, size_t portlen)
{
  if(addr->sockaddr.in.sin_family == AF_INET) {
    struct sockaddr_in* sa = (struct sockaddr_in*)addr;
    log_assert(addrlen >= sizeof(*sa));
    if(inet_ntop(sa->sin_family, &sa->sin_addr, host,
      (socklen_t)hostlen) == 0) {
      log_hex("inet_ntop error: address", &sa->sin_addr,
        sizeof(sa->sin_addr));
      return 0;
    }
    snprintf(port, portlen, "%u", (unsigned)ntohs(sa->sin_port));
  } else if(addr->sockaddr.in.sin_family == AF_INET6) {
    struct sockaddr_in6* sa6 = (struct sockaddr_in6*)addr;
    log_assert(addrlen >= sizeof(*sa6));
    if(inet_ntop(sa6->sin6_family, &sa6->sin6_addr, host,
      (socklen_t)hostlen) == 0) {
      log_hex("inet_ntop error: address", &sa6->sin6_addr,
        sizeof(sa6->sin6_addr));
      return 0;
    }
    snprintf(port, portlen, "%u", (unsigned)ntohs(sa6->sin6_port));
  }
  return 1;
}

/** doq store the blocked packet when write has blocked */
static void
doq_store_blocked_pkt(struct comm_point* c, struct doq_pkt_addr* paddr,
  uint32_t ecn)
{
  if(c->doq_socket->have_blocked_pkt)
    return; /* should not happen that we write when there is
    already a blocked write, but if so, drop it. */
  if(sldns_buffer_limit(c->doq_socket->pkt_buf) >
    sldns_buffer_capacity(c->doq_socket->blocked_pkt))
    return; /* impossibly large, drop packet. impossible because
    pkt_buf and blocked_pkt are the same size. */
  c->doq_socket->have_blocked_pkt = 1;
  c->doq_socket->blocked_pkt_pi.ecn = ecn;
  memcpy(c->doq_socket->blocked_paddr, paddr,
    sizeof(*c->doq_socket->blocked_paddr));
  sldns_buffer_clear(c->doq_socket->blocked_pkt);
  sldns_buffer_write(c->doq_socket->blocked_pkt,
    sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_limit(c->doq_socket->pkt_buf));
  sldns_buffer_flip(c->doq_socket->blocked_pkt);
}

void
doq_send_pkt(struct comm_point* c, struct doq_pkt_addr* paddr, uint32_t ecn)
{
  struct msghdr msg;
  struct iovec iov[1];
  union {
    struct cmsghdr hdr;
    char buf[256];
  } control;
  ssize_t ret;
  iov[0].iov_base = sldns_buffer_begin(c->doq_socket->pkt_buf);
  iov[0].iov_len = sldns_buffer_limit(c->doq_socket->pkt_buf);
  memset(&msg, 0, sizeof(msg));
  msg.msg_name = (void*)&paddr->addr;
  msg.msg_namelen = paddr->addrlen;
  msg.msg_iov = iov;
  msg.msg_iovlen = 1;
  msg.msg_control = control.buf;
#ifndef S_SPLINT_S
  msg.msg_controllen = sizeof(control.buf);
#endif /* S_SPLINT_S */
  msg.msg_flags = 0;

  doq_set_localaddr_cmsg(&msg, sizeof(control.buf), &paddr->localaddr,
    paddr->localaddrlen, paddr->ifindex);
  doq_set_ecn(c->fd, paddr->addr.sockaddr.in.sin_family, ecn);

  for(;;) {
    ret = sendmsg(c->fd, &msg, MSG_DONTWAIT);
    if(ret == -1 && errno == EINTR)
      continue;
    break;
  }
  if(ret == -1) {
#ifndef USE_WINSOCK
    if(errno == EAGAIN ||
#  ifdef EWOULDBLOCK
      errno == EWOULDBLOCK ||
#  endif
      errno == ENOBUFS)
#else
    if(WSAGetLastError() == WSAEINPROGRESS ||
      WSAGetLastError() == WSAENOBUFS ||
      WSAGetLastError() == WSAEWOULDBLOCK)
#endif
    {
      /* udp send has blocked */
      doq_store_blocked_pkt(c, paddr, ecn);
      return;
    }
    if(!udp_send_errno_needs_log((void*)&paddr->addr,
      paddr->addrlen))
      return;
    if(verbosity >= VERB_OPS) {
      char host[256], port[32];
      if(doq_print_addr_port(&paddr->addr, paddr->addrlen,
        host, sizeof(host), port, sizeof(port))) {
        verbose(VERB_OPS, "doq sendmsg to %s %s "
          "failed: %s", host, port,
          strerror(errno));
      } else {
        verbose(VERB_OPS, "doq sendmsg failed: %s",
          strerror(errno));
      }
    }
    return;
  } else if(ret != (ssize_t)sldns_buffer_limit(c->doq_socket->pkt_buf)) {
    char host[256], port[32];
    if(doq_print_addr_port(&paddr->addr, paddr->addrlen, host,
      sizeof(host), port, sizeof(port))) {
      log_err("doq sendmsg to %s %s failed: "
        "sent %d in place of %d bytes", 
        host, port, (int)ret,
        (int)sldns_buffer_limit(c->doq_socket->pkt_buf));
    } else {
      log_err("doq sendmsg failed: "
        "sent %d in place of %d bytes", 
        (int)ret, (int)sldns_buffer_limit(c->doq_socket->pkt_buf));
    }
    return;
  }
}

/** fetch port number */
static int
doq_sockaddr_get_port(struct doq_addr_storage* addr)
{
  if(addr->sockaddr.in.sin_family == AF_INET) {
    struct sockaddr_in* sa = (struct sockaddr_in*)addr;
    return ntohs(sa->sin_port);
  } else if(addr->sockaddr.in.sin_family == AF_INET6) {
    struct sockaddr_in6* sa6 = (struct sockaddr_in6*)addr;
    return ntohs(sa6->sin6_port);
  }
  return 0;
}

/** get local address from ancillary data headers */
static int
doq_get_localaddr_cmsg(struct comm_point* c, struct doq_pkt_addr* paddr,
  int* pkt_continue, struct msghdr* msg)
{
#ifndef S_SPLINT_S
  struct cmsghdr* cmsg;
#endif /* S_SPLINT_S */

  memset(&paddr->localaddr, 0, sizeof(paddr->localaddr));
#ifndef S_SPLINT_S
  for(cmsg = CMSG_FIRSTHDR(msg); cmsg != NULL;
    cmsg = CMSG_NXTHDR(msg, cmsg)) {
    if( cmsg->cmsg_level == IPPROTO_IPV6 &&
      cmsg->cmsg_type == IPV6_PKTINFO) {
      struct in6_pktinfo* v6info =
        (struct in6_pktinfo*)CMSG_DATA(cmsg);
      struct sockaddr_in6* sa= (struct sockaddr_in6*)
        &paddr->localaddr;
      struct sockaddr_in6* rema = (struct sockaddr_in6*)
        &paddr->addr;
      if(rema->sin6_family != AF_INET6) {
        log_err("doq cmsg family mismatch cmsg is ip6");
        *pkt_continue = 1;
        return 0;
      }
      sa->sin6_family = AF_INET6;
      sa->sin6_port = htons(doq_sockaddr_get_port(
        (void*)c->socket->addr));
      paddr->ifindex = v6info->ipi6_ifindex;
      memmove(&sa->sin6_addr, &v6info->ipi6_addr,
        sizeof(struct in6_addr));
      paddr->localaddrlen = sizeof(struct sockaddr_in6);
      break;
#ifdef IP_PKTINFO
    } else if( cmsg->cmsg_level == IPPROTO_IP &&
      cmsg->cmsg_type == IP_PKTINFO) {
      struct in_pktinfo* v4info =
        (struct in_pktinfo*)CMSG_DATA(cmsg);
      struct sockaddr_in* sa= (struct sockaddr_in*)
        &paddr->localaddr;
      struct sockaddr_in* rema = (struct sockaddr_in*)
        &paddr->addr;
      if(rema->sin_family != AF_INET) {
        log_err("doq cmsg family mismatch cmsg is ip4");
        *pkt_continue = 1;
        return 0;
      }
      sa->sin_family = AF_INET;
      sa->sin_port = htons(doq_sockaddr_get_port(
        (void*)c->socket->addr));
      paddr->ifindex = v4info->ipi_ifindex;
      memmove(&sa->sin_addr, &v4info->ipi_addr,
        sizeof(struct in_addr));
      paddr->localaddrlen = sizeof(struct sockaddr_in);
      break;
#elif defined(IP_RECVDSTADDR)
    } else if( cmsg->cmsg_level == IPPROTO_IP &&
      cmsg->cmsg_type == IP_RECVDSTADDR) {
      struct sockaddr_in* sa= (struct sockaddr_in*)
        &paddr->localaddr;
      struct sockaddr_in* rema = (struct sockaddr_in*)
        &paddr->addr;
      if(rema->sin_family != AF_INET) {
        log_err("doq cmsg family mismatch cmsg is ip4");
        *pkt_continue = 1;
        return 0;
      }
      sa->sin_family = AF_INET;
      sa->sin_port = htons(doq_sockaddr_get_port(
        (void*)c->socket->addr));
      paddr->ifindex = 0;
      memmove(&sa.sin_addr, CMSG_DATA(cmsg),
        sizeof(struct in_addr));
      paddr->localaddrlen = sizeof(struct sockaddr_in);
      break;
#endif /* IP_PKTINFO or IP_RECVDSTADDR */
    }
  }
#endif /* S_SPLINT_S */

return 1;
}

/** get packet ecn information */
static uint32_t
msghdr_get_ecn(struct msghdr* msg, int family)
{
#ifndef S_SPLINT_S
  struct cmsghdr* cmsg;
  if(family == AF_INET6) {
    for(cmsg = CMSG_FIRSTHDR(msg); cmsg != NULL;
      cmsg = CMSG_NXTHDR(msg, cmsg)) {
      if(cmsg->cmsg_level == IPPROTO_IPV6 &&
        cmsg->cmsg_type == IPV6_TCLASS &&
        cmsg->cmsg_len != 0) {
        uint8_t* ecn = (uint8_t*)CMSG_DATA(cmsg);
        return *ecn;
      }
    }
    return 0;
  }
  for(cmsg = CMSG_FIRSTHDR(msg); cmsg != NULL;
    cmsg = CMSG_NXTHDR(msg, cmsg)) {
    if(cmsg->cmsg_level == IPPROTO_IP &&
      cmsg->cmsg_type == IP_TOS &&
      cmsg->cmsg_len != 0) {
      uint8_t* ecn = (uint8_t*)CMSG_DATA(cmsg);
      return *ecn;
    }
  }
#endif /* S_SPLINT_S */
  return 0;
}

/** receive packet for DoQ on UDP. get ancillary data for addresses,
 * return false if failed and the callback can stop receiving UDP packets
 * if pkt_continue is false. */
static int
doq_recv(struct comm_point* c, struct doq_pkt_addr* paddr, int* pkt_continue,
  struct ngtcp2_pkt_info* pi)
{
  struct msghdr msg;
  struct iovec iov[1];
  ssize_t rcv;
  union {
    struct cmsghdr hdr;
    char buf[256];
  } ancil;

  msg.msg_name = &paddr->addr;
  msg.msg_namelen = (socklen_t)sizeof(paddr->addr);
  iov[0].iov_base = sldns_buffer_begin(c->doq_socket->pkt_buf);
  iov[0].iov_len = sldns_buffer_remaining(c->doq_socket->pkt_buf);
  msg.msg_iov = iov;
  msg.msg_iovlen = 1;
  msg.msg_control = ancil.buf;
#ifndef S_SPLINT_S
  msg.msg_controllen = sizeof(ancil.buf);
#endif /* S_SPLINT_S */
  msg.msg_flags = 0;

  rcv = recvmsg(c->fd, &msg, MSG_DONTWAIT);
  if(rcv == -1) {
    if(errno != EAGAIN && errno != EINTR
      && udp_recv_needs_log(errno)) {
      log_err("recvmsg failed for doq: %s", strerror(errno));
    }
    *pkt_continue = 0;
    return 0;
  }

  paddr->addrlen = msg.msg_namelen;
  sldns_buffer_skip(c->doq_socket->pkt_buf, rcv);
  sldns_buffer_flip(c->doq_socket->pkt_buf);
  if(!doq_get_localaddr_cmsg(c, paddr, pkt_continue, &msg))
    return 0;
  pi->ecn = msghdr_get_ecn(&msg, paddr->addr.sockaddr.in.sin_family);
  return 1;
}

/** send the version negotiation for doq. scid and dcid are flipped around
 * to send back to the client. */
static void
doq_send_version_negotiation(struct comm_point* c, struct doq_pkt_addr* paddr,
  const uint8_t* dcid, size_t dcidlen, const uint8_t* scid,
  size_t scidlen)
{
  uint32_t versions[2];
  size_t versions_len = 0;
  ngtcp2_ssize ret;
  uint8_t unused_random;

  /* fill the array with supported versions */
  versions[0] = NGTCP2_PROTO_VER_V1;
  versions_len = 1;
  unused_random = ub_random_max(c->doq_socket->rnd, 256);
  sldns_buffer_clear(c->doq_socket->pkt_buf);
  ret = ngtcp2_pkt_write_version_negotiation(
    sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_capacity(c->doq_socket->pkt_buf), unused_random,
    dcid, dcidlen, scid, scidlen, versions, versions_len);
  if(ret < 0) {
    log_err("ngtcp2_pkt_write_version_negotiation failed: %s",
      ngtcp2_strerror(ret));
    return;
  }
  sldns_buffer_set_position(c->doq_socket->pkt_buf, ret);
  sldns_buffer_flip(c->doq_socket->pkt_buf);
  doq_send_pkt(c, paddr, 0);
}

/** Find the doq_conn object by remote address and dcid */
static struct doq_conn*
doq_conn_find(struct doq_table* table, struct doq_addr_storage* addr,
  socklen_t addrlen, struct doq_addr_storage* localaddr,
  socklen_t localaddrlen, int ifindex, const uint8_t* dcid,
  size_t dcidlen)
{
  struct rbnode_type* node;
  struct doq_conn key;
  memset(&key.node, 0, sizeof(key.node));
  key.node.key = &key;
  memmove(&key.key.paddr.addr, addr, addrlen);
  key.key.paddr.addrlen = addrlen;
  memmove(&key.key.paddr.localaddr, localaddr, localaddrlen);
  key.key.paddr.localaddrlen = localaddrlen;
  key.key.paddr.ifindex = ifindex;
  key.key.dcid = (void*)dcid;
  key.key.dcidlen = dcidlen;
  node = rbtree_search(table->conn_tree, &key);
  if(node)
    return (struct doq_conn*)node->key;
  return NULL;
}

/** find the doq_con by the connection id */
static struct doq_conn*
doq_conn_find_by_id(struct doq_table* table, const uint8_t* dcid,
  size_t dcidlen)
{
  struct doq_conid* conid;
  lock_rw_rdlock(&table->conid_lock);
  conid = doq_conid_find(table, dcid, dcidlen);
  if(conid) {
    /* make a copy of the key */
    struct doq_conn* conn;
    struct doq_conn_key key = conid->key;
    uint8_t cid[NGTCP2_MAX_CIDLEN];
    log_assert(conid->key.dcidlen <= NGTCP2_MAX_CIDLEN);
    memcpy(cid, conid->key.dcid, conid->key.dcidlen);
    key.dcid = cid;
    lock_rw_unlock(&table->conid_lock);

    /* now that the conid lock is released, look up the conn */
    lock_rw_rdlock(&table->lock);
    conn = doq_conn_find(table, &key.paddr.addr,
      key.paddr.addrlen, &key.paddr.localaddr,
      key.paddr.localaddrlen, key.paddr.ifindex, key.dcid,
      key.dcidlen);
    if(!conn) {
      /* The connection got deleted between the conid lookup
       * and the connection lock grab, it no longer exists,
       * so return null. */
      lock_rw_unlock(&table->lock);
      return NULL;
    }
    lock_basic_lock(&conn->lock);
    if(conn->is_deleted) {
      lock_rw_unlock(&table->lock);
      lock_basic_unlock(&conn->lock);
      return NULL;
    }
    lock_rw_unlock(&table->lock);
    return conn;
  }
  lock_rw_unlock(&table->conid_lock);
  return NULL;
}

/** Find the doq_conn, by addr or by connection id */
static struct doq_conn*
doq_conn_find_by_addr_or_cid(struct doq_table* table,
  struct doq_pkt_addr* paddr, const uint8_t* dcid, size_t dcidlen)
{
  struct doq_conn* conn;
  lock_rw_rdlock(&table->lock);
  conn = doq_conn_find(table, &paddr->addr, paddr->addrlen,
    &paddr->localaddr, paddr->localaddrlen, paddr->ifindex,
    dcid, dcidlen);
  if(conn && conn->is_deleted) {
    conn = NULL;
  }
  if(conn) {
    lock_basic_lock(&conn->lock);
    lock_rw_unlock(&table->lock);
    verbose(VERB_ALGO, "doq: found connection by address, dcid");
  } else {
    lock_rw_unlock(&table->lock);
    conn = doq_conn_find_by_id(table, dcid, dcidlen);
    if(conn) {
      verbose(VERB_ALGO, "doq: found connection by dcid");
    }
  }
  return conn;
}

/** decode doq packet header, false on handled or failure, true to continue
 * to process the packet */
static int
doq_decode_pkt_header_negotiate(struct comm_point* c,
  struct doq_pkt_addr* paddr, struct doq_conn** conn)
{
#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
  struct ngtcp2_version_cid vc;
#else
  uint32_t version;
  const uint8_t *dcid, *scid;
  size_t dcidlen, scidlen;
#endif
  int rv;

#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
  rv = ngtcp2_pkt_decode_version_cid(&vc,
    sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_limit(c->doq_socket->pkt_buf),
    c->doq_socket->sv_scidlen);
#else
  rv = ngtcp2_pkt_decode_version_cid(&version, &dcid, &dcidlen,
    &scid, &scidlen, sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_limit(c->doq_socket->pkt_buf), c->doq_socket->sv_scidlen);
#endif
  if(rv != 0) {
    if(rv == NGTCP2_ERR_VERSION_NEGOTIATION) {
      /* send the version negotiation */
      doq_send_version_negotiation(c, paddr,
#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
      vc.scid, vc.scidlen, vc.dcid, vc.dcidlen
#else
      scid, scidlen, dcid, dcidlen
#endif
      );
      return 0;
    }
    verbose(VERB_ALGO, "doq: could not decode version "
      "and CID from QUIC packet header: %s",
      ngtcp2_strerror(rv));
    return 0;
  }

  if(verbosity >= VERB_ALGO) {
    verbose(VERB_ALGO, "ngtcp2_pkt_decode_version_cid packet has "
      "QUIC protocol version %u", (unsigned)
#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
      vc.
#endif
      version
      );
    log_hex("dcid",
#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
      (void*)vc.dcid, vc.dcidlen
#else
      (void*)dcid, dcidlen
#endif
      );
    log_hex("scid",
#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
      (void*)vc.scid, vc.scidlen
#else
      (void*)scid, scidlen
#endif
      );
  }
  *conn = doq_conn_find_by_addr_or_cid(c->doq_socket->table, paddr,
#ifdef HAVE_STRUCT_NGTCP2_VERSION_CID
    vc.dcid, vc.dcidlen
#else
    dcid, dcidlen
#endif
    );
  if(*conn)
    (*conn)->doq_socket = c->doq_socket;
  return 1;
}

/** fill cid structure with random data */
static void doq_cid_randfill(struct ngtcp2_cid* cid, size_t datalen,
  struct ub_randstate* rnd)
{
  uint8_t buf[32];
  if(datalen > sizeof(buf))
    datalen = sizeof(buf);
  doq_fill_rand(rnd, buf, datalen);
  ngtcp2_cid_init(cid, buf, datalen);
}

/** send retry packet for doq connection. */
static void
doq_send_retry(struct comm_point* c, struct doq_pkt_addr* paddr,
  struct ngtcp2_pkt_hd* hd)
{
  char host[256], port[32];
  struct ngtcp2_cid scid;
  uint8_t token[NGTCP2_CRYPTO_MAX_RETRY_TOKENLEN];
  ngtcp2_tstamp ts;
  ngtcp2_ssize tokenlen, ret;

  if(!doq_print_addr_port(&paddr->addr, paddr->addrlen, host,
    sizeof(host), port, sizeof(port))) {
    log_err("doq_send_retry failed");
    return;
  }
  verbose(VERB_ALGO, "doq: sending retry packet to %s %s", host, port);

  /* the server chosen source connection ID */
  scid.datalen = c->doq_socket->sv_scidlen;
  doq_cid_randfill(&scid, scid.datalen, c->doq_socket->rnd);

  ts = doq_get_timestamp_nanosec();

  tokenlen = ngtcp2_crypto_generate_retry_token(token,
    c->doq_socket->static_secret, c->doq_socket->static_secret_len,
    hd->version, (void*)&paddr->addr, paddr->addrlen, &scid,
    &hd->dcid, ts);
  if(tokenlen < 0) {
    log_err("ngtcp2_crypto_generate_retry_token failed: %s",
      ngtcp2_strerror(tokenlen));
    return;
  }

  sldns_buffer_clear(c->doq_socket->pkt_buf);
  ret = ngtcp2_crypto_write_retry(sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_capacity(c->doq_socket->pkt_buf), hd->version,
    &hd->scid, &scid, &hd->dcid, token, tokenlen);
  if(ret < 0) {
    log_err("ngtcp2_crypto_write_retry failed: %s",
      ngtcp2_strerror(ret));
    return;
  }
  sldns_buffer_set_position(c->doq_socket->pkt_buf, ret);
  sldns_buffer_flip(c->doq_socket->pkt_buf);
  doq_send_pkt(c, paddr, 0);
}

/** doq send stateless connection close */
static void
doq_send_stateless_connection_close(struct comm_point* c,
  struct doq_pkt_addr* paddr, struct ngtcp2_pkt_hd* hd,
  uint64_t error_code)
{
  ngtcp2_ssize ret;
  sldns_buffer_clear(c->doq_socket->pkt_buf);
  ret = ngtcp2_crypto_write_connection_close(
    sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_capacity(c->doq_socket->pkt_buf), hd->version, &hd->scid,
    &hd->dcid, error_code, NULL, 0);
  if(ret < 0) {
    log_err("ngtcp2_crypto_write_connection_close failed: %s",
      ngtcp2_strerror(ret));
    return;
  }
  sldns_buffer_set_position(c->doq_socket->pkt_buf, ret);
  sldns_buffer_flip(c->doq_socket->pkt_buf);
  doq_send_pkt(c, paddr, 0);
}

/** doq verify retry token, false on failure */
static int
doq_verify_retry_token(struct comm_point* c, struct doq_pkt_addr* paddr,
  struct ngtcp2_cid* ocid, struct ngtcp2_pkt_hd* hd)
{
  char host[256], port[32];
  ngtcp2_tstamp ts;
  if(!doq_print_addr_port(&paddr->addr, paddr->addrlen, host,
    sizeof(host), port, sizeof(port))) {
    log_err("doq_verify_retry_token failed");
    return 0;
  }
  ts = doq_get_timestamp_nanosec();
  verbose(VERB_ALGO, "doq: verifying retry token from %s %s", host,
    port);
  if(ngtcp2_crypto_verify_retry_token(ocid,
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
    hd->token, hd->tokenlen,
#else
    hd->token.base, hd->token.len,
#endif
    c->doq_socket->static_secret,
    c->doq_socket->static_secret_len, hd->version,
    (void*)&paddr->addr, paddr->addrlen, &hd->dcid,
    10*NGTCP2_SECONDS, ts) != 0) {
    verbose(VERB_ALGO, "doq: could not verify retry token "
      "from %s %s", host, port);
    return 0;
  }
  verbose(VERB_ALGO, "doq: verified retry token from %s %s", host, port);
  return 1;
}

/** doq verify token, false on failure */
static int
doq_verify_token(struct comm_point* c, struct doq_pkt_addr* paddr,
  struct ngtcp2_pkt_hd* hd)
{
  char host[256], port[32];
  ngtcp2_tstamp ts;
  if(!doq_print_addr_port(&paddr->addr, paddr->addrlen, host,
    sizeof(host), port, sizeof(port))) {
    log_err("doq_verify_token failed");
    return 0;
  }
  ts = doq_get_timestamp_nanosec();
  verbose(VERB_ALGO, "doq: verifying token from %s %s", host, port);
  if(ngtcp2_crypto_verify_regular_token(
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
    hd->token, hd->tokenlen,
#else
    hd->token.base, hd->token.len,
#endif
    c->doq_socket->static_secret, c->doq_socket->static_secret_len,
    (void*)&paddr->addr, paddr->addrlen, 3600*NGTCP2_SECONDS,
    ts) != 0) {
    verbose(VERB_ALGO, "doq: could not verify token from %s %s",
      host, port);
    return 0;
  }
  verbose(VERB_ALGO, "doq: verified token from %s %s", host, port);
  return 1;
}

/** delete and remove from the lookup tree the doq_conn connection */
static void
doq_delete_connection(struct comm_point* c, struct doq_conn* conn)
{
  struct doq_conn copy;
  uint8_t cid[NGTCP2_MAX_CIDLEN];
  rbnode_type* node;
  if(!conn)
    return;
  /* Copy the key and set it deleted. */
  conn->is_deleted = 1;
  doq_conn_write_disable(conn);
  copy.key = conn->key;
  log_assert(conn->key.dcidlen <= NGTCP2_MAX_CIDLEN);
  memcpy(cid, conn->key.dcid, conn->key.dcidlen);
  copy.key.dcid = cid;
  copy.node.key = &copy;
  lock_basic_unlock(&conn->lock);

  /* Now get the table lock to delete it from the tree */
  lock_rw_wrlock(&c->doq_socket->table->lock);
  node = rbtree_delete(c->doq_socket->table->conn_tree, copy.node.key);
  if(node) {
    conn = (struct doq_conn*)node->key;
    lock_basic_lock(&conn->lock);
    doq_conn_write_list_remove(c->doq_socket->table, conn);
    if(conn->timer.timer_in_list) {
      /* Remove timer from list first, because finding the
       * rbnode element of the setlist of same timeouts
       * needs tree lookup. Edit the tree structure after
       * that lookup. */
      doq_timer_list_remove(c->doq_socket->table,
        &conn->timer);
    }
    if(conn->timer.timer_in_tree)
      doq_timer_tree_remove(c->doq_socket->table,
        &conn->timer);
  }
  lock_rw_unlock(&c->doq_socket->table->lock);
  if(node) {
    lock_basic_unlock(&conn->lock);
    doq_table_quic_size_subtract(c->doq_socket->table,
      sizeof(*conn)+conn->key.dcidlen);
    doq_conn_delete(conn, c->doq_socket->table);
  }
}

/** create and setup a new doq connection, to a new destination, or with
 * a new dcid. It has a new set of streams. It is inserted in the lookup tree.
 * Returns NULL on failure. */
static struct doq_conn*
doq_setup_new_conn(struct comm_point* c, struct doq_pkt_addr* paddr,
  struct ngtcp2_pkt_hd* hd, struct ngtcp2_cid* ocid)
{
  struct doq_conn* conn;
  if(!doq_table_quic_size_available(c->doq_socket->table,
    c->doq_socket->cfg, sizeof(*conn)+hd->dcid.datalen
    + sizeof(struct doq_stream)
    + 100 /* estimated input query */
    + 1200 /* estimated output query */)) {
    verbose(VERB_ALGO, "doq: no mem available for new connection");
    doq_send_stateless_connection_close(c, paddr, hd,
      NGTCP2_CONNECTION_REFUSED);
    return NULL;
  }
  conn = doq_conn_create(c, paddr, hd->dcid.data, hd->dcid.datalen,
    hd->version);
  if(!conn) {
    log_err("doq: could not allocate doq_conn");
    return NULL;
  }
  lock_rw_wrlock(&c->doq_socket->table->lock);
  lock_basic_lock(&conn->lock);
  if(!rbtree_insert(c->doq_socket->table->conn_tree, &conn->node)) {
    lock_rw_unlock(&c->doq_socket->table->lock);
    log_err("doq: duplicate connection");
    /* conn has no entry in writelist, and no timer yet. */
    lock_basic_unlock(&conn->lock);
    doq_conn_delete(conn, c->doq_socket->table);
    return NULL;
  }
  lock_rw_unlock(&c->doq_socket->table->lock);
  doq_table_quic_size_add(c->doq_socket->table,
    sizeof(*conn)+conn->key.dcidlen);
  verbose(VERB_ALGO, "doq: created new connection");

  /* the scid and dcid switch meaning from the accepted client
   * connection to the server connection. The 'source' and 'destination'
   * meaning is reversed. */
  if(!doq_conn_setup(conn, hd->scid.data, hd->scid.datalen,
    (ocid?ocid->data:NULL), (ocid?ocid->datalen:0),
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
    hd->token, hd->tokenlen
#else
    hd->token.base, hd->token.len
#endif
    )) {
    log_err("doq: could not set up connection");
    doq_delete_connection(c, conn);
    return NULL;
  }
  return conn;
}

/** perform doq address validation */
static int
doq_address_validation(struct comm_point* c, struct doq_pkt_addr* paddr,
  struct ngtcp2_pkt_hd* hd, struct ngtcp2_cid* ocid,
  struct ngtcp2_cid** pocid)
{
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
  const uint8_t* token = hd->token;
  size_t tokenlen = hd->tokenlen;
#else
  const uint8_t* token = hd->token.base;
  size_t tokenlen = hd->token.len;
#endif
  verbose(VERB_ALGO, "doq stateless address validation");

  if(tokenlen == 0 || token == NULL) {
    doq_send_retry(c, paddr, hd);
    return 0;
  }
  if(token[0] != NGTCP2_CRYPTO_TOKEN_MAGIC_RETRY &&
    hd->dcid.datalen < NGTCP2_MIN_INITIAL_DCIDLEN) {
    doq_send_stateless_connection_close(c, paddr, hd,
      NGTCP2_INVALID_TOKEN);
    return 0;
  }
  if(token[0] == NGTCP2_CRYPTO_TOKEN_MAGIC_RETRY) {
    if(!doq_verify_retry_token(c, paddr, ocid, hd)) {
      doq_send_stateless_connection_close(c, paddr, hd,
        NGTCP2_INVALID_TOKEN);
      return 0;
    }
    *pocid = ocid;
  } else if(token[0] == NGTCP2_CRYPTO_TOKEN_MAGIC_REGULAR) {
    if(!doq_verify_token(c, paddr, hd)) {
      doq_send_retry(c, paddr, hd);
      return 0;
    }
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
    hd->token = NULL;
    hd->tokenlen = 0;
#else
    hd->token.base = NULL;
    hd->token.len = 0;
#endif
  } else {
    verbose(VERB_ALGO, "doq address validation: unrecognised "
      "token in hd.token.base with magic byte 0x%2.2x",
      (int)token[0]);
    if(c->doq_socket->validate_addr) {
      doq_send_retry(c, paddr, hd);
      return 0;
    }
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
    hd->token = NULL;
    hd->tokenlen = 0;
#else
    hd->token.base = NULL;
    hd->token.len = 0;
#endif
  }
  return 1;
}

/** the doq accept, returns false if no further processing of content */
static int
doq_accept(struct comm_point* c, struct doq_pkt_addr* paddr,
  struct doq_conn** conn, struct ngtcp2_pkt_info* pi)
{
  int rv;
  struct ngtcp2_pkt_hd hd;
  struct ngtcp2_cid ocid, *pocid=NULL;
  int err_retry;
  memset(&hd, 0, sizeof(hd));
  rv = ngtcp2_accept(&hd, sldns_buffer_begin(c->doq_socket->pkt_buf),
    sldns_buffer_limit(c->doq_socket->pkt_buf));
  if(rv != 0) {
    if(rv == NGTCP2_ERR_RETRY) {
      doq_send_retry(c, paddr, &hd);
      return 0;
    }
    log_err("doq: initial packet failed, ngtcp2_accept failed: %s",
      ngtcp2_strerror(rv));
    return 0;
  }
  if(c->doq_socket->validate_addr ||
#ifdef HAVE_STRUCT_NGTCP2_PKT_HD_TOKENLEN
    hd.tokenlen
#else
    hd.token.len
#endif
    ) {
    if(!doq_address_validation(c, paddr, &hd, &ocid, &pocid))
      return 0;
  }
  *conn = doq_setup_new_conn(c, paddr, &hd, pocid);
  if(!*conn)
    return 0;
  (*conn)->doq_socket = c->doq_socket;
  if(!doq_conn_recv(c, paddr, *conn, pi, &err_retry, NULL)) {
    if(err_retry)
      doq_send_retry(c, paddr, &hd);
    doq_delete_connection(c, *conn);
    *conn = NULL;
    return 0;
  }
  return 1;
}

/** doq pickup a timer to wait for for the worker. If any timer exists. */
static void
doq_pickup_timer(struct comm_point* c)
{
  struct doq_timer* t;
  struct timeval tv;
  int have_time = 0;
  memset(&tv, 0, sizeof(tv));

  lock_rw_wrlock(&c->doq_socket->table->lock);
  RBTREE_FOR(t, struct doq_timer*, c->doq_socket->table->timer_tree) {
    if(t->worker_doq_socket == NULL ||
      t->worker_doq_socket == c->doq_socket) {
      /* pick up this element */
      t->worker_doq_socket = c->doq_socket;
      have_time = 1;
      memcpy(&tv, &t->time, sizeof(tv));
      break;
    }
  }
  lock_rw_unlock(&c->doq_socket->table->lock);

  if(have_time) {
    struct timeval rel;
    timeval_subtract(&rel, &tv, c->doq_socket->now_tv);
    comm_timer_set(c->doq_socket->timer, &rel);
    memcpy(&c->doq_socket->marked_time, &tv,
      sizeof(c->doq_socket->marked_time));
    verbose(VERB_ALGO, "doq pickup timer at %d.%6.6d in %d.%6.6d",
      (int)tv.tv_sec, (int)tv.tv_usec, (int)rel.tv_sec,
      (int)rel.tv_usec);
  } else {
    if(comm_timer_is_set(c->doq_socket->timer))
      comm_timer_disable(c->doq_socket->timer);
    memset(&c->doq_socket->marked_time, 0,
      sizeof(c->doq_socket->marked_time));
    verbose(VERB_ALGO, "doq timer disabled");
  }
}

/** doq done with connection, release locks and setup timer and write */
static void
doq_done_setup_timer_and_write(struct comm_point* c, struct doq_conn* conn)
{
  struct doq_conn copy;
  uint8_t cid[NGTCP2_MAX_CIDLEN];
  rbnode_type* node;
  struct timeval new_tv;
  int write_change = 0, timer_change = 0;

  /* No longer in callbacks, so the pointer to doq_socket is back
   * to NULL. */
  conn->doq_socket = NULL;

  if(doq_conn_check_timer(conn, &new_tv))
    timer_change = 1;
  if( (conn->write_interest && !conn->on_write_list) ||
    (!conn->write_interest && conn->on_write_list))
    write_change = 1;

  if(!timer_change && !write_change) {
    /* Nothing to do. */
    lock_basic_unlock(&conn->lock);
    return;
  }

  /* The table lock is needed to change the write list and timer tree.
   * So the connection lock is release and then the connection is
   * looked up again. */
  copy.key = conn->key;
  log_assert(conn->key.dcidlen <= NGTCP2_MAX_CIDLEN);
  memcpy(cid, conn->key.dcid, conn->key.dcidlen);
  copy.key.dcid = cid;
  copy.node.key = &copy;
  lock_basic_unlock(&conn->lock);

  lock_rw_wrlock(&c->doq_socket->table->lock);
  node = rbtree_search(c->doq_socket->table->conn_tree, copy.node.key);
  if(!node) {
    lock_rw_unlock(&c->doq_socket->table->lock);
    /* Must have been deleted in the mean time. */
    return;
  }
  conn = (struct doq_conn*)node->key;
  lock_basic_lock(&conn->lock);
  if(conn->is_deleted) {
    /* It is deleted now. */
    lock_rw_unlock(&c->doq_socket->table->lock);
    lock_basic_unlock(&conn->lock);
    return;
  }

  if(write_change) {
    /* Edit the write lists, we are holding the table.lock and can
     * edit the list first,last and also prev,next and on_list
     * elements in the doq_conn structures. */
    doq_conn_set_write_list(c->doq_socket->table, conn);
  }
  if(timer_change) {
    doq_timer_set(c->doq_socket->table, &conn->timer,
      c->doq_socket, &new_tv);
  }
  lock_rw_unlock(&c->doq_socket->table->lock);
  lock_basic_unlock(&conn->lock);
}

/** doq done with connection callbacks, release locks and setup write */
static void
doq_done_with_conn_cb(struct comm_point* c, struct doq_conn* conn)
{
  struct doq_conn copy;
  uint8_t cid[NGTCP2_MAX_CIDLEN];
  rbnode_type* node;

  /* no longer in callbacks, so the pointer to doq_socket is back
   * to NULL. */
  conn->doq_socket = NULL;

  if( (conn->write_interest && conn->on_write_list) ||
    (!conn->write_interest && !conn->on_write_list)) {
    /* The connection already has the required write list
     * status. */
    lock_basic_unlock(&conn->lock);
    return;
  }

  /* To edit the write list of connections we have to hold the table
   * lock, so we release the connection and then look it up again. */
  copy.key = conn->key;
  log_assert(conn->key.dcidlen <= NGTCP2_MAX_CIDLEN);
  memcpy(cid, conn->key.dcid, conn->key.dcidlen);
  copy.key.dcid = cid;
  copy.node.key = &copy;
  lock_basic_unlock(&conn->lock);

  lock_rw_wrlock(&c->doq_socket->table->lock);
  node = rbtree_search(c->doq_socket->table->conn_tree, copy.node.key);
  if(!node) {
    lock_rw_unlock(&c->doq_socket->table->lock);
    /* must have been deleted in the mean time */
    return;
  }
  conn = (struct doq_conn*)node->key;
  lock_basic_lock(&conn->lock);
  if(conn->is_deleted) {
    /* it is deleted now. */
    lock_rw_unlock(&c->doq_socket->table->lock);
    lock_basic_unlock(&conn->lock);
    return;
  }

  /* edit the write lists, we are holding the table.lock and can
   * edit the list first,last and also prev,next and on_list elements
   * in the doq_conn structures. */
  doq_conn_set_write_list(c->doq_socket->table, conn);
  lock_rw_unlock(&c->doq_socket->table->lock);
  lock_basic_unlock(&conn->lock);
}

/** doq count the length of the write list */
static size_t
doq_write_list_length(struct comm_point* c)
{
  size_t count = 0;
  struct doq_conn* conn;
  lock_rw_rdlock(&c->doq_socket->table->lock);
  conn = c->doq_socket->table->write_list_first;
  while(conn) {
    count++;
    conn = conn->write_next;
  }
  lock_rw_unlock(&c->doq_socket->table->lock);
  return count;
}

/** doq pop the first element from the write list to have write events */
static struct doq_conn*
doq_pop_write_conn(struct comm_point* c)
{
  struct doq_conn* conn;
  lock_rw_wrlock(&c->doq_socket->table->lock);
  conn = doq_table_pop_first(c->doq_socket->table);
  while(conn && conn->is_deleted) {
    lock_basic_unlock(&conn->lock);
    conn = doq_table_pop_first(c->doq_socket->table);
  }
  lock_rw_unlock(&c->doq_socket->table->lock);
  if(conn)
    conn->doq_socket = c->doq_socket;
  return conn;
}

/** doq the connection is done with write callbacks, release it. */
static void
doq_done_with_write_cb(struct comm_point* c, struct doq_conn* conn,
  int delete_it)
{
  if(delete_it) {
    doq_delete_connection(c, conn);
    return;
  }
  doq_done_setup_timer_and_write(c, conn);
}

/** see if the doq socket wants to write packets */
static int
doq_socket_want_write(struct comm_point* c)
{
  int want_write = 0;
  if(c->doq_socket->have_blocked_pkt)
    return 1;
  lock_rw_rdlock(&c->doq_socket->table->lock);
  if(c->doq_socket->table->write_list_first)
    want_write = 1;
  lock_rw_unlock(&c->doq_socket->table->lock);
  return want_write;
}

/** enable write event for the doq server socket fd */
static void
doq_socket_write_enable(struct comm_point* c)
{
  verbose(VERB_ALGO, "doq socket want write");
  if(c->doq_socket->event_has_write)
    return;
  comm_point_listen_for_rw(c, 1, 1);
  c->doq_socket->event_has_write = 1;
}

/** disable write event for the doq server socket fd */
static void
doq_socket_write_disable(struct comm_point* c)
{
  verbose(VERB_ALGO, "doq socket want no write");
  if(!c->doq_socket->event_has_write)
    return;
  comm_point_listen_for_rw(c, 1, 0);
  c->doq_socket->event_has_write = 0;
}

/** write blocked packet, if possible. returns false if failed, again. */
static int
doq_write_blocked_pkt(struct comm_point* c)
{
  struct doq_pkt_addr paddr;
  if(!c->doq_socket->have_blocked_pkt)
    return 1;
  c->doq_socket->have_blocked_pkt = 0;
  if(sldns_buffer_limit(c->doq_socket->blocked_pkt) >
    sldns_buffer_remaining(c->doq_socket->pkt_buf))
    return 1; /* impossibly large, drop it.
    impossible since pkt_buf is same size as blocked_pkt buf. */
  sldns_buffer_clear(c->doq_socket->pkt_buf);
  sldns_buffer_write(c->doq_socket->pkt_buf,
    sldns_buffer_begin(c->doq_socket->blocked_pkt),
    sldns_buffer_limit(c->doq_socket->blocked_pkt));
  sldns_buffer_flip(c->doq_socket->pkt_buf);
  memcpy(&paddr, c->doq_socket->blocked_paddr, sizeof(paddr));
  doq_send_pkt(c, &paddr, c->doq_socket->blocked_pkt_pi.ecn);
  if(c->doq_socket->have_blocked_pkt)
    return 0;
  return 1;
}

/** doq find a timer that timeouted and return the conn, locked. */
static struct doq_conn*
doq_timer_timeout_conn(struct doq_server_socket* doq_socket)
{
  struct doq_conn* conn = NULL;
  struct rbnode_type* node;
  lock_rw_wrlock(&doq_socket->table->lock);
  node = rbtree_first(doq_socket->table->timer_tree);
  if(node && node != RBTREE_NULL) {
    struct doq_timer* t = (struct doq_timer*)node;
    conn = t->conn;

    /* If now < timer then no further timeouts in tree. */
    if(timeval_smaller(doq_socket->now_tv, &t->time)) {
      lock_rw_unlock(&doq_socket->table->lock);
      return NULL;
    }

    lock_basic_lock(&conn->lock);
    conn->doq_socket = doq_socket;

    /* Now that the timer is fired, remove it. */
    doq_timer_unset(doq_socket->table, t);
    lock_rw_unlock(&doq_socket->table->lock);
    return conn;
  }
  lock_rw_unlock(&doq_socket->table->lock);
  return NULL;
}

/** doq timer erase the marker that said which timer the worker uses. */
static void
doq_timer_erase_marker(struct doq_server_socket* doq_socket)
{
  struct doq_timer* t;
  lock_rw_wrlock(&doq_socket->table->lock);
  t = doq_timer_find_time(doq_socket->table, &doq_socket->marked_time);
  if(t && t->worker_doq_socket == doq_socket)
    t->worker_doq_socket = NULL;
  lock_rw_unlock(&doq_socket->table->lock);
  memset(&doq_socket->marked_time, 0, sizeof(doq_socket->marked_time));
}

void
doq_timer_cb(void* arg)
{
  struct doq_server_socket* doq_socket = (struct doq_server_socket*)arg;
  struct doq_conn* conn;
  verbose(VERB_ALGO, "doq timer callback");

  doq_timer_erase_marker(doq_socket);

  while((conn = doq_timer_timeout_conn(doq_socket)) != NULL) {
    if(conn->is_deleted ||
#ifdef HAVE_NGTCP2_CONN_IN_CLOSING_PERIOD
      ngtcp2_conn_in_closing_period(conn->conn) ||
#else
      ngtcp2_conn_is_in_closing_period(conn->conn) ||
#endif
#ifdef HAVE_NGTCP2_CONN_IN_DRAINING_PERIOD
      ngtcp2_conn_in_draining_period(conn->conn)
#else
      ngtcp2_conn_is_in_draining_period(conn->conn)
#endif
      ) {
      if(verbosity >= VERB_ALGO) {
        char remotestr[256];
        addr_to_str((void*)&conn->key.paddr.addr,
          conn->key.paddr.addrlen, remotestr,
          sizeof(remotestr));
        verbose(VERB_ALGO, "doq conn %s is deleted "
          "after timeout", remotestr);
      }
      doq_delete_connection(doq_socket->cp, conn);
      continue;
    }
    if(!doq_conn_handle_timeout(conn))
      doq_delete_connection(doq_socket->cp, conn);
    else doq_done_setup_timer_and_write(doq_socket->cp, conn);
  }

  if(doq_socket_want_write(doq_socket->cp))
    doq_socket_write_enable(doq_socket->cp);
  else doq_socket_write_disable(doq_socket->cp);
  doq_pickup_timer(doq_socket->cp);
}

void
comm_point_doq_callback(int fd, short event, void* arg)
{
  struct comm_point* c;
  struct doq_pkt_addr paddr;
  int i, pkt_continue, err_drop;
  struct doq_conn* conn;
  struct ngtcp2_pkt_info pi;
  size_t count, num_len;

  c = (struct comm_point*)arg;
  log_assert(c->type == comm_doq);

  log_assert(c && c->doq_socket->pkt_buf && c->fd == fd);
  ub_comm_base_now(c->ev->base);

  /* see if there is a blocked packet, and send that if possible.
   * do not attempt to read yet, even if possible, that would just
   * push more answers in reply to those read packets onto the list
   * of written replies. First attempt to clear the write content out.
   * That keeps the memory usage from bloating up. */
  if(c->doq_socket->have_blocked_pkt) {
    if(!doq_write_blocked_pkt(c)) {
      /* this write has also blocked, attempt to write
       * later. Make sure the event listens to write
       * events. */
      if(!c->doq_socket->event_has_write)
        doq_socket_write_enable(c);
      doq_pickup_timer(c);
      return;
    }
  }

  /* see if there is write interest */
  count = 0;
  num_len = doq_write_list_length(c);
  while((conn = doq_pop_write_conn(c)) != NULL) {
    if(conn->is_deleted ||
#ifdef HAVE_NGTCP2_CONN_IN_CLOSING_PERIOD
      ngtcp2_conn_in_closing_period(conn->conn) ||
#else
      ngtcp2_conn_is_in_closing_period(conn->conn) ||
#endif
#ifdef HAVE_NGTCP2_CONN_IN_DRAINING_PERIOD
      ngtcp2_conn_in_draining_period(conn->conn)
#else
      ngtcp2_conn_is_in_draining_period(conn->conn)
#endif
      ) {
      conn->doq_socket = NULL;
      lock_basic_unlock(&conn->lock);
      if(c->doq_socket->have_blocked_pkt) {
        if(!c->doq_socket->event_has_write)
          doq_socket_write_enable(c);
        doq_pickup_timer(c);
        return;
      }
      if(++count > num_len*2)
        break;
      continue;
    }
    if(verbosity >= VERB_ALGO) {
      char remotestr[256];
      addr_to_str((void*)&conn->key.paddr.addr,
        conn->key.paddr.addrlen, remotestr,
        sizeof(remotestr));
      verbose(VERB_ALGO, "doq write connection %s %d",
        remotestr, doq_sockaddr_get_port(
        &conn->key.paddr.addr));
    }
    if(doq_conn_write_streams(c, conn, &err_drop))
      err_drop = 0;
    doq_done_with_write_cb(c, conn, err_drop);
    if(c->doq_socket->have_blocked_pkt) {
      if(!c->doq_socket->event_has_write)
        doq_socket_write_enable(c);
      doq_pickup_timer(c);
      return;
    }
    /* Stop overly long write lists that are created
     * while we are processing. Do those next time there
     * is a write callback. Stops long loops, and keeps
     * fair for other events. */
    if(++count > num_len*2)
      break;
  }

  /* check for data to read */
  if((event&UB_EV_READ)!=0)
    for(i=0; i<NUM_UDP_PER_SELECT; i++) {
    /* there may be a blocked write packet and if so, stop
     * reading because the reply cannot get written. The
     * blocked packet could be written during the conn_recv
     * handling of replies, or for a connection close. */
    if(c->doq_socket->have_blocked_pkt) {
      if(!c->doq_socket->event_has_write)
        doq_socket_write_enable(c);
      doq_pickup_timer(c);
      return;
    }
    sldns_buffer_clear(c->doq_socket->pkt_buf);
    doq_pkt_addr_init(&paddr);
    log_assert(fd != -1);
    log_assert(sldns_buffer_remaining(c->doq_socket->pkt_buf) > 0);
    if(!doq_recv(c, &paddr, &pkt_continue, &pi)) {
      if(pkt_continue)
        continue;
      break;
    }

    /* handle incoming packet from remote addr to localaddr */
    if(verbosity >= VERB_ALGO) {
      char remotestr[256], localstr[256];
      addr_to_str((void*)&paddr.addr, paddr.addrlen,
        remotestr, sizeof(remotestr));
      addr_to_str((void*)&paddr.localaddr,
        paddr.localaddrlen, localstr,
        sizeof(localstr));
      log_info("incoming doq packet from %s port %d on "
        "%s port %d ifindex %d",
        remotestr, doq_sockaddr_get_port(&paddr.addr),
        localstr,
        doq_sockaddr_get_port(&paddr.localaddr),
        paddr.ifindex);
      log_info("doq_recv length %d ecn 0x%x",
        (int)sldns_buffer_limit(c->doq_socket->pkt_buf),
        (int)pi.ecn);
    }

    if(sldns_buffer_limit(c->doq_socket->pkt_buf) == 0)
      continue;

    conn = NULL;
    if(!doq_decode_pkt_header_negotiate(c, &paddr, &conn))
      continue;
    if(!conn) {
      if(!doq_accept(c, &paddr, &conn, &pi))
        continue;
      if(!doq_conn_write_streams(c, conn, NULL)) {
        doq_delete_connection(c, conn);
        continue;
      }
      doq_done_setup_timer_and_write(c, conn);
      continue;
    }
    if(
#ifdef HAVE_NGTCP2_CONN_IN_CLOSING_PERIOD
      ngtcp2_conn_in_closing_period(conn->conn)
#else
      ngtcp2_conn_is_in_closing_period(conn->conn)
#endif
      ) {
      if(!doq_conn_send_close(c, conn)) {
        doq_delete_connection(c, conn);
      } else {
        doq_done_setup_timer_and_write(c, conn);
      }
      continue;
    }
    if(
#ifdef HAVE_NGTCP2_CONN_IN_DRAINING_PERIOD
      ngtcp2_conn_in_draining_period(conn->conn)
#else
      ngtcp2_conn_is_in_draining_period(conn->conn)
#endif
      ) {
      doq_done_setup_timer_and_write(c, conn);
      continue;
    }
    if(!doq_conn_recv(c, &paddr, conn, &pi, NULL, &err_drop)) {
      /* The receive failed, and if it also failed to send
       * a close, drop the connection. That means it is not
       * in the closing period. */
      if(err_drop) {
        doq_delete_connection(c, conn);
      } else {
        doq_done_setup_timer_and_write(c, conn);
      }
      continue;
    }
    if(!doq_conn_write_streams(c, conn, &err_drop)) {
      if(err_drop) {
        doq_delete_connection(c, conn);
      } else {
        doq_done_setup_timer_and_write(c, conn);
      }
      continue;
    }
    doq_done_setup_timer_and_write(c, conn);
  }

  /* see if we want to have more write events */
  verbose(VERB_ALGO, "doq check write enable");
  if(doq_socket_want_write(c))
    doq_socket_write_enable(c);
  else doq_socket_write_disable(c);
  doq_pickup_timer(c);
}

/** create new doq server socket structure */
static struct doq_server_socket*
doq_server_socket_create(struct doq_table* table, struct ub_randstate* rnd,
  const void* quic_sslctx, struct comm_point* c, struct comm_base* base,
  struct config_file* cfg)
{
  size_t doq_buffer_size = 4096; /* bytes buffer size, for one packet. */
  struct doq_server_socket* doq_socket;
  doq_socket = calloc(1, sizeof(*doq_socket));
  if(!doq_socket) {
    return NULL;
  }
  doq_socket->table = table;
  doq_socket->rnd = rnd;
  doq_socket->validate_addr = 1;
  /* the doq_socket has its own copy of the static secret, as
   * well as other config values, so that they do not need table.lock */
  doq_socket->static_secret_len = table->static_secret_len;
  doq_socket->static_secret = memdup(table->static_secret,
    table->static_secret_len);
  if(!doq_socket->static_secret) {
    free(doq_socket);
    return NULL;
  }
  doq_socket->ctx = (SSL_CTX*)quic_sslctx;
  doq_socket->idle_timeout = table->idle_timeout;
  doq_socket->sv_scidlen = table->sv_scidlen;
  doq_socket->cp = c;
  doq_socket->pkt_buf = sldns_buffer_new(doq_buffer_size);
  if(!doq_socket->pkt_buf) {
    free(doq_socket->static_secret);
    free(doq_socket);
    return NULL;
  }
  doq_socket->blocked_pkt = sldns_buffer_new(
    sldns_buffer_capacity(doq_socket->pkt_buf));
  if(!doq_socket->pkt_buf) {
    free(doq_socket->static_secret);
    sldns_buffer_free(doq_socket->pkt_buf);
    free(doq_socket);
    return NULL;
  }
  doq_socket->blocked_paddr = calloc(1,
    sizeof(*doq_socket->blocked_paddr));
  if(!doq_socket->blocked_paddr) {
    free(doq_socket->static_secret);
    sldns_buffer_free(doq_socket->pkt_buf);
    sldns_buffer_free(doq_socket->blocked_pkt);
    free(doq_socket);
    return NULL;
  }
  doq_socket->timer = comm_timer_create(base, doq_timer_cb, doq_socket);
  if(!doq_socket->timer) {
    free(doq_socket->static_secret);
    sldns_buffer_free(doq_socket->pkt_buf);
    sldns_buffer_free(doq_socket->blocked_pkt);
    free(doq_socket->blocked_paddr);
    free(doq_socket);
    return NULL;
  }
  memset(&doq_socket->marked_time, 0, sizeof(doq_socket->marked_time));
  comm_base_timept(base, &doq_socket->now_tt, &doq_socket->now_tv);
  doq_socket->cfg = cfg;
  return doq_socket;
}

/** delete doq server socket structure */
static void
doq_server_socket_delete(struct doq_server_socket* doq_socket)
{
  if(!doq_socket)
    return;
  free(doq_socket->static_secret);
#ifndef HAVE_NGTCP2_CRYPTO_QUICTLS_CONFIGURE_SERVER_CONTEXT
  free(doq_socket->quic_method);
#endif
  sldns_buffer_free(doq_socket->pkt_buf);
  sldns_buffer_free(doq_socket->blocked_pkt);
  free(doq_socket->blocked_paddr);
  comm_timer_delete(doq_socket->timer);
  free(doq_socket);
}

/** find repinfo in the doq table */
static struct doq_conn*
doq_lookup_repinfo(struct doq_table* table, struct comm_reply* repinfo)
{
  struct doq_conn* conn;
  struct doq_conn_key key;
  doq_conn_key_from_repinfo(&key, repinfo);
  lock_rw_rdlock(&table->lock);
  conn = doq_conn_find(table, &key.paddr.addr,
    key.paddr.addrlen, &key.paddr.localaddr,
    key.paddr.localaddrlen, key.paddr.ifindex, key.dcid,
    key.dcidlen);
  if(conn) {
    lock_basic_lock(&conn->lock);
    lock_rw_unlock(&table->lock);
    return conn;
  }
  lock_rw_unlock(&table->lock);
  return NULL;
}

/** doq find connection and stream. From inside callbacks from worker. */
static int
doq_lookup_conn_stream(struct comm_reply* repinfo, struct comm_point* c,
  struct doq_conn** conn, struct doq_stream** stream)
{
  log_assert(c->doq_socket);
  if(c->doq_socket->current_conn) {
    *conn = c->doq_socket->current_conn;
  } else {
    *conn = doq_lookup_repinfo(c->doq_socket->table, repinfo);
    if((*conn) && (*conn)->is_deleted) {
      lock_basic_unlock(&(*conn)->lock);
      *conn = NULL;
    }
    if(*conn) {
      (*conn)->doq_socket = c->doq_socket;
    }
  }
  if(!*conn) {
    *stream = NULL;
    return 0;
  }
  *stream = doq_stream_find(*conn, repinfo->doq_streamid);
  if(!*stream) {
    if(!c->doq_socket->current_conn) {
      /* Not inside callbacks, we have our own lock on conn.
       * Release it. */
      lock_basic_unlock(&(*conn)->lock);
    }
    return 0;
  }
  if((*stream)->is_closed) {
    /* stream is closed, ignore reply or drop */
    if(!c->doq_socket->current_conn) {
      /* Not inside callbacks, we have our own lock on conn.
       * Release it. */
      lock_basic_unlock(&(*conn)->lock);
    }
    return 0;
  }
  return 1;
}

/** doq send a reply from a comm reply */
static void
doq_socket_send_reply(struct comm_reply* repinfo)
{
  struct doq_conn* conn;
  struct doq_stream* stream;
  log_assert(repinfo->c->type == comm_doq);
  if(!doq_lookup_conn_stream(repinfo, repinfo->c, &conn, &stream)) {
    verbose(VERB_ALGO, "doq: send_reply but %s is gone",
      (conn?"stream":"connection"));
    /* No stream, it may have been closed. */
    /* Drop the reply, it cannot be sent. */
    return;
  }
  if(!doq_stream_send_reply(conn, stream, repinfo->c->buffer))
    doq_stream_close(conn, stream, 1);
  if(!repinfo->c->doq_socket->current_conn) {
    /* Not inside callbacks, we have our own lock on conn.
     * Release it. */
    doq_done_with_conn_cb(repinfo->c, conn);
    /* since we sent a reply, or closed it, the assumption is
     * that there is something to write, so enable write event.
     * It waits until the write event happens to write the
     * streams with answers, this allows some answers to be
     * answered before the event loop reaches the doq fd, in
     * repinfo->c->fd, and that collates answers. That would
     * not happen if we write doq packets right now. */
    doq_socket_write_enable(repinfo->c);
  }
}

/** doq drop a reply from a comm reply */
static void
doq_socket_drop_reply(struct comm_reply* repinfo)
{
  struct doq_conn* conn;
  struct doq_stream* stream;
  log_assert(repinfo->c->type == comm_doq);
  if(!doq_lookup_conn_stream(repinfo, repinfo->c, &conn, &stream)) {
    verbose(VERB_ALGO, "doq: drop_reply but %s is gone",
      (conn?"stream":"connection"));
    /* The connection or stream is already gone. */
    return;
  }
  doq_stream_close(conn, stream, 1);
  if(!repinfo->c->doq_socket->current_conn) {
    /* Not inside callbacks, we have our own lock on conn.
     * Release it. */
    doq_done_with_conn_cb(repinfo->c, conn);
    doq_socket_write_enable(repinfo->c);
  }
}
#endif /* HAVE_NGTCP2 */

int adjusted_tcp_timeout(struct comm_point* c)
{
  if(c->tcp_timeout_msec < TCP_QUERY_TIMEOUT_MINIMUM)
    return TCP_QUERY_TIMEOUT_MINIMUM;
  return c->tcp_timeout_msec;
}

/** Use a new tcp handler for new query fd, set to read query */
static void
setup_tcp_handler(struct comm_point* c, int fd, int cur, int max)
{
  int handler_usage;
  log_assert(c->type == comm_tcp || c->type == comm_http);
  log_assert(c->fd == -1);
  sldns_buffer_clear(c->buffer);
#ifdef USE_DNSCRYPT
  if (c->dnscrypt)
    sldns_buffer_clear(c->dnscrypt_buffer);
#endif
  c->tcp_is_reading = 1;
  c->tcp_byte_count = 0;
  c->tcp_keepalive = 0;
  /* if more than half the tcp handlers are in use, use a shorter
   * timeout for this TCP connection, we need to make space for
   * other connections to be able to get attention */
  /* If > 50% TCP handler structures in use, set timeout to 1/100th
   *  configured value.
   * If > 65%TCP handler structures in use, set to 1/500th configured
   *  value.
   * If > 80% TCP handler structures in use, set to 0.
   *
   * If the timeout to use falls below 200 milliseconds, an actual
   * timeout of 200ms is used.
   */
  handler_usage = (cur * 100) / max;
  if(handler_usage > 50 && handler_usage <= 65)
    c->tcp_timeout_msec /= 100;
  else if (handler_usage > 65 && handler_usage <= 80)
    c->tcp_timeout_msec /= 500;
  else if (handler_usage > 80)
    c->tcp_timeout_msec = 0;
  comm_point_start_listening(c, fd, adjusted_tcp_timeout(c));
}

void comm_base_handle_slow_accept(int ATTR_UNUSED(fd),
  short ATTR_UNUSED(event), void* arg)
{
  struct comm_base* b = (struct comm_base*)arg;
  /* timeout for the slow accept, re-enable accepts again */
  if(b->start_accept) {
    verbose(VERB_ALGO, "wait is over, slow accept disabled");
    fptr_ok(fptr_whitelist_start_accept(b->start_accept));
    (*b->start_accept)(b->cb_arg);
    b->eb->slow_accept_enabled = 0;
  }
}

int comm_point_perform_accept(struct comm_point* c,
  struct sockaddr_storage* addr, socklen_t* addrlen)
{
  int new_fd;
  *addrlen = (socklen_t)sizeof(*addr);
#ifndef HAVE_ACCEPT4
  new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen);
#else
  /* SOCK_NONBLOCK saves extra calls to fcntl for the same result */
  new_fd = accept4(c->fd, (struct sockaddr*)addr, addrlen, SOCK_NONBLOCK);
#endif
  if(new_fd == -1) {
#ifndef USE_WINSOCK
    /* EINTR is signal interrupt. others are closed connection. */
    if( errno == EINTR || errno == EAGAIN
#ifdef EWOULDBLOCK
      || errno == EWOULDBLOCK
#endif
#ifdef ECONNABORTED
      || errno == ECONNABORTED
#endif
#ifdef EPROTO
      || errno == EPROTO
#endif /* EPROTO */
      )
      return -1;
#if defined(ENFILE) && defined(EMFILE)
    if(errno == ENFILE || errno == EMFILE) {
      /* out of file descriptors, likely outside of our
       * control. stop accept() calls for some time */
      if(c->ev->base->stop_accept) {
        struct comm_base* b = c->ev->base;
        struct timeval tv;
        verbose(VERB_ALGO, "out of file descriptors: "
          "slow accept");
        ub_comm_base_now(b);
        if(b->eb->last_slow_log+SLOW_LOG_TIME <=
          b->eb->secs) {
          b->eb->last_slow_log = b->eb->secs;
          verbose(VERB_OPS, "accept failed, "
            "slow down accept for %d "
            "msec: %s",
            NETEVENT_SLOW_ACCEPT_TIME,
            sock_strerror(errno));
        }
        b->eb->slow_accept_enabled = 1;
        fptr_ok(fptr_whitelist_stop_accept(
          b->stop_accept));
        (*b->stop_accept)(b->cb_arg);
        /* set timeout, no mallocs */
        tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000;
        tv.tv_usec = (NETEVENT_SLOW_ACCEPT_TIME%1000)*1000;
        b->eb->slow_accept = ub_event_new(b->eb->base,
          -1, UB_EV_TIMEOUT,
          comm_base_handle_slow_accept, b);
        if(b->eb->slow_accept == NULL) {
          /* we do not want to log here, because
           * that would spam the logfiles.
           * error: "event_base_set failed." */
        }
        else if(ub_event_add(b->eb->slow_accept, &tv)
          != 0) {
          /* we do not want to log here,
           * error: "event_add failed." */
        }
      } else {
        log_err("accept, with no slow down, "
          "failed: %s", sock_strerror(errno));
      }
      return -1;
    }
#endif
#else /* USE_WINSOCK */
    if(WSAGetLastError() == WSAEINPROGRESS ||
      WSAGetLastError() == WSAECONNRESET)
      return -1;
    if(WSAGetLastError() == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
      return -1;
    }
#endif
    log_err_addr("accept failed", sock_strerror(errno), addr,
      *addrlen);
    return -1;
  }
  if(c->tcp_conn_limit && c->type == comm_tcp_accept) {
    c->tcl_addr = tcl_addr_lookup(c->tcp_conn_limit, addr, *addrlen);
    if(!tcl_new_connection(c->tcl_addr)) {
      if(verbosity >= 3)
        log_err_addr("accept rejected",
        "connection limit exceeded", addr, *addrlen);
      sock_close(new_fd);
      return -1;
    }
  }
#ifndef HAVE_ACCEPT4
  fd_set_nonblock(new_fd);
#endif
  return new_fd;
}

#ifdef USE_WINSOCK
static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp),
#ifdef HAVE_BIO_SET_CALLBACK_EX
  size_t ATTR_UNUSED(len),
#endif
        int ATTR_UNUSED(argi), long argl,
#ifndef HAVE_BIO_SET_CALLBACK_EX
  long retvalue
#else
  int retvalue, size_t* ATTR_UNUSED(processed)
#endif
  )
{
  int wsa_err = WSAGetLastError(); /* store errcode before it is gone */
  verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper,
    (oper&BIO_CB_RETURN)?"return":"before",
    (oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"),
    wsa_err==WSAEWOULDBLOCK?"wsawb":"");
  /* on windows, check if previous operation caused EWOULDBLOCK */
  if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) ||
    (oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) {
    if(wsa_err == WSAEWOULDBLOCK)
      ub_winsock_tcp_wouldblock((struct ub_event*)
        BIO_get_callback_arg(b), UB_EV_READ);
  }
  if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) ||
    (oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) {
    if(wsa_err == WSAEWOULDBLOCK)
      ub_winsock_tcp_wouldblock((struct ub_event*)
        BIO_get_callback_arg(b), UB_EV_WRITE);
  }
  /* return original return value */
  return retvalue;
}

/** set win bio callbacks for nonblocking operations */
void
comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl)
{
  SSL* ssl = (SSL*)thessl;
  /* set them both just in case, but usually they are the same BIO */
#ifdef HAVE_BIO_SET_CALLBACK_EX
  BIO_set_callback_ex(SSL_get_rbio(ssl), &win_bio_cb);
#else
  BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb);
#endif
  BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)c->ev->ev);
#ifdef HAVE_BIO_SET_CALLBACK_EX
  BIO_set_callback_ex(SSL_get_wbio(ssl), &win_bio_cb);
#else
  BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb);
#endif
  BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)c->ev->ev);
}
#endif

#ifdef HAVE_NGHTTP2
/** Create http2 session server.  Per connection, after TCP accepted.*/
static int http2_session_server_create(struct http2_session* h2_session)
{
  log_assert(h2_session->callbacks);
  h2_session->is_drop = 0;
  if(nghttp2_session_server_new(&h2_session->session,
      h2_session->callbacks,
    h2_session) == NGHTTP2_ERR_NOMEM) {
    log_err("failed to create nghttp2 session server");
    return 0;
  }

  return 1;
}

/** Submit http2 setting to session. Once per session. */
static int http2_submit_settings(struct http2_session* h2_session)
{
  int ret;
  nghttp2_settings_entry settings[1] = {
    {NGHTTP2_SETTINGS_MAX_CONCURRENT_STREAMS,
     h2_session->c->http2_max_streams}};

  ret = nghttp2_submit_settings(h2_session->session, NGHTTP2_FLAG_NONE,
    settings, 1);
  if(ret) {
    verbose(VERB_QUERY, "http2: submit_settings failed, "
      "error: %s", nghttp2_strerror(ret));
    return 0;
  }
  return 1;
}
#endif /* HAVE_NGHTTP2 */

#ifdef HAVE_NGHTTP2
/** Delete http2 stream. After session delete or stream close callback */
static void http2_stream_delete(struct http2_session* h2_session,
  struct http2_stream* h2_stream)
{
  if(h2_stream->mesh_state) {
    mesh_state_remove_reply(h2_stream->mesh, h2_stream->mesh_state,
      h2_session->c);
    h2_stream->mesh_state = NULL;
  }
  http2_req_stream_clear(h2_stream);
  free(h2_stream);
}
#endif /* HAVE_NGHTTP2 */

/** delete http2 session server. After closing connection. */
static void http2_session_server_delete(struct http2_session* h2_session)
{
#ifdef HAVE_NGHTTP2
  struct http2_stream* h2_stream, *next;
  nghttp2_session_del(h2_session->session); /* NULL input is fine */
  h2_session->session = NULL;
  for(h2_stream = h2_session->first_stream; h2_stream;) {
    next = h2_stream->next;
    http2_stream_delete(h2_session, h2_stream);
    h2_stream = next;
  }
  h2_session->first_stream = NULL;
  h2_session->is_drop = 0;
  h2_session->postpone_drop = 0;
  h2_session->c->h2_stream = NULL;
#endif
  (void)h2_session;
}

void
comm_point_tcp_accept_callback(int fd, short event, void* arg)
{
  struct comm_point* c = (struct comm_point*)arg, *c_hdl;
  int new_fd;
  log_assert(c->type == comm_tcp_accept);
  if(!(event & UB_EV_READ)) {
    log_info("ignoring tcp accept event %d", (int)event);
    return;
  }
  ub_comm_base_now(c->ev->base);
  /* find free tcp handler. */
  if(!c->tcp_free) {
    log_warn("accepted too many tcp, connections full");
    return;
  }
  /* accept incoming connection. */
  c_hdl = c->tcp_free;
  /* Should not happen: inconsistent tcp_free state in
   * accept_callback. */
  log_assert(c_hdl->is_in_tcp_free);
  /* clear leftover flags from previous use, and then set the
   * correct event base for the event structure for libevent */
  ub_event_free(c_hdl->ev->ev);
  c_hdl->ev->ev = NULL;
  if((c_hdl->type == comm_tcp && c_hdl->tcp_req_info) ||
    c_hdl->type == comm_local || c_hdl->type == comm_raw)
    c_hdl->tcp_do_toggle_rw = 0;
  else  c_hdl->tcp_do_toggle_rw = 1;

  if(c_hdl->type == comm_http) {
#ifdef HAVE_NGHTTP2
    if(!c_hdl->h2_session ||
      !http2_session_server_create(c_hdl->h2_session)) {
      log_warn("failed to create nghttp2");
      return;
    }
    if(!c_hdl->h2_session ||
      !http2_submit_settings(c_hdl->h2_session)) {
      log_warn("failed to submit http2 settings");
      if(c_hdl->h2_session)
        http2_session_server_delete(c_hdl->h2_session);
      return;
    }
    if(!c->ssl) {
      c_hdl->tcp_do_toggle_rw = 0;
      c_hdl->use_h2 = 1;
    }
#endif
    c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1,
      UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT,
      comm_point_http_handle_callback, c_hdl);
  } else {
    c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1,
      UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT,
      comm_point_tcp_handle_callback, c_hdl);
  }
  if(!c_hdl->ev->ev) {
    log_warn("could not ub_event_new, dropped tcp");
#ifdef HAVE_NGHTTP2
    if(c_hdl->type == comm_http && c_hdl->h2_session)
      http2_session_server_delete(c_hdl->h2_session);
#endif
    return;
  }
  log_assert(fd != -1);
  (void)fd;
  new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.remote_addr,
    &c_hdl->repinfo.remote_addrlen);
  if(new_fd == -1) {
#ifdef HAVE_NGHTTP2
    if(c_hdl->type == comm_http && c_hdl->h2_session)
      http2_session_server_delete(c_hdl->h2_session);
#endif
    return;
  }
  /* Copy remote_address to client_address.
   * Simplest way/time for streams to do that. */
  c_hdl->repinfo.client_addrlen = c_hdl->repinfo.remote_addrlen;
  memmove(&c_hdl->repinfo.client_addr,
    &c_hdl->repinfo.remote_addr,
    c_hdl->repinfo.remote_addrlen);
  if(c->ssl) {
    c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd);
    if(!c_hdl->ssl) {
      c_hdl->fd = new_fd;
      comm_point_close(c_hdl);
      return;
    }
    c_hdl->ssl_shake_state = comm_ssl_shake_read;
#ifdef USE_WINSOCK
    comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl);
#endif
  }

  /* Paranoia: Check that the state has not changed from above: */
  /* Should not happen: tcp_free state changed within accept_callback. */
  log_assert(c_hdl == c->tcp_free);
  log_assert(c_hdl->is_in_tcp_free);
  /* grab the tcp handler buffers */
  c->cur_tcp_count++;
  c->tcp_free = c_hdl->tcp_free;
  c_hdl->tcp_free = NULL;
  c_hdl->is_in_tcp_free = 0;
  if(!c->tcp_free) {
    /* stop accepting incoming queries for now. */
    comm_point_stop_listening(c);
  }
  setup_tcp_handler(c_hdl, new_fd, c->cur_tcp_count, c->max_tcp_count);
}

/** Make tcp handler free for next assignment */
static void
reclaim_tcp_handler(struct comm_point* c)
{
  log_assert(c->type == comm_tcp);
  if(c->ssl) {
#ifdef HAVE_SSL
    SSL_shutdown(c->ssl);
    SSL_free(c->ssl);
    c->ssl = NULL;
#endif
  }
  comm_point_close(c);
  if(c->tcp_parent && !c->is_in_tcp_free) {
    /* Should not happen: bad tcp_free state in reclaim_tcp. */
    log_assert(c->tcp_free == NULL);
    log_assert(c->tcp_parent->cur_tcp_count > 0);
    c->tcp_parent->cur_tcp_count--;
    c->tcp_free = c->tcp_parent->tcp_free;
    c->tcp_parent->tcp_free = c;
    c->is_in_tcp_free = 1;
    if(!c->tcp_free) {
      /* re-enable listening on accept socket */
      comm_point_start_listening(c->tcp_parent, -1, -1);
    }
  }
  c->tcp_more_read_again = NULL;
  c->tcp_more_write_again = NULL;
  c->tcp_byte_count = 0;
  c->pp2_header_state = pp2_header_none;
  sldns_buffer_clear(c->buffer);
}

/** do the callback when writing is done */
static void
tcp_callback_writer(struct comm_point* c)
{
  log_assert(c->type == comm_tcp);
  if(!c->tcp_write_and_read) {
    sldns_buffer_clear(c->buffer);
    c->tcp_byte_count = 0;
  }
  if(c->tcp_do_toggle_rw)
    c->tcp_is_reading = 1;
  /* switch from listening(write) to listening(read) */
  if(c->tcp_req_info) {
    tcp_req_info_handle_writedone(c->tcp_req_info);
  } else {
    comm_point_stop_listening(c);
    if(c->tcp_write_and_read) {
      fptr_ok(fptr_whitelist_comm_point(c->callback));
      if( (*c->callback)(c, c->cb_arg, NETEVENT_PKT_WRITTEN,
        &c->repinfo) ) {
        comm_point_start_listening(c, -1,
          adjusted_tcp_timeout(c));
      }
    } else {
      comm_point_start_listening(c, -1,
          adjusted_tcp_timeout(c));
    }
  }
}

/** do the callback when reading is done */
static void
tcp_callback_reader(struct comm_point* c)
{
  log_assert(c->type == comm_tcp || c->type == comm_local);
  sldns_buffer_flip(c->buffer);
  if(c->tcp_do_toggle_rw)
    c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  if(c->tcp_req_info) {
    tcp_req_info_handle_readdone(c->tcp_req_info);
  } else {
    if(c->type == comm_tcp)
      comm_point_stop_listening(c);
    fptr_ok(fptr_whitelist_comm_point(c->callback));
    if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
      comm_point_start_listening(c, -1,
          adjusted_tcp_timeout(c));
    }
  }
}

#ifdef HAVE_SSL
/** true if the ssl handshake error has to be squelched from the logs */
int
squelch_err_ssl_handshake(unsigned long err)
{
  if(verbosity >= VERB_QUERY)
    return 0; /* only squelch on low verbosity */
  if(ERR_GET_LIB(err) == ERR_LIB_SSL &&
    (ERR_GET_REASON(err) == SSL_R_HTTPS_PROXY_REQUEST ||
     ERR_GET_REASON(err) == SSL_R_HTTP_REQUEST ||
     ERR_GET_REASON(err) == SSL_R_WRONG_VERSION_NUMBER ||
     ERR_GET_REASON(err) == SSL_R_SSLV3_ALERT_BAD_CERTIFICATE
#ifdef SSL_F_TLS_POST_PROCESS_CLIENT_HELLO
     || ERR_GET_REASON(err) == SSL_R_NO_SHARED_CIPHER
#endif
#ifdef SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO
     || ERR_GET_REASON(err) == SSL_R_UNKNOWN_PROTOCOL
     || ERR_GET_REASON(err) == SSL_R_UNSUPPORTED_PROTOCOL
#  ifdef SSL_R_VERSION_TOO_LOW
     || ERR_GET_REASON(err) == SSL_R_VERSION_TOO_LOW
#  endif
#endif
    ))
    return 1;
  return 0;
}
#endif /* HAVE_SSL */

/** continue ssl handshake */
#ifdef HAVE_SSL
static int
ssl_handshake(struct comm_point* c)
{
  int r;
  if(c->ssl_shake_state == comm_ssl_shake_hs_read) {
    /* read condition satisfied back to writing */
    comm_point_listen_for_rw(c, 0, 1);
    c->ssl_shake_state = comm_ssl_shake_none;
    return 1;
  }
  if(c->ssl_shake_state == comm_ssl_shake_hs_write) {
    /* write condition satisfied, back to reading */
    comm_point_listen_for_rw(c, 1, 0);
    c->ssl_shake_state = comm_ssl_shake_none;
    return 1;
  }

  ERR_clear_error();
  r = SSL_do_handshake(c->ssl);
  if(r != 1) {
    int want = SSL_get_error(c->ssl, r);
    if(want == SSL_ERROR_WANT_READ) {
      if(c->ssl_shake_state == comm_ssl_shake_read)
        return 1;
      c->ssl_shake_state = comm_ssl_shake_read;
      comm_point_listen_for_rw(c, 1, 0);
      return 1;
    } else if(want == SSL_ERROR_WANT_WRITE) {
      if(c->ssl_shake_state == comm_ssl_shake_write)
        return 1;
      c->ssl_shake_state = comm_ssl_shake_write;
      comm_point_listen_for_rw(c, 0, 1);
      return 1;
    } else if(r == 0) {
      return 0; /* closed */
    } else if(want == SSL_ERROR_SYSCALL) {
      /* SYSCALL and errno==0 means closed uncleanly */
#ifdef EPIPE
      if(errno == EPIPE && verbosity < 2)
        return 0; /* silence 'broken pipe' */
#endif
#ifdef ECONNRESET
      if(errno == ECONNRESET && verbosity < 2)
        return 0; /* silence reset by peer */
#endif
      if(!tcp_connect_errno_needs_log(
        (struct sockaddr*)&c->repinfo.remote_addr,
        c->repinfo.remote_addrlen))
        return 0; /* silence connect failures that
        show up because after connect this is the
        first system call that accesses the socket */
      if(errno != 0)
        log_err("SSL_handshake syscall: %s",
          strerror(errno));
      return 0;
    } else {
      unsigned long err = ERR_get_error();
      if(!squelch_err_ssl_handshake(err)) {
        long vr;
        log_crypto_err_io_code("ssl handshake failed",
          want, err);
        if((vr=SSL_get_verify_result(c->ssl)) != 0)
          log_err("ssl handshake cert error: %s",
            X509_verify_cert_error_string(
            vr));
        log_addr(VERB_OPS, "ssl handshake failed",
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
      }
      return 0;
    }
  }
  /* this is where peer verification could take place */
  if((SSL_get_verify_mode(c->ssl)&SSL_VERIFY_PEER)) {
    /* verification */
    if(SSL_get_verify_result(c->ssl) == X509_V_OK) {
#ifdef HAVE_SSL_GET1_PEER_CERTIFICATE
      X509* x = SSL_get1_peer_certificate(c->ssl);
#else
      X509* x = SSL_get_peer_certificate(c->ssl);
#endif
      if(!x) {
        log_addr(VERB_ALGO, "SSL connection failed: "
          "no certificate",
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
        return 0;
      }
      log_cert(VERB_ALGO, "peer certificate", x);
#ifdef HAVE_SSL_GET0_PEERNAME
      if(SSL_get0_peername(c->ssl)) {
        char buf[255];
        snprintf(buf, sizeof(buf), "SSL connection "
          "to %s authenticated",
          SSL_get0_peername(c->ssl));
        log_addr(VERB_ALGO, buf, &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
      } else {
#endif
        log_addr(VERB_ALGO, "SSL connection "
          "authenticated", &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
#ifdef HAVE_SSL_GET0_PEERNAME
      }
#endif
      X509_free(x);
    } else {
#ifdef HAVE_SSL_GET1_PEER_CERTIFICATE
      X509* x = SSL_get1_peer_certificate(c->ssl);
#else
      X509* x = SSL_get_peer_certificate(c->ssl);
#endif
      if(x) {
        log_cert(VERB_ALGO, "peer certificate", x);
        X509_free(x);
      }
      log_addr(VERB_ALGO, "SSL connection failed: "
        "failed to authenticate",
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
      return 0;
    }
  } else {
    /* unauthenticated, the verify peer flag was not set
     * in c->ssl when the ssl object was created from ssl_ctx */
    log_addr(VERB_ALGO, "SSL connection", &c->repinfo.remote_addr,
      c->repinfo.remote_addrlen);
  }

#ifdef HAVE_SSL_GET0_ALPN_SELECTED
  /* check if http2 use is negotiated */
  if(c->type == comm_http && c->h2_session) {
    const unsigned char *alpn;
    unsigned int alpnlen = 0;
    SSL_get0_alpn_selected(c->ssl, &alpn, &alpnlen);
    if(alpnlen == 2 && memcmp("h2", alpn, 2) == 0) {
      /* connection upgraded to HTTP2 */
      c->tcp_do_toggle_rw = 0;
      c->use_h2 = 1;
    } else {
      verbose(VERB_ALGO, "client doesn't support HTTP/2");
      return 0;
    }
  }
#endif

  /* setup listen rw correctly */
  if(c->tcp_is_reading) {
    if(c->ssl_shake_state != comm_ssl_shake_read)
      comm_point_listen_for_rw(c, 1, 0);
  } else {
    comm_point_listen_for_rw(c, 0, 1);
  }
  c->ssl_shake_state = comm_ssl_shake_none;
  return 1;
}
#endif /* HAVE_SSL */

/** ssl read callback on TCP */
static int
ssl_handle_read(struct comm_point* c)
{
#ifdef HAVE_SSL
  int r;
  if(c->ssl_shake_state != comm_ssl_shake_none) {
    if(!ssl_handshake(c))
      return 0;
    if(c->ssl_shake_state != comm_ssl_shake_none)
      return 1;
  }
  if(c->pp2_enabled && c->pp2_header_state != pp2_header_done) {
    struct pp2_header* header = NULL;
    size_t want_read_size = 0;
    size_t current_read_size = 0;
    if(c->pp2_header_state == pp2_header_none) {
      want_read_size = PP2_HEADER_SIZE;
      if(sldns_buffer_remaining(c->buffer)<want_read_size) {
        log_err_addr("proxy_protocol: not enough "
          "buffer size to read PROXYv2 header", "",
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
        return 0;
      }
      verbose(VERB_ALGO, "proxy_protocol: reading fixed "
        "part of PROXYv2 header (len %lu)",
        (unsigned long)want_read_size);
      current_read_size = want_read_size;
      if(c->tcp_byte_count < current_read_size) {
        ERR_clear_error();
        if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(
          c->buffer, c->tcp_byte_count),
          current_read_size -
          c->tcp_byte_count)) <= 0) {
          int want = SSL_get_error(c->ssl, r);
          if(want == SSL_ERROR_ZERO_RETURN) {
            if(c->tcp_req_info)
              return tcp_req_info_handle_read_close(c->tcp_req_info);
            return 0; /* shutdown, closed */
          } else if(want == SSL_ERROR_WANT_READ) {
#ifdef USE_WINSOCK
            ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
#endif
            return 1; /* read more later */
          } else if(want == SSL_ERROR_WANT_WRITE) {
            c->ssl_shake_state = comm_ssl_shake_hs_write;
            comm_point_listen_for_rw(c, 0, 1);
            return 1;
          } else if(want == SSL_ERROR_SYSCALL) {
#ifdef ECONNRESET
            if(errno == ECONNRESET && verbosity < 2)
              return 0; /* silence reset by peer */
#endif
            if(errno != 0)
              log_err("SSL_read syscall: %s",
                strerror(errno));
            return 0;
          }
          log_crypto_err_io("could not SSL_read",
            want);
          return 0;
        }
        c->tcp_byte_count += r;
        sldns_buffer_skip(c->buffer, r);
        if(c->tcp_byte_count != current_read_size) return 1;
        c->pp2_header_state = pp2_header_init;
      }
    }
    if(c->pp2_header_state == pp2_header_init) {
      int err;
      err = pp2_read_header(
        sldns_buffer_begin(c->buffer),
        sldns_buffer_limit(c->buffer));
      if(err) {
        log_err("proxy_protocol: could not parse "
          "PROXYv2 header (%s)",
          pp_lookup_error(err));
        return 0;
      }
      header = (struct pp2_header*)sldns_buffer_begin(c->buffer);
      want_read_size = ntohs(header->len);
      if(sldns_buffer_limit(c->buffer) <
        PP2_HEADER_SIZE + want_read_size) {
        log_err_addr("proxy_protocol: not enough "
          "buffer size to read PROXYv2 header", "",
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
        return 0;
      }
      verbose(VERB_ALGO, "proxy_protocol: reading variable "
        "part of PROXYv2 header (len %lu)",
        (unsigned long)want_read_size);
      current_read_size = PP2_HEADER_SIZE + want_read_size;
      if(want_read_size == 0) {
        /* nothing more to read; header is complete */
        c->pp2_header_state = pp2_header_done;
      } else if(c->tcp_byte_count < current_read_size) {
        ERR_clear_error();
        if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(
          c->buffer, c->tcp_byte_count),
          current_read_size -
          c->tcp_byte_count)) <= 0) {
          int want = SSL_get_error(c->ssl, r);
          if(want == SSL_ERROR_ZERO_RETURN) {
            if(c->tcp_req_info)
              return tcp_req_info_handle_read_close(c->tcp_req_info);
            return 0; /* shutdown, closed */
          } else if(want == SSL_ERROR_WANT_READ) {
#ifdef USE_WINSOCK
            ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
#endif
            return 1; /* read more later */
          } else if(want == SSL_ERROR_WANT_WRITE) {
            c->ssl_shake_state = comm_ssl_shake_hs_write;
            comm_point_listen_for_rw(c, 0, 1);
            return 1;
          } else if(want == SSL_ERROR_SYSCALL) {
#ifdef ECONNRESET
            if(errno == ECONNRESET && verbosity < 2)
              return 0; /* silence reset by peer */
#endif
            if(errno != 0)
              log_err("SSL_read syscall: %s",
                strerror(errno));
            return 0;
          }
          log_crypto_err_io("could not SSL_read",
            want);
          return 0;
        }
        c->tcp_byte_count += r;
        sldns_buffer_skip(c->buffer, r);
        if(c->tcp_byte_count != current_read_size) return 1;
        c->pp2_header_state = pp2_header_done;
      }
    }
    if(c->pp2_header_state != pp2_header_done || !header) {
      log_err_addr("proxy_protocol: wrong state for the "
        "PROXYv2 header", "", &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
      return 0;
    }
    sldns_buffer_flip(c->buffer);
    if(!consume_pp2_header(c->buffer, &c->repinfo, 1)) {
      log_err_addr("proxy_protocol: could not consume "
        "PROXYv2 header", "", &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
      return 0;
    }
    verbose(VERB_ALGO, "proxy_protocol: successful read of "
      "PROXYv2 header");
    /* Clear and reset the buffer to read the following
     * DNS packet(s). */
    sldns_buffer_clear(c->buffer);
    c->tcp_byte_count = 0;
    return 1;
  }
  if(c->tcp_byte_count < sizeof(uint16_t)) {
    /* read length bytes */
    ERR_clear_error();
    if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(c->buffer,
      c->tcp_byte_count), (int)(sizeof(uint16_t) -
      c->tcp_byte_count))) <= 0) {
      int want = SSL_get_error(c->ssl, r);
      if(want == SSL_ERROR_ZERO_RETURN) {
        if(c->tcp_req_info)
          return tcp_req_info_handle_read_close(c->tcp_req_info);
        return 0; /* shutdown, closed */
      } else if(want == SSL_ERROR_WANT_READ) {
#ifdef USE_WINSOCK
        ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
#endif
        return 1; /* read more later */
      } else if(want == SSL_ERROR_WANT_WRITE) {
        c->ssl_shake_state = comm_ssl_shake_hs_write;
        comm_point_listen_for_rw(c, 0, 1);
        return 1;
      } else if(want == SSL_ERROR_SYSCALL) {
#ifdef ECONNRESET
        if(errno == ECONNRESET && verbosity < 2)
          return 0; /* silence reset by peer */
#endif
        if(errno != 0)
          log_err("SSL_read syscall: %s",
            strerror(errno));
        return 0;
      }
      log_crypto_err_io("could not SSL_read", want);
      return 0;
    }
    c->tcp_byte_count += r;
    if(c->tcp_byte_count < sizeof(uint16_t))
      return 1;
    if(sldns_buffer_read_u16_at(c->buffer, 0) >
      sldns_buffer_capacity(c->buffer)) {
      verbose(VERB_QUERY, "ssl: dropped larger than buffer");
      return 0;
    }
    sldns_buffer_set_limit(c->buffer,
      sldns_buffer_read_u16_at(c->buffer, 0));
    if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
      verbose(VERB_QUERY, "ssl: dropped bogus too short.");
      return 0;
    }
    sldns_buffer_skip(c->buffer, (ssize_t)(c->tcp_byte_count-sizeof(uint16_t)));
    verbose(VERB_ALGO, "Reading ssl tcp query of length %d",
      (int)sldns_buffer_limit(c->buffer));
  }
  if(sldns_buffer_remaining(c->buffer) > 0) {
    ERR_clear_error();
    r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
      (int)sldns_buffer_remaining(c->buffer));
    if(r <= 0) {
      int want = SSL_get_error(c->ssl, r);
      if(want == SSL_ERROR_ZERO_RETURN) {
        if(c->tcp_req_info)
          return tcp_req_info_handle_read_close(c->tcp_req_info);
        return 0; /* shutdown, closed */
      } else if(want == SSL_ERROR_WANT_READ) {
#ifdef USE_WINSOCK
        ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
#endif
        return 1; /* read more later */
      } else if(want == SSL_ERROR_WANT_WRITE) {
        c->ssl_shake_state = comm_ssl_shake_hs_write;
        comm_point_listen_for_rw(c, 0, 1);
        return 1;
      } else if(want == SSL_ERROR_SYSCALL) {
#ifdef ECONNRESET
        if(errno == ECONNRESET && verbosity < 2)
          return 0; /* silence reset by peer */
#endif
        if(errno != 0)
          log_err("SSL_read syscall: %s",
            strerror(errno));
        return 0;
      }
      log_crypto_err_io("could not SSL_read", want);
      return 0;
    }
    sldns_buffer_skip(c->buffer, (ssize_t)r);
  }
  if(sldns_buffer_remaining(c->buffer) <= 0) {
    tcp_callback_reader(c);
  }
  return 1;
#else
  (void)c;
  return 0;
#endif /* HAVE_SSL */
}

/** ssl write callback on TCP */
static int
ssl_handle_write(struct comm_point* c)
{
#ifdef HAVE_SSL
  int r;
  if(c->ssl_shake_state != comm_ssl_shake_none) {
    if(!ssl_handshake(c))
      return 0;
    if(c->ssl_shake_state != comm_ssl_shake_none)
      return 1;
  }
  /* ignore return, if fails we may simply block */
  (void)SSL_set_mode(c->ssl, (long)SSL_MODE_ENABLE_PARTIAL_WRITE);
  if((c->tcp_write_and_read?c->tcp_write_byte_count:c->tcp_byte_count) < sizeof(uint16_t)) {
    uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(c->buffer));
    ERR_clear_error();
    if(c->tcp_write_and_read) {
      if(c->tcp_write_pkt_len + 2 < LDNS_RR_BUF_SIZE) {
        /* combine the tcp length and the query for
         * write, this emulates writev */
        uint8_t buf[LDNS_RR_BUF_SIZE];
        memmove(buf, &len, sizeof(uint16_t));
        memmove(buf+sizeof(uint16_t),
          c->tcp_write_pkt,
          c->tcp_write_pkt_len);
        r = SSL_write(c->ssl,
          (void*)(buf+c->tcp_write_byte_count),
          c->tcp_write_pkt_len + 2 -
          c->tcp_write_byte_count);
      } else {
        r = SSL_write(c->ssl,
          (void*)(((uint8_t*)&len)+c->tcp_write_byte_count),
          (int)(sizeof(uint16_t)-c->tcp_write_byte_count));
      }
    } else if(sizeof(uint16_t)+sldns_buffer_remaining(c->buffer) <
      LDNS_RR_BUF_SIZE) {
      /* combine the tcp length and the query for write,
       * this emulates writev */
      uint8_t buf[LDNS_RR_BUF_SIZE];
      memmove(buf, &len, sizeof(uint16_t));
      memmove(buf+sizeof(uint16_t),
        sldns_buffer_current(c->buffer),
        sldns_buffer_remaining(c->buffer));
      r = SSL_write(c->ssl, (void*)(buf+c->tcp_byte_count),
        (int)(sizeof(uint16_t)+
        sldns_buffer_remaining(c->buffer)
        - c->tcp_byte_count));
    } else {
      r = SSL_write(c->ssl,
        (void*)(((uint8_t*)&len)+c->tcp_byte_count),
        (int)(sizeof(uint16_t)-c->tcp_byte_count));
    }
    if(r <= 0) {
      int want = SSL_get_error(c->ssl, r);
      if(want == SSL_ERROR_ZERO_RETURN) {
        return 0; /* closed */
      } else if(want == SSL_ERROR_WANT_READ) {
        c->ssl_shake_state = comm_ssl_shake_hs_read;
        comm_point_listen_for_rw(c, 1, 0);
        return 1; /* wait for read condition */
      } else if(want == SSL_ERROR_WANT_WRITE) {
#ifdef USE_WINSOCK
        ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
#endif
        return 1; /* write more later */
      } else if(want == SSL_ERROR_SYSCALL) {
#ifdef EPIPE
        if(errno == EPIPE && verbosity < 2)
          return 0; /* silence 'broken pipe' */
#endif
        if(errno != 0)
          log_err("SSL_write syscall: %s",
            strerror(errno));
        return 0;
      }
      log_crypto_err_io("could not SSL_write", want);
      return 0;
    }
    if(c->tcp_write_and_read) {
      c->tcp_write_byte_count += r;
      if(c->tcp_write_byte_count < sizeof(uint16_t))
        return 1;
    } else {
      c->tcp_byte_count += r;
      if(c->tcp_byte_count < sizeof(uint16_t))
        return 1;
      sldns_buffer_set_position(c->buffer, c->tcp_byte_count -
        sizeof(uint16_t));
    }
    if((!c->tcp_write_and_read && sldns_buffer_remaining(c->buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
      tcp_callback_writer(c);
      return 1;
    }
  }
  log_assert(c->tcp_write_and_read || sldns_buffer_remaining(c->buffer) > 0);
  log_assert(!c->tcp_write_and_read || c->tcp_write_byte_count < c->tcp_write_pkt_len + 2);
  ERR_clear_error();
  if(c->tcp_write_and_read) {
    r = SSL_write(c->ssl, (void*)(c->tcp_write_pkt + c->tcp_write_byte_count - 2),
      (int)(c->tcp_write_pkt_len + 2 - c->tcp_write_byte_count));
  } else {
    r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
      (int)sldns_buffer_remaining(c->buffer));
  }
  if(r <= 0) {
    int want = SSL_get_error(c->ssl, r);
    if(want == SSL_ERROR_ZERO_RETURN) {
      return 0; /* closed */
    } else if(want == SSL_ERROR_WANT_READ) {
      c->ssl_shake_state = comm_ssl_shake_hs_read;
      comm_point_listen_for_rw(c, 1, 0);
      return 1; /* wait for read condition */
    } else if(want == SSL_ERROR_WANT_WRITE) {
#ifdef USE_WINSOCK
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
#endif
      return 1; /* write more later */
    } else if(want == SSL_ERROR_SYSCALL) {
#ifdef EPIPE
      if(errno == EPIPE && verbosity < 2)
        return 0; /* silence 'broken pipe' */
#endif
      if(errno != 0)
        log_err("SSL_write syscall: %s",
          strerror(errno));
      return 0;
    }
    log_crypto_err_io("could not SSL_write", want);
    return 0;
  }
  if(c->tcp_write_and_read) {
    c->tcp_write_byte_count += r;
  } else {
    sldns_buffer_skip(c->buffer, (ssize_t)r);
  }

  if((!c->tcp_write_and_read && sldns_buffer_remaining(c->buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
    tcp_callback_writer(c);
  }
  return 1;
#else
  (void)c;
  return 0;
#endif /* HAVE_SSL */
}

/** handle ssl tcp connection with dns contents */
static int
ssl_handle_it(struct comm_point* c, int is_write)
{
  /* handle case where renegotiation wants read during write call
   * or write during read calls */
  if(is_write && c->ssl_shake_state == comm_ssl_shake_hs_write)
    return ssl_handle_read(c);
  else if(!is_write && c->ssl_shake_state == comm_ssl_shake_hs_read)
    return ssl_handle_write(c);
  /* handle read events for read operation and write events for a
   * write operation */
  else if(!is_write)
    return ssl_handle_read(c);
  return ssl_handle_write(c);
}

/**
 * Handle tcp reading callback.
 * @param fd: file descriptor of socket.
 * @param c: comm point to read from into buffer.
 * @param short_ok: if true, very short packets are OK (for comm_local).
 * @return: 0 on error
 */
static int
comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok)
{
  ssize_t r;
  int recv_initial = 0;
  log_assert(c->type == comm_tcp || c->type == comm_local);
  if(c->ssl)
    return ssl_handle_it(c, 0);
  if(!c->tcp_is_reading && !c->tcp_write_and_read)
    return 0;

  log_assert(fd != -1);
  if(c->pp2_enabled && c->pp2_header_state != pp2_header_done) {
    struct pp2_header* header = NULL;
    size_t want_read_size = 0;
    size_t current_read_size = 0;
    if(c->pp2_header_state == pp2_header_none) {
      want_read_size = PP2_HEADER_SIZE;
      if(sldns_buffer_remaining(c->buffer)<want_read_size) {
        log_err_addr("proxy_protocol: not enough "
          "buffer size to read PROXYv2 header", "",
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
        return 0;
      }
      verbose(VERB_ALGO, "proxy_protocol: reading fixed "
        "part of PROXYv2 header (len %lu)",
        (unsigned long)want_read_size);
      current_read_size = want_read_size;
      if(c->tcp_byte_count < current_read_size) {
        r = recv(fd, (void*)sldns_buffer_at(c->buffer,
          c->tcp_byte_count),
          current_read_size-c->tcp_byte_count, MSG_DONTWAIT);
        if(r == 0) {
          if(c->tcp_req_info)
            return tcp_req_info_handle_read_close(c->tcp_req_info);
          return 0;
        } else if(r == -1) {
          goto recv_error_initial;
        }
        c->tcp_byte_count += r;
        sldns_buffer_skip(c->buffer, r);
        if(c->tcp_byte_count != current_read_size) return 1;
        c->pp2_header_state = pp2_header_init;
      }
    }
    if(c->pp2_header_state == pp2_header_init) {
      int err;
      err = pp2_read_header(
        sldns_buffer_begin(c->buffer),
        sldns_buffer_limit(c->buffer));
      if(err) {
        log_err("proxy_protocol: could not parse "
          "PROXYv2 header (%s)",
          pp_lookup_error(err));
        return 0;
      }
      header = (struct pp2_header*)sldns_buffer_begin(c->buffer);
      want_read_size = ntohs(header->len);
      if(sldns_buffer_limit(c->buffer) <
        PP2_HEADER_SIZE + want_read_size) {
        log_err_addr("proxy_protocol: not enough "
          "buffer size to read PROXYv2 header", "",
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
        return 0;
      }
      verbose(VERB_ALGO, "proxy_protocol: reading variable "
        "part of PROXYv2 header (len %lu)",
        (unsigned long)want_read_size);
      current_read_size = PP2_HEADER_SIZE + want_read_size;
      if(want_read_size == 0) {
        /* nothing more to read; header is complete */
        c->pp2_header_state = pp2_header_done;
      } else if(c->tcp_byte_count < current_read_size) {
        r = recv(fd, (void*)sldns_buffer_at(c->buffer,
          c->tcp_byte_count),
          current_read_size-c->tcp_byte_count, MSG_DONTWAIT);
        if(r == 0) {
          if(c->tcp_req_info)
            return tcp_req_info_handle_read_close(c->tcp_req_info);
          return 0;
        } else if(r == -1) {
          goto recv_error;
        }
        c->tcp_byte_count += r;
        sldns_buffer_skip(c->buffer, r);
        if(c->tcp_byte_count != current_read_size) return 1;
        c->pp2_header_state = pp2_header_done;
      }
    }
    if(c->pp2_header_state != pp2_header_done || !header) {
      log_err_addr("proxy_protocol: wrong state for the "
        "PROXYv2 header", "", &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
      return 0;
    }
    sldns_buffer_flip(c->buffer);
    if(!consume_pp2_header(c->buffer, &c->repinfo, 1)) {
      log_err_addr("proxy_protocol: could not consume "
        "PROXYv2 header", "", &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
      return 0;
    }
    verbose(VERB_ALGO, "proxy_protocol: successful read of "
      "PROXYv2 header");
    /* Clear and reset the buffer to read the following
        * DNS packet(s). */
    sldns_buffer_clear(c->buffer);
    c->tcp_byte_count = 0;
    return 1;
  }

  if(c->tcp_byte_count < sizeof(uint16_t)) {
    /* read length bytes */
    r = recv(fd,(void*)sldns_buffer_at(c->buffer,c->tcp_byte_count),
      sizeof(uint16_t)-c->tcp_byte_count, MSG_DONTWAIT);
    if(r == 0) {
      if(c->tcp_req_info)
        return tcp_req_info_handle_read_close(c->tcp_req_info);
      return 0;
    } else if(r == -1) {
      if(c->pp2_enabled) goto recv_error;
      goto recv_error_initial;
    }
    c->tcp_byte_count += r;
    if(c->tcp_byte_count != sizeof(uint16_t))
      return 1;
    if(sldns_buffer_read_u16_at(c->buffer, 0) >
      sldns_buffer_capacity(c->buffer)) {
      verbose(VERB_QUERY, "tcp: dropped larger than buffer");
      return 0;
    }
    sldns_buffer_set_limit(c->buffer,
      sldns_buffer_read_u16_at(c->buffer, 0));
    if(!short_ok &&
      sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
      verbose(VERB_QUERY, "tcp: dropped bogus too short.");
      return 0;
    }
    verbose(VERB_ALGO, "Reading tcp query of length %d",
      (int)sldns_buffer_limit(c->buffer));
  }

  if(sldns_buffer_remaining(c->buffer) == 0)
    log_err("in comm_point_tcp_handle_read buffer_remaining is "
      "not > 0 as expected, continuing with (harmless) 0 "
      "length recv");
  r = recv(fd, (void*)sldns_buffer_current(c->buffer),
    sldns_buffer_remaining(c->buffer), MSG_DONTWAIT);
  if(r == 0) {
    if(c->tcp_req_info)
      return tcp_req_info_handle_read_close(c->tcp_req_info);
    return 0;
  } else if(r == -1) {
    goto recv_error;
  }
  sldns_buffer_skip(c->buffer, r);
  if(sldns_buffer_remaining(c->buffer) <= 0) {
    tcp_callback_reader(c);
  }
  return 1;

recv_error_initial:
  recv_initial = 1;
recv_error:
#ifndef USE_WINSOCK
  if(errno == EINTR || errno == EAGAIN)
    return 1;
#ifdef ECONNRESET
  if(errno == ECONNRESET && verbosity < 2)
    return 0; /* silence reset by peer */
#endif
  if(recv_initial) {
#ifdef ECONNREFUSED
    if(errno == ECONNREFUSED && verbosity < 2)
      return 0; /* silence reset by peer */
#endif
#ifdef ENETUNREACH
    if(errno == ENETUNREACH && verbosity < 2)
      return 0; /* silence it */
#endif
#ifdef EHOSTDOWN
    if(errno == EHOSTDOWN && verbosity < 2)
      return 0; /* silence it */
#endif
#ifdef EHOSTUNREACH
    if(errno == EHOSTUNREACH && verbosity < 2)
      return 0; /* silence it */
#endif
#ifdef ENETDOWN
    if(errno == ENETDOWN && verbosity < 2)
      return 0; /* silence it */
#endif
#ifdef EACCES
    if(errno == EACCES && verbosity < 2)
      return 0; /* silence it */
#endif
#ifdef ENOTCONN
    if(errno == ENOTCONN) {
      log_err_addr("read (in tcp initial) failed and this "
        "could be because TCP Fast Open is "
        "enabled [--disable-tfo-client "
        "--disable-tfo-server] but does not "
        "work", sock_strerror(errno),
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
      return 0;
    }
#endif
  }
#else /* USE_WINSOCK */
  if(recv_initial) {
    if(WSAGetLastError() == WSAECONNREFUSED && verbosity < 2)
      return 0;
    if(WSAGetLastError() == WSAEHOSTDOWN && verbosity < 2)
      return 0;
    if(WSAGetLastError() == WSAEHOSTUNREACH && verbosity < 2)
      return 0;
    if(WSAGetLastError() == WSAENETDOWN && verbosity < 2)
      return 0;
    if(WSAGetLastError() == WSAENETUNREACH && verbosity < 2)
      return 0;
  }
  if(WSAGetLastError() == WSAECONNRESET)
    return 0;
  if(WSAGetLastError() == WSAEINPROGRESS)
    return 1;
  if(WSAGetLastError() == WSAEWOULDBLOCK) {
    ub_winsock_tcp_wouldblock(c->ev->ev,
      UB_EV_READ);
    return 1;
  }
#endif
  log_err_addr((recv_initial?"read (in tcp initial)":"read (in tcp)"),
    sock_strerror(errno), &c->repinfo.remote_addr,
    c->repinfo.remote_addrlen);
  return 0;
}

/**
 * Handle tcp writing callback.
 * @param fd: file descriptor of socket.
 * @param c: comm point to write buffer out of.
 * @return: 0 on error
 */
static int
comm_point_tcp_handle_write(int fd, struct comm_point* c)
{
  ssize_t r;
  struct sldns_buffer *buffer;
  log_assert(c->type == comm_tcp);
#ifdef USE_DNSCRYPT
  buffer = c->dnscrypt_buffer;
#else
  buffer = c->buffer;
#endif
  if(c->tcp_is_reading && !c->ssl && !c->tcp_write_and_read)
    return 0;
  log_assert(fd != -1);
  if(((!c->tcp_write_and_read && c->tcp_byte_count == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == 0)) && c->tcp_check_nb_connect) {
    /* check for pending error from nonblocking connect */
    /* from Stevens, unix network programming, vol1, 3rd ed, p450*/
    int error = 0;
    socklen_t len = (socklen_t)sizeof(error);
    if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
      &len) < 0){
#ifndef USE_WINSOCK
      error = errno; /* on solaris errno is error */
#else /* USE_WINSOCK */
      error = WSAGetLastError();
#endif
    }
#ifndef USE_WINSOCK
#if defined(EINPROGRESS) && defined(EWOULDBLOCK)
    if(error == EINPROGRESS || error == EWOULDBLOCK)
      return 1; /* try again later */
    else
#endif
    if(error != 0 && verbosity < 2)
      return 0; /* silence lots of chatter in the logs */
                else if(error != 0) {
      log_err_addr("tcp connect", strerror(error),
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
#else /* USE_WINSOCK */
    /* examine error */
    if(error == WSAEINPROGRESS)
      return 1;
    else if(error == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
      return 1;
    } else if(error != 0 && verbosity < 2)
      return 0;
    else if(error != 0) {
      log_err_addr("tcp connect", wsa_strerror(error),
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
#endif /* USE_WINSOCK */
      return 0;
    }
  }
  if(c->ssl)
    return ssl_handle_it(c, 1);

#ifdef USE_MSG_FASTOPEN
  /* Only try this on first use of a connection that uses tfo,
     otherwise fall through to normal write */
  /* Also, TFO support on WINDOWS not implemented at the moment */
  if(c->tcp_do_fastopen == 1) {
    /* this form of sendmsg() does both a connect() and send() so need to
       look for various flavours of error*/
    uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(buffer));
    struct msghdr msg;
    struct iovec iov[2];
    c->tcp_do_fastopen = 0;
    memset(&msg, 0, sizeof(msg));
    if(c->tcp_write_and_read) {
      iov[0].iov_base = (uint8_t*)&len + c->tcp_write_byte_count;
      iov[0].iov_len = sizeof(uint16_t) - c->tcp_write_byte_count;
      iov[1].iov_base = c->tcp_write_pkt;
      iov[1].iov_len = c->tcp_write_pkt_len;
    } else {
      iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
      iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
      iov[1].iov_base = sldns_buffer_begin(buffer);
      iov[1].iov_len = sldns_buffer_limit(buffer);
    }
    log_assert(iov[0].iov_len > 0);
    msg.msg_name = &c->repinfo.remote_addr;
    msg.msg_namelen = c->repinfo.remote_addrlen;
    msg.msg_iov = iov;
    msg.msg_iovlen = 2;
    r = sendmsg(fd, &msg, MSG_FASTOPEN);
    if (r == -1) {
#if defined(EINPROGRESS) && defined(EWOULDBLOCK)
      /* Handshake is underway, maybe because no TFO cookie available.
         Come back to write the message*/
      if(errno == EINPROGRESS || errno == EWOULDBLOCK)
        return 1;
#endif
      if(errno == EINTR || errno == EAGAIN)
        return 1;
      /* Not handling EISCONN here as shouldn't ever hit that case.*/
      if(errno != EPIPE
#ifdef EOPNOTSUPP
        /* if /proc/sys/net/ipv4/tcp_fastopen is
         * disabled on Linux, sendmsg may return
         * 'Operation not supported', if so
         * fallthrough to ordinary connect. */
        && errno != EOPNOTSUPP
#endif
        && errno != 0) {
        if(verbosity < 2)
          return 0; /* silence lots of chatter in the logs */
        log_err_addr("tcp sendmsg", strerror(errno),
          &c->repinfo.remote_addr,
          c->repinfo.remote_addrlen);
        return 0;
      }
      verbose(VERB_ALGO, "tcp sendmsg for fastopen failed (with %s), try normal connect", strerror(errno));
      /* fallthrough to nonFASTOPEN
       * (MSG_FASTOPEN on Linux 3 produces EPIPE)
       * we need to perform connect() */
      if(connect(fd, (struct sockaddr *)&c->repinfo.remote_addr,
        c->repinfo.remote_addrlen) == -1) {
#ifdef EINPROGRESS
        if(errno == EINPROGRESS)
          return 1; /* wait until connect done*/
#endif
#ifdef USE_WINSOCK
        if(WSAGetLastError() == WSAEINPROGRESS ||
          WSAGetLastError() == WSAEWOULDBLOCK)
          return 1; /* wait until connect done*/
#endif
        if(tcp_connect_errno_needs_log(
          (struct sockaddr *)&c->repinfo.remote_addr,
          c->repinfo.remote_addrlen)) {
          log_err_addr("outgoing tcp: connect after EPIPE for fastopen",
            strerror(errno),
            &c->repinfo.remote_addr,
            c->repinfo.remote_addrlen);
        }
        return 0;
      }

    } else {
      if(c->tcp_write_and_read) {
        c->tcp_write_byte_count += r;
        if(c->tcp_write_byte_count < sizeof(uint16_t))
          return 1;
      } else {
        c->tcp_byte_count += r;
        if(c->tcp_byte_count < sizeof(uint16_t))
          return 1;
        sldns_buffer_set_position(buffer, c->tcp_byte_count -
          sizeof(uint16_t));
      }
      if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
        tcp_callback_writer(c);
        return 1;
      }
    }
  }
#endif /* USE_MSG_FASTOPEN */

  if((c->tcp_write_and_read?c->tcp_write_byte_count:c->tcp_byte_count) < sizeof(uint16_t)) {
    uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(buffer));
#ifdef HAVE_WRITEV
    struct iovec iov[2];
    if(c->tcp_write_and_read) {
      iov[0].iov_base = (uint8_t*)&len + c->tcp_write_byte_count;
      iov[0].iov_len = sizeof(uint16_t) - c->tcp_write_byte_count;
      iov[1].iov_base = c->tcp_write_pkt;
      iov[1].iov_len = c->tcp_write_pkt_len;
    } else {
      iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
      iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
      iov[1].iov_base = sldns_buffer_begin(buffer);
      iov[1].iov_len = sldns_buffer_limit(buffer);
    }
    log_assert(iov[0].iov_len > 0);
    r = writev(fd, iov, 2);
#else /* HAVE_WRITEV */
    if(c->tcp_write_and_read) {
      r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_write_byte_count),
        sizeof(uint16_t)-c->tcp_write_byte_count, 0);
    } else {
      r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count),
        sizeof(uint16_t)-c->tcp_byte_count, 0);
    }
#endif /* HAVE_WRITEV */
    if(r == -1) {
#ifndef USE_WINSOCK
#  ifdef EPIPE
                  if(errno == EPIPE && verbosity < 2)
                          return 0; /* silence 'broken pipe' */
  #endif
      if(errno == EINTR || errno == EAGAIN)
        return 1;
#ifdef ECONNRESET
      if(errno == ECONNRESET && verbosity < 2)
        return 0; /* silence reset by peer */
#endif
#  ifdef HAVE_WRITEV
      log_err_addr("tcp writev", strerror(errno),
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
#  else /* HAVE_WRITEV */
      log_err_addr("tcp send s", strerror(errno),
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
#  endif /* HAVE_WRITEV */
#else
      if(WSAGetLastError() == WSAENOTCONN)
        return 1;
      if(WSAGetLastError() == WSAEINPROGRESS)
        return 1;
      if(WSAGetLastError() == WSAEWOULDBLOCK) {
        ub_winsock_tcp_wouldblock(c->ev->ev,
          UB_EV_WRITE);
        return 1;
      }
      if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
        return 0; /* silence reset by peer */
      log_err_addr("tcp send s",
        wsa_strerror(WSAGetLastError()),
        &c->repinfo.remote_addr,
        c->repinfo.remote_addrlen);
#endif
      return 0;
    }
    if(c->tcp_write_and_read) {
      c->tcp_write_byte_count += r;
      if(c->tcp_write_byte_count < sizeof(uint16_t))
        return 1;
    } else {
      c->tcp_byte_count += r;
      if(c->tcp_byte_count < sizeof(uint16_t))
        return 1;
      sldns_buffer_set_position(buffer, c->tcp_byte_count -
        sizeof(uint16_t));
    }
    if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
      tcp_callback_writer(c);
      return 1;
    }
  }
  log_assert(c->tcp_write_and_read || sldns_buffer_remaining(buffer) > 0);
  log_assert(!c->tcp_write_and_read || c->tcp_write_byte_count < c->tcp_write_pkt_len + 2);
  if(c->tcp_write_and_read) {
    r = send(fd, (void*)(c->tcp_write_pkt + c->tcp_write_byte_count - 2),
      c->tcp_write_pkt_len + 2 - c->tcp_write_byte_count, 0);
  } else {
    r = send(fd, (void*)sldns_buffer_current(buffer),
      sldns_buffer_remaining(buffer), 0);
  }
  if(r == -1) {
#ifndef USE_WINSOCK
    if(errno == EINTR || errno == EAGAIN)
      return 1;
#ifdef ECONNRESET
    if(errno == ECONNRESET && verbosity < 2)
      return 0; /* silence reset by peer */
#endif
#else
    if(WSAGetLastError() == WSAEINPROGRESS)
      return 1;
    if(WSAGetLastError() == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
      return 1;
    }
    if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
      return 0; /* silence reset by peer */
#endif
    log_err_addr("tcp send r", sock_strerror(errno),
      &c->repinfo.remote_addr,
      c->repinfo.remote_addrlen);
    return 0;
  }
  if(c->tcp_write_and_read) {
    c->tcp_write_byte_count += r;
  } else {
    sldns_buffer_skip(buffer, r);
  }

  if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
    tcp_callback_writer(c);
  }

  return 1;
}

/** read again to drain buffers when there could be more to read, returns 0
 * on failure which means the comm point is closed. */
static int
tcp_req_info_read_again(int fd, struct comm_point* c)
{
  while(c->tcp_req_info->read_again) {
    int r;
    c->tcp_req_info->read_again = 0;
    if(c->tcp_is_reading)
      r = comm_point_tcp_handle_read(fd, c, 0);
    else  r = comm_point_tcp_handle_write(fd, c);
    if(!r) {
      reclaim_tcp_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
      return 0;
    }
  }
  return 1;
}

/** read again to drain buffers when there could be more to read */
static void
tcp_more_read_again(int fd, struct comm_point* c)
{
  /* if the packet is done, but another one could be waiting on
   * the connection, the callback signals this, and we try again */
  /* this continues until the read routines get EAGAIN or so,
   * and thus does not call the callback, and the bool is 0 */
  int* moreread = c->tcp_more_read_again;
  while(moreread && *moreread) {
    *moreread = 0;
    if(!comm_point_tcp_handle_read(fd, c, 0)) {
      reclaim_tcp_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
      return;
    }
  }
}

/** write again to fill up when there could be more to write */
static void
tcp_more_write_again(int fd, struct comm_point* c)
{
  /* if the packet is done, but another is waiting to be written,
   * the callback signals it and we try again. */
  /* this continues until the write routines get EAGAIN or so,
   * and thus does not call the callback, and the bool is 0 */
  int* morewrite = c->tcp_more_write_again;
  while(morewrite && *morewrite) {
    *morewrite = 0;
    if(!comm_point_tcp_handle_write(fd, c)) {
      reclaim_tcp_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
      return;
    }
  }
}

void
comm_point_tcp_handle_callback(int fd, short event, void* arg)
{
  struct comm_point* c = (struct comm_point*)arg;
  log_assert(c->type == comm_tcp);
  ub_comm_base_now(c->ev->base);

  if(c->fd == -1 || c->fd != fd)
    return; /* duplicate event, but commpoint closed. */

#ifdef USE_DNSCRYPT
  /* Initialize if this is a dnscrypt socket */
  if(c->tcp_parent) {
    c->dnscrypt = c->tcp_parent->dnscrypt;
  }
  if(c->dnscrypt && c->dnscrypt_buffer == c->buffer) {
    c->dnscrypt_buffer = sldns_buffer_new(sldns_buffer_capacity(c->buffer));
    if(!c->dnscrypt_buffer) {
      log_err("Could not allocate dnscrypt buffer");
      reclaim_tcp_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
      return;
    }
  }
#endif

  if((event&UB_EV_TIMEOUT)) {
    verbose(VERB_QUERY, "tcp took too long, dropped");
    reclaim_tcp_handler(c);
    if(!c->tcp_do_close) {
      fptr_ok(fptr_whitelist_comm_point(c->callback));
      (void)(*c->callback)(c, c->cb_arg,
        NETEVENT_TIMEOUT, NULL);
    }
    return;
  }
  if((event&UB_EV_READ)
#ifdef USE_MSG_FASTOPEN
    && !(c->tcp_do_fastopen && (event&UB_EV_WRITE))
#endif
    ) {
    int has_tcpq = (c->tcp_req_info != NULL);
    int* moreread = c->tcp_more_read_again;
    if(!comm_point_tcp_handle_read(fd, c, 0)) {
      reclaim_tcp_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
      return;
    }
    if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again) {
      if(!tcp_req_info_read_again(fd, c))
        return;
    }
    if(moreread && *moreread)
      tcp_more_read_again(fd, c);
    return;
  }
  if((event&UB_EV_WRITE)) {
    int has_tcpq = (c->tcp_req_info != NULL);
    int* morewrite = c->tcp_more_write_again;
    if(!comm_point_tcp_handle_write(fd, c)) {
      reclaim_tcp_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
      return;
    }
    if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again) {
      if(!tcp_req_info_read_again(fd, c))
        return;
    }
    if(morewrite && *morewrite)
      tcp_more_write_again(fd, c);
    return;
  }
  log_err("Ignored event %d for tcphdl.", event);
}

/** Make http handler free for next assignment */
static void
reclaim_http_handler(struct comm_point* c)
{
  log_assert(c->type == comm_http);
  if(c->ssl) {
#ifdef HAVE_SSL
    SSL_shutdown(c->ssl);
    SSL_free(c->ssl);
    c->ssl = NULL;
#endif
  }
  comm_point_close(c);
  if(c->tcp_parent && !c->is_in_tcp_free) {
    /* Should not happen: bad tcp_free state in reclaim_http. */
    log_assert(c->tcp_free == NULL);
    log_assert(c->tcp_parent->cur_tcp_count > 0);
    c->tcp_parent->cur_tcp_count--;
    c->tcp_free = c->tcp_parent->tcp_free;
    c->tcp_parent->tcp_free = c;
    c->is_in_tcp_free = 1;
    if(!c->tcp_free) {
      /* re-enable listening on accept socket */
      comm_point_start_listening(c->tcp_parent, -1, -1);
    }
  }
}

/** read more data for http (with ssl) */
static int
ssl_http_read_more(struct comm_point* c)
{
#ifdef HAVE_SSL
  int r;
  log_assert(sldns_buffer_remaining(c->buffer) > 0);
  ERR_clear_error();
  r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
    (int)sldns_buffer_remaining(c->buffer));
  if(r <= 0) {
    int want = SSL_get_error(c->ssl, r);
    if(want == SSL_ERROR_ZERO_RETURN) {
      return 0; /* shutdown, closed */
    } else if(want == SSL_ERROR_WANT_READ) {
      return 1; /* read more later */
    } else if(want == SSL_ERROR_WANT_WRITE) {
      c->ssl_shake_state = comm_ssl_shake_hs_write;
      comm_point_listen_for_rw(c, 0, 1);
      return 1;
    } else if(want == SSL_ERROR_SYSCALL) {
#ifdef ECONNRESET
      if(errno == ECONNRESET && verbosity < 2)
        return 0; /* silence reset by peer */
#endif
      if(errno != 0)
        log_err("SSL_read syscall: %s",
          strerror(errno));
      return 0;
    }
    log_crypto_err_io("could not SSL_read", want);
    return 0;
  }
  verbose(VERB_ALGO, "ssl http read more skip to %d + %d",
    (int)sldns_buffer_position(c->buffer), (int)r);
  sldns_buffer_skip(c->buffer, (ssize_t)r);
  return 1;
#else
  (void)c;
  return 0;
#endif /* HAVE_SSL */
}

/** read more data for http */
static int
http_read_more(int fd, struct comm_point* c)
{
  ssize_t r;
  log_assert(sldns_buffer_remaining(c->buffer) > 0);
  r = recv(fd, (void*)sldns_buffer_current(c->buffer),
    sldns_buffer_remaining(c->buffer), MSG_DONTWAIT);
  if(r == 0) {
    return 0;
  } else if(r == -1) {
#ifndef USE_WINSOCK
    if(errno == EINTR || errno == EAGAIN)
      return 1;
#else /* USE_WINSOCK */
    if(WSAGetLastError() == WSAECONNRESET)
      return 0;
    if(WSAGetLastError() == WSAEINPROGRESS)
      return 1;
    if(WSAGetLastError() == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
      return 1;
    }
#endif
    log_err_addr("read (in http r)", sock_strerror(errno),
      &c->repinfo.remote_addr, c->repinfo.remote_addrlen);
    return 0;
  }
  verbose(VERB_ALGO, "http read more skip to %d + %d",
    (int)sldns_buffer_position(c->buffer), (int)r);
  sldns_buffer_skip(c->buffer, r);
  return 1;
}

/** return true if http header has been read (one line complete) */
static int
http_header_done(sldns_buffer* buf)
{
  size_t i;
  for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
    /* there was a \r before the \n, but we ignore that */
    if((char)sldns_buffer_read_u8_at(buf, i) == '\n')
      return 1;
  }
  return 0;
}

/** return character string into buffer for header line, moves buffer
 * past that line and puts zero terminator into linefeed-newline */
static char*
http_header_line(sldns_buffer* buf)
{
  char* result = (char*)sldns_buffer_current(buf);
  size_t i;
  for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
    /* terminate the string on the \r */
    if((char)sldns_buffer_read_u8_at(buf, i) == '\r')
      sldns_buffer_write_u8_at(buf, i, 0);
    /* terminate on the \n and skip past the it and done */
    if((char)sldns_buffer_read_u8_at(buf, i) == '\n') {
      sldns_buffer_write_u8_at(buf, i, 0);
      sldns_buffer_set_position(buf, i+1);
      return result;
    }
  }
  return NULL;
}

/** move unread buffer to start and clear rest for putting the rest into it */
static void
http_moveover_buffer(sldns_buffer* buf)
{
  size_t pos = sldns_buffer_position(buf);
  size_t len = sldns_buffer_remaining(buf);
  sldns_buffer_clear(buf);
  memmove(sldns_buffer_begin(buf), sldns_buffer_at(buf, pos), len);
  sldns_buffer_set_position(buf, len);
}

/** a http header is complete, process it */
static int
http_process_initial_header(struct comm_point* c)
{
  char* line = http_header_line(c->buffer);
  if(!line) return 1;
  verbose(VERB_ALGO, "http header: %s", line);
  if(strncasecmp(line, "HTTP/1.1 ", 9) == 0) {
    /* check returncode */
    if(line[9] != '2') {
      verbose(VERB_ALGO, "http bad status %s", line+9);
      return 0;
    }
  } else if(strncasecmp(line, "Content-Length: ", 16) == 0) {
    if(!c->http_is_chunked)
      c->tcp_byte_count = (size_t)atoi(line+16);
  } else if(strncasecmp(line, "Transfer-Encoding: chunked", 19+7) == 0) {
    c->tcp_byte_count = 0;
    c->http_is_chunked = 1;
  } else if(line[0] == 0) {
    /* end of initial headers */
    c->http_in_headers = 0;
    if(c->http_is_chunked)
      c->http_in_chunk_headers = 1;
    /* remove header text from front of buffer
     * the buffer is going to be used to return the data segment
     * itself and we don't want the header to get returned
     * prepended with it */
    http_moveover_buffer(c->buffer);
    sldns_buffer_flip(c->buffer);
    return 1;
  }
  /* ignore other headers */
  return 1;
}

/** a chunk header is complete, process it, return 0=fail, 1=continue next
 * header line, 2=done with chunked transfer*/
static int
http_process_chunk_header(struct comm_point* c)
{
  char* line = http_header_line(c->buffer);
  if(!line) return 1;
  if(c->http_in_chunk_headers == 3) {
    verbose(VERB_ALGO, "http chunk trailer: %s", line);
    /* are we done ? */
    if(line[0] == 0 && c->tcp_byte_count == 0) {
      /* callback of http reader when NETEVENT_DONE,
       * end of data, with no data in buffer */
      sldns_buffer_set_position(c->buffer, 0);
      sldns_buffer_set_limit(c->buffer, 0);
      fptr_ok(fptr_whitelist_comm_point(c->callback));
      (void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
      /* return that we are done */
      return 2;
    }
    if(line[0] == 0) {
      /* continue with header of the next chunk */
      c->http_in_chunk_headers = 1;
      /* remove header text from front of buffer */
      http_moveover_buffer(c->buffer);
      sldns_buffer_flip(c->buffer);
      return 1;
    }
    /* ignore further trail headers */
    return 1;
  }
  verbose(VERB_ALGO, "http chunk header: %s", line);
  if(c->http_in_chunk_headers == 1) {
    /* read chunked start line */
    char* end = NULL;
    c->tcp_byte_count = (size_t)strtol(line, &end, 16);
    if(end == line)
      return 0;
    c->http_in_chunk_headers = 0;
    /* remove header text from front of buffer */
    http_moveover_buffer(c->buffer);
    sldns_buffer_flip(c->buffer);
    if(c->tcp_byte_count == 0) {
      /* done with chunks, process chunk_trailer lines */
      c->http_in_chunk_headers = 3;
    }
    return 1;
  }
  /* ignore other headers */
  return 1;
}

/** handle nonchunked data segment, 0=fail, 1=wait */
static int
http_nonchunk_segment(struct comm_point* c)
{
  /* c->buffer at position..limit has new data we read in.
   * the buffer itself is full of nonchunked data.
   * we are looking to read tcp_byte_count more data
   * and then the transfer is done. */
  size_t remainbufferlen;
  size_t got_now = sldns_buffer_limit(c->buffer);
  if(c->tcp_byte_count <= got_now) {
    /* done, this is the last data fragment */
    c->http_stored = 0;
    sldns_buffer_set_position(c->buffer, 0);
    fptr_ok(fptr_whitelist_comm_point(c->callback));
    (void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
    return 1;
  }
  /* if we have the buffer space,
   * read more data collected into the buffer */
  remainbufferlen = sldns_buffer_capacity(c->buffer) -
    sldns_buffer_limit(c->buffer);
  if(remainbufferlen+got_now >= c->tcp_byte_count ||
    remainbufferlen >= (size_t)(c->ssl?16384:2048)) {
    size_t total = sldns_buffer_limit(c->buffer);
    sldns_buffer_clear(c->buffer);
    sldns_buffer_set_position(c->buffer, total);
    c->http_stored = total;
    /* return and wait to read more */
    return 1;
  }
  /* call callback with this data amount, then
   * wait for more */
  c->tcp_byte_count -= got_now;
  c->http_stored = 0;
  sldns_buffer_set_position(c->buffer, 0);
  fptr_ok(fptr_whitelist_comm_point(c->callback));
  (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
  /* c->callback has to buffer_clear(c->buffer). */
  /* return and wait to read more */
  return 1;
}

/** handle chunked data segment, return 0=fail, 1=wait, 2=process more */
static int
http_chunked_segment(struct comm_point* c)
{
  /* the c->buffer has from position..limit new data we read. */
  /* the current chunk has length tcp_byte_count.
   * once we read that read more chunk headers.
   */
  size_t remainbufferlen;
  size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
  verbose(VERB_ALGO, "http_chunked_segment: got now %d, tcpbytcount %d, http_stored %d, buffer pos %d, buffer limit %d", (int)got_now, (int)c->tcp_byte_count, (int)c->http_stored, (int)sldns_buffer_position(c->buffer), (int)sldns_buffer_limit(c->buffer));
  if(c->tcp_byte_count <= got_now) {
    /* the chunk has completed (with perhaps some extra data
     * from next chunk header and next chunk) */
    /* save too much info into temp buffer */
    size_t fraglen;
    struct comm_reply repinfo;
    c->http_stored = 0;
    sldns_buffer_skip(c->buffer, (ssize_t)c->tcp_byte_count);
    sldns_buffer_clear(c->http_temp);
    sldns_buffer_write(c->http_temp,
      sldns_buffer_current(c->buffer),
      sldns_buffer_remaining(c->buffer));
    sldns_buffer_flip(c->http_temp);

    /* callback with this fragment */
    fraglen = sldns_buffer_position(c->buffer);
    sldns_buffer_set_position(c->buffer, 0);
    sldns_buffer_set_limit(c->buffer, fraglen);
    repinfo = c->repinfo;
    fptr_ok(fptr_whitelist_comm_point(c->callback));
    (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &repinfo);
    /* c->callback has to buffer_clear(). */

    /* is commpoint deleted? */
    if(!repinfo.c) {
      return 1;
    }
    /* copy waiting info */
    sldns_buffer_clear(c->buffer);
    sldns_buffer_write(c->buffer,
      sldns_buffer_begin(c->http_temp),
      sldns_buffer_remaining(c->http_temp));
    sldns_buffer_flip(c->buffer);
    /* process end of chunk trailer header lines, until
     * an empty line */
    c->http_in_chunk_headers = 3;
    /* process more data in buffer (if any) */
    return 2;
  }
  c->tcp_byte_count -= got_now;

  /* if we have the buffer space,
   * read more data collected into the buffer */
  remainbufferlen = sldns_buffer_capacity(c->buffer) -
    sldns_buffer_limit(c->buffer);
  if(remainbufferlen >= c->tcp_byte_count ||
    remainbufferlen >= 2048) {
    size_t total = sldns_buffer_limit(c->buffer);
    sldns_buffer_clear(c->buffer);
    sldns_buffer_set_position(c->buffer, total);
    c->http_stored = total;
    /* return and wait to read more */
    return 1;
  }

  /* callback of http reader for a new part of the data */
  c->http_stored = 0;
  sldns_buffer_set_position(c->buffer, 0);
  fptr_ok(fptr_whitelist_comm_point(c->callback));
  (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
  /* c->callback has to buffer_clear(c->buffer). */
  /* return and wait to read more */
  return 1;
}

#ifdef HAVE_NGHTTP2
/** Create new http2 session. Called when creating handling comm point. */
static struct http2_session* http2_session_create(struct comm_point* c)
{
  struct http2_session* session = calloc(1, sizeof(*session));
  if(!session) {
    log_err("malloc failure while creating http2 session");
    return NULL;
  }
  session->c = c;

  return session;
}
#endif

/** Delete http2 session. After closing connection or on error */
static void http2_session_delete(struct http2_session* h2_session)
{
#ifdef HAVE_NGHTTP2
  if(h2_session->callbacks)
    nghttp2_session_callbacks_del(h2_session->callbacks);
  free(h2_session);
#else
  (void)h2_session;
#endif
}

#ifdef HAVE_NGHTTP2
struct http2_stream* http2_stream_create(int32_t stream_id)
{
  struct http2_stream* h2_stream = calloc(1, sizeof(*h2_stream));
  if(!h2_stream) {
    log_err("malloc failure while creating http2 stream");
    return NULL;
  }
  h2_stream->stream_id = stream_id;
  return h2_stream;
}
#endif

void http2_stream_add_meshstate(struct http2_stream* h2_stream,
  struct mesh_area* mesh, struct mesh_state* m)
{
  h2_stream->mesh = mesh;
  h2_stream->mesh_state = m;
}

void http2_stream_remove_mesh_state(struct http2_stream* h2_stream)
{
  if(!h2_stream)
    return;
  h2_stream->mesh_state = NULL;
}

#ifdef HAVE_NGHTTP2
void http2_session_add_stream(struct http2_session* h2_session,
  struct http2_stream* h2_stream)
{
  if(h2_session->first_stream)
    h2_session->first_stream->prev = h2_stream;
  h2_stream->next = h2_session->first_stream;
  h2_session->first_stream = h2_stream;
}

/** remove stream from session linked list. After stream close callback or
 * closing connection */
static void http2_session_remove_stream(struct http2_session* h2_session,
  struct http2_stream* h2_stream)
{
  if(h2_stream->prev)
    h2_stream->prev->next = h2_stream->next;
  else
    h2_session->first_stream = h2_stream->next;
  if(h2_stream->next)
    h2_stream->next->prev = h2_stream->prev;

}

int http2_stream_close_cb(nghttp2_session* ATTR_UNUSED(session),
  int32_t stream_id, uint32_t ATTR_UNUSED(error_code), void* cb_arg)
{
  struct http2_stream* h2_stream;
  struct http2_session* h2_session = (struct http2_session*)cb_arg;
  if(!(h2_stream = nghttp2_session_get_stream_user_data(
    h2_session->session, stream_id))) {
    return 0;
  }
  http2_session_remove_stream(h2_session, h2_stream);
  http2_stream_delete(h2_session, h2_stream);
  return 0;
}

ssize_t http2_recv_cb(nghttp2_session* ATTR_UNUSED(session), uint8_t* buf,
  size_t len, int ATTR_UNUSED(flags), void* cb_arg)
{
  struct http2_session* h2_session = (struct http2_session*)cb_arg;
  ssize_t ret;

  log_assert(h2_session->c->type == comm_http);
  log_assert(h2_session->c->h2_session);
  if(++h2_session->reads_count > h2_session->c->http2_max_streams) {
    /* We are somewhat arbitrarily capping the amount of
     * consecutive reads on the HTTP2 session to the number of max
     * allowed streams.
     * When we reach the cap, error out with NGHTTP2_ERR_WOULDBLOCK
     * to signal nghttp2_session_recv() to stop reading for now. */
    h2_session->reads_count = 0;
    return NGHTTP2_ERR_WOULDBLOCK;
  }

#ifdef HAVE_SSL
  if(h2_session->c->ssl) {
    int r;
    ERR_clear_error();
    r = SSL_read(h2_session->c->ssl, buf, len);
    if(r <= 0) {
      int want = SSL_get_error(h2_session->c->ssl, r);
      if(want == SSL_ERROR_ZERO_RETURN) {
        return NGHTTP2_ERR_EOF;
      } else if(want == SSL_ERROR_WANT_READ) {
        return NGHTTP2_ERR_WOULDBLOCK;
      } else if(want == SSL_ERROR_WANT_WRITE) {
        h2_session->c->ssl_shake_state = comm_ssl_shake_hs_write;
        comm_point_listen_for_rw(h2_session->c, 0, 1);
        return NGHTTP2_ERR_WOULDBLOCK;
      } else if(want == SSL_ERROR_SYSCALL) {
#ifdef ECONNRESET
        if(errno == ECONNRESET && verbosity < 2)
          return NGHTTP2_ERR_CALLBACK_FAILURE;
#endif
        if(errno != 0)
          log_err("SSL_read syscall: %s",
            strerror(errno));
        return NGHTTP2_ERR_CALLBACK_FAILURE;
      }
      log_crypto_err_io("could not SSL_read", want);
      return NGHTTP2_ERR_CALLBACK_FAILURE;
    }
    return r;
  }
#endif /* HAVE_SSL */

  ret = recv(h2_session->c->fd, (void*)buf, len, MSG_DONTWAIT);
  if(ret == 0) {
    return NGHTTP2_ERR_EOF;
  } else if(ret < 0) {
#ifndef USE_WINSOCK
    if(errno == EINTR || errno == EAGAIN)
      return NGHTTP2_ERR_WOULDBLOCK;
#ifdef ECONNRESET
    if(errno == ECONNRESET && verbosity < 2)
      return NGHTTP2_ERR_CALLBACK_FAILURE;
#endif
    log_err_addr("could not http2 recv: %s", strerror(errno),
      &h2_session->c->repinfo.remote_addr,
      h2_session->c->repinfo.remote_addrlen);
#else /* USE_WINSOCK */
    if(WSAGetLastError() == WSAECONNRESET)
      return NGHTTP2_ERR_CALLBACK_FAILURE;
    if(WSAGetLastError() == WSAEINPROGRESS)
      return NGHTTP2_ERR_WOULDBLOCK;
    if(WSAGetLastError() == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(h2_session->c->ev->ev,
        UB_EV_READ);
      return NGHTTP2_ERR_WOULDBLOCK;
    }
    log_err_addr("could not http2 recv: %s",
      wsa_strerror(WSAGetLastError()),
      &h2_session->c->repinfo.remote_addr,
      h2_session->c->repinfo.remote_addrlen);
#endif
    return NGHTTP2_ERR_CALLBACK_FAILURE;
  }
  return ret;
}
#endif /* HAVE_NGHTTP2 */

/** Handle http2 read */
static int
comm_point_http2_handle_read(int ATTR_UNUSED(fd), struct comm_point* c)
{
#ifdef HAVE_NGHTTP2
  int ret;
  log_assert(c->h2_session);

  /* reading until recv cb returns NGHTTP2_ERR_WOULDBLOCK */
  ret = nghttp2_session_recv(c->h2_session->session);
  if(ret) {
    if(ret != NGHTTP2_ERR_EOF &&
      ret != NGHTTP2_ERR_CALLBACK_FAILURE) {
      char a[256];
      addr_to_str(&c->repinfo.remote_addr,
        c->repinfo.remote_addrlen, a, sizeof(a));
      verbose(VERB_QUERY, "http2: session_recv from %s failed, "
        "error: %s", a, nghttp2_strerror(ret));
    }
    return 0;
  }
  if(nghttp2_session_want_write(c->h2_session->session)) {
    c->tcp_is_reading = 0;
    comm_point_stop_listening(c);
    comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
  } else if(!nghttp2_session_want_read(c->h2_session->session))
    return 0; /* connection can be closed */
  return 1;
#else
  (void)c;
  return 0;
#endif
}

/**
 * Handle http reading callback.
 * @param fd: file descriptor of socket.
 * @param c: comm point to read from into buffer.
 * @return: 0 on error
 */
static int
comm_point_http_handle_read(int fd, struct comm_point* c)
{
  log_assert(c->type == comm_http);
  log_assert(fd != -1);

  /* if we are in ssl handshake, handle SSL handshake */
#ifdef HAVE_SSL
  if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
    if(!ssl_handshake(c))
      return 0;
    if(c->ssl_shake_state != comm_ssl_shake_none)
      return 1;
  }
#endif /* HAVE_SSL */

  if(!c->tcp_is_reading)
    return 1;

  if(c->use_h2) {
    return comm_point_http2_handle_read(fd, c);
  }

  /* http version is <= http/1.1 */

  if(c->http_min_version >= http_version_2) {
    /* HTTP/2 failed, not allowed to use lower version. */
    return 0;
  }

  /* read more data */
  if(c->ssl) {
    if(!ssl_http_read_more(c))
      return 0;
  } else {
    if(!http_read_more(fd, c))
      return 0;
  }

  if(c->http_stored >= sldns_buffer_position(c->buffer)) {
    /* read did not work but we wanted more data, there is
     * no bytes to process now. */
    return 1;
  }
  sldns_buffer_flip(c->buffer);
  /* if we are partway in a segment of data, position us at the point
   * where we left off previously */
  if(c->http_stored < sldns_buffer_limit(c->buffer))
    sldns_buffer_set_position(c->buffer, c->http_stored);
  else  sldns_buffer_set_position(c->buffer, sldns_buffer_limit(c->buffer));

  while(sldns_buffer_remaining(c->buffer) > 0) {
    /* Handle HTTP/1.x data */
    /* if we are reading headers, read more headers */
    if(c->http_in_headers || c->http_in_chunk_headers) {
      /* if header is done, process the header */
      if(!http_header_done(c->buffer)) {
        /* copy remaining data to front of buffer
         * and set rest for writing into it */
        http_moveover_buffer(c->buffer);
        /* return and wait to read more */
        return 1;
      }
      if(!c->http_in_chunk_headers) {
        /* process initial headers */
        if(!http_process_initial_header(c))
          return 0;
      } else {
        /* process chunk headers */
        int r = http_process_chunk_header(c);
        if(r == 0) return 0;
        if(r == 2) return 1; /* done */
        /* r == 1, continue */
      }
      /* see if we have more to process */
      continue;
    }

    if(!c->http_is_chunked) {
      /* if we are reading nonchunks, process that*/
      return http_nonchunk_segment(c);
    } else {
      /* if we are reading chunks, read the chunk */
      int r = http_chunked_segment(c);
      if(r == 0) return 0;
      if(r == 1) return 1;
      continue;
    }
  }
  /* broke out of the loop; could not process header instead need
   * to read more */
  /* moveover any remaining data and read more data */
  http_moveover_buffer(c->buffer);
  /* return and wait to read more */
  return 1;
}

/** check pending connect for http */
static int
http_check_connect(int fd, struct comm_point* c)
{
  /* check for pending error from nonblocking connect */
  /* from Stevens, unix network programming, vol1, 3rd ed, p450*/
  int error = 0;
  socklen_t len = (socklen_t)sizeof(error);
  if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
    &len) < 0){
#ifndef USE_WINSOCK
    error = errno; /* on solaris errno is error */
#else /* USE_WINSOCK */
    error = WSAGetLastError();
#endif
  }
#ifndef USE_WINSOCK
#if defined(EINPROGRESS) && defined(EWOULDBLOCK)
  if(error == EINPROGRESS || error == EWOULDBLOCK)
    return 1; /* try again later */
  else
#endif
  if(error != 0 && verbosity < 2)
    return 0; /* silence lots of chatter in the logs */
  else if(error != 0) {
    log_err_addr("http connect", strerror(error),
      &c->repinfo.remote_addr, c->repinfo.remote_addrlen);
#else /* USE_WINSOCK */
  /* examine error */
  if(error == WSAEINPROGRESS)
    return 1;
  else if(error == WSAEWOULDBLOCK) {
    ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
    return 1;
  } else if(error != 0 && verbosity < 2)
    return 0;
  else if(error != 0) {
    log_err_addr("http connect", wsa_strerror(error),
      &c->repinfo.remote_addr, c->repinfo.remote_addrlen);
#endif /* USE_WINSOCK */
    return 0;
  }
  /* keep on processing this socket */
  return 2;
}

/** write more data for http (with ssl) */
static int
ssl_http_write_more(struct comm_point* c)
{
#ifdef HAVE_SSL
  int r;
  log_assert(sldns_buffer_remaining(c->buffer) > 0);
  ERR_clear_error();
  r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
    (int)sldns_buffer_remaining(c->buffer));
  if(r <= 0) {
    int want = SSL_get_error(c->ssl, r);
    if(want == SSL_ERROR_ZERO_RETURN) {
      return 0; /* closed */
    } else if(want == SSL_ERROR_WANT_READ) {
      c->ssl_shake_state = comm_ssl_shake_hs_read;
      comm_point_listen_for_rw(c, 1, 0);
      return 1; /* wait for read condition */
    } else if(want == SSL_ERROR_WANT_WRITE) {
      return 1; /* write more later */
    } else if(want == SSL_ERROR_SYSCALL) {
#ifdef EPIPE
      if(errno == EPIPE && verbosity < 2)
        return 0; /* silence 'broken pipe' */
#endif
      if(errno != 0)
        log_err("SSL_write syscall: %s",
          strerror(errno));
      return 0;
    }
    log_crypto_err_io("could not SSL_write", want);
    return 0;
  }
  sldns_buffer_skip(c->buffer, (ssize_t)r);
  return 1;
#else
  (void)c;
  return 0;
#endif /* HAVE_SSL */
}

/** write more data for http */
static int
http_write_more(int fd, struct comm_point* c)
{
  ssize_t r;
  log_assert(sldns_buffer_remaining(c->buffer) > 0);
  r = send(fd, (void*)sldns_buffer_current(c->buffer),
    sldns_buffer_remaining(c->buffer), 0);
  if(r == -1) {
#ifndef USE_WINSOCK
    if(errno == EINTR || errno == EAGAIN)
      return 1;
#else
    if(WSAGetLastError() == WSAEINPROGRESS)
      return 1;
    if(WSAGetLastError() == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
      return 1;
    }
#endif
    log_err_addr("http send r", sock_strerror(errno),
      &c->repinfo.remote_addr, c->repinfo.remote_addrlen);
    return 0;
  }
  sldns_buffer_skip(c->buffer, r);
  return 1;
}

#ifdef HAVE_NGHTTP2
ssize_t http2_send_cb(nghttp2_session* ATTR_UNUSED(session), const uint8_t* buf,
  size_t len, int ATTR_UNUSED(flags), void* cb_arg)
{
  ssize_t ret;
  struct http2_session* h2_session = (struct http2_session*)cb_arg;
  log_assert(h2_session->c->type == comm_http);
  log_assert(h2_session->c->h2_session);

#ifdef HAVE_SSL
  if(h2_session->c->ssl) {
    int r;
    ERR_clear_error();
    r = SSL_write(h2_session->c->ssl, buf, len);
    if(r <= 0) {
      int want = SSL_get_error(h2_session->c->ssl, r);
      if(want == SSL_ERROR_ZERO_RETURN) {
        return NGHTTP2_ERR_CALLBACK_FAILURE;
      } else if(want == SSL_ERROR_WANT_READ) {
        h2_session->c->ssl_shake_state = comm_ssl_shake_hs_read;
        comm_point_listen_for_rw(h2_session->c, 1, 0);
        return NGHTTP2_ERR_WOULDBLOCK;
      } else if(want == SSL_ERROR_WANT_WRITE) {
        return NGHTTP2_ERR_WOULDBLOCK;
      } else if(want == SSL_ERROR_SYSCALL) {
#ifdef EPIPE
        if(errno == EPIPE && verbosity < 2)
          return NGHTTP2_ERR_CALLBACK_FAILURE;
#endif
        if(errno != 0)
          log_err("SSL_write syscall: %s",
            strerror(errno));
        return NGHTTP2_ERR_CALLBACK_FAILURE;
      }
      log_crypto_err_io("could not SSL_write", want);
      return NGHTTP2_ERR_CALLBACK_FAILURE;
    }
    return r;
  }
#endif /* HAVE_SSL */

  ret = send(h2_session->c->fd, (void*)buf, len, 0);
  if(ret == 0) {
    return NGHTTP2_ERR_CALLBACK_FAILURE;
  } else if(ret < 0) {
#ifndef USE_WINSOCK
    if(errno == EINTR || errno == EAGAIN)
      return NGHTTP2_ERR_WOULDBLOCK;
#ifdef EPIPE
    if(errno == EPIPE && verbosity < 2)
      return NGHTTP2_ERR_CALLBACK_FAILURE;
#endif
#ifdef ECONNRESET
    if(errno == ECONNRESET && verbosity < 2)
      return NGHTTP2_ERR_CALLBACK_FAILURE;
#endif
    log_err_addr("could not http2 write: %s", strerror(errno),
      &h2_session->c->repinfo.remote_addr,
      h2_session->c->repinfo.remote_addrlen);
#else /* USE_WINSOCK */
    if(WSAGetLastError() == WSAENOTCONN)
      return NGHTTP2_ERR_WOULDBLOCK;
    if(WSAGetLastError() == WSAEINPROGRESS)
      return NGHTTP2_ERR_WOULDBLOCK;
    if(WSAGetLastError() == WSAEWOULDBLOCK) {
      ub_winsock_tcp_wouldblock(h2_session->c->ev->ev,
        UB_EV_WRITE);
      return NGHTTP2_ERR_WOULDBLOCK;
    }
    if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
      return NGHTTP2_ERR_CALLBACK_FAILURE;
    log_err_addr("could not http2 write: %s",
      wsa_strerror(WSAGetLastError()),
      &h2_session->c->repinfo.remote_addr,
      h2_session->c->repinfo.remote_addrlen);
#endif
    return NGHTTP2_ERR_CALLBACK_FAILURE;
  }
  return ret;
}
#endif /* HAVE_NGHTTP2 */

/** Handle http2 writing */
static int
comm_point_http2_handle_write(int ATTR_UNUSED(fd), struct comm_point* c)
{
#ifdef HAVE_NGHTTP2
  int ret;
  log_assert(c->h2_session);

  ret = nghttp2_session_send(c->h2_session->session);
  if(ret) {
    verbose(VERB_QUERY, "http2: session_send failed, "
      "error: %s", nghttp2_strerror(ret));
    return 0;
  }

  if(nghttp2_session_want_read(c->h2_session->session)) {
    c->tcp_is_reading = 1;
    comm_point_stop_listening(c);
    comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
  } else if(!nghttp2_session_want_write(c->h2_session->session))
    return 0; /* connection can be closed */
  return 1;
#else
  (void)c;
  return 0;
#endif
}

/**
 * Handle http writing callback.
 * @param fd: file descriptor of socket.
 * @param c: comm point to write buffer out of.
 * @return: 0 on error
 */
static int
comm_point_http_handle_write(int fd, struct comm_point* c)
{
  log_assert(c->type == comm_http);
  log_assert(fd != -1);

  /* check pending connect errors, if that fails, we wait for more,
   * or we can continue to write contents */
  if(c->tcp_check_nb_connect) {
    int r = http_check_connect(fd, c);
    if(r == 0) return 0;
    if(r == 1) return 1;
    c->tcp_check_nb_connect = 0;
  }
  /* if we are in ssl handshake, handle SSL handshake */
#ifdef HAVE_SSL
  if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
    if(!ssl_handshake(c))
      return 0;
    if(c->ssl_shake_state != comm_ssl_shake_none)
      return 1;
  }
#endif /* HAVE_SSL */
  if(c->tcp_is_reading)
    return 1;

  if(c->use_h2) {
    return comm_point_http2_handle_write(fd, c);
  }

  /* http version is <= http/1.1 */

  if(c->http_min_version >= http_version_2) {
    /* HTTP/2 failed, not allowed to use lower version. */
    return 0;
  }

  /* if we are writing, write more */
  if(c->ssl) {
    if(!ssl_http_write_more(c))
      return 0;
  } else {
    if(!http_write_more(fd, c))
      return 0;
  }

  /* we write a single buffer contents, that can contain
   * the http request, and then flip to read the results */
  /* see if write is done */
  if(sldns_buffer_remaining(c->buffer) == 0) {
    sldns_buffer_clear(c->buffer);
    if(c->tcp_do_toggle_rw)
      c->tcp_is_reading = 1;
    c->tcp_byte_count = 0;
    /* switch from listening(write) to listening(read) */
    comm_point_stop_listening(c);
    comm_point_start_listening(c, -1, -1);
  }
  return 1;
}

void
comm_point_http_handle_callback(int fd, short event, void* arg)
{
  struct comm_point* c = (struct comm_point*)arg;
  log_assert(c->type == comm_http);
  ub_comm_base_now(c->ev->base);

  if((event&UB_EV_TIMEOUT)) {
    verbose(VERB_QUERY, "http took too long, dropped");
    reclaim_http_handler(c);
    if(!c->tcp_do_close) {
      fptr_ok(fptr_whitelist_comm_point(c->callback));
      (void)(*c->callback)(c, c->cb_arg,
        NETEVENT_TIMEOUT, NULL);
    }
    return;
  }
  if((event&UB_EV_READ)) {
    if(!comm_point_http_handle_read(fd, c)) {
      reclaim_http_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
    }
    return;
  }
  if((event&UB_EV_WRITE)) {
    if(!comm_point_http_handle_write(fd, c)) {
      reclaim_http_handler(c);
      if(!c->tcp_do_close) {
        fptr_ok(fptr_whitelist_comm_point(
          c->callback));
        (void)(*c->callback)(c, c->cb_arg,
          NETEVENT_CLOSED, NULL);
      }
    }
    return;
  }
  log_err("Ignored event %d for httphdl.", event);
}

void comm_point_local_handle_callback(int fd, short event, void* arg)
{
  struct comm_point* c = (struct comm_point*)arg;
  log_assert(c->type == comm_local);
  ub_comm_base_now(c->ev->base);

  if((event&UB_EV_READ)) {
    if(!comm_point_tcp_handle_read(fd, c, 1)) {
      fptr_ok(fptr_whitelist_comm_point(c->callback));
      (void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED,
        NULL);
    }
    return;
  }
  log_err("Ignored event %d for localhdl.", event);
}

void comm_point_raw_handle_callback(int ATTR_UNUSED(fd),
  short event, void* arg)
{
  struct comm_point* c = (struct comm_point*)arg;
  int err = NETEVENT_NOERROR;
  log_assert(c->type == comm_raw);
  ub_comm_base_now(c->ev->base);

  if((event&UB_EV_TIMEOUT))
    err = NETEVENT_TIMEOUT;
  fptr_ok(fptr_whitelist_comm_point_raw(c->callback));
  (void)(*c->callback)(c, c->cb_arg, err, NULL);
}

struct comm_point*
comm_point_create_udp(struct comm_base *base, int fd, sldns_buffer* buffer,
  int pp2_enabled, comm_point_callback_type* callback,
  void* callback_arg, struct unbound_socket* socket)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = fd;
  c->buffer = buffer;
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_udp;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 0;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = buffer;
#endif
  c->inuse = 0;
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->socket = socket;
  c->pp2_enabled = pp2_enabled;
  c->pp2_header_state = pp2_header_none;
  evbits = UB_EV_READ | UB_EV_PERSIST;
  /* ub_event stuff */
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_udp_callback, c);
  if(c->ev->ev == NULL) {
    log_err("could not baseset udp event");
    comm_point_delete(c);
    return NULL;
  }
  if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
    log_err("could not add udp event");
    comm_point_delete(c);
    return NULL;
  }
  c->event_added = 1;
  return c;
}

#if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
struct comm_point*
comm_point_create_udp_ancil(struct comm_base *base, int fd,
  sldns_buffer* buffer, int pp2_enabled,
  comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = fd;
  c->buffer = buffer;
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_udp;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = buffer;
#endif
  c->inuse = 0;
  c->tcp_do_toggle_rw = 0;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->socket = socket;
  c->pp2_enabled = pp2_enabled;
  c->pp2_header_state = pp2_header_none;
  evbits = UB_EV_READ | UB_EV_PERSIST;
  /* ub_event stuff */
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_udp_ancil_callback, c);
  if(c->ev->ev == NULL) {
    log_err("could not baseset udp event");
    comm_point_delete(c);
    return NULL;
  }
  if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
    log_err("could not add udp event");
    comm_point_delete(c);
    return NULL;
  }
  c->event_added = 1;
  return c;
}
#endif

struct comm_point*
comm_point_create_doq(struct comm_base *base, int fd, sldns_buffer* buffer,
  comm_point_callback_type* callback, void* callback_arg,
  struct unbound_socket* socket, struct doq_table* table,
  struct ub_randstate* rnd, const void* quic_sslctx,
  struct config_file* cfg)
{
#ifdef HAVE_NGTCP2
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = fd;
  c->buffer = buffer;
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_doq;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 0;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = NULL;
#endif
  c->doq_socket = doq_server_socket_create(table, rnd, quic_sslctx, c,
    base, cfg);
  if(!c->doq_socket) {
    log_err("could not create doq comm_point");
    comm_point_delete(c);
    return NULL;
  }
  c->inuse = 0;
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->socket = socket;
  c->pp2_enabled = 0;
  c->pp2_header_state = pp2_header_none;
  evbits = UB_EV_READ | UB_EV_PERSIST;
  /* ub_event stuff */
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_doq_callback, c);
  if(c->ev->ev == NULL) {
    log_err("could not baseset udp event");
    comm_point_delete(c);
    return NULL;
  }
  if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
    log_err("could not add udp event");
    comm_point_delete(c);
    return NULL;
  }
  c->event_added = 1;
  return c;
#else
  /* no libngtcp2, so no QUIC support */
  (void)base;
  (void)buffer;
  (void)callback;
  (void)callback_arg;
  (void)socket;
  (void)rnd;
  (void)table;
  (void)quic_sslctx;
  (void)cfg;
  sock_close(fd);
  return NULL;
#endif /* HAVE_NGTCP2 */
}

static struct comm_point*
comm_point_create_tcp_handler(struct comm_base *base,
  struct comm_point* parent, size_t bufsize,
  struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
  void* callback_arg, struct unbound_socket* socket)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = -1;
  c->buffer = sldns_buffer_new(bufsize);
  if(!c->buffer) {
    free(c->ev);
    free(c);
    return NULL;
  }
  c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
  if(!c->timeout) {
    sldns_buffer_free(c->buffer);
    free(c->ev);
    free(c);
    return NULL;
  }
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = parent;
  c->tcp_timeout_msec = parent->tcp_timeout_msec;
  c->tcp_conn_limit = parent->tcp_conn_limit;
  c->tcl_addr = NULL;
  c->tcp_keepalive = 0;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_tcp;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 1;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  /* We don't know just yet if this is a dnscrypt channel. Allocation
   * will be done when handling the callback. */
  c->dnscrypt_buffer = c->buffer;
#endif
  c->repinfo.c = c;
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->socket = socket;
  c->pp2_enabled = parent->pp2_enabled;
  c->pp2_header_state = pp2_header_none;
  if(spoolbuf) {
    c->tcp_req_info = tcp_req_info_create(spoolbuf);
    if(!c->tcp_req_info) {
      log_err("could not create tcp commpoint");
      sldns_buffer_free(c->buffer);
      free(c->timeout);
      free(c->ev);
      free(c);
      return NULL;
    }
    c->tcp_req_info->cp = c;
    c->tcp_do_close = 1;
    c->tcp_do_toggle_rw = 0;
  }
  /* add to parent free list */
  c->tcp_free = parent->tcp_free;
  parent->tcp_free = c;
  c->is_in_tcp_free = 1;
  /* ub_event stuff */
  evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_tcp_handle_callback, c);
  if(c->ev->ev == NULL)
  {
    log_err("could not basetset tcphdl event");
    parent->tcp_free = c->tcp_free;
    tcp_req_info_delete(c->tcp_req_info);
    sldns_buffer_free(c->buffer);
    free(c->timeout);
    free(c->ev);
    free(c);
    return NULL;
  }
  return c;
}

static struct comm_point*
comm_point_create_http_handler(struct comm_base *base,
  struct comm_point* parent, size_t bufsize, int harden_large_queries,
  uint32_t http_max_streams, char* http_endpoint,
  comm_point_callback_type* callback, void* callback_arg,
  struct unbound_socket* socket)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = -1;
  c->buffer = sldns_buffer_new(bufsize);
  if(!c->buffer) {
    free(c->ev);
    free(c);
    return NULL;
  }
  c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
  if(!c->timeout) {
    sldns_buffer_free(c->buffer);
    free(c->ev);
    free(c);
    return NULL;
  }
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = parent;
  c->tcp_timeout_msec = parent->tcp_timeout_msec;
  c->tcp_conn_limit = parent->tcp_conn_limit;
  c->tcl_addr = NULL;
  c->tcp_keepalive = 0;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_http;
  c->tcp_do_close = 1;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 1; /* will be set to 0 after http2 upgrade */
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = NULL;
#endif
  c->repinfo.c = c;
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->socket = socket;
  c->pp2_enabled = 0;
  c->pp2_header_state = pp2_header_none;

  c->http_min_version = http_version_2;
  c->http2_stream_max_qbuffer_size = bufsize;
  if(harden_large_queries && bufsize > 512)
    c->http2_stream_max_qbuffer_size = 512;
  c->http2_max_streams = http_max_streams;
  if(!(c->http_endpoint = strdup(http_endpoint))) {
    log_err("could not strdup http_endpoint");
    sldns_buffer_free(c->buffer);
    free(c->timeout);
    free(c->ev);
    free(c);
    return NULL;
  }
  c->use_h2 = 0;
#ifdef HAVE_NGHTTP2
  if(!(c->h2_session = http2_session_create(c))) {
    log_err("could not create http2 session");
    free(c->http_endpoint);
    sldns_buffer_free(c->buffer);
    free(c->timeout);
    free(c->ev);
    free(c);
    return NULL;
  }
  if(!(c->h2_session->callbacks = http2_req_callbacks_create())) {
    log_err("could not create http2 callbacks");
    http2_session_delete(c->h2_session);
    free(c->http_endpoint);
    sldns_buffer_free(c->buffer);
    free(c->timeout);
    free(c->ev);
    free(c);
    return NULL;
  }
#endif

  /* add to parent free list */
  c->tcp_free = parent->tcp_free;
  parent->tcp_free = c;
  c->is_in_tcp_free = 1;
  /* ub_event stuff */
  evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_http_handle_callback, c);
  if(c->ev->ev == NULL)
  {
    log_err("could not set http handler event");
    parent->tcp_free = c->tcp_free;
    http2_session_delete(c->h2_session);
    sldns_buffer_free(c->buffer);
    free(c->timeout);
    free(c->ev);
    free(c);
    return NULL;
  }
  return c;
}

struct comm_point*
comm_point_create_tcp(struct comm_base *base, int fd, int num,
  int idle_timeout, int harden_large_queries,
  uint32_t http_max_streams, char* http_endpoint,
  struct tcl_list* tcp_conn_limit, size_t bufsize,
  struct sldns_buffer* spoolbuf, enum listen_type port_type,
  int pp2_enabled, comm_point_callback_type* callback,
  void* callback_arg, struct unbound_socket* socket)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  int i;
  /* first allocate the TCP accept listener */
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = fd;
  c->buffer = NULL;
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_timeout_msec = idle_timeout;
  c->tcp_conn_limit = tcp_conn_limit;
  c->tcl_addr = NULL;
  c->tcp_keepalive = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = num;
  c->cur_tcp_count = 0;
  c->tcp_handlers = (struct comm_point**)calloc((size_t)num,
    sizeof(struct comm_point*));
  if(!c->tcp_handlers) {
    free(c->ev);
    free(c);
    return NULL;
  }
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_tcp_accept;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 0;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = NULL;
#endif
  c->callback = NULL;
  c->cb_arg = NULL;
  c->socket = socket;
  c->pp2_enabled = (port_type==listen_type_http?0:pp2_enabled);
  c->pp2_header_state = pp2_header_none;
  evbits = UB_EV_READ | UB_EV_PERSIST;
  /* ub_event stuff */
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_tcp_accept_callback, c);
  if(c->ev->ev == NULL) {
    log_err("could not baseset tcpacc event");
    comm_point_delete(c);
    return NULL;
  }
  if (ub_event_add(c->ev->ev, c->timeout) != 0) {
    log_err("could not add tcpacc event");
    comm_point_delete(c);
    return NULL;
  }
  c->event_added = 1;
  /* now prealloc the handlers */
  for(i=0; i<num; i++) {
    if(port_type == listen_type_tcp ||
      port_type == listen_type_ssl ||
      port_type == listen_type_tcp_dnscrypt) {
      c->tcp_handlers[i] = comm_point_create_tcp_handler(base,
        c, bufsize, spoolbuf, callback, callback_arg, socket);
    } else if(port_type == listen_type_http) {
      c->tcp_handlers[i] = comm_point_create_http_handler(
        base, c, bufsize, harden_large_queries,
        http_max_streams, http_endpoint,
        callback, callback_arg, socket);
    }
    else {
      log_err("could not create tcp handler, unknown listen "
        "type");
      return NULL;
    }
    if(!c->tcp_handlers[i]) {
      comm_point_delete(c);
      return NULL;
    }
  }

  return c;
}

struct comm_point*
comm_point_create_tcp_out(struct comm_base *base, size_t bufsize,
        comm_point_callback_type* callback, void* callback_arg)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = -1;
  c->buffer = sldns_buffer_new(bufsize);
  if(!c->buffer) {
    free(c->ev);
    free(c);
    return NULL;
  }
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_timeout_msec = TCP_QUERY_TIMEOUT;
  c->tcp_conn_limit = NULL;
  c->tcl_addr = NULL;
  c->tcp_keepalive = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_tcp;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 1;
  c->tcp_check_nb_connect = 1;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 1;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = c->buffer;
#endif
  c->repinfo.c = c;
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->pp2_enabled = 0;
  c->pp2_header_state = pp2_header_none;
  evbits = UB_EV_PERSIST | UB_EV_WRITE;
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_tcp_handle_callback, c);
  if(c->ev->ev == NULL)
  {
    log_err("could not baseset tcpout event");
    sldns_buffer_free(c->buffer);
    free(c->ev);
    free(c);
    return NULL;
  }

  return c;
}

struct comm_point*
comm_point_create_http_out(struct comm_base *base, size_t bufsize,
        comm_point_callback_type* callback, void* callback_arg,
  sldns_buffer* temp)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = -1;
  c->buffer = sldns_buffer_new(bufsize);
  if(!c->buffer) {
    free(c->ev);
    free(c);
    return NULL;
  }
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_http;
  c->tcp_do_close = 0;
  c->do_not_close = 0;
  c->tcp_do_toggle_rw = 1;
  c->tcp_check_nb_connect = 1;
  c->http_in_headers = 1;
  c->http_in_chunk_headers = 0;
  c->http_is_chunked = 0;
  c->http_temp = temp;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 1;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = c->buffer;
#endif
  c->repinfo.c = c;
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->pp2_enabled = 0;
  c->pp2_header_state = pp2_header_none;
  evbits = UB_EV_PERSIST | UB_EV_WRITE;
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_http_handle_callback, c);
  if(c->ev->ev == NULL)
  {
    log_err("could not baseset tcpout event");
#ifdef HAVE_SSL
    SSL_free(c->ssl);
#endif
    sldns_buffer_free(c->buffer);
    free(c->ev);
    free(c);
    return NULL;
  }

  return c;
}

struct comm_point*
comm_point_create_local(struct comm_base *base, int fd, size_t bufsize,
        comm_point_callback_type* callback, void* callback_arg)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = fd;
  c->buffer = sldns_buffer_new(bufsize);
  if(!c->buffer) {
    free(c->ev);
    free(c);
    return NULL;
  }
  c->timeout = NULL;
  c->tcp_is_reading = 1;
  c->tcp_byte_count = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_local;
  c->tcp_do_close = 0;
  c->do_not_close = 1;
  c->tcp_do_toggle_rw = 0;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = c->buffer;
#endif
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->pp2_enabled = 0;
  c->pp2_header_state = pp2_header_none;
  /* ub_event stuff */
  evbits = UB_EV_PERSIST | UB_EV_READ;
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_local_handle_callback, c);
  if(c->ev->ev == NULL) {
    log_err("could not baseset localhdl event");
    free(c->ev);
    free(c);
    return NULL;
  }
  if (ub_event_add(c->ev->ev, c->timeout) != 0) {
    log_err("could not add localhdl event");
    ub_event_free(c->ev->ev);
    free(c->ev);
    free(c);
    return NULL;
  }
  c->event_added = 1;
  return c;
}

struct comm_point*
comm_point_create_raw(struct comm_base* base, int fd, int writing,
  comm_point_callback_type* callback, void* callback_arg)
{
  struct comm_point* c = (struct comm_point*)calloc(1,
    sizeof(struct comm_point));
  short evbits;
  if(!c)
    return NULL;
  c->ev = (struct internal_event*)calloc(1,
    sizeof(struct internal_event));
  if(!c->ev) {
    free(c);
    return NULL;
  }
  c->ev->base = base;
  c->fd = fd;
  c->buffer = NULL;
  c->timeout = NULL;
  c->tcp_is_reading = 0;
  c->tcp_byte_count = 0;
  c->tcp_parent = NULL;
  c->max_tcp_count = 0;
  c->cur_tcp_count = 0;
  c->tcp_handlers = NULL;
  c->tcp_free = NULL;
  c->is_in_tcp_free = 0;
  c->type = comm_raw;
  c->tcp_do_close = 0;
  c->do_not_close = 1;
  c->tcp_do_toggle_rw = 0;
  c->tcp_check_nb_connect = 0;
#ifdef USE_MSG_FASTOPEN
  c->tcp_do_fastopen = 0;
#endif
#ifdef USE_DNSCRYPT
  c->dnscrypt = 0;
  c->dnscrypt_buffer = c->buffer;
#endif
  c->callback = callback;
  c->cb_arg = callback_arg;
  c->pp2_enabled = 0;
  c->pp2_header_state = pp2_header_none;
  /* ub_event stuff */
  if(writing)
    evbits = UB_EV_PERSIST | UB_EV_WRITE;
  else  evbits = UB_EV_PERSIST | UB_EV_READ;
  c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
    comm_point_raw_handle_callback, c);
  if(c->ev->ev == NULL) {
    log_err("could not baseset rawhdl event");
    free(c->ev);
    free(c);
    return NULL;
  }
  if (ub_event_add(c->ev->ev, c->timeout) != 0) {
    log_err("could not add rawhdl event");
    ub_event_free(c->ev->ev);
    free(c->ev);
    free(c);
    return NULL;
  }
  c->event_added = 1;
  return c;
}

void
comm_point_close(struct comm_point* c)
{
  if(!c)
    return;
  if(c->fd != -1) {
    verbose(5, "comm_point_close of %d: event_del", c->fd);
    if(c->event_added) {
      if(ub_event_del(c->ev->ev) != 0) {
        log_err("could not event_del on close");
      }
      c->event_added = 0;
    }
  }
  tcl_close_connection(c->tcl_addr);
  if(c->tcp_req_info)
    tcp_req_info_clear(c->tcp_req_info);
  if(c->h2_session)
    http2_session_server_delete(c->h2_session);
  /* stop the comm point from reading or writing after it is closed. */
  if(c->tcp_more_read_again && *c->tcp_more_read_again)
    *c->tcp_more_read_again = 0;
  if(c->tcp_more_write_again && *c->tcp_more_write_again)
    *c->tcp_more_write_again = 0;

  /* close fd after removing from event lists, or epoll.. is messed up */
  if(c->fd != -1 && !c->do_not_close) {
#ifdef USE_WINSOCK
    if(c->type == comm_tcp || c->type == comm_http) {
      /* delete sticky events for the fd, it gets closed */
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
      ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
    }
#endif
    verbose(VERB_ALGO, "close fd %d", c->fd);
    sock_close(c->fd);
  }
  c->fd = -1;
}

void
comm_point_delete(struct comm_point* c)
{
  if(!c)
    return;
  if((c->type == comm_tcp || c->type == comm_http) && c->ssl) {
#ifdef HAVE_SSL
    SSL_shutdown(c->ssl);
    SSL_free(c->ssl);
#endif
  }
  if(c->type == comm_http && c->http_endpoint) {
    free(c->http_endpoint);
    c->http_endpoint = NULL;
  }
  comm_point_close(c);
  if(c->tcp_handlers) {
    int i;
    for(i=0; i<c->max_tcp_count; i++)
      comm_point_delete(c->tcp_handlers[i]);
    free(c->tcp_handlers);
  }
  free(c->timeout);
  if(c->type == comm_tcp || c->type == comm_local || c->type == comm_http) {
    sldns_buffer_free(c->buffer);
#ifdef USE_DNSCRYPT
    if(c->dnscrypt && c->dnscrypt_buffer != c->buffer) {
      sldns_buffer_free(c->dnscrypt_buffer);
    }
#endif
    if(c->tcp_req_info) {
      tcp_req_info_delete(c->tcp_req_info);
    }
    if(c->h2_session) {
      http2_session_delete(c->h2_session);
    }
  }
#ifdef HAVE_NGTCP2
  if(c->doq_socket)
    doq_server_socket_delete(c->doq_socket);
#endif
  ub_event_free(c->ev->ev);
  free(c->ev);
  free(c);
}

#ifdef USE_DNSTAP
static void
send_reply_dnstap(struct dt_env* dtenv,
  struct sockaddr* addr, socklen_t addrlen,
  struct sockaddr_storage* client_addr, socklen_t client_addrlen,
  enum comm_point_type type, void* ssl, sldns_buffer* buffer)
{
  log_addr(VERB_ALGO, "from local addr", (void*)addr, addrlen);
  log_addr(VERB_ALGO, "response to client", client_addr, client_addrlen);
  dt_msg_send_client_response(dtenv, client_addr,
    (struct sockaddr_storage*)addr, type, ssl, buffer);
}
#endif

void
comm_point_send_reply(struct comm_reply *repinfo)
{
  struct sldns_buffer* buffer;
  log_assert(repinfo && repinfo->c);
#ifdef USE_DNSCRYPT
  buffer = repinfo->c->dnscrypt_buffer;
  if(!dnsc_handle_uncurved_request(repinfo)) {
    return;
  }
#else
  buffer = repinfo->c->buffer;
#endif
  if(repinfo->c->type == comm_udp) {
    if(repinfo->srctype)
      comm_point_send_udp_msg_if(repinfo->c, buffer,
        (struct sockaddr*)&repinfo->remote_addr,
        repinfo->remote_addrlen, repinfo);
    else
      comm_point_send_udp_msg(repinfo->c, buffer,
        (struct sockaddr*)&repinfo->remote_addr,
        repinfo->remote_addrlen, 0);
#ifdef USE_DNSTAP
    /*
     * sending src (client)/dst (local service) addresses over
     * DNSTAP from udp callback
     */
    if(repinfo->c->dtenv != NULL && repinfo->c->dtenv->log_client_response_messages) {
      send_reply_dnstap(repinfo->c->dtenv,
        repinfo->c->socket->addr,
        repinfo->c->socket->addrlen,
        &repinfo->client_addr, repinfo->client_addrlen,
        repinfo->c->type, repinfo->c->ssl,
        repinfo->c->buffer);
    }
#endif
  } else {
#ifdef USE_DNSTAP
    struct dt_env* dtenv =
#ifdef HAVE_NGTCP2
      repinfo->c->doq_socket
      ?repinfo->c->dtenv:
#endif
      repinfo->c->tcp_parent->dtenv;
    struct sldns_buffer* dtbuffer = repinfo->c->tcp_req_info
      ?repinfo->c->tcp_req_info->spool_buffer
      :repinfo->c->buffer;
#ifdef USE_DNSCRYPT
    if(repinfo->c->dnscrypt && repinfo->is_dnscrypted)
      dtbuffer = repinfo->c->buffer;
#endif
    /*
     * sending src (client)/dst (local service) addresses over
     * DNSTAP from other callbacks
     */
    if(dtenv != NULL && dtenv->log_client_response_messages) {
      send_reply_dnstap(dtenv,
        repinfo->c->socket->addr,
        repinfo->c->socket->addrlen,
        &repinfo->client_addr, repinfo->client_addrlen,
        repinfo->c->type, repinfo->c->ssl,
        dtbuffer);
    }
#endif
    if(repinfo->c->tcp_req_info) {
      tcp_req_info_send_reply(repinfo->c->tcp_req_info);
    } else if(repinfo->c->use_h2) {
      if(!http2_submit_dns_response(repinfo->c->h2_session)) {
        comm_point_drop_reply(repinfo);
        return;
      }
      repinfo->c->h2_stream = NULL;
      repinfo->c->tcp_is_reading = 0;
      comm_point_stop_listening(repinfo->c);
      comm_point_start_listening(repinfo->c, -1,
        adjusted_tcp_timeout(repinfo->c));
      return;
#ifdef HAVE_NGTCP2
    } else if(repinfo->c->doq_socket) {
      doq_socket_send_reply(repinfo);
#endif
    } else {
      comm_point_start_listening(repinfo->c, -1,
        adjusted_tcp_timeout(repinfo->c));
    }
  }
}

void
comm_point_drop_reply(struct comm_reply* repinfo)
{
  if(!repinfo)
    return;
  log_assert(repinfo->c);
  log_assert(repinfo->c->type != comm_tcp_accept);
  if(repinfo->c->type == comm_udp)
    return;
  if(repinfo->c->tcp_req_info)
    repinfo->c->tcp_req_info->is_drop = 1;
  if(repinfo->c->type == comm_http) {
    if(repinfo->c->h2_session) {
      repinfo->c->h2_session->is_drop = 1;
      if(!repinfo->c->h2_session->postpone_drop)
        reclaim_http_handler(repinfo->c);
      return;
    }
    reclaim_http_handler(repinfo->c);
    return;
#ifdef HAVE_NGTCP2
  } else if(repinfo->c->doq_socket) {
    doq_socket_drop_reply(repinfo);
    return;
#endif
  }
  reclaim_tcp_handler(repinfo->c);
}

void
comm_point_stop_listening(struct comm_point* c)
{
  verbose(VERB_ALGO, "comm point stop listening %d", c->fd);
  if(c->event_added) {
    if(ub_event_del(c->ev->ev) != 0) {
      log_err("event_del error to stoplisten");
    }
    c->event_added = 0;
  }
}

void
comm_point_start_listening(struct comm_point* c, int newfd, int msec)
{
  verbose(VERB_ALGO, "comm point start listening %d (%d msec)",
    c->fd==-1?newfd:c->fd, msec);
  if(c->type == comm_tcp_accept && !c->tcp_free) {
    /* no use to start listening no free slots. */
    return;
  }
  if(c->event_added) {
    if(ub_event_del(c->ev->ev) != 0) {
      log_err("event_del error to startlisten");
    }
    c->event_added = 0;
  }
  if(msec != -1 && msec != 0) {
    if(!c->timeout) {
      c->timeout = (struct timeval*)malloc(sizeof(
        struct timeval));
      if(!c->timeout) {
        log_err("cpsl: malloc failed. No net read.");
        return;
      }
    }
    ub_event_add_bits(c->ev->ev, UB_EV_TIMEOUT);
#ifndef S_SPLINT_S /* splint fails on struct timeval. */
    c->timeout->tv_sec = msec/1000;
    c->timeout->tv_usec = (msec%1000)*1000;
#endif /* S_SPLINT_S */
  } else {
    if(msec == 0 || !c->timeout) {
      ub_event_del_bits(c->ev->ev, UB_EV_TIMEOUT);
    }
  }
  if(c->type == comm_tcp || c->type == comm_http) {
    ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
    if(c->tcp_write_and_read) {
      verbose(5, "startlistening %d mode rw", (newfd==-1?c->fd:newfd));
      ub_event_add_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
    } else if(c->tcp_is_reading) {
      verbose(5, "startlistening %d mode r", (newfd==-1?c->fd:newfd));
      ub_event_add_bits(c->ev->ev, UB_EV_READ);
    } else  {
      verbose(5, "startlistening %d mode w", (newfd==-1?c->fd:newfd));
      ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
    }
  }
  if(newfd != -1) {
    if(c->fd != -1 && c->fd != newfd) {
      verbose(5, "cpsl close of fd %d for %d", c->fd, newfd);
      sock_close(c->fd);
    }
    c->fd = newfd;
    ub_event_set_fd(c->ev->ev, c->fd);
  }
  if(ub_event_add(c->ev->ev, msec==0?NULL:c->timeout) != 0) {
    log_err("event_add failed. in cpsl.");
    return;
  }
  c->event_added = 1;
}

void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr)
{
  verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr);
  if(c->event_added) {
    if(ub_event_del(c->ev->ev) != 0) {
      log_err("event_del error to cplf");
    }
    c->event_added = 0;
  }
  if(!c->timeout) {
    ub_event_del_bits(c->ev->ev, UB_EV_TIMEOUT);
  }
  ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
  if(rd) ub_event_add_bits(c->ev->ev, UB_EV_READ);
  if(wr) ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
  if(ub_event_add(c->ev->ev, c->timeout) != 0) {
    log_err("event_add failed. in cplf.");
    return;
  }
  c->event_added = 1;
}

size_t comm_point_get_mem(struct comm_point* c)
{
  size_t s;
  if(!c)
    return 0;
  s = sizeof(*c) + sizeof(*c->ev);
  if(c->timeout)
    s += sizeof(*c->timeout);
  if(c->type == comm_tcp || c->type == comm_local) {
    s += sizeof(*c->buffer) + sldns_buffer_capacity(c->buffer);
#ifdef USE_DNSCRYPT
    s += sizeof(*c->dnscrypt_buffer);
    if(c->buffer != c->dnscrypt_buffer) {
      s += sldns_buffer_capacity(c->dnscrypt_buffer);
    }
#endif
  }
  if(c->type == comm_tcp_accept) {
    int i;
    for(i=0; i<c->max_tcp_count; i++)
      s += comm_point_get_mem(c->tcp_handlers[i]);
  }
  return s;
}

struct comm_timer*
comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg)
{
  struct internal_timer *tm = (struct internal_timer*)calloc(1,
    sizeof(struct internal_timer));
  if(!tm) {
    log_err("malloc failed");
    return NULL;
  }
  tm->super.ev_timer = tm;
  tm->base = base;
  tm->super.callback = cb;
  tm->super.cb_arg = cb_arg;
  tm->ev = ub_event_new(base->eb->base, -1, UB_EV_TIMEOUT,
    comm_timer_callback, &tm->super);
  if(tm->ev == NULL) {
    log_err("timer_create: event_base_set failed.");
    free(tm);
    return NULL;
  }
  return &tm->super;
}

void
comm_timer_disable(struct comm_timer* timer)
{
  if(!timer)
    return;
  ub_timer_del(timer->ev_timer->ev);
  timer->ev_timer->enabled = 0;
}

void
comm_timer_set(struct comm_timer* timer, struct timeval* tv)
{
  log_assert(tv);
  if(timer->ev_timer->enabled)
    comm_timer_disable(timer);
  if(ub_timer_add(timer->ev_timer->ev, timer->ev_timer->base->eb->base,
    comm_timer_callback, timer, tv) != 0)
    log_err("comm_timer_set: evtimer_add failed.");
  timer->ev_timer->enabled = 1;
}

void
comm_timer_delete(struct comm_timer* timer)
{
  if(!timer)
    return;
  comm_timer_disable(timer);
  /* Free the sub struct timer->ev_timer derived from the super struct timer.
   * i.e. assert(timer == timer->ev_timer)
   */
  ub_event_free(timer->ev_timer->ev);
  free(timer->ev_timer);
}

void
comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg)
{
  struct comm_timer* tm = (struct comm_timer*)arg;
  if(!(event&UB_EV_TIMEOUT))
    return;
  ub_comm_base_now(tm->ev_timer->base);
  tm->ev_timer->enabled = 0;
  fptr_ok(fptr_whitelist_comm_timer(tm->callback));
  (*tm->callback)(tm->cb_arg);
}

int
comm_timer_is_set(struct comm_timer* timer)
{
  return (int)timer->ev_timer->enabled;
}

size_t
comm_timer_get_mem(struct comm_timer* timer)
{
  if(!timer) return 0;
  return sizeof(struct internal_timer);
}

struct comm_signal*
comm_signal_create(struct comm_base* base,
        void (*callback)(int, void*), void* cb_arg)
{
  struct comm_signal* com = (struct comm_signal*)malloc(
    sizeof(struct comm_signal));
  if(!com) {
    log_err("malloc failed");
    return NULL;
  }
  com->base = base;
  com->callback = callback;
  com->cb_arg = cb_arg;
  com->ev_signal = NULL;
  return com;
}

void
comm_signal_callback(int sig, short event, void* arg)
{
  struct comm_signal* comsig = (struct comm_signal*)arg;
  if(!(event & UB_EV_SIGNAL))
    return;
  ub_comm_base_now(comsig->base);
  fptr_ok(fptr_whitelist_comm_signal(comsig->callback));
  (*comsig->callback)(sig, comsig->cb_arg);
}

int
comm_signal_bind(struct comm_signal* comsig, int sig)
{
  struct internal_signal* entry = (struct internal_signal*)calloc(1,
    sizeof(struct internal_signal));
  if(!entry) {
    log_err("malloc failed");
    return 0;
  }
  log_assert(comsig);
  /* add signal event */
  entry->ev = ub_signal_new(comsig->base->eb->base, sig,
    comm_signal_callback, comsig);
  if(entry->ev == NULL) {
    log_err("Could not create signal event");
    free(entry);
    return 0;
  }
  if(ub_signal_add(entry->ev, NULL) != 0) {
    log_err("Could not add signal handler");
    ub_event_free(entry->ev);
    free(entry);
    return 0;
  }
  /* link into list */
  entry->next = comsig->ev_signal;
  comsig->ev_signal = entry;
  return 1;
}

void
comm_signal_delete(struct comm_signal* comsig)
{
  struct internal_signal* p, *np;
  if(!comsig)
    return;
  p=comsig->ev_signal;
  while(p) {
    np = p->next;
    ub_signal_del(p->ev);
    ub_event_free(p->ev);
    free(p);
    p = np;
  }
  free(comsig);
}

Coverage Report

Created: 2025-10-10 06:37