tor/src/or/circuitbuild.c

/* Copyright (c) 2001 Matej Pfajfar.
 * Copyright (c) 2001-2004, Roger Dingledine.
 * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
 * Copyright (c) 2007-2011, The Tor Project, Inc. */
/* See LICENSE for licensing information */

/**
 * \file circuitbuild.c
 * \brief The actual details of building circuits.
 **/

#define CIRCUIT_PRIVATE

#include "or.h"
#include "circuitbuild.h"
#include "circuitlist.h"
#include "circuituse.h"
#include "config.h"
#include "connection.h"
#include "connection_edge.h"
#include "connection_or.h"
#include "control.h"
#include "directory.h"
#include "main.h"
#include "networkstatus.h"
#include "nodelist.h"
#include "onion.h"
#include "policies.h"
#include "relay.h"
#include "rephist.h"
#include "router.h"
#include "routerlist.h"
#include "routerparse.h"
#include "crypto.h"
#undef log
#include <math.h>

#ifndef MIN
#define MIN(a,b) ((a)<(b)?(a):(b))
#endif

#define CBT_BIN_TO_MS(bin) ((bin)*CBT_BIN_WIDTH + (CBT_BIN_WIDTH/2))

/********* START VARIABLES **********/
/** Global list of circuit build times */
// FIXME: Add this as a member for entry_guard_t instead of global?
// Then we could do per-guard statistics, as guards are likely to
// vary in their own latency. The downside of this is that guards
// can change frequently, so we'd be building a lot more circuits
// most likely.
circuit_build_times_t circ_times;

/** A global list of all circuits at this hop. */
extern circuit_t *global_circuitlist;

/** An entry_guard_t represents our information about a chosen long-term
 * first hop, known as a "helper" node in the literature. We can't just
 * use a node_t, since we want to remember these even when we
 * don't have any directory info. */
typedef struct {
  char nickname[MAX_NICKNAME_LEN+1];
  char identity[DIGEST_LEN];
  time_t chosen_on_date; /**< Approximately when was this guard added?
                          * "0" if we don't know. */
  char *chosen_by_version; /**< What tor version added this guard? NULL
                            * if we don't know. */
  unsigned int made_contact : 1; /**< 0 if we have never connected to this
                                  * router, 1 if we have. */
  unsigned int can_retry : 1; /**< Should we retry connecting to this entry,
                               * in spite of having it marked as unreachable?*/
  time_t bad_since; /**< 0 if this guard is currently usable, or the time at
                      * which it was observed to become (according to the
                      * directory or the user configuration) unusable. */
  time_t unreachable_since; /**< 0 if we can connect to this guard, or the
                             * time at which we first noticed we couldn't
                             * connect to it. */
  time_t last_attempted; /**< 0 if we can connect to this guard, or the time
                          * at which we last failed to connect to it. */
} entry_guard_t;

/** A list of our chosen entry guards. */
static smartlist_t *entry_guards = NULL;
/** A value of 1 means that the entry_guards list has changed
 * and those changes need to be flushed to disk. */
static int entry_guards_dirty = 0;

/** If set, we're running the unit tests: we should avoid clobbering
 * our state file or accessing get_options() or get_or_state() */
static int unit_tests = 0;

/********* END VARIABLES ************/

static int circuit_deliver_create_cell(circuit_t *circ,
                                       uint8_t cell_type, const char *payload);
static int onion_pick_cpath_exit(origin_circuit_t *circ, extend_info_t *exit);
static crypt_path_t *onion_next_hop_in_cpath(crypt_path_t *cpath);
static int onion_extend_cpath(origin_circuit_t *circ);
static int count_acceptable_nodes(smartlist_t *routers);
static int onion_append_hop(crypt_path_t **head_ptr, extend_info_t *choice);

static void entry_guards_changed(void);

static int
circuit_build_times_disabled(void)
{
  if (unit_tests) {
    return 0;
  } else {
    int consensus_disabled = networkstatus_get_param(NULL, "cbtdisabled",
                                                     0, 0, 1);
    int config_disabled = !get_options()->LearnCircuitBuildTimeout;
    int dirauth_disabled = get_options()->AuthoritativeDir;
    int state_disabled = (get_or_state()->LastWritten == -1);

    if (consensus_disabled || config_disabled || dirauth_disabled ||
           state_disabled) {
      log_info(LD_CIRC,
               "CircuitBuildTime learning is disabled. "
               "Consensus=%d, Config=%d, AuthDir=%d, StateFile=%d",
               consensus_disabled, config_disabled, dirauth_disabled,
               state_disabled);
      return 1;
    } else {
      return 0;
    }
  }
}

static int32_t
circuit_build_times_max_timeouts(void)
{
  return networkstatus_get_param(NULL, "cbtmaxtimeouts",
                                 CBT_DEFAULT_MAX_RECENT_TIMEOUT_COUNT,
                                 CBT_MIN_MAX_RECENT_TIMEOUT_COUNT,
                                 CBT_MAX_MAX_RECENT_TIMEOUT_COUNT);
}

static int32_t
circuit_build_times_default_num_xm_modes(void)
{
  int32_t num = networkstatus_get_param(NULL, "cbtnummodes",
                                        CBT_DEFAULT_NUM_XM_MODES,
                                        CBT_MIN_NUM_XM_MODES,
                                        CBT_MAX_NUM_XM_MODES);
  return num;
}

static int32_t
circuit_build_times_min_circs_to_observe(void)
{
  int32_t num = networkstatus_get_param(NULL, "cbtmincircs",
                                        CBT_DEFAULT_MIN_CIRCUITS_TO_OBSERVE,
                                        CBT_MIN_MIN_CIRCUITS_TO_OBSERVE,
                                        CBT_MAX_MIN_CIRCUITS_TO_OBSERVE);
  return num;
}

/** Return true iff <b>cbt</b> has recorded enough build times that we
 * want to start acting on the timeout it implies. */
int
circuit_build_times_enough_to_compute(circuit_build_times_t *cbt)
{
  return cbt->total_build_times >= circuit_build_times_min_circs_to_observe();
}

double
circuit_build_times_quantile_cutoff(void)
{
  int32_t num = networkstatus_get_param(NULL, "cbtquantile",
                                        CBT_DEFAULT_QUANTILE_CUTOFF,
                                        CBT_MIN_QUANTILE_CUTOFF,
                                        CBT_MAX_QUANTILE_CUTOFF);
  return num/100.0;
}

int
circuit_build_times_get_bw_scale(networkstatus_t *ns)
{
  return networkstatus_get_param(ns, "bwweightscale",
                                 BW_WEIGHT_SCALE,
                                 BW_MIN_WEIGHT_SCALE,
                                 BW_MAX_WEIGHT_SCALE);
}

static double
circuit_build_times_close_quantile(void)
{
  int32_t param;
  /* Cast is safe - circuit_build_times_quantile_cutoff() is capped */
  int32_t min = (int)tor_lround(100*circuit_build_times_quantile_cutoff());
  param = networkstatus_get_param(NULL, "cbtclosequantile",
             CBT_DEFAULT_CLOSE_QUANTILE,
             CBT_MIN_CLOSE_QUANTILE,
             CBT_MAX_CLOSE_QUANTILE);
  if (param < min) {
    log_warn(LD_DIR, "Consensus parameter cbtclosequantile is "
             "too small, raising to %d", min);
    param = min;
  }
  return param / 100.0;
}

static int32_t
circuit_build_times_test_frequency(void)
{
  int32_t num = networkstatus_get_param(NULL, "cbttestfreq",
                                        CBT_DEFAULT_TEST_FREQUENCY,
                                        CBT_MIN_TEST_FREQUENCY,
                                        CBT_MAX_TEST_FREQUENCY);
  return num;
}

static int32_t
circuit_build_times_min_timeout(void)
{
  int32_t num = networkstatus_get_param(NULL, "cbtmintimeout",
                                        CBT_DEFAULT_TIMEOUT_MIN_VALUE,
                                        CBT_MIN_TIMEOUT_MIN_VALUE,
                                        CBT_MAX_TIMEOUT_MIN_VALUE);
  return num;
}

int32_t
circuit_build_times_initial_timeout(void)
{
  int32_t min = circuit_build_times_min_timeout();
  int32_t param = networkstatus_get_param(NULL, "cbtinitialtimeout",
                                          CBT_DEFAULT_TIMEOUT_INITIAL_VALUE,
                                          CBT_MIN_TIMEOUT_INITIAL_VALUE,
                                          CBT_MAX_TIMEOUT_INITIAL_VALUE);
  if (param < min) {
    log_warn(LD_DIR, "Consensus parameter cbtinitialtimeout is too small, "
             "raising to %d", min);
    param = min;
  }
  return param;
}

static int32_t
circuit_build_times_recent_circuit_count(networkstatus_t *ns)
{
  return networkstatus_get_param(ns, "cbtrecentcount",
                                 CBT_DEFAULT_RECENT_CIRCUITS,
                                 CBT_MIN_RECENT_CIRCUITS,
                                 CBT_MAX_RECENT_CIRCUITS);
}

/**
 * This function is called when we get a consensus update.
 *
 * It checks to see if we have changed any consensus parameters
 * that require reallocation or discard of previous stats.
 */
void
circuit_build_times_new_consensus_params(circuit_build_times_t *cbt,
                                         networkstatus_t *ns)
{
  int32_t num = circuit_build_times_recent_circuit_count(ns);

  if (num > 0 && num != cbt->liveness.num_recent_circs) {
    int8_t *recent_circs;
    log_notice(LD_CIRC, "Changing recent timeout size from %d to %d",
               cbt->liveness.num_recent_circs, num);

    tor_assert(cbt->liveness.timeouts_after_firsthop);

    /*
     * Technically this is a circular array that we are reallocating
     * and memcopying. However, since it only consists of either 1s
     * or 0s, and is only used in a statistical test to determine when
     * we should discard our history after a sufficient number of 1's
     * have been reached, it is fine if order is not preserved or
     * elements are lost.
     *
     * cbtrecentcount should only be changing in cases of severe network
     * distress anyway, so memory correctness here is paramount over
     * doing acrobatics to preserve the array.
     */
    recent_circs = tor_malloc_zero(sizeof(int8_t)*num);
    memcpy(recent_circs, cbt->liveness.timeouts_after_firsthop,
           sizeof(int8_t)*MIN(num, cbt->liveness.num_recent_circs));

    // Adjust the index if it needs it.
    if (num < cbt->liveness.num_recent_circs) {
      cbt->liveness.after_firsthop_idx = MIN(num-1,
              cbt->liveness.after_firsthop_idx);
    }

    tor_free(cbt->liveness.timeouts_after_firsthop);
    cbt->liveness.timeouts_after_firsthop = recent_circs;
    cbt->liveness.num_recent_circs = num;
  }
}

/** Make a note that we're running unit tests (rather than running Tor
 * itself), so we avoid clobbering our state file. */
void
circuitbuild_running_unit_tests(void)
{
  unit_tests = 1;
}

/**
 * Return the initial default or configured timeout in milliseconds
 */
static double
circuit_build_times_get_initial_timeout(void)
{
  double timeout;
  if (!unit_tests && get_options()->CircuitBuildTimeout) {
    timeout = get_options()->CircuitBuildTimeout*1000;
    if (timeout < circuit_build_times_min_timeout()) {
      log_warn(LD_CIRC, "Config CircuitBuildTimeout too low. Setting to %ds",
               circuit_build_times_min_timeout()/1000);
      timeout = circuit_build_times_min_timeout();
    }
  } else {
    timeout = circuit_build_times_initial_timeout();
  }
  return timeout;
}

/**
 * Reset the build time state.
 *
 * Leave estimated parameters, timeout and network liveness intact
 * for future use.
 */
void
circuit_build_times_reset(circuit_build_times_t *cbt)
{
  memset(cbt->circuit_build_times, 0, sizeof(cbt->circuit_build_times));
  cbt->total_build_times = 0;
  cbt->build_times_idx = 0;
  cbt->have_computed_timeout = 0;
}

/**
 * Initialize the buildtimes structure for first use.
 *
 * Sets the initial timeout values based on either the config setting,
 * the consensus param, or the default (CBT_DEFAULT_TIMEOUT_INITIAL_VALUE).
 */
void
circuit_build_times_init(circuit_build_times_t *cbt)
{
  memset(cbt, 0, sizeof(*cbt));
  cbt->liveness.num_recent_circs =
      circuit_build_times_recent_circuit_count(NULL);
  cbt->liveness.timeouts_after_firsthop = tor_malloc_zero(sizeof(int8_t)*
                                      cbt->liveness.num_recent_circs);
  cbt->close_ms = cbt->timeout_ms = circuit_build_times_get_initial_timeout();
  control_event_buildtimeout_set(cbt, BUILDTIMEOUT_SET_EVENT_RESET);
}

#if 0
/**
 * Rewind our build time history by n positions.
 */
static void
circuit_build_times_rewind_history(circuit_build_times_t *cbt, int n)
{
  int i = 0;

  cbt->build_times_idx -= n;
  cbt->build_times_idx %= CBT_NCIRCUITS_TO_OBSERVE;

  for (i = 0; i < n; i++) {
    cbt->circuit_build_times[(i+cbt->build_times_idx)
                             %CBT_NCIRCUITS_TO_OBSERVE]=0;
  }

  if (cbt->total_build_times > n) {
    cbt->total_build_times -= n;
  } else {
    cbt->total_build_times = 0;
  }

  log_info(LD_CIRC,
          "Rewound history by %d places. Current index: %d. "
          "Total: %d", n, cbt->build_times_idx, cbt->total_build_times);
}
#endif

/**
 * Add a new build time value <b>time</b> to the set of build times. Time
 * units are milliseconds.
 *
 * circuit_build_times <b>cbt</a> is a circular array, so loop around when
 * array is full.
 */
int
circuit_build_times_add_time(circuit_build_times_t *cbt, build_time_t time)
{
  if (time <= 0 || time > CBT_BUILD_TIME_MAX) {
    log_warn(LD_BUG, "Circuit build time is too large (%u)."
                      "This is probably a bug.", time);
    tor_fragile_assert();
    return -1;
  }

  log_debug(LD_CIRC, "Adding circuit build time %u", time);

  cbt->circuit_build_times[cbt->build_times_idx] = time;
  cbt->build_times_idx = (cbt->build_times_idx + 1) % CBT_NCIRCUITS_TO_OBSERVE;
  if (cbt->total_build_times < CBT_NCIRCUITS_TO_OBSERVE)
    cbt->total_build_times++;

  if ((cbt->total_build_times % CBT_SAVE_STATE_EVERY) == 0) {
    /* Save state every n circuit builds */
    if (!unit_tests && !get_options()->AvoidDiskWrites)
      or_state_mark_dirty(get_or_state(), 0);
  }

  return 0;
}

/**
 * Return maximum circuit build time
 */
static build_time_t
circuit_build_times_max(circuit_build_times_t *cbt)
{
  int i = 0;
  build_time_t max_build_time = 0;
  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] > max_build_time
            && cbt->circuit_build_times[i] != CBT_BUILD_ABANDONED)
      max_build_time = cbt->circuit_build_times[i];
  }
  return max_build_time;
}

#if 0
/** Return minimum circuit build time */
build_time_t
circuit_build_times_min(circuit_build_times_t *cbt)
{
  int i = 0;
  build_time_t min_build_time = CBT_BUILD_TIME_MAX;
  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] && /* 0 <-> uninitialized */
        cbt->circuit_build_times[i] < min_build_time)
      min_build_time = cbt->circuit_build_times[i];
  }
  if (min_build_time == CBT_BUILD_TIME_MAX) {
    log_warn(LD_CIRC, "No build times less than CBT_BUILD_TIME_MAX!");
  }
  return min_build_time;
}
#endif

/**
 * Calculate and return a histogram for the set of build times.
 *
 * Returns an allocated array of histrogram bins representing
 * the frequency of index*CBT_BIN_WIDTH millisecond
 * build times. Also outputs the number of bins in nbins.
 *
 * The return value must be freed by the caller.
 */
static uint32_t *
circuit_build_times_create_histogram(circuit_build_times_t *cbt,
                                     build_time_t *nbins)
{
  uint32_t *histogram;
  build_time_t max_build_time = circuit_build_times_max(cbt);
  int i, c;

  *nbins = 1 + (max_build_time / CBT_BIN_WIDTH);
  histogram = tor_malloc_zero(*nbins * sizeof(build_time_t));

  // calculate histogram
  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] == 0
            || cbt->circuit_build_times[i] == CBT_BUILD_ABANDONED)
      continue; /* 0 <-> uninitialized */

    c = (cbt->circuit_build_times[i] / CBT_BIN_WIDTH);
    histogram[c]++;
  }

  return histogram;
}

/**
 * Return the Pareto start-of-curve parameter Xm.
 *
 * Because we are not a true Pareto curve, we compute this as the
 * weighted average of the N=3 most frequent build time bins.
 */
static build_time_t
circuit_build_times_get_xm(circuit_build_times_t *cbt)
{
  build_time_t i, nbins;
  build_time_t *nth_max_bin;
  int32_t bin_counts=0;
  build_time_t ret = 0;
  uint32_t *histogram = circuit_build_times_create_histogram(cbt, &nbins);
  int n=0;
  int num_modes = circuit_build_times_default_num_xm_modes();

  // Only use one mode if < 1000 buildtimes. Not enough data
  // for multiple.
  if (cbt->total_build_times < CBT_NCIRCUITS_TO_OBSERVE)
    num_modes = 1;

  nth_max_bin = (build_time_t*)tor_malloc_zero(num_modes*sizeof(build_time_t));

  for (i = 0; i < nbins; i++) {
    if (histogram[i] >= histogram[nth_max_bin[0]]) {
      nth_max_bin[0] = i;
    }

    for (n = 1; n < num_modes; n++) {
      if (histogram[i] >= histogram[nth_max_bin[n]] &&
           (!histogram[nth_max_bin[n-1]]
               || histogram[i] < histogram[nth_max_bin[n-1]])) {
        nth_max_bin[n] = i;
      }
    }
  }

  for (n = 0; n < num_modes; n++) {
    bin_counts += histogram[nth_max_bin[n]];
    ret += CBT_BIN_TO_MS(nth_max_bin[n])*histogram[nth_max_bin[n]];
    log_info(LD_CIRC, "Xm mode #%d: %u %u", n, CBT_BIN_TO_MS(nth_max_bin[n]),
             histogram[nth_max_bin[n]]);
  }

  ret /= bin_counts;
  tor_free(histogram);
  tor_free(nth_max_bin);

  return ret;
}

/**
 * Output a histogram of current circuit build times to
 * the or_state_t state structure.
 */
void
circuit_build_times_update_state(circuit_build_times_t *cbt,
                                 or_state_t *state)
{
  uint32_t *histogram;
  build_time_t i = 0;
  build_time_t nbins = 0;
  config_line_t **next, *line;

  histogram = circuit_build_times_create_histogram(cbt, &nbins);
  // write to state
  config_free_lines(state->BuildtimeHistogram);
  next = &state->BuildtimeHistogram;
  *next = NULL;

  state->TotalBuildTimes = cbt->total_build_times;
  state->CircuitBuildAbandonedCount = 0;

  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] == CBT_BUILD_ABANDONED)
      state->CircuitBuildAbandonedCount++;
  }

  for (i = 0; i < nbins; i++) {
    // compress the histogram by skipping the blanks
    if (histogram[i] == 0) continue;
    *next = line = tor_malloc_zero(sizeof(config_line_t));
    line->key = tor_strdup("CircuitBuildTimeBin");
    line->value = tor_malloc(25);
    tor_snprintf(line->value, 25, "%d %d",
            CBT_BIN_TO_MS(i), histogram[i]);
    next = &(line->next);
  }

  if (!unit_tests) {
    if (!get_options()->AvoidDiskWrites)
      or_state_mark_dirty(get_or_state(), 0);
  }

  tor_free(histogram);
}

/**
 * Shuffle the build times array.
 *
 * Stolen from http://en.wikipedia.org/wiki/Fisher\u2013Yates_shuffle
 */
static void
circuit_build_times_shuffle_and_store_array(circuit_build_times_t *cbt,
                                            build_time_t *raw_times,
                                            int num_times)
{
  int n = num_times;
  if (num_times > CBT_NCIRCUITS_TO_OBSERVE) {
    log_notice(LD_CIRC, "Decreasing circuit_build_times size from %d to %d",
               num_times, CBT_NCIRCUITS_TO_OBSERVE);
  }

  /* This code can only be run on a compact array */
  while (n-- > 1) {
    int k = crypto_rand_int(n + 1); /* 0 <= k <= n. */
    build_time_t tmp = raw_times[k];
    raw_times[k] = raw_times[n];
    raw_times[n] = tmp;
  }

  /* Since the times are now shuffled, take a random CBT_NCIRCUITS_TO_OBSERVE
   * subset (ie the first CBT_NCIRCUITS_TO_OBSERVE values) */
  for (n = 0; n < MIN(num_times, CBT_NCIRCUITS_TO_OBSERVE); n++) {
    circuit_build_times_add_time(cbt, raw_times[n]);
  }
}

/**
 * Filter old synthetic timeouts that were created before the
 * new right-censored Pareto calculation was deployed.
 *
 * Once all clients before 0.2.1.13-alpha are gone, this code
 * will be unused.
 */
static int
circuit_build_times_filter_timeouts(circuit_build_times_t *cbt)
{
  int num_filtered=0, i=0;
  double timeout_rate = 0;
  build_time_t max_timeout = 0;

  timeout_rate = circuit_build_times_timeout_rate(cbt);
  max_timeout = (build_time_t)cbt->close_ms;

  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] > max_timeout) {
      build_time_t replaced = cbt->circuit_build_times[i];
      num_filtered++;
      cbt->circuit_build_times[i] = CBT_BUILD_ABANDONED;

      log_debug(LD_CIRC, "Replaced timeout %d with %d", replaced,
               cbt->circuit_build_times[i]);
    }
  }

  log_info(LD_CIRC,
           "We had %d timeouts out of %d build times, "
           "and filtered %d above the max of %u",
          (int)(cbt->total_build_times*timeout_rate),
          cbt->total_build_times, num_filtered, max_timeout);

  return num_filtered;
}

/**
 * Load histogram from <b>state</b>, shuffling the resulting array
 * after we do so. Use this result to estimate parameters and
 * calculate the timeout.
 *
 * Return -1 on error.
 */
int
circuit_build_times_parse_state(circuit_build_times_t *cbt,
                                or_state_t *state)
{
  int tot_values = 0;
  uint32_t loaded_cnt = 0, N = 0;
  config_line_t *line;
  unsigned int i;
  build_time_t *loaded_times;
  int err = 0;
  circuit_build_times_init(cbt);

  if (circuit_build_times_disabled()) {
    return 0;
  }

  /* build_time_t 0 means uninitialized */
  loaded_times = tor_malloc_zero(sizeof(build_time_t)*state->TotalBuildTimes);

  for (line = state->BuildtimeHistogram; line; line = line->next) {
    smartlist_t *args = smartlist_create();
    smartlist_split_string(args, line->value, " ",
                           SPLIT_SKIP_SPACE|SPLIT_IGNORE_BLANK, 0);
    if (smartlist_len(args) < 2) {
      log_warn(LD_GENERAL, "Unable to parse circuit build times: "
                           "Too few arguments to CircuitBuildTime");
      err = 1;
      SMARTLIST_FOREACH(args, char*, cp, tor_free(cp));
      smartlist_free(args);
      break;
    } else {
      const char *ms_str = smartlist_get(args,0);
      const char *count_str = smartlist_get(args,1);
      uint32_t count, k;
      build_time_t ms;
      int ok;
      ms = (build_time_t)tor_parse_ulong(ms_str, 0, 0,
                                         CBT_BUILD_TIME_MAX, &ok, NULL);
      if (!ok) {
        log_warn(LD_GENERAL, "Unable to parse circuit build times: "
                             "Unparsable bin number");
        err = 1;
        SMARTLIST_FOREACH(args, char*, cp, tor_free(cp));
        smartlist_free(args);
        break;
      }
      count = (uint32_t)tor_parse_ulong(count_str, 0, 0,
                                        UINT32_MAX, &ok, NULL);
      if (!ok) {
        log_warn(LD_GENERAL, "Unable to parse circuit build times: "
                             "Unparsable bin count");
        err = 1;
        SMARTLIST_FOREACH(args, char*, cp, tor_free(cp));
        smartlist_free(args);
        break;
      }

      if (loaded_cnt+count+state->CircuitBuildAbandonedCount
            > state->TotalBuildTimes) {
        log_warn(LD_CIRC,
                 "Too many build times in state file. "
                 "Stopping short before %d",
                 loaded_cnt+count);
        SMARTLIST_FOREACH(args, char*, cp, tor_free(cp));
        smartlist_free(args);
        break;
      }

      for (k = 0; k < count; k++) {
        loaded_times[loaded_cnt++] = ms;
      }
      N++;
      SMARTLIST_FOREACH(args, char*, cp, tor_free(cp));
      smartlist_free(args);
    }
  }

  log_info(LD_CIRC,
           "Adding %d timeouts.", state->CircuitBuildAbandonedCount);
  for (i=0; i < state->CircuitBuildAbandonedCount; i++) {
    loaded_times[loaded_cnt++] = CBT_BUILD_ABANDONED;
  }

  if (loaded_cnt != state->TotalBuildTimes) {
    log_warn(LD_CIRC,
            "Corrupt state file? Build times count mismatch. "
            "Read %d times, but file says %d", loaded_cnt,
            state->TotalBuildTimes);
    err = 1;
    circuit_build_times_reset(cbt);
    goto done;
  }

  circuit_build_times_shuffle_and_store_array(cbt, loaded_times, loaded_cnt);

  /* Verify that we didn't overwrite any indexes */
  for (i=0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (!cbt->circuit_build_times[i])
      break;
    tot_values++;
  }
  log_info(LD_CIRC,
           "Loaded %d/%d values from %d lines in circuit time histogram",
           tot_values, cbt->total_build_times, N);

  if (cbt->total_build_times != tot_values
        || cbt->total_build_times > CBT_NCIRCUITS_TO_OBSERVE) {
    log_warn(LD_CIRC,
            "Corrupt state file? Shuffled build times mismatch. "
            "Read %d times, but file says %d", tot_values,
            state->TotalBuildTimes);
    err = 1;
    circuit_build_times_reset(cbt);
    goto done;
  }

  circuit_build_times_set_timeout(cbt);

  if (!state->CircuitBuildAbandonedCount && cbt->total_build_times) {
    circuit_build_times_filter_timeouts(cbt);
  }

 done:
  tor_free(loaded_times);
  return err ? -1 : 0;
}

/**
 * Estimates the Xm and Alpha parameters using
 * http://en.wikipedia.org/wiki/Pareto_distribution#Parameter_estimation
 *
 * The notable difference is that we use mode instead of min to estimate Xm.
 * This is because our distribution is frechet-like. We claim this is
 * an acceptable approximation because we are only concerned with the
 * accuracy of the CDF of the tail.
 */
int
circuit_build_times_update_alpha(circuit_build_times_t *cbt)
{
  build_time_t *x=cbt->circuit_build_times;
  double a = 0;
  int n=0,i=0,abandoned_count=0;
  build_time_t max_time=0;

  /* http://en.wikipedia.org/wiki/Pareto_distribution#Parameter_estimation */
  /* We sort of cheat here and make our samples slightly more pareto-like
   * and less frechet-like. */
  cbt->Xm = circuit_build_times_get_xm(cbt);

  tor_assert(cbt->Xm > 0);

  for (i=0; i< CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (!x[i]) {
      continue;
    }

    if (x[i] < cbt->Xm) {
      a += tor_mathlog(cbt->Xm);
    } else if (x[i] == CBT_BUILD_ABANDONED) {
      abandoned_count++;
    } else {
      a += tor_mathlog(x[i]);
      if (x[i] > max_time)
        max_time = x[i];
    }
    n++;
  }

  /*
   * We are erring and asserting here because this can only happen
   * in codepaths other than startup. The startup state parsing code
   * performs this same check, and resets state if it hits it. If we
   * hit it at runtime, something serious has gone wrong.
   */
  if (n!=cbt->total_build_times) {
    log_err(LD_CIRC, "Discrepancy in build times count: %d vs %d", n,
            cbt->total_build_times);
  }
  tor_assert(n==cbt->total_build_times);

  if (max_time <= 0) {
    /* This can happen if Xm is actually the *maximum* value in the set.
     * It can also happen if we've abandoned every single circuit somehow.
     * In either case, tell the caller not to compute a new build timeout. */
    log_warn(LD_BUG,
             "Could not determine largest build time (%d). "
             "Xm is %dms and we've abandoned %d out of %d circuits.", max_time,
             cbt->Xm, abandoned_count, n);
    return 0;
  }

  a += abandoned_count*tor_mathlog(max_time);

  a -= n*tor_mathlog(cbt->Xm);
  // Estimator comes from Eq #4 in:
  // "Bayesian estimation based on trimmed samples from Pareto populations"
  // by Arturo J. Fernández. We are right-censored only.
  a = (n-abandoned_count)/a;

  cbt->alpha = a;

  return 1;
}

/**
 * This is the Pareto Quantile Function. It calculates the point x
 * in the distribution such that F(x) = quantile (ie quantile*100%
 * of the mass of the density function is below x on the curve).
 *
 * We use it to calculate the timeout and also to generate synthetic
 * values of time for circuits that timeout before completion.
 *
 * See http://en.wikipedia.org/wiki/Quantile_function,
 * http://en.wikipedia.org/wiki/Inverse_transform_sampling and
 * http://en.wikipedia.org/wiki/Pareto_distribution#Generating_a_
 *     random_sample_from_Pareto_distribution
 * That's right. I'll cite wikipedia all day long.
 *
 * Return value is in milliseconds.
 */
double
circuit_build_times_calculate_timeout(circuit_build_times_t *cbt,
                                      double quantile)
{
  double ret;
  tor_assert(quantile >= 0);
  tor_assert(1.0-quantile > 0);
  tor_assert(cbt->Xm > 0);

  ret = cbt->Xm/pow(1.0-quantile,1.0/cbt->alpha);
  if (ret > INT32_MAX) {
    ret = INT32_MAX;
  }
  tor_assert(ret > 0);
  return ret;
}

/** Pareto CDF */
double
circuit_build_times_cdf(circuit_build_times_t *cbt, double x)
{
  double ret;
  tor_assert(cbt->Xm > 0);
  ret = 1.0-pow(cbt->Xm/x,cbt->alpha);
  tor_assert(0 <= ret && ret <= 1.0);
  return ret;
}

/**
 * Generate a synthetic time using our distribution parameters.
 *
 * The return value will be within the [q_lo, q_hi) quantile points
 * on the CDF.
 */
build_time_t
circuit_build_times_generate_sample(circuit_build_times_t *cbt,
                                    double q_lo, double q_hi)
{
  double randval = crypto_rand_double();
  build_time_t ret;
  double u;

  /* Generate between [q_lo, q_hi) */
  /*XXXX This is what nextafter is supposed to be for; we should use it on the
   * platforms that support it. */
  q_hi -= 1.0/(INT32_MAX);

  tor_assert(q_lo >= 0);
  tor_assert(q_hi < 1);
  tor_assert(q_lo < q_hi);

  u = q_lo + (q_hi-q_lo)*randval;

  tor_assert(0 <= u && u < 1.0);
  /* circuit_build_times_calculate_timeout returns <= INT32_MAX */
  ret = (build_time_t)
    tor_lround(circuit_build_times_calculate_timeout(cbt, u));
  tor_assert(ret > 0);
  return ret;
}

/**
 * Estimate an initial alpha parameter by solving the quantile
 * function with a quantile point and a specific timeout value.
 */
void
circuit_build_times_initial_alpha(circuit_build_times_t *cbt,
                                  double quantile, double timeout_ms)
{
  // Q(u) = Xm/((1-u)^(1/a))
  // Q(0.8) = Xm/((1-0.8))^(1/a)) = CircBuildTimeout
  // CircBuildTimeout = Xm/((1-0.8))^(1/a))
  // CircBuildTimeout = Xm*((1-0.8))^(-1/a))
  // ln(CircBuildTimeout) = ln(Xm)+ln(((1-0.8)))*(-1/a)
  // -ln(1-0.8)/(ln(CircBuildTimeout)-ln(Xm))=a
  tor_assert(quantile >= 0);
  tor_assert(cbt->Xm > 0);
  cbt->alpha = tor_mathlog(1.0-quantile)/
    (tor_mathlog(cbt->Xm)-tor_mathlog(timeout_ms));
  tor_assert(cbt->alpha > 0);
}

/**
 * Returns true if we need circuits to be built
 */
int
circuit_build_times_needs_circuits(circuit_build_times_t *cbt)
{
  /* Return true if < MIN_CIRCUITS_TO_OBSERVE */
  return !circuit_build_times_enough_to_compute(cbt);
}

/**
 * Returns true if we should build a timeout test circuit
 * right now.
 */
int
circuit_build_times_needs_circuits_now(circuit_build_times_t *cbt)
{
  return circuit_build_times_needs_circuits(cbt) &&
    approx_time()-cbt->last_circ_at > circuit_build_times_test_frequency();
}

/**
 * Called to indicate that the network showed some signs of liveness,
 * i.e. we received a cell.
 *
 * This is used by circuit_build_times_network_check_live() to decide
 * if we should record the circuit build timeout or not.
 *
 * This function is called every time we receive a cell. Avoid
 * syscalls, events, and other high-intensity work.
 */
void
circuit_build_times_network_is_live(circuit_build_times_t *cbt)
{
  time_t now = approx_time();
  if (cbt->liveness.nonlive_timeouts > 0) {
    log_notice(LD_CIRC,
               "Tor now sees network activity. Restoring circuit build "
               "timeout recording. Network was down for %d seconds "
               "during %d circuit attempts.",
               (int)(now - cbt->liveness.network_last_live),
               cbt->liveness.nonlive_timeouts);
  }
  cbt->liveness.network_last_live = now;
  cbt->liveness.nonlive_timeouts = 0;
}

/**
 * Called to indicate that we completed a circuit. Because this circuit
 * succeeded, it doesn't count as a timeout-after-the-first-hop.
 *
 * This is used by circuit_build_times_network_check_changed() to determine
 * if we had too many recent timeouts and need to reset our learned timeout
 * to something higher.
 */
void
circuit_build_times_network_circ_success(circuit_build_times_t *cbt)
{
  cbt->liveness.timeouts_after_firsthop[cbt->liveness.after_firsthop_idx] = 0;
  cbt->liveness.after_firsthop_idx++;
  cbt->liveness.after_firsthop_idx %= cbt->liveness.num_recent_circs;
}

/**
 * A circuit just timed out. If it failed after the first hop, record it
 * in our history for later deciding if the network speed has changed.
 *
 * This is used by circuit_build_times_network_check_changed() to determine
 * if we had too many recent timeouts and need to reset our learned timeout
 * to something higher.
 */
static void
circuit_build_times_network_timeout(circuit_build_times_t *cbt,
                                    int did_onehop)
{
  if (did_onehop) {
    cbt->liveness.timeouts_after_firsthop[cbt->liveness.after_firsthop_idx]=1;
    cbt->liveness.after_firsthop_idx++;
    cbt->liveness.after_firsthop_idx %= cbt->liveness.num_recent_circs;
  }
}

/**
 * A circuit was just forcibly closed. If there has been no recent network
 * activity at all, but this circuit was launched back when we thought the
 * network was live, increment the number of "nonlive" circuit timeouts.
 *
 * This is used by circuit_build_times_network_check_live() to decide
 * if we should record the circuit build timeout or not.
 */
static void
circuit_build_times_network_close(circuit_build_times_t *cbt,
                                    int did_onehop, time_t start_time)
{
  time_t now = time(NULL);
  /*
   * Check if this is a timeout that was for a circuit that spent its
   * entire existence during a time where we have had no network activity.
   */
  if (cbt->liveness.network_last_live < start_time) {
    if (did_onehop) {
      char last_live_buf[ISO_TIME_LEN+1];
      char start_time_buf[ISO_TIME_LEN+1];
      char now_buf[ISO_TIME_LEN+1];
      format_local_iso_time(last_live_buf, cbt->liveness.network_last_live);
      format_local_iso_time(start_time_buf, start_time);
      format_local_iso_time(now_buf, now);
      log_warn(LD_BUG,
               "Circuit somehow completed a hop while the network was "
               "not live. Network was last live at %s, but circuit launched "
               "at %s. It's now %s.", last_live_buf, start_time_buf,
               now_buf);
    }
    cbt->liveness.nonlive_timeouts++;
    if (cbt->liveness.nonlive_timeouts == 1) {
      log_notice(LD_CIRC,
                 "Tor has not observed any network activity for the past %d "
                 "seconds. Disabling circuit build timeout code.",
                 (int)(now - cbt->liveness.network_last_live));
    } else {
      log_info(LD_CIRC,
             "Got non-live timeout. Current count is: %d",
             cbt->liveness.nonlive_timeouts);
    }
  }
}

/**
 * When the network is not live, we do not record circuit build times.
 *
 * The network is considered not live if there has been at least one
 * circuit build that began and ended (had its close_ms measurement
 * period expire) since we last received a cell.
 *
 * Also has the side effect of rewinding the circuit time history
 * in the case of recent liveness changes.
 */
int
circuit_build_times_network_check_live(circuit_build_times_t *cbt)
{
  if (cbt->liveness.nonlive_timeouts > 0) {
    return 0;
  }

  return 1;
}

/**
 * Returns true if we have seen more than MAX_RECENT_TIMEOUT_COUNT of
 * the past RECENT_CIRCUITS time out after the first hop. Used to detect
 * if the network connection has changed significantly, and if so,
 * resets our circuit build timeout to the default.
 *
 * Also resets the entire timeout history in this case and causes us
 * to restart the process of building test circuits and estimating a
 * new timeout.
 */
int
circuit_build_times_network_check_changed(circuit_build_times_t *cbt)
{
  int total_build_times = cbt->total_build_times;
  int timeout_count=0;
  int i;

  /* how many of our recent circuits made it to the first hop but then
   * timed out? */
  for (i = 0; i < cbt->liveness.num_recent_circs; i++) {
    timeout_count += cbt->liveness.timeouts_after_firsthop[i];
  }

  /* If 80% of our recent circuits are timing out after the first hop,
   * we need to re-estimate a new initial alpha and timeout. */
  if (timeout_count < circuit_build_times_max_timeouts()) {
    return 0;
  }

  circuit_build_times_reset(cbt);
  memset(cbt->liveness.timeouts_after_firsthop, 0,
          sizeof(*cbt->liveness.timeouts_after_firsthop)*
          cbt->liveness.num_recent_circs);
  cbt->liveness.after_firsthop_idx = 0;

  /* Check to see if this has happened before. If so, double the timeout
   * to give people on abysmally bad network connections a shot at access */
  if (cbt->timeout_ms >= circuit_build_times_get_initial_timeout()) {
    if (cbt->timeout_ms > INT32_MAX/2 || cbt->close_ms > INT32_MAX/2) {
      log_warn(LD_CIRC, "Insanely large circuit build timeout value. "
              "(timeout = %lfmsec, close = %lfmsec)",
               cbt->timeout_ms, cbt->close_ms);
    } else {
      cbt->timeout_ms *= 2;
      cbt->close_ms *= 2;
    }
  } else {
    cbt->close_ms = cbt->timeout_ms
                  = circuit_build_times_get_initial_timeout();
  }

  control_event_buildtimeout_set(cbt, BUILDTIMEOUT_SET_EVENT_RESET);

  log_notice(LD_CIRC,
            "Network connection speed appears to have changed. Resetting "
            "timeout to %lds after %d timeouts and %d buildtimes.",
            tor_lround(cbt->timeout_ms/1000), timeout_count,
            total_build_times);

  return 1;
}

/**
 * Count the number of timeouts in a set of cbt data.
 */
double
circuit_build_times_timeout_rate(const circuit_build_times_t *cbt)
{
  int i=0,timeouts=0;
  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] >= cbt->timeout_ms) {
       timeouts++;
    }
  }

  if (!cbt->total_build_times)
    return 0;

  return ((double)timeouts)/cbt->total_build_times;
}

/**
 * Count the number of closed circuits in a set of cbt data.
 */
double
circuit_build_times_close_rate(const circuit_build_times_t *cbt)
{
  int i=0,closed=0;
  for (i = 0; i < CBT_NCIRCUITS_TO_OBSERVE; i++) {
    if (cbt->circuit_build_times[i] == CBT_BUILD_ABANDONED) {
       closed++;
    }
  }

  if (!cbt->total_build_times)
    return 0;

  return ((double)closed)/cbt->total_build_times;
}

/**
 * Store a timeout as a synthetic value.
 *
 * Returns true if the store was successful and we should possibly
 * update our timeout estimate.
 */
int
circuit_build_times_count_close(circuit_build_times_t *cbt,
                                int did_onehop,
                                time_t start_time)
{
  if (circuit_build_times_disabled()) {
    cbt->close_ms = cbt->timeout_ms
                  = circuit_build_times_get_initial_timeout();
    return 0;
  }

  /* Record this force-close to help determine if the network is dead */
  circuit_build_times_network_close(cbt, did_onehop, start_time);

  /* Only count timeouts if network is live.. */
  if (!circuit_build_times_network_check_live(cbt)) {
    return 0;
  }

  circuit_build_times_add_time(cbt, CBT_BUILD_ABANDONED);
  return 1;
}

/**
 * Update timeout counts to determine if we need to expire
 * our build time history due to excessive timeouts.
 *
 * We do not record any actual time values at this stage;
 * we are only interested in recording the fact that a timeout
 * happened. We record the time values via
 * circuit_build_times_count_close() and circuit_build_times_add_time().
 */
void
circuit_build_times_count_timeout(circuit_build_times_t *cbt,
                                  int did_onehop)
{
  if (circuit_build_times_disabled()) {
    cbt->close_ms = cbt->timeout_ms
                  = circuit_build_times_get_initial_timeout();
    return;
  }

  /* Register the fact that a timeout just occurred. */
  circuit_build_times_network_timeout(cbt, did_onehop);

  /* If there are a ton of timeouts, we should reset
   * the circuit build timeout. */
  circuit_build_times_network_check_changed(cbt);
}

/**
 * Estimate a new timeout based on history and set our timeout
 * variable accordingly.
 */
static int
circuit_build_times_set_timeout_worker(circuit_build_times_t *cbt)
{
  build_time_t max_time;
  if (!circuit_build_times_enough_to_compute(cbt))
    return 0;

  if (!circuit_build_times_update_alpha(cbt))
    return 0;

  cbt->timeout_ms = circuit_build_times_calculate_timeout(cbt,
                                circuit_build_times_quantile_cutoff());

  cbt->close_ms = circuit_build_times_calculate_timeout(cbt,
                                circuit_build_times_close_quantile());

  max_time = circuit_build_times_max(cbt);

  /* Sometimes really fast guard nodes give us such a steep curve
   * that this ends up being not that much greater than timeout_ms.
   * Make it be at least 1 min to handle this case. */
  cbt->close_ms = MAX(cbt->close_ms, circuit_build_times_initial_timeout());

  if (cbt->timeout_ms > max_time) {
    log_notice(LD_CIRC,
               "Circuit build timeout of %dms is beyond the maximum build "
               "time we have ever observed. Capping it to %dms.",
               (int)cbt->timeout_ms, max_time);
    cbt->timeout_ms = max_time;
  }

  if (max_time < INT32_MAX/2 && cbt->close_ms > 2*max_time) {
    log_notice(LD_CIRC,
               "Circuit build measurement period of %dms is more than twice "
               "the maximum build time we have ever observed. Capping it to "
               "%dms.", (int)cbt->close_ms, 2*max_time);
    cbt->close_ms = 2*max_time;
  }

  cbt->have_computed_timeout = 1;
  return 1;
}

/**
 * Exposed function to compute a new timeout. Dispatches events and
 * also filters out extremely high timeout values.
 */
void
circuit_build_times_set_timeout(circuit_build_times_t *cbt)
{
  long prev_timeout = tor_lround(cbt->timeout_ms/1000);
  double timeout_rate;

  if (!circuit_build_times_set_timeout_worker(cbt))
    return;

  if (cbt->timeout_ms < circuit_build_times_min_timeout()) {
    log_warn(LD_CIRC, "Set buildtimeout to low value %lfms. Setting to %dms",
             cbt->timeout_ms, circuit_build_times_min_timeout());
    cbt->timeout_ms = circuit_build_times_min_timeout();
    if (cbt->close_ms < cbt->timeout_ms) {
      /* This shouldn't happen because of MAX() in timeout_worker above,
       * but doing it just in case */
      cbt->close_ms = circuit_build_times_initial_timeout();
    }
  }

  control_event_buildtimeout_set(cbt, BUILDTIMEOUT_SET_EVENT_COMPUTED);

  timeout_rate = circuit_build_times_timeout_rate(cbt);

  if (prev_timeout > tor_lround(cbt->timeout_ms/1000)) {
    log_notice(LD_CIRC,
               "Based on %d circuit times, it looks like we don't need to "
               "wait so long for circuits to finish. We will now assume a "
               "circuit is too slow to use after waiting %ld seconds.",
               cbt->total_build_times,
               tor_lround(cbt->timeout_ms/1000));
    log_info(LD_CIRC,
             "Circuit timeout data: %lfms, %lfms, Xm: %d, a: %lf, r: %lf",
             cbt->timeout_ms, cbt->close_ms, cbt->Xm, cbt->alpha,
             timeout_rate);
  } else if (prev_timeout < tor_lround(cbt->timeout_ms/1000)) {
    log_notice(LD_CIRC,
               "Based on %d circuit times, it looks like we need to wait "
               "longer for circuits to finish. We will now assume a "
               "circuit is too slow to use after waiting %ld seconds.",
               cbt->total_build_times,
               tor_lround(cbt->timeout_ms/1000));
    log_info(LD_CIRC,
             "Circuit timeout data: %lfms, %lfms, Xm: %d, a: %lf, r: %lf",
             cbt->timeout_ms, cbt->close_ms, cbt->Xm, cbt->alpha,
             timeout_rate);
  } else {
    log_info(LD_CIRC,
             "Set circuit build timeout to %lds (%lfms, %lfms, Xm: %d, a: %lf,"
             " r: %lf) based on %d circuit times",
             tor_lround(cbt->timeout_ms/1000),
             cbt->timeout_ms, cbt->close_ms, cbt->Xm, cbt->alpha, timeout_rate,
             cbt->total_build_times);
  }
}

/** Iterate over values of circ_id, starting from conn-\>next_circ_id,
 * and with the high bit specified by conn-\>circ_id_type, until we get
 * a circ_id that is not in use by any other circuit on that conn.
 *
 * Return it, or 0 if can't get a unique circ_id.
 */
static circid_t
get_unique_circ_id_by_conn(or_connection_t *conn)
{
  circid_t test_circ_id;
  circid_t attempts=0;
  circid_t high_bit;

  tor_assert(conn);
  if (conn->circ_id_type == CIRC_ID_TYPE_NEITHER) {
    log_warn(LD_BUG, "Trying to pick a circuit ID for a connection from "
             "a client with no identity.");
    return 0;
  }
  high_bit = (conn->circ_id_type == CIRC_ID_TYPE_HIGHER) ? 1<<15 : 0;
  do {
    /* Sequentially iterate over test_circ_id=1...1<<15-1 until we find a
     * circID such that (high_bit|test_circ_id) is not already used. */
    test_circ_id = conn->next_circ_id++;
    if (test_circ_id == 0 || test_circ_id >= 1<<15) {
      test_circ_id = 1;
      conn->next_circ_id = 2;
    }
    if (++attempts > 1<<15) {
      /* Make sure we don't loop forever if all circ_id's are used. This
       * matters because it's an external DoS opportunity.
       */
      log_warn(LD_CIRC,"No unused circ IDs. Failing.");
      return 0;
    }
    test_circ_id |= high_bit;
  } while (circuit_id_in_use_on_orconn(test_circ_id, conn));
  return test_circ_id;
}

/** If <b>verbose</b> is false, allocate and return a comma-separated list of
 * the currently built elements of circuit_t.  If <b>verbose</b> is true, also
 * list information about link status in a more verbose format using spaces.
 * If <b>verbose_names</b> is false, give nicknames for Named routers and hex
 * digests for others; if <b>verbose_names</b> is true, use $DIGEST=Name style
 * names.
 */
static char *
circuit_list_path_impl(origin_circuit_t *circ, int verbose, int verbose_names)
{
  crypt_path_t *hop;
  smartlist_t *elements;
  const char *states[] = {"closed", "waiting for keys", "open"};
  char *s;

  elements = smartlist_create();

  if (verbose) {
    const char *nickname = build_state_get_exit_nickname(circ->build_state);
    char *cp;
    tor_asprintf(&cp, "%s%s circ (length %d%s%s):",
                 circ->build_state->is_internal ? "internal" : "exit",
                 circ->build_state->need_uptime ? " (high-uptime)" : "",
                 circ->build_state->desired_path_len,
                 circ->_base.state == CIRCUIT_STATE_OPEN ? "" : ", exit ",
                 circ->_base.state == CIRCUIT_STATE_OPEN ? "" :
                 (nickname?nickname:"*unnamed*"));
    smartlist_add(elements, cp);
  }

  hop = circ->cpath;
  do {
    char *elt;
    const char *id;
    const node_t *node;
    if (!hop)
      break;
    if (!verbose && hop->state != CPATH_STATE_OPEN)
      break;
    if (!hop->extend_info)
      break;
    id = hop->extend_info->identity_digest;
    if (verbose_names) {
      elt = tor_malloc(MAX_VERBOSE_NICKNAME_LEN+1);
      if ((node = node_get_by_id(id))) {
        node_get_verbose_nickname(node, elt);
      } else if (is_legal_nickname(hop->extend_info->nickname)) {
        elt[0] = '$';
        base16_encode(elt+1, HEX_DIGEST_LEN+1, id, DIGEST_LEN);
        elt[HEX_DIGEST_LEN+1]= '~';
        strlcpy(elt+HEX_DIGEST_LEN+2,
                hop->extend_info->nickname, MAX_NICKNAME_LEN+1);
      } else {
        elt[0] = '$';
        base16_encode(elt+1, HEX_DIGEST_LEN+1, id, DIGEST_LEN);
      }
    } else { /* ! verbose_names */
      node = node_get_by_id(id);
      if (node && node_is_named(node)) {
        elt = tor_strdup(node_get_nickname(node));
      } else {
        elt = tor_malloc(HEX_DIGEST_LEN+2);
        elt[0] = '$';
        base16_encode(elt+1, HEX_DIGEST_LEN+1, id, DIGEST_LEN);
      }
    }
    tor_assert(elt);
    if (verbose) {
      size_t len = strlen(elt)+2+strlen(states[hop->state])+1;
      char *v = tor_malloc(len);
      tor_assert(hop->state <= 2);
      tor_snprintf(v,len,"%s(%s)",elt,states[hop->state]);
      smartlist_add(elements, v);
      tor_free(elt);
    } else {
      smartlist_add(elements, elt);
    }
    hop = hop->next;
  } while (hop != circ->cpath);

  s = smartlist_join_strings(elements, verbose?" ":",", 0, NULL);
  SMARTLIST_FOREACH(elements, char*, cp, tor_free(cp));
  smartlist_free(elements);
  return s;
}

/** If <b>verbose</b> is false, allocate and return a comma-separated
 * list of the currently built elements of circuit_t.  If
 * <b>verbose</b> is true, also list information about link status in
 * a more verbose format using spaces.
 */
char *
circuit_list_path(origin_circuit_t *circ, int verbose)
{
  return circuit_list_path_impl(circ, verbose, 0);
}

/** Allocate and return a comma-separated list of the currently built elements
 * of circuit_t, giving each as a verbose nickname.
 */
char *
circuit_list_path_for_controller(origin_circuit_t *circ)
{
  return circuit_list_path_impl(circ, 0, 1);
}

/** Log, at severity <b>severity</b>, the nicknames of each router in
 * circ's cpath. Also log the length of the cpath, and the intended
 * exit point.
 */
void
circuit_log_path(int severity, unsigned int domain, origin_circuit_t *circ)
{
  char *s = circuit_list_path(circ,1);
  tor_log(severity,domain,"%s",s);
  tor_free(s);
}

/** Tell the rep(utation)hist(ory) module about the status of the links
 * in circ.  Hops that have become OPEN are marked as successfully
 * extended; the _first_ hop that isn't open (if any) is marked as
 * unable to extend.
 */
/* XXXX Someday we should learn from OR circuits too. */
void
circuit_rep_hist_note_result(origin_circuit_t *circ)
{
  crypt_path_t *hop;
  const char *prev_digest = NULL;
  hop = circ->cpath;
  if (!hop) /* circuit hasn't started building yet. */
    return;
  if (server_mode(get_options())) {
    const routerinfo_t *me = router_get_my_routerinfo();
    if (!me)
      return;
    prev_digest = me->cache_info.identity_digest;
  }
  do {
    const node_t *node = node_get_by_id(hop->extend_info->identity_digest);
    if (node) { /* Why do we check this?  We know the identity. -NM XXXX */
      if (prev_digest) {
        if (hop->state == CPATH_STATE_OPEN)
          rep_hist_note_extend_succeeded(prev_digest, node->identity);
        else {
          rep_hist_note_extend_failed(prev_digest, node->identity);
          break;
        }
      }
      prev_digest = node->identity;
    } else {
      prev_digest = NULL;
    }
    hop=hop->next;
  } while (hop!=circ->cpath);
}

/** Pick all the entries in our cpath. Stop and return 0 when we're
 * happy, or return -1 if an error occurs. */
static int
onion_populate_cpath(origin_circuit_t *circ)
{
  int r;
 again:
  r = onion_extend_cpath(circ);
  if (r < 0) {
    log_info(LD_CIRC,"Generating cpath hop failed.");
    return -1;
  }
  if (r == 0)
    goto again;
  return 0; /* if r == 1 */
}

/** Create and return a new origin circuit. Initialize its purpose and
 * build-state based on our arguments.  The <b>flags</b> argument is a
 * bitfield of CIRCLAUNCH_* flags. */
origin_circuit_t *
origin_circuit_init(uint8_t purpose, int flags)
{
  /* sets circ->p_circ_id and circ->p_conn */
  origin_circuit_t *circ = origin_circuit_new();
  circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OR_WAIT);
  circ->build_state = tor_malloc_zero(sizeof(cpath_build_state_t));
  circ->build_state->onehop_tunnel =
    ((flags & CIRCLAUNCH_ONEHOP_TUNNEL) ? 1 : 0);
  circ->build_state->need_uptime =
    ((flags & CIRCLAUNCH_NEED_UPTIME) ? 1 : 0);
  circ->build_state->need_capacity =
    ((flags & CIRCLAUNCH_NEED_CAPACITY) ? 1 : 0);
  circ->build_state->is_internal =
    ((flags & CIRCLAUNCH_IS_INTERNAL) ? 1 : 0);
  circ->_base.purpose = purpose;
  return circ;
}

/** Build a new circuit for <b>purpose</b>. If <b>exit</b>
 * is defined, then use that as your exit router, else choose a suitable
 * exit node.
 *
 * Also launch a connection to the first OR in the chosen path, if
 * it's not open already.
 */
origin_circuit_t *
circuit_establish_circuit(uint8_t purpose, extend_info_t *exit, int flags)
{
  origin_circuit_t *circ;
  int err_reason = 0;

  circ = origin_circuit_init(purpose, flags);

  if (onion_pick_cpath_exit(circ, exit) < 0 ||
      onion_populate_cpath(circ) < 0) {
    circuit_mark_for_close(TO_CIRCUIT(circ), END_CIRC_REASON_NOPATH);
    return NULL;
  }

  control_event_circuit_status(circ, CIRC_EVENT_LAUNCHED, 0);

  if ((err_reason = circuit_handle_first_hop(circ)) < 0) {
    circuit_mark_for_close(TO_CIRCUIT(circ), -err_reason);
    return NULL;
  }
  return circ;
}

/** Start establishing the first hop of our circuit. Figure out what
 * OR we should connect to, and if necessary start the connection to
 * it. If we're already connected, then send the 'create' cell.
 * Return 0 for ok, -reason if circ should be marked-for-close. */
int
circuit_handle_first_hop(origin_circuit_t *circ)
{
  crypt_path_t *firsthop;
  or_connection_t *n_conn;
  int err_reason = 0;
  const char *msg = NULL;
  int should_launch = 0;

  firsthop = onion_next_hop_in_cpath(circ->cpath);
  tor_assert(firsthop);
  tor_assert(firsthop->extend_info);

  /* now see if we're already connected to the first OR in 'route' */
  log_debug(LD_CIRC,"Looking for firsthop '%s:%u'",
            fmt_addr(&firsthop->extend_info->addr),
            firsthop->extend_info->port);

  n_conn = connection_or_get_for_extend(firsthop->extend_info->identity_digest,
                                        &firsthop->extend_info->addr,
                                        &msg,
                                        &should_launch);

  if (!n_conn) {
    /* not currently connected in a useful way. */
    const char *name = strlen(firsthop->extend_info->nickname) ?
      firsthop->extend_info->nickname : fmt_addr(&firsthop->extend_info->addr);
    log_info(LD_CIRC, "Next router is %s: %s ",
             safe_str_client(name), msg?msg:"???");
    circ->_base.n_hop = extend_info_dup(firsthop->extend_info);

    if (should_launch) {
      if (circ->build_state->onehop_tunnel)
        control_event_bootstrap(BOOTSTRAP_STATUS_CONN_DIR, 0);
      n_conn = connection_or_connect(&firsthop->extend_info->addr,
                                     firsthop->extend_info->port,
                                     firsthop->extend_info->identity_digest);
      if (!n_conn) { /* connect failed, forget the whole thing */
        log_info(LD_CIRC,"connect to firsthop failed. Closing.");
        return -END_CIRC_REASON_CONNECTFAILED;
      }
    }

    log_debug(LD_CIRC,"connecting in progress (or finished). Good.");
    /* return success. The onion/circuit/etc will be taken care of
     * automatically (may already have been) whenever n_conn reaches
     * OR_CONN_STATE_OPEN.
     */
    return 0;
  } else { /* it's already open. use it. */
    tor_assert(!circ->_base.n_hop);
    circ->_base.n_conn = n_conn;
    log_debug(LD_CIRC,"Conn open. Delivering first onion skin.");
    if ((err_reason = circuit_send_next_onion_skin(circ)) < 0) {
      log_info(LD_CIRC,"circuit_send_next_onion_skin failed.");
      return err_reason;
    }
  }
  return 0;
}

/** Find any circuits that are waiting on <b>or_conn</b> to become
 * open and get them to send their create cells forward.
 *
 * Status is 1 if connect succeeded, or 0 if connect failed.
 */
void
circuit_n_conn_done(or_connection_t *or_conn, int status)
{
  smartlist_t *pending_circs;
  int err_reason = 0;

  log_debug(LD_CIRC,"or_conn to %s/%s, status=%d",
            or_conn->nickname ? or_conn->nickname : "NULL",
            or_conn->_base.address, status);

  pending_circs = smartlist_create();
  circuit_get_all_pending_on_or_conn(pending_circs, or_conn);

  SMARTLIST_FOREACH_BEGIN(pending_circs, circuit_t *, circ)
    {
      /* These checks are redundant wrt get_all_pending_on_or_conn, but I'm
       * leaving them in in case it's possible for the status of a circuit to
       * change as we're going down the list. */
      if (circ->marked_for_close || circ->n_conn || !circ->n_hop ||
          circ->state != CIRCUIT_STATE_OR_WAIT)
        continue;

      if (tor_digest_is_zero(circ->n_hop->identity_digest)) {
        /* Look at addr/port. This is an unkeyed connection. */
        if (!tor_addr_eq(&circ->n_hop->addr, &or_conn->_base.addr) ||
            circ->n_hop->port != or_conn->_base.port)
          continue;
      } else {
        /* We expected a key. See if it's the right one. */
        if (memcmp(or_conn->identity_digest,
                   circ->n_hop->identity_digest, DIGEST_LEN))
          continue;
      }
      if (!status) { /* or_conn failed; close circ */
        log_info(LD_CIRC,"or_conn failed. Closing circ.");
        circuit_mark_for_close(circ, END_CIRC_REASON_OR_CONN_CLOSED);
        continue;
      }
      log_debug(LD_CIRC, "Found circ, sending create cell.");
      /* circuit_deliver_create_cell will set n_circ_id and add us to
       * orconn_circuid_circuit_map, so we don't need to call
       * set_circid_orconn here. */
      circ->n_conn = or_conn;
      extend_info_free(circ->n_hop);
      circ->n_hop = NULL;

      if (CIRCUIT_IS_ORIGIN(circ)) {
        if ((err_reason =
             circuit_send_next_onion_skin(TO_ORIGIN_CIRCUIT(circ))) < 0) {
          log_info(LD_CIRC,
                   "send_next_onion_skin failed; circuit marked for closing.");
          circuit_mark_for_close(circ, -err_reason);
          continue;
          /* XXX could this be bad, eg if next_onion_skin failed because conn
           *     died? */
        }
      } else {
        /* pull the create cell out of circ->onionskin, and send it */
        tor_assert(circ->n_conn_onionskin);
        if (circuit_deliver_create_cell(circ,CELL_CREATE,
                                        circ->n_conn_onionskin)<0) {
          circuit_mark_for_close(circ, END_CIRC_REASON_RESOURCELIMIT);
          continue;
        }
        tor_free(circ->n_conn_onionskin);
        circuit_set_state(circ, CIRCUIT_STATE_OPEN);
      }
    }
  SMARTLIST_FOREACH_END(circ);

  smartlist_free(pending_circs);
}

/** Find a new circid that isn't currently in use on the circ->n_conn
 * for the outgoing
 * circuit <b>circ</b>, and deliver a cell of type <b>cell_type</b>
 * (either CELL_CREATE or CELL_CREATE_FAST) with payload <b>payload</b>
 * to this circuit.
 * Return -1 if we failed to find a suitable circid, else return 0.
 */
static int
circuit_deliver_create_cell(circuit_t *circ, uint8_t cell_type,
                            const char *payload)
{
  cell_t cell;
  circid_t id;

  tor_assert(circ);
  tor_assert(circ->n_conn);
  tor_assert(payload);
  tor_assert(cell_type == CELL_CREATE || cell_type == CELL_CREATE_FAST);

  id = get_unique_circ_id_by_conn(circ->n_conn);
  if (!id) {
    log_warn(LD_CIRC,"failed to get unique circID.");
    return -1;
  }
  log_debug(LD_CIRC,"Chosen circID %u.", id);
  circuit_set_n_circid_orconn(circ, id, circ->n_conn);

  memset(&cell, 0, sizeof(cell_t));
  cell.command = cell_type;
  cell.circ_id = circ->n_circ_id;

  memcpy(cell.payload, payload, ONIONSKIN_CHALLENGE_LEN);
  append_cell_to_circuit_queue(circ, circ->n_conn, &cell,
                               CELL_DIRECTION_OUT, 0);

  if (CIRCUIT_IS_ORIGIN(circ)) {
    /* mark it so it gets better rate limiting treatment. */
    circ->n_conn->client_used = time(NULL);
  }

  return 0;
}

/** We've decided to start our reachability testing. If all
 * is set, log this to the user. Return 1 if we did, or 0 if
 * we chose not to log anything. */
int
inform_testing_reachability(void)
{
  char dirbuf[128];
  const routerinfo_t *me = router_get_my_routerinfo();
  if (!me)
    return 0;
  control_event_server_status(LOG_NOTICE,
                              "CHECKING_REACHABILITY ORADDRESS=%s:%d",
                              me->address, me->or_port);
  if (me->dir_port) {
    tor_snprintf(dirbuf, sizeof(dirbuf), " and DirPort %s:%d",
                 me->address, me->dir_port);
    control_event_server_status(LOG_NOTICE,
                                "CHECKING_REACHABILITY DIRADDRESS=%s:%d",
                                me->address, me->dir_port);
  }
  log_notice(LD_OR, "Now checking whether ORPort %s:%d%s %s reachable... "
                         "(this may take up to %d minutes -- look for log "
                         "messages indicating success)",
      me->address, me->or_port,
      me->dir_port ? dirbuf : "",
      me->dir_port ? "are" : "is",
      TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT/60);

  return 1;
}

/** Return true iff we should send a create_fast cell to start building a given
 * circuit */
static INLINE int
should_use_create_fast_for_circuit(origin_circuit_t *circ)
{
  or_options_t *options = get_options();
  tor_assert(circ->cpath);
  tor_assert(circ->cpath->extend_info);

  if (!circ->cpath->extend_info->onion_key)
    return 1; /* our hand is forced: only a create_fast will work. */
  if (!options->FastFirstHopPK)
    return 0; /* we prefer to avoid create_fast */
  if (server_mode(options)) {
    /* We're a server, and we know an onion key. We can choose.
     * Prefer to blend in. */
    return 0;
  }

  return 1;
}

/** Return true if <b>circ</b> is the type of circuit we want to count
 * timeouts from. In particular, we want it to have not completed yet
 * (already completing indicates we cannibalized it), and we want it to
 * have exactly three hops.
 */
int
circuit_timeout_want_to_count_circ(origin_circuit_t *circ)
{
  return !circ->has_opened
          && circ->build_state->desired_path_len == DEFAULT_ROUTE_LEN;
}

/** This is the backbone function for building circuits.
 *
 * If circ's first hop is closed, then we need to build a create
 * cell and send it forward.
 *
 * Otherwise, we need to build a relay extend cell and send it
 * forward.
 *
 * Return -reason if we want to tear down circ, else return 0.
 */
int
circuit_send_next_onion_skin(origin_circuit_t *circ)
{
  crypt_path_t *hop;
  const node_t *node;
  char payload[2+4+DIGEST_LEN+ONIONSKIN_CHALLENGE_LEN];
  char *onionskin;
  size_t payload_len;

  tor_assert(circ);

  if (circ->cpath->state == CPATH_STATE_CLOSED) {
    int fast;
    uint8_t cell_type;
    log_debug(LD_CIRC,"First skin; sending create cell.");
    if (circ->build_state->onehop_tunnel)
      control_event_bootstrap(BOOTSTRAP_STATUS_ONEHOP_CREATE, 0);
    else
      control_event_bootstrap(BOOTSTRAP_STATUS_CIRCUIT_CREATE, 0);

    node = node_get_by_id(circ->_base.n_conn->identity_digest);
    fast = should_use_create_fast_for_circuit(circ);
    if (!fast) {
      /* We are an OR and we know the right onion key: we should
       * send an old slow create cell.
       */
      cell_type = CELL_CREATE;
      if (onion_skin_create(circ->cpath->extend_info->onion_key,
                            &(circ->cpath->dh_handshake_state),
                            payload) < 0) {
        log_warn(LD_CIRC,"onion_skin_create (first hop) failed.");
        return - END_CIRC_REASON_INTERNAL;
      }
      note_request("cell: create", 1);
    } else {
      /* We are not an OR, and we're building the first hop of a circuit to a
       * new OR: we can be speedy and use CREATE_FAST to save an RSA operation
       * and a DH operation. */
      cell_type = CELL_CREATE_FAST;
      memset(payload, 0, sizeof(payload));
      crypto_rand((char*) circ->cpath->fast_handshake_state,
                  sizeof(circ->cpath->fast_handshake_state));
      memcpy(payload, circ->cpath->fast_handshake_state,
             sizeof(circ->cpath->fast_handshake_state));
      note_request("cell: create fast", 1);
    }

    if (circuit_deliver_create_cell(TO_CIRCUIT(circ), cell_type, payload) < 0)
      return - END_CIRC_REASON_RESOURCELIMIT;

    circ->cpath->state = CPATH_STATE_AWAITING_KEYS;
    circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_BUILDING);
    log_info(LD_CIRC,"First hop: finished sending %s cell to '%s'",
             fast ? "CREATE_FAST" : "CREATE",
             node ? node_get_nickname(node) : "<unnamed>");
  } else {
    tor_assert(circ->cpath->state == CPATH_STATE_OPEN);
    tor_assert(circ->_base.state == CIRCUIT_STATE_BUILDING);
    log_debug(LD_CIRC,"starting to send subsequent skin.");
    hop = onion_next_hop_in_cpath(circ->cpath);
    if (!hop) {
      /* done building the circuit. whew. */
      circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OPEN);
      if (circuit_timeout_want_to_count_circ(circ)) {
        struct timeval end;
        long timediff;
        tor_gettimeofday(&end);
        timediff = tv_mdiff(&circ->_base.timestamp_created, &end);

        /*
         * If the circuit build time is much greater than we would have cut
         * it off at, we probably had a suspend event along this codepath,
         * and we should discard the value.
         */
        if (timediff < 0 || timediff > 2*circ_times.close_ms+1000) {
          log_notice(LD_CIRC, "Strange value for circuit build time: %ldmsec. "
                              "Assuming clock jump. Purpose %d", timediff,
                              circ->_base.purpose);
        } else if (!circuit_build_times_disabled()) {
          /* Only count circuit times if the network is live */
          if (circuit_build_times_network_check_live(&circ_times)) {
            circuit_build_times_add_time(&circ_times, (build_time_t)timediff);
            circuit_build_times_set_timeout(&circ_times);
          }

          if (circ->_base.purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) {
            circuit_build_times_network_circ_success(&circ_times);
          }
        }
      }
      log_info(LD_CIRC,"circuit built!");
      circuit_reset_failure_count(0);
      if (circ->build_state->onehop_tunnel)
        control_event_bootstrap(BOOTSTRAP_STATUS_REQUESTING_STATUS, 0);
      if (!can_complete_circuit && !circ->build_state->onehop_tunnel) {
        or_options_t *options = get_options();
        can_complete_circuit=1;
        /* FFFF Log a count of known routers here */
        log_notice(LD_GENERAL,
            "Tor has successfully opened a circuit. "
            "Looks like client functionality is working.");
        control_event_bootstrap(BOOTSTRAP_STATUS_DONE, 0);
        control_event_client_status(LOG_NOTICE, "CIRCUIT_ESTABLISHED");
        if (server_mode(options) && !check_whether_orport_reachable()) {
          inform_testing_reachability();
          consider_testing_reachability(1, 1);
        }
      }
      circuit_rep_hist_note_result(circ);
      circuit_has_opened(circ); /* do other actions as necessary */

      /* We're done with measurement circuits here. Just close them */
      if (circ->_base.purpose == CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT)
        circuit_mark_for_close(TO_CIRCUIT(circ), END_CIRC_REASON_FINISHED);
      return 0;
    }

    if (tor_addr_family(&hop->extend_info->addr) != AF_INET) {
      log_warn(LD_BUG, "Trying to extend to a non-IPv4 address.");
      return - END_CIRC_REASON_INTERNAL;
    }

    set_uint32(payload, tor_addr_to_ipv4n(&hop->extend_info->addr));
    set_uint16(payload+4, htons(hop->extend_info->port));

    onionskin = payload+2+4;
    memcpy(payload+2+4+ONIONSKIN_CHALLENGE_LEN,
           hop->extend_info->identity_digest, DIGEST_LEN);
    payload_len = 2+4+ONIONSKIN_CHALLENGE_LEN+DIGEST_LEN;

    if (onion_skin_create(hop->extend_info->onion_key,
                          &(hop->dh_handshake_state), onionskin) < 0) {
      log_warn(LD_CIRC,"onion_skin_create failed.");
      return - END_CIRC_REASON_INTERNAL;
    }

    log_info(LD_CIRC,"Sending extend relay cell.");
    note_request("cell: extend", 1);
    /* send it to hop->prev, because it will transfer
     * it to a create cell and then send to hop */
    if (relay_send_command_from_edge(0, TO_CIRCUIT(circ),
                                     RELAY_COMMAND_EXTEND,
                                     payload, payload_len, hop->prev) < 0)
      return 0; /* circuit is closed */

    hop->state = CPATH_STATE_AWAITING_KEYS;
  }
  return 0;
}

/** Our clock just jumped by <b>seconds_elapsed</b>. Assume
 * something has also gone wrong with our network: notify the user,
 * and abandon all not-yet-used circuits. */
void
circuit_note_clock_jumped(int seconds_elapsed)
{
  int severity = server_mode(get_options()) ? LOG_WARN : LOG_NOTICE;
  tor_log(severity, LD_GENERAL, "Your system clock just jumped %d seconds %s; "
      "assuming established circuits no longer work.",
      seconds_elapsed >=0 ? seconds_elapsed : -seconds_elapsed,
      seconds_elapsed >=0 ? "forward" : "backward");
  control_event_general_status(LOG_WARN, "CLOCK_JUMPED TIME=%d",
                               seconds_elapsed);
  can_complete_circuit=0; /* so it'll log when it works again */
  control_event_client_status(severity, "CIRCUIT_NOT_ESTABLISHED REASON=%s",
                              "CLOCK_JUMPED");
  circuit_mark_all_unused_circs();
  circuit_expire_all_dirty_circs();
}

/** Take the 'extend' <b>cell</b>, pull out addr/port plus the onion
 * skin and identity digest for the next hop. If we're already connected,
 * pass the onion skin to the next hop using a create cell; otherwise
 * launch a new OR connection, and <b>circ</b> will notice when the
 * connection succeeds or fails.
 *
 * Return -1 if we want to warn and tear down the circuit, else return 0.
 */
int
circuit_extend(cell_t *cell, circuit_t *circ)
{
  or_connection_t *n_conn;
  relay_header_t rh;
  char *onionskin;
  char *id_digest=NULL;
  uint32_t n_addr32;
  uint16_t n_port;
  tor_addr_t n_addr;
  const char *msg = NULL;
  int should_launch = 0;

  if (circ->n_conn) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "n_conn already set. Bug/attack. Closing.");
    return -1;
  }
  if (circ->n_hop) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "conn to next hop already launched. Bug/attack. Closing.");
    return -1;
  }

  if (!server_mode(get_options())) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "Got an extend cell, but running as a client. Closing.");
    return -1;
  }

  relay_header_unpack(&rh, cell->payload);

  if (rh.length < 4+2+ONIONSKIN_CHALLENGE_LEN+DIGEST_LEN) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "Wrong length %d on extend cell. Closing circuit.",
           rh.length);
    return -1;
  }

  n_addr32 = ntohl(get_uint32(cell->payload+RELAY_HEADER_SIZE));
  n_port = ntohs(get_uint16(cell->payload+RELAY_HEADER_SIZE+4));
  onionskin = (char*) cell->payload+RELAY_HEADER_SIZE+4+2;
  id_digest = (char*) cell->payload+RELAY_HEADER_SIZE+4+2+
    ONIONSKIN_CHALLENGE_LEN;
  tor_addr_from_ipv4h(&n_addr, n_addr32);

  if (!n_port || !n_addr32) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "Client asked me to extend to zero destination port or addr.");
    return -1;
  }

  /* Check if they asked us for 0000..0000. We support using
   * an empty fingerprint for the first hop (e.g. for a bridge relay),
   * but we don't want to let people send us extend cells for empty
   * fingerprints -- a) because it opens the user up to a mitm attack,
   * and b) because it lets an attacker force the relay to hold open a
   * new TLS connection for each extend request. */
  if (tor_digest_is_zero(id_digest)) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "Client asked me to extend without specifying an id_digest.");
    return -1;
  }

  /* Next, check if we're being asked to connect to the hop that the
   * extend cell came from. There isn't any reason for that, and it can
   * assist circular-path attacks. */
  if (!memcmp(id_digest, TO_OR_CIRCUIT(circ)->p_conn->identity_digest,
              DIGEST_LEN)) {
    log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
           "Client asked me to extend back to the previous hop.");
    return -1;
  }

  n_conn = connection_or_get_for_extend(id_digest,
                                        &n_addr,
                                        &msg,
                                        &should_launch);

  if (!n_conn) {
    log_debug(LD_CIRC|LD_OR,"Next router (%s:%d): %s",
              fmt_addr(&n_addr), (int)n_port, msg?msg:"????");

    circ->n_hop = extend_info_alloc(NULL /*nickname*/,
                                    id_digest,
                                    NULL /*onion_key*/,
                                    &n_addr, n_port);

    circ->n_conn_onionskin = tor_malloc(ONIONSKIN_CHALLENGE_LEN);
    memcpy(circ->n_conn_onionskin, onionskin, ONIONSKIN_CHALLENGE_LEN);
    circuit_set_state(circ, CIRCUIT_STATE_OR_WAIT);

    if (should_launch) {
      /* we should try to open a connection */
      n_conn = connection_or_connect(&n_addr, n_port, id_digest);
      if (!n_conn) {
        log_info(LD_CIRC,"Launching n_conn failed. Closing circuit.");
        circuit_mark_for_close(circ, END_CIRC_REASON_CONNECTFAILED);
        return 0;
      }
      log_debug(LD_CIRC,"connecting in progress (or finished). Good.");
    }
    /* return success. The onion/circuit/etc will be taken care of
     * automatically (may already have been) whenever n_conn reaches
     * OR_CONN_STATE_OPEN.
     */
    return 0;
  }

  tor_assert(!circ->n_hop); /* Connection is already established. */
  circ->n_conn = n_conn;
  log_debug(LD_CIRC,"n_conn is %s:%u",
            n_conn->_base.address,n_conn->_base.port);

  if (circuit_deliver_create_cell(circ, CELL_CREATE, onionskin) < 0)
    return -1;
  return 0;
}

/** Initialize cpath-\>{f|b}_{crypto|digest} from the key material in
 * key_data.  key_data must contain CPATH_KEY_MATERIAL bytes, which are
 * used as follows:
 *   - 20 to initialize f_digest
 *   - 20 to initialize b_digest
 *   - 16 to key f_crypto
 *   - 16 to key b_crypto
 *
 * (If 'reverse' is true, then f_XX and b_XX are swapped.)
 */
int
circuit_init_cpath_crypto(crypt_path_t *cpath, const char *key_data,
                          int reverse)
{
  crypto_digest_env_t *tmp_digest;
  crypto_cipher_env_t *tmp_crypto;

  tor_assert(cpath);
  tor_assert(key_data);
  tor_assert(!(cpath->f_crypto || cpath->b_crypto ||
             cpath->f_digest || cpath->b_digest));

  cpath->f_digest = crypto_new_digest_env();
  crypto_digest_add_bytes(cpath->f_digest, key_data, DIGEST_LEN);
  cpath->b_digest = crypto_new_digest_env();
  crypto_digest_add_bytes(cpath->b_digest, key_data+DIGEST_LEN, DIGEST_LEN);

  if (!(cpath->f_crypto =
        crypto_create_init_cipher(key_data+(2*DIGEST_LEN),1))) {
    log_warn(LD_BUG,"Forward cipher initialization failed.");
    return -1;
  }
  if (!(cpath->b_crypto =
        crypto_create_init_cipher(key_data+(2*DIGEST_LEN)+CIPHER_KEY_LEN,0))) {
    log_warn(LD_BUG,"Backward cipher initialization failed.");
    return -1;
  }

  if (reverse) {
    tmp_digest = cpath->f_digest;
    cpath->f_digest = cpath->b_digest;
    cpath->b_digest = tmp_digest;
    tmp_crypto = cpath->f_crypto;
    cpath->f_crypto = cpath->b_crypto;
    cpath->b_crypto = tmp_crypto;
  }

  return 0;
}

/** A created or extended cell came back to us on the circuit, and it included
 * <b>reply</b> as its body.  (If <b>reply_type</b> is CELL_CREATED, the body
 * contains (the second DH key, plus KH).  If <b>reply_type</b> is
 * CELL_CREATED_FAST, the body contains a secret y and a hash H(x|y).)
 *
 * Calculate the appropriate keys and digests, make sure KH is
 * correct, and initialize this hop of the cpath.
 *
 * Return - reason if we want to mark circ for close, else return 0.
 */
int
circuit_finish_handshake(origin_circuit_t *circ, uint8_t reply_type,
                         const uint8_t *reply)
{
  char keys[CPATH_KEY_MATERIAL_LEN];
  crypt_path_t *hop;

  if (circ->cpath->state == CPATH_STATE_AWAITING_KEYS)
    hop = circ->cpath;
  else {
    hop = onion_next_hop_in_cpath(circ->cpath);
    if (!hop) { /* got an extended when we're all done? */
      log_warn(LD_PROTOCOL,"got extended when circ already built? Closing.");
      return - END_CIRC_REASON_TORPROTOCOL;
    }
  }
  tor_assert(hop->state == CPATH_STATE_AWAITING_KEYS);

  if (reply_type == CELL_CREATED && hop->dh_handshake_state) {
    if (onion_skin_client_handshake(hop->dh_handshake_state, (char*)reply,keys,
                                    DIGEST_LEN*2+CIPHER_KEY_LEN*2) < 0) {
      log_warn(LD_CIRC,"onion_skin_client_handshake failed.");
      return -END_CIRC_REASON_TORPROTOCOL;
    }
    /* Remember hash of g^xy */
    memcpy(hop->handshake_digest, reply+DH_KEY_LEN, DIGEST_LEN);
  } else if (reply_type == CELL_CREATED_FAST && !hop->dh_handshake_state) {
    if (fast_client_handshake(hop->fast_handshake_state, reply,
                              (uint8_t*)keys,
                              DIGEST_LEN*2+CIPHER_KEY_LEN*2) < 0) {
      log_warn(LD_CIRC,"fast_client_handshake failed.");
      return -END_CIRC_REASON_TORPROTOCOL;
    }
    memcpy(hop->handshake_digest, reply+DIGEST_LEN, DIGEST_LEN);
  } else {
    log_warn(LD_PROTOCOL,"CREATED cell type did not match CREATE cell type.");
    return -END_CIRC_REASON_TORPROTOCOL;
  }

  crypto_dh_free(hop->dh_handshake_state); /* don't need it anymore */
  hop->dh_handshake_state = NULL;

  memset(hop->fast_handshake_state, 0, sizeof(hop->fast_handshake_state));

  if (circuit_init_cpath_crypto(hop, keys, 0)<0) {
    return -END_CIRC_REASON_TORPROTOCOL;
  }

  hop->state = CPATH_STATE_OPEN;
  log_info(LD_CIRC,"Finished building %scircuit hop:",
           (reply_type == CELL_CREATED_FAST) ? "fast " : "");
  circuit_log_path(LOG_INFO,LD_CIRC,circ);
  control_event_circuit_status(circ, CIRC_EVENT_EXTENDED, 0);

  return 0;
}

/** We received a relay truncated cell on circ.
 *
 * Since we don't ask for truncates currently, getting a truncated
 * means that a connection broke or an extend failed. For now,
 * just give up: for circ to close, and return 0.
 */
int
circuit_truncated(origin_circuit_t *circ, crypt_path_t *layer)
{
//  crypt_path_t *victim;
//  connection_t *stream;

  tor_assert(circ);
  tor_assert(layer);

  /* XXX Since we don't ask for truncates currently, getting a truncated
   *     means that a connection broke or an extend failed. For now,
   *     just give up.
   */
  circuit_mark_for_close(TO_CIRCUIT(circ),
          END_CIRC_REASON_FLAG_REMOTE|END_CIRC_REASON_OR_CONN_CLOSED);
  return 0;

#if 0
  while (layer->next != circ->cpath) {
    /* we need to clear out layer->next */
    victim = layer->next;
    log_debug(LD_CIRC, "Killing a layer of the cpath.");

    for (stream = circ->p_streams; stream; stream=stream->next_stream) {
      if (stream->cpath_layer == victim) {
        log_info(LD_APP, "Marking stream %d for close because of truncate.",
                 stream->stream_id);
        /* no need to send 'end' relay cells,
         * because the other side's already dead
         */
        connection_mark_unattached_ap(stream, END_STREAM_REASON_DESTROY);
      }
    }

    layer->next = victim->next;
    circuit_free_cpath_node(victim);
  }

  log_info(LD_CIRC, "finished");
  return 0;
#endif
}

/** Given a response payload and keys, initialize, then send a created
 * cell back.
 */
int
onionskin_answer(or_circuit_t *circ, uint8_t cell_type, const char *payload,
                 const char *keys)
{
  cell_t cell;
  crypt_path_t *tmp_cpath;

  tmp_cpath = tor_malloc_zero(sizeof(crypt_path_t));
  tmp_cpath->magic = CRYPT_PATH_MAGIC;

  memset(&cell, 0, sizeof(cell_t));
  cell.command = cell_type;
  cell.circ_id = circ->p_circ_id;

  circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OPEN);

  memcpy(cell.payload, payload,
         cell_type == CELL_CREATED ? ONIONSKIN_REPLY_LEN : DIGEST_LEN*2);

  log_debug(LD_CIRC,"init digest forward 0x%.8x, backward 0x%.8x.",
            (unsigned int)get_uint32(keys),
            (unsigned int)get_uint32(keys+20));
  if (circuit_init_cpath_crypto(tmp_cpath, keys, 0)<0) {
    log_warn(LD_BUG,"Circuit initialization failed");
    tor_free(tmp_cpath);
    return -1;
  }
  circ->n_digest = tmp_cpath->f_digest;
  circ->n_crypto = tmp_cpath->f_crypto;
  circ->p_digest = tmp_cpath->b_digest;
  circ->p_crypto = tmp_cpath->b_crypto;
  tmp_cpath->magic = 0;
  tor_free(tmp_cpath);

  if (cell_type == CELL_CREATED)
    memcpy(circ->handshake_digest, cell.payload+DH_KEY_LEN, DIGEST_LEN);
  else
    memcpy(circ->handshake_digest, cell.payload+DIGEST_LEN, DIGEST_LEN);

  circ->is_first_hop = (cell_type == CELL_CREATED_FAST);

  append_cell_to_circuit_queue(TO_CIRCUIT(circ),
                               circ->p_conn, &cell, CELL_DIRECTION_IN, 0);
  log_debug(LD_CIRC,"Finished sending 'created' cell.");

  if (!is_local_addr(&circ->p_conn->_base.addr) &&
      !connection_or_nonopen_was_started_here(circ->p_conn)) {
    /* record that we could process create cells from a non-local conn
     * that we didn't initiate; presumably this means that create cells
     * can reach us too. */
    router_orport_found_reachable();
  }

  return 0;
}

/** Choose a length for a circuit of purpose <b>purpose</b>.
 * Default length is 3 + the number of endpoints that would give something
 * away. If the routerlist <b>routers</b> doesn't have enough routers
 * to handle the desired path length, return as large a path length as
 * is feasible, except if it's less than 2, in which case return -1.
 */
static int
new_route_len(uint8_t purpose, extend_info_t *exit,
              smartlist_t *nodes)
{
  int num_acceptable_routers;
  int routelen;

  tor_assert(nodes);

  routelen = DEFAULT_ROUTE_LEN;
  if (exit &&
      purpose != CIRCUIT_PURPOSE_TESTING &&
      purpose != CIRCUIT_PURPOSE_S_ESTABLISH_INTRO)
    routelen++;

  num_acceptable_routers = count_acceptable_nodes(nodes);

  log_debug(LD_CIRC,"Chosen route length %d (%d/%d routers suitable).",
            routelen, num_acceptable_routers, smartlist_len(nodes));

  if (num_acceptable_routers < 2) {
    log_info(LD_CIRC,
             "Not enough acceptable routers (%d). Discarding this circuit.",
             num_acceptable_routers);
    return -1;
  }

  if (num_acceptable_routers < routelen) {
    log_info(LD_CIRC,"Not enough routers: cutting routelen from %d to %d.",
             routelen, num_acceptable_routers);
    routelen = num_acceptable_routers;
  }

  return routelen;
}

/** Return a newly allocated list of uint16_t * for each predicted port not
 * handled by a current circuit. */
static smartlist_t *
circuit_get_unhandled_ports(time_t now)
{
  smartlist_t *dest = rep_hist_get_predicted_ports(now);
  circuit_remove_handled_ports(dest);
  return dest;
}

/** Return 1 if we already have circuits present or on the way for
 * all anticipated ports. Return 0 if we should make more.
 *
 * If we're returning 0, set need_uptime and need_capacity to
 * indicate any requirements that the unhandled ports have.
 */
int
circuit_all_predicted_ports_handled(time_t now, int *need_uptime,
                                    int *need_capacity)
{
  int i, enough;
  uint16_t *port;
  smartlist_t *sl = circuit_get_unhandled_ports(now);
  smartlist_t *LongLivedServices = get_options()->LongLivedPorts;
  tor_assert(need_uptime);
  tor_assert(need_capacity);
  // Always predict need_capacity
  *need_capacity = 1;
  enough = (smartlist_len(sl) == 0);
  for (i = 0; i < smartlist_len(sl); ++i) {
    port = smartlist_get(sl, i);
    if (smartlist_string_num_isin(LongLivedServices, *port))
      *need_uptime = 1;
    tor_free(port);
  }
  smartlist_free(sl);
  return enough;
}

/** Return 1 if <b>node</b> can handle one or more of the ports in
 * <b>needed_ports</b>, else return 0.
 */
static int
node_handles_some_port(const node_t *node, smartlist_t *needed_ports)
{ /* XXXX MOVE */
  int i;
  uint16_t port;

  for (i = 0; i < smartlist_len(needed_ports); ++i) {
    addr_policy_result_t r;
    /* alignment issues aren't a worry for this dereference, since
       needed_ports is explicitly a smartlist of uint16_t's */
    port = *(uint16_t *)smartlist_get(needed_ports, i);
    tor_assert(port);
    if (node)
      r = compare_addr_to_node_policy(0, port, node);
    else
      continue;
    if (r != ADDR_POLICY_REJECTED && r != ADDR_POLICY_PROBABLY_REJECTED)
      return 1;
  }
  return 0;
}

/** Return true iff <b>conn</b> needs another general circuit to be
 * built. */
static int
ap_stream_wants_exit_attention(connection_t *conn)
{
  if (conn->type == CONN_TYPE_AP &&
      conn->state == AP_CONN_STATE_CIRCUIT_WAIT &&
      !conn->marked_for_close &&
      !(TO_EDGE_CONN(conn)->want_onehop) && /* ignore one-hop streams */
      !(TO_EDGE_CONN(conn)->use_begindir) && /* ignore targeted dir fetches */
      !(TO_EDGE_CONN(conn)->chosen_exit_name) && /* ignore defined streams */
      !connection_edge_is_rendezvous_stream(TO_EDGE_CONN(conn)) &&
      !circuit_stream_is_being_handled(TO_EDGE_CONN(conn), 0,
                                       MIN_CIRCUITS_HANDLING_STREAM))
    return 1;
  return 0;
}

/** Return a pointer to a suitable router to be the exit node for the
 * general-purpose circuit we're about to build.
 *
 * Look through the connection array, and choose a router that maximizes
 * the number of pending streams that can exit from this router.
 *
 * Return NULL if we can't find any suitable routers.
 */
static const node_t *
choose_good_exit_server_general(int need_uptime, int need_capacity)
{
  int *n_supported;
  int n_pending_connections = 0;
  smartlist_t *connections;
  int best_support = -1;
  int n_best_support=0;
  or_options_t *options = get_options();
  const smartlist_t *the_nodes;
  const node_t *node=NULL;

  connections = get_connection_array();

  /* Count how many connections are waiting for a circuit to be built.
   * We use this for log messages now, but in the future we may depend on it.
   */
  SMARTLIST_FOREACH(connections, connection_t *, conn,
  {
    if (ap_stream_wants_exit_attention(conn))
      ++n_pending_connections;
  });
//  log_fn(LOG_DEBUG, "Choosing exit node; %d connections are pending",
//         n_pending_connections);
  /* Now we count, for each of the routers in the directory, how many
   * of the pending connections could possibly exit from that
   * router (n_supported[i]). (We can't be sure about cases where we
   * don't know the IP address of the pending connection.)
   *
   * -1 means "Don't use this router at all."
   */
  the_nodes = nodelist_get_list();
  n_supported = tor_malloc(sizeof(int)*smartlist_len(the_nodes));
  SMARTLIST_FOREACH_BEGIN(the_nodes, const node_t *, node) {
    const int i = node_sl_idx;
    if (router_digest_is_me(node->identity)) {
      n_supported[i] = -1;
//      log_fn(LOG_DEBUG,"Skipping node %s -- it's me.", router->nickname);
      /* XXX there's probably a reverse predecessor attack here, but
       * it's slow. should we take this out? -RD
       */
      continue;
    }
    if (!node_has_descriptor(node))
      continue;
    if (!node->is_running || node->is_bad_exit) {
      n_supported[i] = -1;
      continue; /* skip routers that are known to be down or bad exits */
    }
    if (node_is_unreliable(node, need_uptime, need_capacity, 0) &&
        (!options->ExitNodes ||
         !routerset_contains_node(options->ExitNodes, node))) {
      /* FFFF Someday, differentiate between a routerset that names
       * routers, and a routerset that names countries, and only do this
       * check if they've asked for specific exit relays. Or if the country
       * they ask for is rare. Or something. */
      n_supported[i] = -1;
      continue; /* skip routers that are not suitable, unless we have
                 * ExitNodes set, in which case we asked for it */
    }
    if (!(node->is_valid || options->_AllowInvalid & ALLOW_INVALID_EXIT)) {
      /* if it's invalid and we don't want it */
      n_supported[i] = -1;
//      log_fn(LOG_DEBUG,"Skipping node %s (index %d) -- invalid router.",
//             router->nickname, i);
      continue; /* skip invalid routers */
    }
    if (options->ExcludeSingleHopRelays &&
        node_allows_single_hop_exits(node)) {
      n_supported[i] = -1;
      continue;
    }
    if (node_exit_policy_rejects_all(node)) {
      n_supported[i] = -1;
//      log_fn(LOG_DEBUG,"Skipping node %s (index %d) -- it rejects all.",
//             router->nickname, i);
      continue; /* skip routers that reject all */
    }
    n_supported[i] = 0;
    /* iterate over connections */
    SMARTLIST_FOREACH_BEGIN(connections, connection_t *, conn) {
      if (!ap_stream_wants_exit_attention(conn))
        continue; /* Skip everything but APs in CIRCUIT_WAIT */
      if (connection_ap_can_use_exit(TO_EDGE_CONN(conn), node, 1)) {
        ++n_supported[i];
//        log_fn(LOG_DEBUG,"%s is supported. n_supported[%d] now %d.",
//               router->nickname, i, n_supported[i]);
      } else {
//        log_fn(LOG_DEBUG,"%s (index %d) would reject this stream.",
//               router->nickname, i);
      }
    } SMARTLIST_FOREACH_END(conn);
    if (n_pending_connections > 0 && n_supported[i] == 0) {
      /* Leave best_support at -1 if that's where it is, so we can
       * distinguish it later. */
      continue;
    }
    if (n_supported[i] > best_support) {
      /* If this router is better than previous ones, remember its index
       * and goodness, and start counting how many routers are this good. */
      best_support = n_supported[i]; n_best_support=1;
//      log_fn(LOG_DEBUG,"%s is new best supported option so far.",
//             router->nickname);
    } else if (n_supported[i] == best_support) {
      /* If this router is _as good_ as the best one, just increment the
       * count of equally good routers.*/
      ++n_best_support;
    }
  } SMARTLIST_FOREACH_END(node);
  log_info(LD_CIRC,
           "Found %d servers that might support %d/%d pending connections.",
           n_best_support, best_support >= 0 ? best_support : 0,
           n_pending_connections);

  /* If any routers definitely support any pending connections, choose one
   * at random. */
  if (best_support > 0) {
    smartlist_t *supporting = smartlist_create(), *use = smartlist_create();

    SMARTLIST_FOREACH(the_nodes, const node_t *, node, {
      if (n_supported[node_sl_idx] == best_support)
        smartlist_add(supporting, (void*)node);
    });

    routersets_get_node_disjunction(use, supporting, options->ExitNodes,
                               options->_ExcludeExitNodesUnion, 1);
    if (smartlist_len(use) == 0 && options->ExitNodes &&
        !options->StrictNodes) { /* give up on exitnodes and try again */
      routersets_get_node_disjunction(use, supporting, NULL,
                                 options->_ExcludeExitNodesUnion, 1);
    }
    node = node_sl_choose_by_bandwidth(use, WEIGHT_FOR_EXIT);
    smartlist_free(use);
    smartlist_free(supporting);
  } else {
    /* Either there are no pending connections, or no routers even seem to
     * possibly support any of them.  Choose a router at random that satisfies
     * at least one predicted exit port. */

    int attempt;
    smartlist_t *needed_ports, *supporting, *use;

    if (best_support == -1) {
      if (need_uptime || need_capacity) {
        log_info(LD_CIRC,
                 "We couldn't find any live%s%s routers; falling back "
                 "to list of all routers.",
                 need_capacity?", fast":"",
                 need_uptime?", stable":"");
        tor_free(n_supported);
        return choose_good_exit_server_general(0, 0);
      }
      log_notice(LD_CIRC, "All routers are down or won't exit%s -- "
                 "choosing a doomed exit at random.",
                 options->_ExcludeExitNodesUnion ? " or are Excluded" : "");
    }
    supporting = smartlist_create();
    use = smartlist_create();
    needed_ports = circuit_get_unhandled_ports(time(NULL));
    for (attempt = 0; attempt < 2; attempt++) {
      /* try once to pick only from routers that satisfy a needed port,
       * then if there are none, pick from any that support exiting. */
      SMARTLIST_FOREACH_BEGIN(the_nodes, const node_t *, node) {
        if (!node_has_descriptor(node))
          continue;
        if (n_supported[node_sl_idx] != -1 &&
            (attempt || node_handles_some_port(node, needed_ports))) {
//          log_fn(LOG_DEBUG,"Try %d: '%s' is a possibility.",
//                 try, router->nickname);
          smartlist_add(supporting, (void*)node);
        }
      } SMARTLIST_FOREACH_END(node);

      routersets_get_node_disjunction(use, supporting, options->ExitNodes,
                                 options->_ExcludeExitNodesUnion, 1);
      if (smartlist_len(use) == 0 && options->ExitNodes &&
          !options->StrictNodes) { /* give up on exitnodes and try again */
        routersets_get_node_disjunction(use, supporting, NULL,
                                   options->_ExcludeExitNodesUnion, 1);
      }
      /* FFF sometimes the above results in null, when the requested
       * exit node is considered down by the consensus. we should pick
       * it anyway, since the user asked for it. */
      node = node_sl_choose_by_bandwidth(use, WEIGHT_FOR_EXIT);
      if (node)
        break;
      smartlist_clear(supporting);
      smartlist_clear(use);
    }
    SMARTLIST_FOREACH(needed_ports, uint16_t *, cp, tor_free(cp));
    smartlist_free(needed_ports);
    smartlist_free(use);
    smartlist_free(supporting);
  }

  tor_free(n_supported);
  if (node) {
    log_info(LD_CIRC, "Chose exit server '%s'", node_get_nickname(node));
    return node;
  }
  if (options->ExitNodes && options->StrictNodes) {
    log_warn(LD_CIRC,
             "No specified exit routers seem to be running, and "
             "StrictNodes is set: can't choose an exit.");
  }
  return NULL;
}

/** Return a pointer to a suitable router to be the exit node for the
 * circuit of purpose <b>purpose</b> that we're about to build (or NULL
 * if no router is suitable).
 *
 * For general-purpose circuits, pass it off to
 * choose_good_exit_server_general()
 *
 * For client-side rendezvous circuits, choose a random node, weighted
 * toward the preferences in 'options'.
 */
static const node_t *
choose_good_exit_server(uint8_t purpose,
                        int need_uptime, int need_capacity, int is_internal)
{
  or_options_t *options = get_options();
  router_crn_flags_t flags = CRN_NEED_DESC;
  if (need_uptime)
    flags |= CRN_NEED_UPTIME;
  if (need_capacity)
    flags |= CRN_NEED_CAPACITY;

  switch (purpose) {
    case CIRCUIT_PURPOSE_C_GENERAL:
      if (options->_AllowInvalid & ALLOW_INVALID_MIDDLE)
        flags |= CRN_ALLOW_INVALID;
      if (is_internal) /* pick it like a middle hop */
        return router_choose_random_node(NULL, options->ExcludeNodes, flags);
      else
        return choose_good_exit_server_general(need_uptime,need_capacity);
    case CIRCUIT_PURPOSE_C_ESTABLISH_REND:
      if (options->_AllowInvalid & ALLOW_INVALID_RENDEZVOUS)
        flags |= CRN_ALLOW_INVALID;
      return router_choose_random_node(NULL, options->ExcludeNodes, flags);
  }
  log_warn(LD_BUG,"Unhandled purpose %d", purpose);
  tor_fragile_assert();
  return NULL;
}

/** Log a warning if the user specified an exit for the circuit that
 * has been excluded from use by ExcludeNodes or ExcludeExitNodes. */
static void
warn_if_last_router_excluded(origin_circuit_t *circ, const extend_info_t *exit)
{
  or_options_t *options = get_options();
  routerset_t *rs = options->ExcludeNodes;
  const char *description;
  int domain = LD_CIRC;
  uint8_t purpose = circ->_base.purpose;

  if (circ->build_state->onehop_tunnel)
    return;

  switch (purpose)
    {
    default:
    case CIRCUIT_PURPOSE_OR:
    case CIRCUIT_PURPOSE_INTRO_POINT:
    case CIRCUIT_PURPOSE_REND_POINT_WAITING:
    case CIRCUIT_PURPOSE_REND_ESTABLISHED:
      log_warn(LD_BUG, "Called on non-origin circuit (purpose %d)",
               (int)purpose);
      return;
    case CIRCUIT_PURPOSE_C_GENERAL:
      if (circ->build_state->is_internal)
        return;
      description = "Requested exit node";
      rs = options->_ExcludeExitNodesUnion;
      break;
    case CIRCUIT_PURPOSE_C_INTRODUCING:
    case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT:
    case CIRCUIT_PURPOSE_C_INTRODUCE_ACKED:
    case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO:
    case CIRCUIT_PURPOSE_S_CONNECT_REND:
    case CIRCUIT_PURPOSE_S_REND_JOINED:
    case CIRCUIT_PURPOSE_TESTING:
      return;
    case CIRCUIT_PURPOSE_C_ESTABLISH_REND:
    case CIRCUIT_PURPOSE_C_REND_READY:
    case CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED:
    case CIRCUIT_PURPOSE_C_REND_JOINED:
      description = "Chosen rendezvous point";
      domain = LD_BUG;
      break;
    case CIRCUIT_PURPOSE_CONTROLLER:
      rs = options->_ExcludeExitNodesUnion;
      description = "Controller-selected circuit target";
      break;
    }

  if (routerset_contains_extendinfo(rs, exit)) {
    log_fn(LOG_WARN, domain, "%s '%s' is in ExcludeNodes%s. Using anyway "
           "(circuit purpose %d).",
           description,exit->nickname,
           rs==options->ExcludeNodes?"":" or ExcludeExitNodes",
           (int)purpose);
    circuit_log_path(LOG_WARN, domain, circ);
  }

  return;
}

/** Decide a suitable length for circ's cpath, and pick an exit
 * router (or use <b>exit</b> if provided). Store these in the
 * cpath. Return 0 if ok, -1 if circuit should be closed. */
static int
onion_pick_cpath_exit(origin_circuit_t *circ, extend_info_t *exit)
{
  cpath_build_state_t *state = circ->build_state;

  if (state->onehop_tunnel) {
    log_debug(LD_CIRC, "Launching a one-hop circuit for dir tunnel.");
    state->desired_path_len = 1;
  } else {
    int r = new_route_len(circ->_base.purpose, exit, nodelist_get_list());
    if (r < 1) /* must be at least 1 */
      return -1;
    state->desired_path_len = r;
  }

  if (exit) { /* the circuit-builder pre-requested one */
    warn_if_last_router_excluded(circ, exit);
    log_info(LD_CIRC,"Using requested exit node '%s'", exit->nickname);
    exit = extend_info_dup(exit);
  } else { /* we have to decide one */
    const node_t *node =
      choose_good_exit_server(circ->_base.purpose, state->need_uptime,
                              state->need_capacity, state->is_internal);
    if (!node) {
      log_warn(LD_CIRC,"failed to choose an exit server");
      return -1;
    }
    exit = extend_info_from_node(node);
    tor_assert(exit);
  }
  state->chosen_exit = exit;
  return 0;
}

/** Give <b>circ</b> a new exit destination to <b>exit</b>, and add a
 * hop to the cpath reflecting this. Don't send the next extend cell --
 * the caller will do this if it wants to.
 */
int
circuit_append_new_exit(origin_circuit_t *circ, extend_info_t *exit)
{
  cpath_build_state_t *state;
  tor_assert(exit);
  tor_assert(circ);

  state = circ->build_state;
  tor_assert(state);
  extend_info_free(state->chosen_exit);
  state->chosen_exit = extend_info_dup(exit);

  ++circ->build_state->desired_path_len;
  onion_append_hop(&circ->cpath, exit);
  return 0;
}

/** Take an open <b>circ</b>, and add a new hop at the end, based on
 * <b>info</b>. Set its state back to CIRCUIT_STATE_BUILDING, and then
 * send the next extend cell to begin connecting to that hop.
 */
int
circuit_extend_to_new_exit(origin_circuit_t *circ, extend_info_t *exit)
{
  int err_reason = 0;
  warn_if_last_router_excluded(circ, exit);
  circuit_append_new_exit(circ, exit);
  circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_BUILDING);
  if ((err_reason = circuit_send_next_onion_skin(circ))<0) {
    log_warn(LD_CIRC, "Couldn't extend circuit to new point '%s'.",
             exit->nickname);
    circuit_mark_for_close(TO_CIRCUIT(circ), -err_reason);
    return -1;
  }
  return 0;
}

/** Return the number of routers in <b>routers</b> that are currently up
 * and available for building circuits through.
 */
static int
count_acceptable_nodes(smartlist_t *nodes)
{
  int num=0;

  SMARTLIST_FOREACH_BEGIN(nodes, const node_t *, node) {
    //    log_debug(LD_CIRC,
//              "Contemplating whether router %d (%s) is a new option.",
//              i, r->nickname);
    if (! node->is_running)
//      log_debug(LD_CIRC,"Nope, the directory says %d is not running.",i);
      continue;
    if (! node->is_valid)
//      log_debug(LD_CIRC,"Nope, the directory says %d is not valid.",i);
      continue;
    if (! node_has_descriptor(node))
      continue;
      /* XXX This clause makes us count incorrectly: if AllowInvalidRouters
       * allows this node in some places, then we're getting an inaccurate
       * count. For now, be conservative and don't count it. But later we
       * should try to be smarter. */
    ++num;
  } SMARTLIST_FOREACH_END(node);

//    log_debug(LD_CIRC,"I like %d. num_acceptable_routers now %d.",i, num);

  return num;
}

/** Add <b>new_hop</b> to the end of the doubly-linked-list <b>head_ptr</b>.
 * This function is used to extend cpath by another hop.
 */
void
onion_append_to_cpath(crypt_path_t **head_ptr, crypt_path_t *new_hop)
{
  if (*head_ptr) {
    new_hop->next = (*head_ptr);
    new_hop->prev = (*head_ptr)->prev;
    (*head_ptr)->prev->next = new_hop;
    (*head_ptr)->prev = new_hop;
  } else {
    *head_ptr = new_hop;
    new_hop->prev = new_hop->next = new_hop;
  }
}

/** A helper function used by onion_extend_cpath(). Use <b>purpose</b>
 * and <b>state</b> and the cpath <b>head</b> (currently populated only
 * to length <b>cur_len</b> to decide a suitable middle hop for a
 * circuit. In particular, make sure we don't pick the exit node or its
 * family, and make sure we don't duplicate any previous nodes or their
 * families. */
static const node_t *
choose_good_middle_server(uint8_t purpose,
                          cpath_build_state_t *state,
                          crypt_path_t *head,
                          int cur_len)
{
  int i;
  const node_t *r, *choice;
  crypt_path_t *cpath;
  smartlist_t *excluded;
  or_options_t *options = get_options();
  router_crn_flags_t flags = CRN_NEED_DESC;
  tor_assert(_CIRCUIT_PURPOSE_MIN <= purpose &&
             purpose <= _CIRCUIT_PURPOSE_MAX);

  log_debug(LD_CIRC, "Contemplating intermediate hop: random choice.");
  excluded = smartlist_create();
  if ((r = build_state_get_exit_node(state))) {
    smartlist_add(excluded, (void*) r);
    nodelist_add_node_family(excluded, r);
  }
  for (i = 0, cpath = head; i < cur_len; ++i, cpath=cpath->next) {
    if ((r = node_get_by_id(cpath->extend_info->identity_digest))) {
      smartlist_add(excluded, (void*)r);
      nodelist_add_node_family(excluded, r);
    }
  }

  if (state->need_uptime)
    flags |= CRN_NEED_UPTIME;
  if (state->need_capacity)
    flags |= CRN_NEED_CAPACITY;
  if (options->_AllowInvalid & ALLOW_INVALID_MIDDLE)
    flags |= CRN_ALLOW_INVALID;
  choice = router_choose_random_node(excluded, options->ExcludeNodes, flags);
  smartlist_free(excluded);
  return choice;
}

/** Pick a good entry server for the circuit to be built according to
 * <b>state</b>.  Don't reuse a chosen exit (if any), don't use this
 * router (if we're an OR), and respect firewall settings; if we're
 * configured to use entry guards, return one.
 *
 * If <b>state</b> is NULL, we're choosing a router to serve as an entry
 * guard, not for any particular circuit.
 */
static const node_t *
choose_good_entry_server(uint8_t purpose, cpath_build_state_t *state)
{
  const node_t *choice;
  smartlist_t *excluded;
  or_options_t *options = get_options();
  router_crn_flags_t flags = CRN_NEED_GUARD|CRN_NEED_DESC;
  const node_t *node;

  if (state && options->UseEntryGuards &&
      (purpose != CIRCUIT_PURPOSE_TESTING || options->BridgeRelay)) {
    /* This is request for an entry server to use for a regular circuit,
     * and we use entry guard nodes.  Just return one of the guard nodes.  */
    return choose_random_entry(state);
  }

  excluded = smartlist_create();

  if (state && (node = build_state_get_exit_node(state))) {
    /* Exclude the exit node from the state, if we have one.  Also exclude its
     * family. */
    smartlist_add(excluded, (void*)node);
    nodelist_add_node_family(excluded, node);
  }
  if (firewall_is_fascist_or()) {
    /* Exclude all ORs that we can't reach through our firewall */
    smartlist_t *nodes = nodelist_get_list();
    SMARTLIST_FOREACH(nodes, const node_t *, node, {
      if (!fascist_firewall_allows_node(node))
        smartlist_add(excluded, (void*)node);
    });
  }
  /* and exclude current entry guards and their families, if applicable */
  if (options->UseEntryGuards && entry_guards) {
    SMARTLIST_FOREACH(entry_guards, entry_guard_t *, entry,
      {
        if ((node = node_get_by_id(entry->identity))) {
          smartlist_add(excluded, (void*)node);
          nodelist_add_node_family(excluded, node);
        }
      });
  }

  if (state) {
    if (state->need_uptime)
      flags |= CRN_NEED_UPTIME;
    if (state->need_capacity)
      flags |= CRN_NEED_CAPACITY;
  }
  if (options->_AllowInvalid & ALLOW_INVALID_ENTRY)
    flags |= CRN_ALLOW_INVALID;

  choice = router_choose_random_node(excluded, options->ExcludeNodes, flags);
  smartlist_free(excluded);
  return choice;
}

/** Return the first non-open hop in cpath, or return NULL if all
 * hops are open. */
static crypt_path_t *
onion_next_hop_in_cpath(crypt_path_t *cpath)
{
  crypt_path_t *hop = cpath;
  do {
    if (hop->state != CPATH_STATE_OPEN)
      return hop;
    hop = hop->next;
  } while (hop != cpath);
  return NULL;
}

/** Choose a suitable next hop in the cpath <b>head_ptr</b>,
 * based on <b>state</b>. Append the hop info to head_ptr.
 */
static int
onion_extend_cpath(origin_circuit_t *circ)
{
  uint8_t purpose = circ->_base.purpose;
  cpath_build_state_t *state = circ->build_state;
  int cur_len = circuit_get_cpath_len(circ);
  extend_info_t *info = NULL;

  if (cur_len >= state->desired_path_len) {
    log_debug(LD_CIRC, "Path is complete: %d steps long",
              state->desired_path_len);
    return 1;
  }

  log_debug(LD_CIRC, "Path is %d long; we want %d", cur_len,
            state->desired_path_len);

  if (cur_len == state->desired_path_len - 1) { /* Picking last node */
    info = extend_info_dup(state->chosen_exit);
  } else if (cur_len == 0) { /* picking first node */
    const node_t *r = choose_good_entry_server(purpose, state);
    if (r) {
      info = extend_info_from_node(r);
      tor_assert(info);
    }
  } else {
    const node_t *r =
      choose_good_middle_server(purpose, state, circ->cpath, cur_len);
    if (r) {
      info = extend_info_from_node(r);
      tor_assert(info);
    }
  }

  if (!info) {
    log_warn(LD_CIRC,"Failed to find node for hop %d of our path. Discarding "
             "this circuit.", cur_len);
    return -1;
  }

  log_debug(LD_CIRC,"Chose router %s for hop %d (exit is %s)",
            info->nickname, cur_len+1, build_state_get_exit_nickname(state));

  onion_append_hop(&circ->cpath, info);
  extend_info_free(info);
  return 0;
}

/** Create a new hop, annotate it with information about its
 * corresponding router <b>choice</b>, and append it to the
 * end of the cpath <b>head_ptr</b>. */
static int
onion_append_hop(crypt_path_t **head_ptr, extend_info_t *choice)
{
  crypt_path_t *hop = tor_malloc_zero(sizeof(crypt_path_t));

  /* link hop into the cpath, at the end. */
  onion_append_to_cpath(head_ptr, hop);

  hop->magic = CRYPT_PATH_MAGIC;
  hop->state = CPATH_STATE_CLOSED;

  hop->extend_info = extend_info_dup(choice);

  hop->package_window = circuit_initial_package_window();
  hop->deliver_window = CIRCWINDOW_START;

  return 0;
}

/** Allocate a new extend_info object based on the various arguments. */
extend_info_t *
extend_info_alloc(const char *nickname, const char *digest,
                  crypto_pk_env_t *onion_key,
                  const tor_addr_t *addr, uint16_t port)
{
  extend_info_t *info = tor_malloc_zero(sizeof(extend_info_t));
  memcpy(info->identity_digest, digest, DIGEST_LEN);
  if (nickname)
    strlcpy(info->nickname, nickname, sizeof(info->nickname));
  if (onion_key)
    info->onion_key = crypto_pk_dup_key(onion_key);
  tor_addr_copy(&info->addr, addr);
  info->port = port;
  return info;
}

/** Allocate and return a new extend_info_t that can be used to build a
 * circuit to or through the router <b>r</b>. */
extend_info_t *
extend_info_from_router(const routerinfo_t *r)
{
  tor_addr_t addr;
  tor_assert(r);
  tor_addr_from_ipv4h(&addr, r->addr);
  return extend_info_alloc(r->nickname, r->cache_info.identity_digest,
                           r->onion_pkey, &addr, r->or_port);
}

/** Allocate and return a new extend_info that can be used to build a ircuit
 * to or through the node <b>node</b>.  May return NULL if there is not
 * enough info about <b>node</b> to extend to it--for example, if there
 * is no routerinfo_t or microdesc_t.
 **/
extend_info_t *
extend_info_from_node(const node_t *node)
{
  if (node->ri) {
    return extend_info_from_router(node->ri);
  } else if (node->rs && node->md) {
    tor_addr_t addr;
    tor_addr_from_ipv4h(&addr, node->rs->addr);
    return extend_info_alloc(node->rs->nickname,
                             node->identity,
                             node->md->onion_pkey,
                             &addr,
                             node->rs->or_port);
  } else {
    return NULL;
  }
}

/** Release storage held by an extend_info_t struct. */
void
extend_info_free(extend_info_t *info)
{
  if (!info)
    return;
  crypto_free_pk_env(info->onion_key);
  tor_free(info);
}

/** Allocate and return a new extend_info_t with the same contents as
 * <b>info</b>. */
extend_info_t *
extend_info_dup(extend_info_t *info)
{
  extend_info_t *newinfo;
  tor_assert(info);
  newinfo = tor_malloc(sizeof(extend_info_t));
  memcpy(newinfo, info, sizeof(extend_info_t));
  if (info->onion_key)
    newinfo->onion_key = crypto_pk_dup_key(info->onion_key);
  else
    newinfo->onion_key = NULL;
  return newinfo;
}

/** Return the routerinfo_t for the chosen exit router in <b>state</b>.
 * If there is no chosen exit, or if we don't know the routerinfo_t for
 * the chosen exit, return NULL.
 */
const node_t *
build_state_get_exit_node(cpath_build_state_t *state)
{
  if (!state || !state->chosen_exit)
    return NULL;
  return node_get_by_id(state->chosen_exit->identity_digest);
}

/** Return the nickname for the chosen exit router in <b>state</b>. If
 * there is no chosen exit, or if we don't know the routerinfo_t for the
 * chosen exit, return NULL.
 */
const char *
build_state_get_exit_nickname(cpath_build_state_t *state)
{
  if (!state || !state->chosen_exit)
    return NULL;
  return state->chosen_exit->nickname;
}

/** Check whether the entry guard <b>e</b> is usable, given the directory
 * authorities' opinion about the router (stored in <b>ri</b>) and the user's
 * configuration (in <b>options</b>). Set <b>e</b>-&gt;bad_since
 * accordingly. Return true iff the entry guard's status changes.
 *
 * If it's not usable, set *<b>reason</b> to a static string explaining why.
 */
static int
entry_guard_set_status(entry_guard_t *e, const node_t *node,
                       time_t now, or_options_t *options, const char **reason)
{
  char buf[HEX_DIGEST_LEN+1];
  int changed = 0;

  *reason = NULL;

  /* Do we want to mark this guard as bad? */
  if (!node)
    *reason = "unlisted";
  else if (!node->is_running)
    *reason = "down";
  else if (options->UseBridges && (!node->ri ||
                                   node->ri->purpose != ROUTER_PURPOSE_BRIDGE))
    *reason = "not a bridge";
  else if (!options->UseBridges && !node->is_possible_guard &&
           !routerset_contains_node(options->EntryNodes,node))
    *reason = "not recommended as a guard";
  else if (routerset_contains_node(options->ExcludeNodes, node))
    *reason = "excluded";

  if (*reason && ! e->bad_since) {
    /* Router is newly bad. */
    base16_encode(buf, sizeof(buf), e->identity, DIGEST_LEN);
    log_info(LD_CIRC, "Entry guard %s (%s) is %s: marking as unusable.",
             e->nickname, buf, *reason);

    e->bad_since = now;
    control_event_guard(e->nickname, e->identity, "BAD");
    changed = 1;
  } else if (!*reason && e->bad_since) {
    /* There's nothing wrong with the router any more. */
    base16_encode(buf, sizeof(buf), e->identity, DIGEST_LEN);
    log_info(LD_CIRC, "Entry guard %s (%s) is no longer unusable: "
             "marking as ok.", e->nickname, buf);

    e->bad_since = 0;
    control_event_guard(e->nickname, e->identity, "GOOD");
    changed = 1;
  }
  return changed;
}

/** Return true iff enough time has passed since we last tried to connect
 * to the unreachable guard <b>e</b> that we're willing to try again. */
static int
entry_is_time_to_retry(entry_guard_t *e, time_t now)
{
  long diff;
  if (e->last_attempted < e->unreachable_since)
    return 1;
  diff = now - e->unreachable_since;
  if (diff < 6*60*60)
    return now > (e->last_attempted + 60*60);
  else if (diff < 3*24*60*60)
    return now > (e->last_attempted + 4*60*60);
  else if (diff < 7*24*60*60)
    return now > (e->last_attempted + 18*60*60);
  else
    return now > (e->last_attempted + 36*60*60);
}

/** Return the node corresponding to <b>e</b>, if <b>e</b> is
 * working well enough that we are willing to use it as an entry
 * right now. (Else return NULL.) In particular, it must be
 * - Listed as either up or never yet contacted;
 * - Present in the routerlist;
 * - Listed as 'stable' or 'fast' by the current dirserver consensus,
 *   if demanded by <b>need_uptime</b> or <b>need_capacity</b>
 *   (unless it's a configured EntryNode);
 * - Allowed by our current ReachableORAddresses config option; and
 * - Currently thought to be reachable by us (unless <b>assume_reachable</b>
 *   is true).
 *
 * If the answer is no, set *<b>msg</b> to an explanation of why.
 */
static INLINE const node_t *
entry_is_live(entry_guard_t *e, int need_uptime, int need_capacity,
              int assume_reachable, const char **msg)
{
  const node_t *node;
  or_options_t *options = get_options();
  tor_assert(msg);

  if (e->bad_since) {
    *msg = "bad";
    return NULL;
  }
  /* no good if it's unreachable, unless assume_unreachable or can_retry. */
  if (!assume_reachable && !e->can_retry &&
      e->unreachable_since && !entry_is_time_to_retry(e, time(NULL))) {
    *msg = "unreachable";
    return NULL;
  }
  node = node_get_by_id(e->identity);
  if (!node || !node_has_descriptor(node)) {
    *msg = "no descriptor";
    return NULL;
  }
  if (get_options()->UseBridges) {
    if (node_get_purpose(node) != ROUTER_PURPOSE_BRIDGE) {
      *msg = "not a bridge";
      return NULL;
    }
  } else { /* !get_options()->UseBridges */
    if (node_get_purpose(node) != ROUTER_PURPOSE_GENERAL) {
      *msg = "not general-purpose";
      return NULL;
    }
  }
  if (options->EntryNodes &&
      routerset_contains_node(options->EntryNodes, node)) {
    /* they asked for it, they get it */
    need_uptime = need_capacity = 0;
  }
  if (node_is_unreliable(node, need_uptime, need_capacity, 0)) {
    *msg = "not fast/stable";
    return NULL;
  }
  if (!fascist_firewall_allows_node(node)) {
    *msg = "unreachable by config";
    return NULL;
  }
  return node;
}

/** Return the number of entry guards that we think are usable. */
static int
num_live_entry_guards(void)
{
  int n = 0;
  const char *msg;
  if (! entry_guards)
    return 0;
  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, entry,
    {
      if (entry_is_live(entry, 0, 1, 0, &msg))
        ++n;
    });
  return n;
}

/** If <b>digest</b> matches the identity of any node in the
 * entry_guards list, return that node. Else return NULL. */
static INLINE entry_guard_t *
is_an_entry_guard(const char *digest)
{
  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, entry,
                    if (!memcmp(digest, entry->identity, DIGEST_LEN))
                      return entry;
                   );
  return NULL;
}

/** Dump a description of our list of entry guards to the log at level
 * <b>severity</b>. */
static void
log_entry_guards(int severity)
{
  smartlist_t *elements = smartlist_create();
  char *s;

  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e,
    {
      const char *msg = NULL;
      char *cp;
      if (entry_is_live(e, 0, 1, 0, &msg))
        tor_asprintf(&cp, "%s (up %s)",
                     e->nickname,
                     e->made_contact ? "made-contact" : "never-contacted");
      else
        tor_asprintf(&cp, "%s (%s, %s)",
                     e->nickname, msg,
                     e->made_contact ? "made-contact" : "never-contacted");
      smartlist_add(elements, cp);
    });

  s = smartlist_join_strings(elements, ",", 0, NULL);
  SMARTLIST_FOREACH(elements, char*, cp, tor_free(cp));
  smartlist_free(elements);
  log_fn(severity,LD_CIRC,"%s",s);
  tor_free(s);
}

/** Called when one or more guards that we would previously have used for some
 * purpose are no longer in use because a higher-priority guard has become
 * usable again. */
static void
control_event_guard_deferred(void)
{
  /* XXXX We don't actually have a good way to figure out _how many_ entries
   * are live for some purpose.  We need an entry_is_even_slightly_live()
   * function for this to work right.  NumEntryGuards isn't reliable: if we
   * need guards with weird properties, we can have more than that number
   * live.
   **/
#if 0
  int n = 0;
  const char *msg;
  or_options_t *options = get_options();
  if (!entry_guards)
    return;
  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, entry,
    {
      if (entry_is_live(entry, 0, 1, 0, &msg)) {
        if (n++ == options->NumEntryGuards) {
          control_event_guard(entry->nickname, entry->identity, "DEFERRED");
          return;
        }
      }
    });
#endif
}

/** Add a new (preferably stable and fast) router to our
 * entry_guards list. Return a pointer to the router if we succeed,
 * or NULL if we can't find any more suitable entries.
 *
 * If <b>chosen</b> is defined, use that one, and if it's not
 * already in our entry_guards list, put it at the *beginning*.
 * Else, put the one we pick at the end of the list. */
static const node_t *
add_an_entry_guard(const node_t *chosen, int reset_status)
{
  const node_t *node;
  entry_guard_t *entry;

  if (chosen) {
    node = chosen;
    entry = is_an_entry_guard(node->identity);
    if (entry) {
      if (reset_status) {
        entry->bad_since = 0;
        entry->can_retry = 1;
      }
      return NULL;
    }
  } else {
    node = choose_good_entry_server(CIRCUIT_PURPOSE_C_GENERAL, NULL);
    if (!node)
      return NULL;
  }
  entry = tor_malloc_zero(sizeof(entry_guard_t));
  log_info(LD_CIRC, "Chose '%s' as new entry guard.",
           node_get_nickname(node));
  strlcpy(entry->nickname, node_get_nickname(node), sizeof(entry->nickname));
  memcpy(entry->identity, node->identity, DIGEST_LEN);
  /* Choose expiry time smudged over the past month. The goal here
   * is to a) spread out when Tor clients rotate their guards, so they
   * don't all select them on the same day, and b) avoid leaving a
   * precise timestamp in the state file about when we first picked
   * this guard. For details, see the Jan 2010 or-dev thread. */
  entry->chosen_on_date = time(NULL) - crypto_rand_int(3600*24*30);
  entry->chosen_by_version = tor_strdup(VERSION);
  if (chosen) /* prepend */
    smartlist_insert(entry_guards, 0, entry);
  else /* append */
    smartlist_add(entry_guards, entry);
  control_event_guard(entry->nickname, entry->identity, "NEW");
  control_event_guard_deferred();
  log_entry_guards(LOG_INFO);
  return node;
}

/** If the use of entry guards is configured, choose more entry guards
 * until we have enough in the list. */
static void
pick_entry_guards(or_options_t *options)
{
  int changed = 0;

  tor_assert(entry_guards);

  while (num_live_entry_guards() < options->NumEntryGuards) {
    if (!add_an_entry_guard(NULL, 0))
      break;
    changed = 1;
  }
  if (changed)
    entry_guards_changed();
}

/** How long (in seconds) do we allow an entry guard to be nonfunctional,
 * unlisted, excluded, or otherwise nonusable before we give up on it? */
#define ENTRY_GUARD_REMOVE_AFTER (30*24*60*60)

/** Release all storage held by <b>e</b>. */
static void
entry_guard_free(entry_guard_t *e)
{
  if (!e)
    return;
  tor_free(e->chosen_by_version);
  tor_free(e);
}

/** Remove any entry guard which was selected by an unknown version of Tor,
 * or which was selected by a version of Tor that's known to select
 * entry guards badly. */
static int
remove_obsolete_entry_guards(time_t now)
{
  int changed = 0, i;

  for (i = 0; i < smartlist_len(entry_guards); ++i) {
    entry_guard_t *entry = smartlist_get(entry_guards, i);
    const char *ver = entry->chosen_by_version;
    const char *msg = NULL;
    tor_version_t v;
    int version_is_bad = 0, date_is_bad = 0;
    if (!ver) {
      msg = "does not say what version of Tor it was selected by";
      version_is_bad = 1;
    } else if (tor_version_parse(ver, &v)) {
      msg = "does not seem to be from any recognized version of Tor";
      version_is_bad = 1;
    } else {
      size_t len = strlen(ver)+5;
      char *tor_ver = tor_malloc(len);
      tor_snprintf(tor_ver, len, "Tor %s", ver);
      if ((tor_version_as_new_as(tor_ver, "0.1.0.10-alpha") &&
           !tor_version_as_new_as(tor_ver, "0.1.2.16-dev")) ||
          (tor_version_as_new_as(tor_ver, "0.2.0.0-alpha") &&
           !tor_version_as_new_as(tor_ver, "0.2.0.6-alpha")) ||
          /* above are bug 440; below are bug 1217 */
          (tor_version_as_new_as(tor_ver, "0.2.1.3-alpha") &&
           !tor_version_as_new_as(tor_ver, "0.2.1.23")) ||
          (tor_version_as_new_as(tor_ver, "0.2.2.0-alpha") &&
           !tor_version_as_new_as(tor_ver, "0.2.2.7-alpha"))) {
        msg = "was selected without regard for guard bandwidth";
        version_is_bad = 1;
      }
      tor_free(tor_ver);
    }
    if (!version_is_bad && entry->chosen_on_date + 3600*24*60 < now) {
      /* It's been 2 months since the date listed in our state file. */
      msg = "was selected several months ago";
      date_is_bad = 1;
    }

    if (version_is_bad || date_is_bad) { /* we need to drop it */
      char dbuf[HEX_DIGEST_LEN+1];
      tor_assert(msg);
      base16_encode(dbuf, sizeof(dbuf), entry->identity, DIGEST_LEN);
      log_fn(version_is_bad ? LOG_NOTICE : LOG_INFO, LD_CIRC,
             "Entry guard '%s' (%s) %s. (Version=%s.) Replacing it.",
             entry->nickname, dbuf, msg, ver?escaped(ver):"none");
      control_event_guard(entry->nickname, entry->identity, "DROPPED");
      entry_guard_free(entry);
      smartlist_del_keeporder(entry_guards, i--);
      log_entry_guards(LOG_INFO);
      changed = 1;
    }
  }

  return changed ? 1 : 0;
}

/** Remove all entry guards that have been down or unlisted for so
 * long that we don't think they'll come up again. Return 1 if we
 * removed any, or 0 if we did nothing. */
static int
remove_dead_entry_guards(time_t now)
{
  char dbuf[HEX_DIGEST_LEN+1];
  char tbuf[ISO_TIME_LEN+1];
  int i;
  int changed = 0;

  for (i = 0; i < smartlist_len(entry_guards); ) {
    entry_guard_t *entry = smartlist_get(entry_guards, i);
    if (entry->bad_since &&
        entry->bad_since + ENTRY_GUARD_REMOVE_AFTER < now) {

      base16_encode(dbuf, sizeof(dbuf), entry->identity, DIGEST_LEN);
      format_local_iso_time(tbuf, entry->bad_since);
      log_info(LD_CIRC, "Entry guard '%s' (%s) has been down or unlisted "
               "since %s local time; removing.",
               entry->nickname, dbuf, tbuf);
      control_event_guard(entry->nickname, entry->identity, "DROPPED");
      entry_guard_free(entry);
      smartlist_del_keeporder(entry_guards, i);
      log_entry_guards(LOG_INFO);
      changed = 1;
    } else
      ++i;
  }
  return changed ? 1 : 0;
}

/** A new directory or router-status has arrived; update the down/listed
 * status of the entry guards.
 *
 * An entry is 'down' if the directory lists it as nonrunning.
 * An entry is 'unlisted' if the directory doesn't include it.
 *
 * Don't call this on startup; only on a fresh download. Otherwise we'll
 * think that things are unlisted.
 */
void
entry_guards_compute_status(or_options_t *options, time_t now)
{
  int changed = 0;
  int severity = LOG_DEBUG;
  digestmap_t *reasons;

  if (! entry_guards)
    return;

  if (options->EntryNodes) /* reshuffle the entry guard list if needed */
    entry_nodes_should_be_added();

  reasons = digestmap_new();
  SMARTLIST_FOREACH_BEGIN(entry_guards, entry_guard_t *, entry)
    {
      const node_t *r = node_get_by_id(entry->identity);
      const char *reason = NULL;
      if (entry_guard_set_status(entry, r, now, options, &reason))
        changed = 1;

      if (entry->bad_since)
        tor_assert(reason);
      if (reason)
        digestmap_set(reasons, entry->identity, (char*)reason);
    }
  SMARTLIST_FOREACH_END(entry);

  if (remove_dead_entry_guards(now))
    changed = 1;

  severity = changed ? LOG_DEBUG : LOG_INFO;

  if (changed) {
    SMARTLIST_FOREACH_BEGIN(entry_guards, entry_guard_t *, entry) {
      const char *reason = digestmap_get(reasons, entry->identity);
      const char *live_msg = "";
      const node_t *r = entry_is_live(entry, 0, 1, 0, &live_msg);
      log_info(LD_CIRC, "Summary: Entry '%s' is %s, %s%s%s, and %s%s.",
               entry->nickname,
               entry->unreachable_since ? "unreachable" : "reachable",
               entry->bad_since ? "unusable" : "usable",
               reason ? ", ": "",
               reason ? reason : "",
               r ? "live" : "not live / ",
               r ? "" : live_msg);
    } SMARTLIST_FOREACH_END(entry);
    log_info(LD_CIRC, "    (%d/%d entry guards are usable/new)",
             num_live_entry_guards(), smartlist_len(entry_guards));
    log_entry_guards(LOG_INFO);
    entry_guards_changed();
  }

  digestmap_free(reasons, NULL);
}

/** Called when a connection to an OR with the identity digest <b>digest</b>
 * is established (<b>succeeded</b>==1) or has failed (<b>succeeded</b>==0).
 * If the OR is an entry, change that entry's up/down status.
 * Return 0 normally, or -1 if we want to tear down the new connection.
 *
 * If <b>mark_relay_status</b>, also call router_set_status() on this
 * relay.
 *
 * XXX022 change succeeded and mark_relay_status into 'int flags'.
 */
int
entry_guard_register_connect_status(const char *digest, int succeeded,
                                    int mark_relay_status, time_t now)
{
  int changed = 0;
  int refuse_conn = 0;
  int first_contact = 0;
  entry_guard_t *entry = NULL;
  int idx = -1;
  char buf[HEX_DIGEST_LEN+1];

  if (! entry_guards)
    return 0;

  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e,
    {
      if (!memcmp(e->identity, digest, DIGEST_LEN)) {
        entry = e;
        idx = e_sl_idx;
        break;
      }
    });

  if (!entry)
    return 0;

  base16_encode(buf, sizeof(buf), entry->identity, DIGEST_LEN);

  if (succeeded) {
    if (entry->unreachable_since) {
      log_info(LD_CIRC, "Entry guard '%s' (%s) is now reachable again. Good.",
               entry->nickname, buf);
      entry->can_retry = 0;
      entry->unreachable_since = 0;
      entry->last_attempted = now;
      control_event_guard(entry->nickname, entry->identity, "UP");
      changed = 1;
    }
    if (!entry->made_contact) {
      entry->made_contact = 1;
      first_contact = changed = 1;
    }
  } else { /* ! succeeded */
    if (!entry->made_contact) {
      /* We've never connected to this one. */
      log_info(LD_CIRC,
               "Connection to never-contacted entry guard '%s' (%s) failed. "
               "Removing from the list. %d/%d entry guards usable/new.",
               entry->nickname, buf,
               num_live_entry_guards()-1, smartlist_len(entry_guards)-1);
      control_event_guard(entry->nickname, entry->identity, "DROPPED");
      entry_guard_free(entry);
      smartlist_del_keeporder(entry_guards, idx);
      log_entry_guards(LOG_INFO);
      changed = 1;
    } else if (!entry->unreachable_since) {
      log_info(LD_CIRC, "Unable to connect to entry guard '%s' (%s). "
               "Marking as unreachable.", entry->nickname, buf);
      entry->unreachable_since = entry->last_attempted = now;
      control_event_guard(entry->nickname, entry->identity, "DOWN");
      changed = 1;
      entry->can_retry = 0; /* We gave it an early chance; no good. */
    } else {
      char tbuf[ISO_TIME_LEN+1];
      format_iso_time(tbuf, entry->unreachable_since);
      log_debug(LD_CIRC, "Failed to connect to unreachable entry guard "
                "'%s' (%s).  It has been unreachable since %s.",
                entry->nickname, buf, tbuf);
      entry->last_attempted = now;
      entry->can_retry = 0; /* We gave it an early chance; no good. */
    }
  }

  /* if the caller asked us to, also update the is_running flags for this
   * relay */
  if (mark_relay_status)
    router_set_status(digest, succeeded);

  if (first_contact) {
    /* We've just added a new long-term entry guard. Perhaps the network just
     * came back? We should give our earlier entries another try too,
     * and close this connection so we don't use it before we've given
     * the others a shot. */
    SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e, {
        if (e == entry)
          break;
        if (e->made_contact) {
          const char *msg;
          const node_t *r = entry_is_live(e, 0, 1, 1, &msg);
          if (r && e->unreachable_since) {
            refuse_conn = 1;
            e->can_retry = 1;
          }
        }
      });
    if (refuse_conn) {
      log_info(LD_CIRC,
               "Connected to new entry guard '%s' (%s). Marking earlier "
               "entry guards up. %d/%d entry guards usable/new.",
               entry->nickname, buf,
               num_live_entry_guards(), smartlist_len(entry_guards));
      log_entry_guards(LOG_INFO);
      changed = 1;
    }
  }

  if (changed)
    entry_guards_changed();
  return refuse_conn ? -1 : 0;
}

/** When we try to choose an entry guard, should we parse and add
 * config's EntryNodes first? */
static int should_add_entry_nodes = 0;

/** Called when the value of EntryNodes changes in our configuration. */
void
entry_nodes_should_be_added(void)
{
  log_info(LD_CIRC, "EntryNodes config option set. Putting configured "
           "relays at the front of the entry guard list.");
  should_add_entry_nodes = 1;
}

/** Add all nodes in EntryNodes that aren't currently guard nodes to the list
 * of guard nodes, at the front. */
static void
entry_guards_prepend_from_config(or_options_t *options)
{
  smartlist_t *entry_nodes, *entry_fps;
  smartlist_t *old_entry_guards_on_list, *old_entry_guards_not_on_list;
  tor_assert(entry_guards);

  should_add_entry_nodes = 0;

  if (!options->EntryNodes) {
    /* It's possible that a controller set EntryNodes, thus making
     * should_add_entry_nodes set, then cleared it again, all before the
     * call to choose_random_entry() that triggered us. If so, just return.
     */
    return;
  }

  {
    char *string = routerset_to_string(options->EntryNodes);
    log_info(LD_CIRC,"Adding configured EntryNodes '%s'.", string);
    tor_free(string);
  }

  entry_nodes = smartlist_create();
  entry_fps = smartlist_create();
  old_entry_guards_on_list = smartlist_create();
  old_entry_guards_not_on_list = smartlist_create();

  /* Split entry guards into those on the list and those not. */

  /* Now that we allow countries and IP ranges in EntryNodes, this is
   * potentially an enormous list. It's not so bad though because we
   * only call this function when a) we're making a new circuit, and b)
   * we've called directory_info_has_arrived() or changed our EntryNodes
   * since the last time we made a circuit. */
  routerset_get_all_nodes(entry_nodes, options->EntryNodes, 0);
  SMARTLIST_FOREACH(entry_nodes, const node_t *,node,
                    smartlist_add(entry_fps, (void*)node->identity));

  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e, {
    if (smartlist_digest_isin(entry_fps, e->identity))
      smartlist_add(old_entry_guards_on_list, e);
    else
      smartlist_add(old_entry_guards_not_on_list, e);
  });

  /* Remove all currently configured entry guards from entry_routers. */
  SMARTLIST_FOREACH(entry_nodes, const node_t *, node, {
    if (is_an_entry_guard(node->identity)) {
      SMARTLIST_DEL_CURRENT(entry_nodes, node);
    }
  });

  /* Now build the new entry_guards list. */
  smartlist_clear(entry_guards);
  /* First, the previously configured guards that are in EntryNodes. */
  smartlist_add_all(entry_guards, old_entry_guards_on_list);
  /* Next, the rest of EntryNodes */
  SMARTLIST_FOREACH(entry_nodes, const node_t *, node, {
    add_an_entry_guard(node, 0);
  });
  /* Finally, the remaining previously configured guards that are not in
   * EntryNodes, unless we're strict in which case we drop them */
  if (options->StrictNodes) {
    SMARTLIST_FOREACH(old_entry_guards_not_on_list, entry_guard_t *, e,
                      entry_guard_free(e));
  } else {
    smartlist_add_all(entry_guards, old_entry_guards_not_on_list);
  }

  smartlist_free(entry_nodes);
  smartlist_free(entry_fps);
  smartlist_free(old_entry_guards_on_list);
  smartlist_free(old_entry_guards_not_on_list);
  entry_guards_changed();
}

/** Return 0 if we're fine adding arbitrary routers out of the
 * directory to our entry guard list, or return 1 if we have a
 * list already and we'd prefer to stick to it.
 */
int
entry_list_is_constrained(or_options_t *options)
{
  if (options->EntryNodes)
    return 1;
  if (options->UseBridges)
    return 1;
  return 0;
}

/* Are we dead set against changing our entry guard list, or would we
 * change it if it means keeping Tor usable? */
static int
entry_list_is_totally_static(or_options_t *options)
{
  if (options->EntryNodes && options->StrictNodes)
    return 1;
  if (options->UseBridges)
    return 1;
  return 0;
}

/** Pick a live (up and listed) entry guard from entry_guards. If
 * <b>state</b> is non-NULL, this is for a specific circuit --
 * make sure not to pick this circuit's exit or any node in the
 * exit's family. If <b>state</b> is NULL, we're looking for a random
 * guard (likely a bridge). */
const node_t *
choose_random_entry(cpath_build_state_t *state)
{
  or_options_t *options = get_options();
  smartlist_t *live_entry_guards = smartlist_create();
  smartlist_t *exit_family = smartlist_create();
  const node_t *chosen_exit =
    state?build_state_get_exit_node(state) : NULL;
  const node_t *node = NULL;
  int need_uptime = state ? state->need_uptime : 0;
  int need_capacity = state ? state->need_capacity : 0;
  int preferred_min, consider_exit_family = 0;

  if (chosen_exit) {
    smartlist_add(exit_family, (void*) chosen_exit);
    nodelist_add_node_family(exit_family, chosen_exit);
    consider_exit_family = 1;
  }

  if (!entry_guards)
    entry_guards = smartlist_create();

  if (should_add_entry_nodes)
    entry_guards_prepend_from_config(options);

  if (!entry_list_is_constrained(options) &&
      smartlist_len(entry_guards) < options->NumEntryGuards)
    pick_entry_guards(options);

 retry:
  smartlist_clear(live_entry_guards);
  SMARTLIST_FOREACH_BEGIN(entry_guards, entry_guard_t *, entry) {
      const char *msg;
      node = entry_is_live(entry, need_uptime, need_capacity, 0, &msg);
      if (!node)
        continue; /* down, no point */
      if (consider_exit_family && smartlist_isin(exit_family, node))
        continue; /* avoid relays that are family members of our exit */
      if (options->EntryNodes &&
          !routerset_contains_node(options->EntryNodes, node)) {
        /* We've come to the end of our preferred entry nodes. */
        if (smartlist_len(live_entry_guards))
          goto choose_and_finish; /* only choose from the ones we like */
        if (options->StrictNodes) {
          /* in theory this case should never happen, since
           * entry_guards_prepend_from_config() drops unwanted relays */
          tor_fragile_assert();
        } else {
          log_info(LD_CIRC,
                   "No relays from EntryNodes available. Using others.");
        }
      }
      smartlist_add(live_entry_guards, (void*)node);
      if (!entry->made_contact) {
        /* Always start with the first not-yet-contacted entry
         * guard. Otherwise we might add several new ones, pick
         * the second new one, and now we've expanded our entry
         * guard list without needing to. */
        goto choose_and_finish;
      }
      if (smartlist_len(live_entry_guards) >= options->NumEntryGuards)
        break; /* we have enough */
  } SMARTLIST_FOREACH_END(entry);

  if (entry_list_is_constrained(options)) {
    /* If we prefer the entry nodes we've got, and we have at least
     * one choice, that's great. Use it. */
    preferred_min = 1;
  } else {
    /* Try to have at least 2 choices available. This way we don't
     * get stuck with a single live-but-crummy entry and just keep
     * using him.
     * (We might get 2 live-but-crummy entry guards, but so be it.) */
    preferred_min = 2;
  }

  if (smartlist_len(live_entry_guards) < preferred_min) {
    if (!entry_list_is_totally_static(options)) {
      /* still no? try adding a new entry then */
      /* XXX if guard doesn't imply fast and stable, then we need
       * to tell add_an_entry_guard below what we want, or it might
       * be a long time til we get it. -RD */
      node = add_an_entry_guard(NULL, 0);
      if (node) {
        entry_guards_changed();
        /* XXX we start over here in case the new node we added shares
         * a family with our exit node. There's a chance that we'll just
         * load up on entry guards here, if the network we're using is
         * one big family. Perhaps we should teach add_an_entry_guard()
         * to understand nodes-to-avoid-if-possible? -RD */
        goto retry;
      }
    }
    if (!node && need_uptime) {
      need_uptime = 0; /* try without that requirement */
      goto retry;
    }
    if (!node && need_capacity) {
      /* still no? last attempt, try without requiring capacity */
      need_capacity = 0;
      goto retry;
    }
    if (!node && entry_list_is_constrained(options) && consider_exit_family) {
      /* still no? if we're using bridges or have strictentrynodes
       * set, and our chosen exit is in the same family as all our
       * bridges/entry guards, then be flexible about families. */
      consider_exit_family = 0;
      goto retry;
    }
    /* live_entry_guards may be empty below. Oh well, we tried. */
  }

 choose_and_finish:
  if (entry_list_is_constrained(options)) {
    /* We need to weight by bandwidth, because our bridges or entryguards
     * were not already selected proportional to their bandwidth. */
    node = node_sl_choose_by_bandwidth(live_entry_guards, WEIGHT_FOR_GUARD);
  } else {
    /* We choose uniformly at random here, because choose_good_entry_server()
     * already weights its choices by bandwidth, so we don't want to
     * *double*-weight our guard selection. */
    node = smartlist_choose(live_entry_guards);
  }
  smartlist_free(live_entry_guards);
  smartlist_free(exit_family);
  return node;
}

/** Parse <b>state</b> and learn about the entry guards it describes.
 * If <b>set</b> is true, and there are no errors, replace the global
 * entry_list with what we find.
 * On success, return 0. On failure, alloc into *<b>msg</b> a string
 * describing the error, and return -1.
 */
int
entry_guards_parse_state(or_state_t *state, int set, char **msg)
{
  entry_guard_t *node = NULL;
  smartlist_t *new_entry_guards = smartlist_create();
  config_line_t *line;
  time_t now = time(NULL);
  const char *state_version = state->TorVersion;
  digestmap_t *added_by = digestmap_new();

  *msg = NULL;
  for (line = state->EntryGuards; line; line = line->next) {
    if (!strcasecmp(line->key, "EntryGuard")) {
      smartlist_t *args = smartlist_create();
      node = tor_malloc_zero(sizeof(entry_guard_t));
      /* all entry guards on disk have been contacted */
      node->made_contact = 1;
      smartlist_add(new_entry_guards, node);
      smartlist_split_string(args, line->value, " ",
                             SPLIT_SKIP_SPACE|SPLIT_IGNORE_BLANK, 0);
      if (smartlist_len(args)<2) {
        *msg = tor_strdup("Unable to parse entry nodes: "
                          "Too few arguments to EntryGuard");
      } else if (!is_legal_nickname(smartlist_get(args,0))) {
        *msg = tor_strdup("Unable to parse entry nodes: "
                          "Bad nickname for EntryGuard");
      } else {
        strlcpy(node->nickname, smartlist_get(args,0), MAX_NICKNAME_LEN+1);
        if (base16_decode(node->identity, DIGEST_LEN, smartlist_get(args,1),
                          strlen(smartlist_get(args,1)))<0) {
          *msg = tor_strdup("Unable to parse entry nodes: "
                            "Bad hex digest for EntryGuard");
        }
      }
      SMARTLIST_FOREACH(args, char*, cp, tor_free(cp));
      smartlist_free(args);
      if (*msg)
        break;
    } else if (!strcasecmp(line->key, "EntryGuardDownSince") ||
               !strcasecmp(line->key, "EntryGuardUnlistedSince")) {
      time_t when;
      time_t last_try = 0;
      if (!node) {
        *msg = tor_strdup("Unable to parse entry nodes: "
               "EntryGuardDownSince/UnlistedSince without EntryGuard");
        break;
      }
      if (parse_iso_time(line->value, &when)<0) {
        *msg = tor_strdup("Unable to parse entry nodes: "
                          "Bad time in EntryGuardDownSince/UnlistedSince");
        break;
      }
      if (when > now) {
        /* It's a bad idea to believe info in the future: you can wind
         * up with timeouts that aren't allowed to happen for years. */
        continue;
      }
      if (strlen(line->value) >= ISO_TIME_LEN+ISO_TIME_LEN+1) {
        /* ignore failure */
        (void) parse_iso_time(line->value+ISO_TIME_LEN+1, &last_try);
      }
      if (!strcasecmp(line->key, "EntryGuardDownSince")) {
        node->unreachable_since = when;
        node->last_attempted = last_try;
      } else {
        node->bad_since = when;
      }
    } else if (!strcasecmp(line->key, "EntryGuardAddedBy")) {
      char d[DIGEST_LEN];
      /* format is digest version date */
      if (strlen(line->value) < HEX_DIGEST_LEN+1+1+1+ISO_TIME_LEN) {
        log_warn(LD_BUG, "EntryGuardAddedBy line is not long enough.");
        continue;
      }
      if (base16_decode(d, sizeof(d), line->value, HEX_DIGEST_LEN)<0 ||
          line->value[HEX_DIGEST_LEN] != ' ') {
        log_warn(LD_BUG, "EntryGuardAddedBy line %s does not begin with "
                 "hex digest", escaped(line->value));
        continue;
      }
      digestmap_set(added_by, d, tor_strdup(line->value+HEX_DIGEST_LEN+1));
    } else {
      log_warn(LD_BUG, "Unexpected key %s", line->key);
    }
  }

  SMARTLIST_FOREACH(new_entry_guards, entry_guard_t *, e,
   {
     char *sp;
     char *val = digestmap_get(added_by, e->identity);
     if (val && (sp = strchr(val, ' '))) {
       time_t when;
       *sp++ = '\0';
       if (parse_iso_time(sp, &when)<0) {
         log_warn(LD_BUG, "Can't read time %s in EntryGuardAddedBy", sp);
       } else {
         e->chosen_by_version = tor_strdup(val);
         e->chosen_on_date = when;
       }
     } else {
       if (state_version) {
         e->chosen_by_version = tor_strdup(state_version);
         e->chosen_on_date = time(NULL) - crypto_rand_int(3600*24*30);
       }
     }
   });

  if (*msg || !set) {
    SMARTLIST_FOREACH(new_entry_guards, entry_guard_t *, e,
                      entry_guard_free(e));
    smartlist_free(new_entry_guards);
  } else { /* !err && set */
    if (entry_guards) {
      SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e,
                        entry_guard_free(e));
      smartlist_free(entry_guards);
    }
    entry_guards = new_entry_guards;
    entry_guards_dirty = 0;
    /* XXX022 hand new_entry_guards to this func, and move it up a
     * few lines, so we don't have to re-dirty it */
    if (remove_obsolete_entry_guards(now))
      entry_guards_dirty = 1;
  }
  digestmap_free(added_by, _tor_free);
  return *msg ? -1 : 0;
}

/** Our list of entry guards has changed, or some element of one
 * of our entry guards has changed. Write the changes to disk within
 * the next few minutes.
 */
static void
entry_guards_changed(void)
{
  time_t when;
  entry_guards_dirty = 1;

  /* or_state_save() will call entry_guards_update_state(). */
  when = get_options()->AvoidDiskWrites ? time(NULL) + 3600 : time(NULL)+600;
  or_state_mark_dirty(get_or_state(), when);
}

/** If the entry guard info has not changed, do nothing and return.
 * Otherwise, free the EntryGuards piece of <b>state</b> and create
 * a new one out of the global entry_guards list, and then mark
 * <b>state</b> dirty so it will get saved to disk.
 */
void
entry_guards_update_state(or_state_t *state)
{
  config_line_t **next, *line;
  if (! entry_guards_dirty)
    return;

  config_free_lines(state->EntryGuards);
  next = &state->EntryGuards;
  *next = NULL;
  if (!entry_guards)
    entry_guards = smartlist_create();
  SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e,
    {
      char dbuf[HEX_DIGEST_LEN+1];
      if (!e->made_contact)
        continue; /* don't write this one to disk */
      *next = line = tor_malloc_zero(sizeof(config_line_t));
      line->key = tor_strdup("EntryGuard");
      line->value = tor_malloc(HEX_DIGEST_LEN+MAX_NICKNAME_LEN+2);
      base16_encode(dbuf, sizeof(dbuf), e->identity, DIGEST_LEN);
      tor_snprintf(line->value,HEX_DIGEST_LEN+MAX_NICKNAME_LEN+2,
                   "%s %s", e->nickname, dbuf);
      next = &(line->next);
      if (e->unreachable_since) {
        *next = line = tor_malloc_zero(sizeof(config_line_t));
        line->key = tor_strdup("EntryGuardDownSince");
        line->value = tor_malloc(ISO_TIME_LEN+1+ISO_TIME_LEN+1);
        format_iso_time(line->value, e->unreachable_since);
        if (e->last_attempted) {
          line->value[ISO_TIME_LEN] = ' ';
          format_iso_time(line->value+ISO_TIME_LEN+1, e->last_attempted);
        }
        next = &(line->next);
      }
      if (e->bad_since) {
        *next = line = tor_malloc_zero(sizeof(config_line_t));
        line->key = tor_strdup("EntryGuardUnlistedSince");
        line->value = tor_malloc(ISO_TIME_LEN+1);
        format_iso_time(line->value, e->bad_since);
        next = &(line->next);
      }
      if (e->chosen_on_date && e->chosen_by_version &&
          !strchr(e->chosen_by_version, ' ')) {
        char d[HEX_DIGEST_LEN+1];
        char t[ISO_TIME_LEN+1];
        size_t val_len;
        *next = line = tor_malloc_zero(sizeof(config_line_t));
        line->key = tor_strdup("EntryGuardAddedBy");
        val_len = (HEX_DIGEST_LEN+1+strlen(e->chosen_by_version)
                   +1+ISO_TIME_LEN+1);
        line->value = tor_malloc(val_len);
        base16_encode(d, sizeof(d), e->identity, DIGEST_LEN);
        format_iso_time(t, e->chosen_on_date);
        tor_snprintf(line->value, val_len, "%s %s %s",
                     d, e->chosen_by_version, t);
        next = &(line->next);
      }
    });
  if (!get_options()->AvoidDiskWrites)
    or_state_mark_dirty(get_or_state(), 0);
  entry_guards_dirty = 0;
}

/** If <b>question</b> is the string "entry-guards", then dump
 * to *<b>answer</b> a newly allocated string describing all of
 * the nodes in the global entry_guards list. See control-spec.txt
 * for details.
 * For backward compatibility, we also handle the string "helper-nodes".
 * */
int
getinfo_helper_entry_guards(control_connection_t *conn,
                            const char *question, char **answer,
                            const char **errmsg)
{
  (void) conn;
  (void) errmsg;

  if (!strcmp(question,"entry-guards") ||
      !strcmp(question,"helper-nodes")) {
    smartlist_t *sl = smartlist_create();
    char tbuf[ISO_TIME_LEN+1];
    char nbuf[MAX_VERBOSE_NICKNAME_LEN+1];
    if (!entry_guards)
      entry_guards = smartlist_create();
    SMARTLIST_FOREACH_BEGIN(entry_guards, entry_guard_t *, e) {
        size_t len = MAX_VERBOSE_NICKNAME_LEN+ISO_TIME_LEN+32;
        char *c = tor_malloc(len);
        const char *status = NULL;
        time_t when = 0;
        const node_t *node;

        if (!e->made_contact) {
          status = "never-connected";
        } else if (e->bad_since) {
          when = e->bad_since;
          status = "unusable";
        } else {
          status = "up";
        }

        node = node_get_by_id(e->identity);
        if (node) {
          node_get_verbose_nickname(node, nbuf);
        } else {
          nbuf[0] = '$';
          base16_encode(nbuf+1, sizeof(nbuf)-1, e->identity, DIGEST_LEN);
          /* e->nickname field is not very reliable if we don't know about
           * this router any longer; don't include it. */
        }

        if (when) {
          format_iso_time(tbuf, when);
          tor_snprintf(c, len, "%s %s %s\n", nbuf, status, tbuf);
        } else {
          tor_snprintf(c, len, "%s %s\n", nbuf, status);
        }
        smartlist_add(sl, c);
    } SMARTLIST_FOREACH_END(e);
    *answer = smartlist_join_strings(sl, "", 0, NULL);
    SMARTLIST_FOREACH(sl, char *, c, tor_free(c));
    smartlist_free(sl);
  }
  return 0;
}

/** Information about a configured bridge. Currently this just matches the
 * ones in the torrc file, but one day we may be able to learn about new
 * bridges on our own, and remember them in the state file. */
typedef struct {
  /** Address of the bridge. */
  tor_addr_t addr;
  /** TLS port for the bridge. */
  uint16_t port;
  /** Expected identity digest, or all zero bytes if we don't know what the
   * digest should be. */
  char identity[DIGEST_LEN];
  /** When should we next try to fetch a descriptor for this bridge? */
  download_status_t fetch_status;
} bridge_info_t;

/** A list of configured bridges. Whenever we actually get a descriptor
 * for one, we add it as an entry guard. */
static smartlist_t *bridge_list = NULL;

/** Initialize the bridge list to empty, creating it if needed. */
void
clear_bridge_list(void)
{
  if (!bridge_list)
    bridge_list = smartlist_create();
  SMARTLIST_FOREACH(bridge_list, bridge_info_t *, b, tor_free(b));
  smartlist_clear(bridge_list);
}

/** Return a bridge pointer if <b>ri</b> is one of our known bridges
 * (either by comparing keys if possible, else by comparing addr/port).
 * Else return NULL. */
static bridge_info_t *
get_configured_bridge_by_addr_port_digest(tor_addr_t *addr, uint16_t port,
                                          const char *digest)
{
  if (!bridge_list)
    return NULL;
  SMARTLIST_FOREACH_BEGIN(bridge_list, bridge_info_t *, bridge)
    {
      if (tor_digest_is_zero(bridge->identity) &&
          !tor_addr_compare(&bridge->addr, addr, CMP_EXACT) &&
          bridge->port == port)
        return bridge;
      if (!memcmp(bridge->identity, digest, DIGEST_LEN))
        return bridge;
    }
  SMARTLIST_FOREACH_END(bridge);
  return NULL;
}

/** Wrapper around get_configured_bridge_by_addr_port_digest() to look
 * it up via router descriptor <b>ri</b>. */
static bridge_info_t *
get_configured_bridge_by_routerinfo(const routerinfo_t *ri)
{
  tor_addr_t addr;
  tor_addr_from_ipv4h(&addr, ri->addr);
  return get_configured_bridge_by_addr_port_digest(&addr,
                              ri->or_port, ri->cache_info.identity_digest);
}

/** Return 1 if <b>ri</b> is one of our known bridges, else 0. */
int
routerinfo_is_a_configured_bridge(const routerinfo_t *ri)
{
  return get_configured_bridge_by_routerinfo(ri) ? 1 : 0;
}

/** We made a connection to a router at <b>addr</b>:<b>port</b>
 * without knowing its digest. Its digest turned out to be <b>digest</b>.
 * If it was a bridge, and we still don't know its digest, record it.
 */
void
learned_router_identity(tor_addr_t *addr, uint16_t port, const char *digest)
{
  bridge_info_t *bridge =
    get_configured_bridge_by_addr_port_digest(addr, port, digest);
  if (bridge && tor_digest_is_zero(bridge->identity)) {
    memcpy(bridge->identity, digest, DIGEST_LEN);
    log_notice(LD_DIR, "Learned fingerprint %s for bridge %s:%d",
               hex_str(digest, DIGEST_LEN), fmt_addr(addr), port);
  }
}

/** Remember a new bridge at <b>addr</b>:<b>port</b>. If <b>digest</b>
 * is set, it tells us the identity key too. */
void
bridge_add_from_config(const tor_addr_t *addr, uint16_t port, char *digest)
{
  bridge_info_t *b = tor_malloc_zero(sizeof(bridge_info_t));
  tor_addr_copy(&b->addr, addr);
  b->port = port;
  if (digest)
    memcpy(b->identity, digest, DIGEST_LEN);
  b->fetch_status.schedule = DL_SCHED_BRIDGE;
  if (!bridge_list)
    bridge_list = smartlist_create();
  smartlist_add(bridge_list, b);
}

/** If <b>digest</b> is one of our known bridges, return it. */
static bridge_info_t *
find_bridge_by_digest(const char *digest)
{
  SMARTLIST_FOREACH(bridge_list, bridge_info_t *, bridge,
    {
      if (!memcmp(bridge->identity, digest, DIGEST_LEN))
        return bridge;
    });
  return NULL;
}

/** We need to ask <b>bridge</b> for its server descriptor. <b>address</b>
 * is a helpful string describing this bridge. */
static void
launch_direct_bridge_descriptor_fetch(bridge_info_t *bridge)
{
  char *address;

  if (connection_get_by_type_addr_port_purpose(
      CONN_TYPE_DIR, &bridge->addr, bridge->port,
      DIR_PURPOSE_FETCH_SERVERDESC))
    return; /* it's already on the way */

  address = tor_dup_addr(&bridge->addr);
  directory_initiate_command(address, &bridge->addr,
                             bridge->port, 0,
                             0, /* does not matter */
                             1, bridge->identity,
                             DIR_PURPOSE_FETCH_SERVERDESC,
                             ROUTER_PURPOSE_BRIDGE,
                             0, "authority.z", NULL, 0, 0);
  tor_free(address);
}

/** Fetching the bridge descriptor from the bridge authority returned a
 * "not found". Fall back to trying a direct fetch. */
void
retry_bridge_descriptor_fetch_directly(const char *digest)
{
  bridge_info_t *bridge = find_bridge_by_digest(digest);
  if (!bridge)
    return; /* not found? oh well. */

  launch_direct_bridge_descriptor_fetch(bridge);
}

/** For each bridge in our list for which we don't currently have a
 * descriptor, fetch a new copy of its descriptor -- either directly
 * from the bridge or via a bridge authority. */
void
fetch_bridge_descriptors(or_options_t *options, time_t now)
{
  int num_bridge_auths = get_n_authorities(BRIDGE_AUTHORITY);
  int ask_bridge_directly;
  int can_use_bridge_authority;

  if (!bridge_list)
    return;

  SMARTLIST_FOREACH_BEGIN(bridge_list, bridge_info_t *, bridge)
    {
      if (!download_status_is_ready(&bridge->fetch_status, now,
                                    IMPOSSIBLE_TO_DOWNLOAD))
        continue; /* don't bother, no need to retry yet */

      /* schedule another fetch as if this one will fail, in case it does */
      download_status_failed(&bridge->fetch_status, 0);

      can_use_bridge_authority = !tor_digest_is_zero(bridge->identity) &&
                                 num_bridge_auths;
      ask_bridge_directly = !can_use_bridge_authority ||
                            !options->UpdateBridgesFromAuthority;
      log_debug(LD_DIR, "ask_bridge_directly=%d (%d, %d, %d)",
                ask_bridge_directly, tor_digest_is_zero(bridge->identity),
                !options->UpdateBridgesFromAuthority, !num_bridge_auths);

      if (ask_bridge_directly &&
          !fascist_firewall_allows_address_or(&bridge->addr, bridge->port)) {
        log_notice(LD_DIR, "Bridge at '%s:%d' isn't reachable by our "
                   "firewall policy. %s.", fmt_addr(&bridge->addr),
                   bridge->port,
                   can_use_bridge_authority ?
                     "Asking bridge authority instead" : "Skipping");
        if (can_use_bridge_authority)
          ask_bridge_directly = 0;
        else
          continue;
      }

      if (ask_bridge_directly) {
        /* we need to ask the bridge itself for its descriptor. */
        launch_direct_bridge_descriptor_fetch(bridge);
      } else {
        /* We have a digest and we want to ask an authority. We could
         * combine all the requests into one, but that may give more
         * hints to the bridge authority than we want to give. */
        char resource[10 + HEX_DIGEST_LEN];
        memcpy(resource, "fp/", 3);
        base16_encode(resource+3, HEX_DIGEST_LEN+1,
                      bridge->identity, DIGEST_LEN);
        memcpy(resource+3+HEX_DIGEST_LEN, ".z", 3);
        log_info(LD_DIR, "Fetching bridge info '%s' from bridge authority.",
                 resource);
        directory_get_from_dirserver(DIR_PURPOSE_FETCH_SERVERDESC,
                ROUTER_PURPOSE_BRIDGE, resource, 0);
      }
    }
  SMARTLIST_FOREACH_END(bridge);
}

/** We just learned a descriptor for a bridge. See if that
 * digest is in our entry guard list, and add it if not. */
void
learned_bridge_descriptor(routerinfo_t *ri, int from_cache)
{
  tor_assert(ri);
  tor_assert(ri->purpose == ROUTER_PURPOSE_BRIDGE);
  if (get_options()->UseBridges) {
    int first = !any_bridge_descriptors_known();
    bridge_info_t *bridge = get_configured_bridge_by_routerinfo(ri);
    time_t now = time(NULL);
    router_set_status(ri->cache_info.identity_digest, 1);

    if (bridge) { /* if we actually want to use this one */
      const node_t *node;
      /* it's here; schedule its re-fetch for a long time from now. */
      if (!from_cache)
        download_status_reset(&bridge->fetch_status);

      node = node_get_by_id(ri->cache_info.identity_digest);
      tor_assert(node);
      add_an_entry_guard(node, 1);
      log_notice(LD_DIR, "new bridge descriptor '%s' (%s)", ri->nickname,
                 from_cache ? "cached" : "fresh");
      /* set entry->made_contact so if it goes down we don't drop it from
       * our entry node list */
      entry_guard_register_connect_status(ri->cache_info.identity_digest,
                                          1, 0, now);
      if (first)
        routerlist_retry_directory_downloads(now);
    }
  }
}

/** Return 1 if any of our entry guards have descriptors that
 * are marked with purpose 'bridge' and are running. Else return 0.
 *
 * We use this function to decide if we're ready to start building
 * circuits through our bridges, or if we need to wait until the
 * directory "server/authority" requests finish. */
int
any_bridge_descriptors_known(void)
{
  tor_assert(get_options()->UseBridges);
  return choose_random_entry(NULL)!=NULL ? 1 : 0;
}

/** Return 1 if there are any directory conns fetching bridge descriptors
 * that aren't marked for close. We use this to guess if we should tell
 * the controller that we have a problem. */
int
any_pending_bridge_descriptor_fetches(void)
{
  smartlist_t *conns = get_connection_array();
  SMARTLIST_FOREACH(conns, connection_t *, conn,
  {
    if (conn->type == CONN_TYPE_DIR &&
        conn->purpose == DIR_PURPOSE_FETCH_SERVERDESC &&
        TO_DIR_CONN(conn)->router_purpose == ROUTER_PURPOSE_BRIDGE &&
        !conn->marked_for_close &&
        conn->linked && !conn->linked_conn->marked_for_close) {
      log_debug(LD_DIR, "found one: %s", conn->address);
      return 1;
    }
  });
  return 0;
}

/** Return 1 if we have at least one descriptor for an entry guard
 * (bridge or member of EntryNodes) and all descriptors we know are
 * down. Else return 0. If <b>act</b> is 1, then mark the down guards
 * up; else just observe and report. */
static int
entries_retry_helper(or_options_t *options, int act)
{
  const node_t *node;
  int any_known = 0;
  int any_running = 0;
  int need_bridges = options->UseBridges != 0;
  if (!entry_guards)
    entry_guards = smartlist_create();
  SMARTLIST_FOREACH_BEGIN(entry_guards, entry_guard_t *, e) {
      node = node_get_by_id(e->identity);
      if (node && node_has_descriptor(node) &&
          node_is_bridge(node) == need_bridges) {
        any_known = 1;
        if (node->is_running)
          any_running = 1; /* some entry is both known and running */
        else if (act) {
          /* Mark all current connections to this OR as unhealthy, since
           * otherwise there could be one that started 30 seconds
           * ago, and in 30 seconds it will time out, causing us to mark
           * the node down and undermine the retry attempt. We mark even
           * the established conns, since if the network just came back
           * we'll want to attach circuits to fresh conns. */
          connection_or_set_bad_connections(node->identity, 1);

          /* mark this entry node for retry */
          router_set_status(node->identity, 1);
          e->can_retry = 1;
          e->bad_since = 0;
        }
      }
  } SMARTLIST_FOREACH_END(e);
  log_debug(LD_DIR, "%d: any_known %d, any_running %d",
            act, any_known, any_running);
  return any_known && !any_running;
}

/** Do we know any descriptors for our bridges / entrynodes, and are
 * all the ones we have descriptors for down? */
int
entries_known_but_down(or_options_t *options)
{
  tor_assert(entry_list_is_constrained(options));
  return entries_retry_helper(options, 0);
}

/** Mark all down known bridges / entrynodes up. */
void
entries_retry_all(or_options_t *options)
{
  tor_assert(entry_list_is_constrained(options));
  entries_retry_helper(options, 1);
}

/** Release all storage held by the list of entry guards and related
 * memory structs. */
void
entry_guards_free_all(void)
{
  if (entry_guards) {
    SMARTLIST_FOREACH(entry_guards, entry_guard_t *, e,
                      entry_guard_free(e));
    smartlist_free(entry_guards);
    entry_guards = NULL;
  }
  clear_bridge_list();
  smartlist_free(bridge_list);
  bridge_list = NULL;
}