Merge remote-tracking branch 'andrea/ticket18640_v3'

12 files changed, 840 insertions, 12 deletions

diff --git a/src/or/config.c b/src/or/config.c
index 541025de16..31bf81877d 100644
--- a/src/or/config.c
+++ b/src/or/config.c
@@ -211,6 +211,7 @@ static config_var_t option_vars_[] = {
   V(CountPrivateBandwidth,       BOOL,     "0"),
   V(DataDirectory,               FILENAME, NULL),
   V(DataDirectoryGroupReadable,  BOOL,     "0"),
+  V(DisableOOSCheck,             BOOL,     "1"),
   V(DisableNetwork,              BOOL,     "0"),
   V(DirAllowPrivateAddresses,    BOOL,     "0"),
   V(TestingAuthDirTimeToLearnReachability, INTERVAL, "30 minutes"),
@@ -1374,6 +1375,35 @@ options_act_reversible(const or_options_t *old_options, char **msg)
       connection_mark_for_close(conn);
     }
   });
+
+  if (set_conn_limit) {
+    /*
+     * If we adjusted the conn limit, recompute the OOS threshold too
+     *
+     * How many possible sockets to keep in reserve?  If we have lots of
+     * possible sockets, keep this below a limit and set ConnLimit_high_thresh
+     * very close to ConnLimit_, but if ConnLimit_ is low, shrink it in
+     * proportion.
+     *
+     * Somewhat arbitrarily, set socks_in_reserve to 5% of ConnLimit_, but
+     * cap it at 64.
+     */
+    int socks_in_reserve = options->ConnLimit_ / 20;
+    if (socks_in_reserve > 64) socks_in_reserve = 64;
+
+    options->ConnLimit_high_thresh = options->ConnLimit_ - socks_in_reserve;
+    options->ConnLimit_low_thresh = (options->ConnLimit_ / 4) * 3;
+    log_info(LD_GENERAL,
+             "Recomputed OOS thresholds: ConnLimit %d, ConnLimit_ %d, "
+             "ConnLimit_high_thresh %d, ConnLimit_low_thresh %d",
+             options->ConnLimit, options->ConnLimit_,
+             options->ConnLimit_high_thresh,
+             options->ConnLimit_low_thresh);
+
+    /* Give the OOS handler a chance with the new thresholds */
+    connection_check_oos(get_n_open_sockets(), 0);
+  }
+
   goto done;
 
  rollback:
diff --git a/src/or/connection.c b/src/or/connection.c
index 68e442df54..9b583f4243 100644
--- a/src/or/connection.c
+++ b/src/or/connection.c
@@ -754,9 +754,9 @@ connection_mark_for_close_(connection_t *conn, int line, const char *file)
  * For all other cases, use connection_mark_and_flush() instead, which
  * checks for or_connection_t properly, instead.  See below.
  */
-void
-connection_mark_for_close_internal_(connection_t *conn,
-                                    int line, const char *file)
+MOCK_IMPL(void,
+connection_mark_for_close_internal_, (connection_t *conn,
+                                      int line, const char *file))
 {
   assert_connection_ok(conn,0);
   tor_assert(line);
@@ -1090,6 +1090,7 @@ connection_listener_new(const struct sockaddr *listensockaddr,
   int start_reading = 0;
   static int global_next_session_group = SESSION_GROUP_FIRST_AUTO;
   tor_addr_t addr;
+  int exhaustion = 0;
 
   if (listensockaddr->sa_family == AF_INET ||
       listensockaddr->sa_family == AF_INET6) {
@@ -1108,6 +1109,11 @@ connection_listener_new(const struct sockaddr *listensockaddr,
       int e = tor_socket_errno(s);
       if (ERRNO_IS_RESOURCE_LIMIT(e)) {
         warn_too_many_conns();
+        /*
+         * We'll call the OOS handler at the error exit, so set the
+         * exhaustion flag for it.
+         */
+        exhaustion = 1;
       } else {
         log_warn(LD_NET, "Socket creation failed: %s",
                  tor_socket_strerror(e));
@@ -1226,6 +1232,11 @@ connection_listener_new(const struct sockaddr *listensockaddr,
       int e = tor_socket_errno(s);
       if (ERRNO_IS_RESOURCE_LIMIT(e)) {
         warn_too_many_conns();
+        /*
+         * We'll call the OOS handler at the error exit, so set the
+         * exhaustion flag for it.
+         */
+        exhaustion = 1;
       } else {
         log_warn(LD_NET,"Socket creation failed: %s.", strerror(e));
       }
@@ -1344,6 +1355,12 @@ connection_listener_new(const struct sockaddr *listensockaddr,
     dnsserv_configure_listener(conn);
   }
 
+  /*
+   * Normal exit; call the OOS handler since connection count just changed;
+   * the exhaustion flag will always be zero here though.
+   */
+  connection_check_oos(get_n_open_sockets(), 0);
+
   return conn;
 
  err:
@@ -1352,6 +1369,9 @@ connection_listener_new(const struct sockaddr *listensockaddr,
   if (conn)
     connection_free(conn);
 
+  /* Call the OOS handler, indicate if we saw an exhaustion-related error */
+  connection_check_oos(get_n_open_sockets(), exhaustion);
+
   return NULL;
 }
 
@@ -1442,21 +1462,34 @@ connection_handle_listener_read(connection_t *conn, int new_type)
   if (!SOCKET_OK(news)) { /* accept() error */
     int e = tor_socket_errno(conn->s);
     if (ERRNO_IS_ACCEPT_EAGAIN(e)) {
-      return 0; /* they hung up before we could accept(). that's fine. */
+      /*
+       * they hung up before we could accept(). that's fine.
+       *
+       * give the OOS handler a chance to run though
+       */
+      connection_check_oos(get_n_open_sockets(), 0);
+      return 0;
     } else if (ERRNO_IS_RESOURCE_LIMIT(e)) {
       warn_too_many_conns();
+      /* Exhaustion; tell the OOS handler */
+      connection_check_oos(get_n_open_sockets(), 1);
       return 0;
     }
     /* else there was a real error. */
     log_warn(LD_NET,"accept() failed: %s. Closing listener.",
              tor_socket_strerror(e));
     connection_mark_for_close(conn);
+    /* Tell the OOS handler about this too */
+    connection_check_oos(get_n_open_sockets(), 0);
     return -1;
   }
   log_debug(LD_NET,
             "Connection accepted on socket %d (child of fd %d).",
             (int)news,(int)conn->s);
 
+  /* We accepted a new conn; run OOS handler */
+  connection_check_oos(get_n_open_sockets(), 0);
+
   if (make_socket_reuseable(news) < 0) {
     if (tor_socket_errno(news) == EINVAL) {
       /* This can happen on OSX if we get a badly timed shutdown. */
@@ -1661,12 +1694,18 @@ connection_connect_sockaddr,(connection_t *conn,
 
   s = tor_open_socket_nonblocking(protocol_family, SOCK_STREAM, proto);
   if (! SOCKET_OK(s)) {
+    /*
+     * Early OOS handler calls; it matters if it's an exhaustion-related
+     * error or not.
+     */
     *socket_error = tor_socket_errno(s);
     if (ERRNO_IS_RESOURCE_LIMIT(*socket_error)) {
       warn_too_many_conns();
+      connection_check_oos(get_n_open_sockets(), 1);
     } else {
       log_warn(LD_NET,"Error creating network socket: %s",
                tor_socket_strerror(*socket_error));
+      connection_check_oos(get_n_open_sockets(), 0);
     }
     return -1;
   }
@@ -1676,6 +1715,13 @@ connection_connect_sockaddr,(connection_t *conn,
              tor_socket_strerror(errno));
   }
 
+  /*
+   * We've got the socket open; give the OOS handler a chance to check
+   * against configuured maximum socket number, but tell it no exhaustion
+   * failure.
+   */
+  connection_check_oos(get_n_open_sockets(), 0);
+
   if (bindaddr && bind(s, bindaddr, bindaddr_len) < 0) {
     *socket_error = tor_socket_errno(s);
     log_warn(LD_NET,"Error binding network socket: %s",
@@ -4454,6 +4500,256 @@ connection_reached_eof(connection_t *conn)
   }
 }
 
+/** Comparator for the two-orconn case in OOS victim sort */
+static int
+oos_victim_comparator_for_orconns(or_connection_t *a, or_connection_t *b)
+{
+  int a_circs, b_circs;
+  /* Fewer circuits == higher priority for OOS kill, sort earlier */
+
+  a_circs = connection_or_get_num_circuits(a);
+  b_circs = connection_or_get_num_circuits(b);
+
+  if (a_circs < b_circs) return -1;
+  else if (b_circs > a_circs) return 1;
+  else return 0;
+}
+
+/** Sort comparator for OOS victims; better targets sort before worse
+ * ones. */
+static int
+oos_victim_comparator(const void **a_v, const void **b_v)
+{
+  connection_t *a = NULL, *b = NULL;
+
+  /* Get connection pointers out */
+
+  a = (connection_t *)(*a_v);
+  b = (connection_t *)(*b_v);
+
+  tor_assert(a != NULL);
+  tor_assert(b != NULL);
+
+  /*
+   * We always prefer orconns as victims currently; we won't even see
+   * these non-orconn cases, but if we do, sort them after orconns.
+   */
+  if (a->type == CONN_TYPE_OR && b->type == CONN_TYPE_OR) {
+    return oos_victim_comparator_for_orconns(TO_OR_CONN(a), TO_OR_CONN(b));
+  } else {
+    /*
+     * One isn't an orconn; if one is, it goes first.  We currently have no
+     * opinions about cases where neither is an orconn.
+     */
+    if (a->type == CONN_TYPE_OR) return -1;
+    else if (b->type == CONN_TYPE_OR) return 1;
+    else return 0;
+  }
+}
+
+/** Pick n victim connections for the OOS handler and return them in a
+ * smartlist.
+ */
+MOCK_IMPL(STATIC smartlist_t *,
+pick_oos_victims, (int n))
+{
+  smartlist_t *eligible = NULL, *victims = NULL;
+  smartlist_t *conns;
+  int conn_counts_by_type[CONN_TYPE_MAX_ + 1], i;
+
+  /*
+   * Big damn assumption (someone improve this someday!):
+   *
+   * Socket exhaustion normally happens on high-volume relays, and so
+   * most of the connections involved are orconns.  We should pick victims
+   * by assembling a list of all orconns, and sorting them in order of
+   * how much 'damage' by some metric we'd be doing by dropping them.
+   *
+   * If we move on from orconns, we should probably think about incoming
+   * directory connections next, or exit connections.  Things we should
+   * probably never kill are controller connections and listeners.
+   *
+   * This function will count how many connections of different types
+   * exist and log it for purposes of gathering data on typical OOS
+   * situations to guide future improvements.
+   */
+
+  /* First, get the connection array */
+  conns = get_connection_array();
+  /*
+   * Iterate it and pick out eligible connection types, and log some stats
+   * along the way.
+   */
+  eligible = smartlist_new();
+  memset(conn_counts_by_type, 0, sizeof(conn_counts_by_type));
+  SMARTLIST_FOREACH_BEGIN(conns, connection_t *, c) {
+    /* Bump the counter */
+    tor_assert(c->type <= CONN_TYPE_MAX_);
+    ++(conn_counts_by_type[c->type]);
+
+    /* Skip anything without a socket we can free */
+    if (!(SOCKET_OK(c->s))) {
+      continue;
+    }
+
+    /* Skip anything we would count as moribund */
+    if (connection_is_moribund(c)) {
+      continue;
+    }
+
+    switch (c->type) {
+      case CONN_TYPE_OR:
+        /* We've got an orconn, it's eligible to be OOSed */
+        smartlist_add(eligible, c);
+        break;
+      default:
+        /* We don't know what to do with it, ignore it */
+        break;
+    }
+  } SMARTLIST_FOREACH_END(c);
+
+  /* Log some stats */
+  if (smartlist_len(conns) > 0) {
+    /* At least one counter must be non-zero */
+    log_info(LD_NET, "Some stats on conn types seen during OOS follow");
+    for (i = CONN_TYPE_MIN_; i <= CONN_TYPE_MAX_; ++i) {
+      /* Did we see any? */
+      if (conn_counts_by_type[i] > 0) {
+        log_info(LD_NET, "%s: %d conns",
+                 conn_type_to_string(i),
+                 conn_counts_by_type[i]);
+      }
+    }
+    log_info(LD_NET, "Done with OOS conn type stats");
+  }
+
+  /* Did we find more eligible targets than we want to kill? */
+  if (smartlist_len(eligible) > n) {
+    /* Sort the list in order of target preference */
+    smartlist_sort(eligible, oos_victim_comparator);
+    /* Pick first n as victims */
+    victims = smartlist_new();
+    for (i = 0; i < n; ++i) {
+      smartlist_add(victims, smartlist_get(eligible, i));
+    }
+    /* Free the original list */
+    smartlist_free(eligible);
+  } else {
+    /* No, we can just call them all victims */
+    victims = eligible;
+  }
+
+  return victims;
+}
+
+/** Kill a list of connections for the OOS handler. */
+MOCK_IMPL(STATIC void,
+kill_conn_list_for_oos, (smartlist_t *conns))
+{
+  if (!conns) return;
+
+  SMARTLIST_FOREACH_BEGIN(conns, connection_t *, c) {
+    /* Make sure the channel layer gets told about orconns */
+    if (c->type == CONN_TYPE_OR) {
+      connection_or_close_for_error(TO_OR_CONN(c), 1);
+    } else {
+      connection_mark_for_close(c);
+    }
+  } SMARTLIST_FOREACH_END(c);
+
+  log_notice(LD_NET,
+             "OOS handler marked %d connections",
+             smartlist_len(conns));
+}
+
+/** Out-of-Sockets handler; n_socks is the current number of open
+ * sockets, and failed is non-zero if a socket exhaustion related
+ * error immediately preceded this call.  This is where to do
+ * circuit-killing heuristics as needed.
+ */
+void
+connection_check_oos(int n_socks, int failed)
+{
+  int target_n_socks = 0, moribund_socks, socks_to_kill;
+  smartlist_t *conns;
+
+  /* Early exit: is OOS checking disabled? */
+  if (get_options()->DisableOOSCheck) {
+    return;
+  }
+
+  /* Sanity-check args */
+  tor_assert(n_socks >= 0);
+
+  /*
+   * Make some log noise; keep it at debug level since this gets a chance
+   * to run on every connection attempt.
+   */
+  log_debug(LD_NET,
+            "Running the OOS handler (%d open sockets, %s)",
+            n_socks, (failed != 0) ? "exhaustion seen" : "no exhaustion");
+
+  /*
+   * Check if we're really handling an OOS condition, and if so decide how
+   * many sockets we want to get down to.  Be sure we check if the threshold
+   * is distinct from zero first; it's possible for this to be called a few
+   * times before we've finished reading the config.
+   */
+  if (n_socks >= get_options()->ConnLimit_high_thresh &&
+      get_options()->ConnLimit_high_thresh != 0 &&
+      get_options()->ConnLimit_ != 0) {
+    /* Try to get down to the low threshold */
+    target_n_socks = get_options()->ConnLimit_low_thresh;
+    log_notice(LD_NET,
+               "Current number of sockets %d is greater than configured "
+               "limit %d; OOS handler trying to get down to %d",
+               n_socks, get_options()->ConnLimit_high_thresh,
+               target_n_socks);
+  } else if (failed) {
+    /*
+     * If we're not at the limit but we hit a socket exhaustion error, try to
+     * drop some (but not as aggressively as ConnLimit_low_threshold, which is
+     * 3/4 of ConnLimit_)
+     */
+    target_n_socks = (n_socks * 9) / 10;
+    log_notice(LD_NET,
+               "We saw socket exhaustion at %d open sockets; OOS handler "
+               "trying to get down to %d",
+               n_socks, target_n_socks);
+  }
+
+  if (target_n_socks > 0) {
+    /*
+     * It's an OOS!
+     *
+     * Count moribund sockets; it's be important that anything we decide
+     * to get rid of here but don't immediately close get counted as moribund
+     * on subsequent invocations so we don't try to kill too many things if
+     * connection_check_oos() gets called multiple times.
+     */
+    moribund_socks = connection_count_moribund();
+
+    if (moribund_socks < n_socks - target_n_socks) {
+      socks_to_kill = n_socks - target_n_socks - moribund_socks;
+
+      conns = pick_oos_victims(socks_to_kill);
+      if (conns) {
+        kill_conn_list_for_oos(conns);
+        log_notice(LD_NET,
+                   "OOS handler killed %d conns", smartlist_len(conns));
+        smartlist_free(conns);
+      } else {
+        log_notice(LD_NET, "OOS handler failed to pick any victim conns");
+      }
+    } else {
+      log_notice(LD_NET,
+                 "Not killing any sockets for OOS because there are %d "
+                 "already moribund, and we only want to eliminate %d",
+                 moribund_socks, n_socks - target_n_socks);
+    }
+  }
+}
+
 /** Log how many bytes are used by buffers of different kinds and sizes. */
 void
 connection_dump_buffer_mem_stats(int severity)
diff --git a/src/or/connection.h b/src/or/connection.h
index f8e0f73246..d25e002fa4 100644
--- a/src/or/connection.h
+++ b/src/or/connection.h
@@ -34,8 +34,8 @@ void connection_about_to_close_connection(connection_t *conn);
 void connection_close_immediate(connection_t *conn);
 void connection_mark_for_close_(connection_t *conn,
                                 int line, const char *file);
-void connection_mark_for_close_internal_(connection_t *conn,
-                                         int line, const char *file);
+MOCK_DECL(void, connection_mark_for_close_internal_,
+          (connection_t *conn, int line, const char *file));
 
 #define connection_mark_for_close(c) \
   connection_mark_for_close_((c), __LINE__, SHORT_FILE__)
@@ -247,6 +247,22 @@ void clock_skew_warning(const connection_t *conn, long apparent_skew,
                         int trusted, log_domain_mask_t domain,
                         const char *received, const char *source);
 
+/** Check if a connection is on the way out so the OOS handler doesn't try
+ * to kill more than it needs. */
+static inline int
+connection_is_moribund(connection_t *conn)
+{
+  if (conn != NULL &&
+      (conn->conn_array_index < 0 ||
+       conn->marked_for_close)) {
+    return 1;
+  } else {
+    return 0;
+  }
+}
+
+void connection_check_oos(int n_socks, int failed);
+
 #ifdef CONNECTION_PRIVATE
 STATIC void connection_free_(connection_t *conn);
 
@@ -265,6 +281,9 @@ MOCK_DECL(STATIC int,connection_connect_sockaddr,
                                              const struct sockaddr *bindaddr,
                                              socklen_t bindaddr_len,
                                              int *socket_error));
+MOCK_DECL(STATIC void, kill_conn_list_for_oos, (smartlist_t *conns));
+MOCK_DECL(STATIC smartlist_t *, pick_oos_victims, (int n));
+
 #endif
 
 #endif
diff --git a/src/or/connection_or.c b/src/or/connection_or.c
index 1f0c4bdef5..72d8e13e90 100644
--- a/src/or/connection_or.c
+++ b/src/or/connection_or.c
@@ -394,8 +394,8 @@ connection_or_change_state(or_connection_t *conn, uint8_t state)
  * be an or_connection_t field, but it got moved to channel_t and we
  * shouldn't maintain two copies. */
 
-int
-connection_or_get_num_circuits(or_connection_t *conn)
+MOCK_IMPL(int,
+connection_or_get_num_circuits, (or_connection_t *conn))
 {
   tor_assert(conn);
 
diff --git a/src/or/connection_or.h b/src/or/connection_or.h
index e2ec47a4f2..2e8c6066cc 100644
--- a/src/or/connection_or.h
+++ b/src/or/connection_or.h
@@ -64,7 +64,7 @@ void connection_or_init_conn_from_address(or_connection_t *conn,
 int connection_or_client_learned_peer_id(or_connection_t *conn,
                                          const uint8_t *peer_id);
 time_t connection_or_client_used(or_connection_t *conn);
-int connection_or_get_num_circuits(or_connection_t *conn);
+MOCK_DECL(int, connection_or_get_num_circuits, (or_connection_t *conn));
 void or_handshake_state_free(or_handshake_state_t *state);
 void or_handshake_state_record_cell(or_connection_t *conn,
                                     or_handshake_state_t *state,
diff --git a/src/or/main.c b/src/or/main.c
index 4dbd9a005b..03c2b7ed58 100644
--- a/src/or/main.c
+++ b/src/or/main.c
@@ -381,8 +381,8 @@ connection_in_array(connection_t *conn)
 /** Set <b>*array</b> to an array of all connections. <b>*array</b> must not
  * be modified.
  */
-smartlist_t *
-get_connection_array(void)
+MOCK_IMPL(smartlist_t *,
+get_connection_array, (void))
 {
   if (!connection_array)
     connection_array = smartlist_new();
@@ -651,6 +651,23 @@ close_closeable_connections(void)
   }
 }
 
+/** Count moribund connections for the OOS handler */
+MOCK_IMPL(int,
+connection_count_moribund, (void))
+{
+  int moribund = 0;
+
+  /*
+   * Count things we'll try to kill when close_closeable_connections()
+   * runs next.
+   */
+  SMARTLIST_FOREACH_BEGIN(closeable_connection_lst, connection_t *, conn) {
+    if (SOCKET_OK(conn->s) && connection_is_moribund(conn)) ++moribund;
+  } SMARTLIST_FOREACH_END(conn);
+
+  return moribund;
+}
+
 /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
  * some data to read. */
 static void
diff --git a/src/or/main.h b/src/or/main.h
index 31a22de424..0220ae3c57 100644
--- a/src/or/main.h
+++ b/src/or/main.h
@@ -25,7 +25,7 @@ int connection_in_array(connection_t *conn);
 void add_connection_to_closeable_list(connection_t *conn);
 int connection_is_on_closeable_list(connection_t *conn);
 
-smartlist_t *get_connection_array(void);
+MOCK_DECL(smartlist_t *, get_connection_array, (void));
 MOCK_DECL(uint64_t,get_bytes_read,(void));
 MOCK_DECL(uint64_t,get_bytes_written,(void));
 
@@ -47,6 +47,8 @@ MOCK_DECL(void,connection_start_writing,(connection_t *conn));
 
 void connection_stop_reading_from_linked_conn(connection_t *conn);
 
+MOCK_DECL(int, connection_count_moribund, (void));
+
 void directory_all_unreachable(time_t now);
 void directory_info_has_arrived(time_t now, int from_cache, int suppress_logs);
 
diff --git a/src/or/or.h b/src/or/or.h
index 43b31c0fdd..be58fa0d0a 100644
--- a/src/or/or.h
+++ b/src/or/or.h
@@ -3699,6 +3699,10 @@ typedef struct {
 
   int ConnLimit; /**< Demanded minimum number of simultaneous connections. */
   int ConnLimit_; /**< Maximum allowed number of simultaneous connections. */
+  int ConnLimit_high_thresh; /**< start trying to lower socket usage if we
+                              *   have this many. */
+  int ConnLimit_low_thresh; /**< try to get down to here after socket
+                             *   exhaustion. */
   int RunAsDaemon; /**< If true, run in the background. (Unix only) */
   int FascistFirewall; /**< Whether to prefer ORs reachable on open ports. */
   smartlist_t *FirewallPorts; /**< Which ports our firewall allows
@@ -4454,6 +4458,9 @@ typedef struct {
    * participate in the protocol. If on (default), a flag is added to the
    * vote indicating participation. */
   int AuthDirSharedRandomness;
+
+  /** If 1, we skip all OOS checks. */
+  int DisableOOSCheck;
 } or_options_t;
 
 /** Persistent state for an onion router, as saved to disk. */
diff --git a/src/test/include.am b/src/test/include.am
index d0bc808877..2e91f7c9dc 100644
--- a/src/test/include.am
+++ b/src/test/include.am
@@ -103,6 +103,7 @@ src_test_test_SOURCES = \
 	src/test/test_microdesc.c \
 	src/test/test_nodelist.c \
 	src/test/test_oom.c \
+	src/test/test_oos.c \
 	src/test/test_options.c \
 	src/test/test_policy.c \
 	src/test/test_procmon.c \
diff --git a/src/test/test.c b/src/test/test.c
index f8610168f6..2f10c7e90b 100644
--- a/src/test/test.c
+++ b/src/test/test.c
@@ -1210,6 +1210,7 @@ struct testgroup_t testgroups[] = {
   { "link-handshake/", link_handshake_tests },
   { "nodelist/", nodelist_tests },
   { "oom/", oom_tests },
+  { "oos/", oos_tests },
   { "options/", options_tests },
   { "policy/" , policy_tests },
   { "procmon/", procmon_tests },
diff --git a/src/test/test.h b/src/test/test.h
index 6744d255f1..c0643e154d 100644
--- a/src/test/test.h
+++ b/src/test/test.h
@@ -202,6 +202,7 @@ extern struct testcase_t logging_tests[];
 extern struct testcase_t microdesc_tests[];
 extern struct testcase_t nodelist_tests[];
 extern struct testcase_t oom_tests[];
+extern struct testcase_t oos_tests[];
 extern struct testcase_t options_tests[];
 extern struct testcase_t policy_tests[];
 extern struct testcase_t procmon_tests[];
diff --git a/src/test/test_oos.c b/src/test/test_oos.c
new file mode 100644
index 0000000000..9451c5c537
--- /dev/null
+++ b/src/test/test_oos.c
@@ -0,0 +1,454 @@
+/* Copyright (c) 2016, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+/* Unit tests for OOS handler */
+
+#define CONNECTION_PRIVATE
+
+#include "or.h"
+#include "config.h"
+#include "connection.h"
+#include "connection_or.h"
+#include "main.h"
+#include "test.h"
+
+static or_options_t mock_options;
+
+static void
+reset_options_mock(void)
+{
+  memset(&mock_options, 0, sizeof(or_options_t));
+}
+
+static const or_options_t *
+mock_get_options(void)
+{
+  return &mock_options;
+}
+
+static int moribund_calls = 0;
+static int moribund_conns = 0;
+
+static int
+mock_connection_count_moribund(void)
+{
+  ++moribund_calls;
+
+  return moribund_conns;
+}
+
+/*
+ * For unit test purposes it's sufficient to tell that
+ * kill_conn_list_for_oos() was called with an approximately
+ * sane argument; it's just the thing we returned from the
+ * mock for pick_oos_victims().
+ */
+
+static int kill_conn_list_calls = 0;
+static int kill_conn_list_killed = 0;
+
+static void
+kill_conn_list_mock(smartlist_t *conns)
+{
+  ++kill_conn_list_calls;
+
+  tt_assert(conns != NULL);
+
+  kill_conn_list_killed += smartlist_len(conns);
+
+ done:
+  return;
+}
+
+static int pick_oos_mock_calls = 0;
+static int pick_oos_mock_fail = 0;
+static int pick_oos_mock_last_n = 0;
+
+static smartlist_t *
+pick_oos_victims_mock(int n)
+{
+  smartlist_t *l;
+  int i;
+
+  ++pick_oos_mock_calls;
+
+  tt_int_op(n, OP_GT, 0);
+
+  if (!pick_oos_mock_fail) {
+    /*
+     * connection_check_oos() just passes the list onto
+     * kill_conn_list_for_oos(); we don't need to simulate
+     * its content for this mock, just its existence, but
+     * we do need to check the parameter.
+     */
+    l = smartlist_new();
+    for (i = 0; i < n; ++i) smartlist_add(l, NULL);
+  } else {
+    l = NULL;
+  }
+
+  pick_oos_mock_last_n = n;
+
+ done:
+  return l;
+}
+
+/** Unit test for the logic in connection_check_oos(), which is concerned
+ * with comparing thresholds and connection counts to decide if an OOS has
+ * occurred and if so, how many connections to try to kill, and then using
+ * pick_oos_victims() and kill_conn_list_for_oos() to carry out its grim
+ * duty.
+ */
+static void
+test_oos_connection_check_oos(void *arg)
+{
+  (void)arg;
+
+  /* Set up mocks */
+  reset_options_mock();
+  /* OOS handling is only sensitive to these fields */
+  mock_options.ConnLimit = 32;
+  mock_options.ConnLimit_ = 64;
+  mock_options.ConnLimit_high_thresh = 60;
+  mock_options.ConnLimit_low_thresh = 50;
+  MOCK(get_options, mock_get_options);
+  moribund_calls = 0;
+  moribund_conns = 0;
+  MOCK(connection_count_moribund, mock_connection_count_moribund);
+  kill_conn_list_calls = 0;
+  kill_conn_list_killed = 0;
+  MOCK(kill_conn_list_for_oos, kill_conn_list_mock);
+  pick_oos_mock_calls = 0;
+  pick_oos_mock_fail = 0;
+  MOCK(pick_oos_victims, pick_oos_victims_mock);
+
+  /* No OOS case */
+  connection_check_oos(50, 0);
+  tt_int_op(moribund_calls, OP_EQ, 0);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 0);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 0);
+
+  /* OOS from socket count, nothing moribund */
+  connection_check_oos(62, 0);
+  tt_int_op(moribund_calls, OP_EQ, 1);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 1);
+  /* 12 == 62 - ConnLimit_low_thresh */
+  tt_int_op(pick_oos_mock_last_n, OP_EQ, 12);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 1);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 12);
+
+  /* OOS from socket count, some are moribund */
+  kill_conn_list_killed = 0;
+  moribund_conns = 5;
+  connection_check_oos(62, 0);
+  tt_int_op(moribund_calls, OP_EQ, 2);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 2);
+  /* 7 == 62 - ConnLimit_low_thresh - moribund_conns */
+  tt_int_op(pick_oos_mock_last_n, OP_EQ, 7);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 2);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 7);
+
+  /* OOS from socket count, but pick fails */
+  kill_conn_list_killed = 0;
+  moribund_conns = 0;
+  pick_oos_mock_fail = 1;
+  connection_check_oos(62, 0);
+  tt_int_op(moribund_calls, OP_EQ, 3);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 3);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 2);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 0);
+  pick_oos_mock_fail = 0;
+
+  /*
+   * OOS from socket count with so many moribund conns
+   * we have none to kill.
+   */
+  kill_conn_list_killed = 0;
+  moribund_conns = 15;
+  connection_check_oos(62, 0);
+  tt_int_op(moribund_calls, OP_EQ, 4);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 3);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 2);
+
+  /*
+   * OOS from socket exhaustion; OOS handler will try to
+   * kill 1/10 (5) of the connections.
+   */
+  kill_conn_list_killed = 0;
+  moribund_conns = 0;
+  connection_check_oos(50, 1);
+  tt_int_op(moribund_calls, OP_EQ, 5);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 4);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 3);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 5);
+
+  /* OOS from socket exhaustion with moribund conns */
+  kill_conn_list_killed = 0;
+  moribund_conns = 2;
+  connection_check_oos(50, 1);
+  tt_int_op(moribund_calls, OP_EQ, 6);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 5);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 4);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 3);
+
+  /* OOS from socket exhaustion with many moribund conns */
+  kill_conn_list_killed = 0;
+  moribund_conns = 7;
+  connection_check_oos(50, 1);
+  tt_int_op(moribund_calls, OP_EQ, 7);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 5);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 4);
+
+  /* OOS with both socket exhaustion and above-threshold */
+  kill_conn_list_killed = 0;
+  moribund_conns = 0;
+  connection_check_oos(62, 1);
+  tt_int_op(moribund_calls, OP_EQ, 8);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 6);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 5);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 12);
+
+  /*
+   * OOS with both socket exhaustion and above-threshold with some
+   * moribund conns
+   */
+  kill_conn_list_killed = 0;
+  moribund_conns = 5;
+  connection_check_oos(62, 1);
+  tt_int_op(moribund_calls, OP_EQ, 9);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 7);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 6);
+  tt_int_op(kill_conn_list_killed, OP_EQ, 7);
+
+  /*
+   * OOS with both socket exhaustion and above-threshold with many
+   * moribund conns
+   */
+  kill_conn_list_killed = 0;
+  moribund_conns = 15;
+  connection_check_oos(62, 1);
+  tt_int_op(moribund_calls, OP_EQ, 10);
+  tt_int_op(pick_oos_mock_calls, OP_EQ, 7);
+  tt_int_op(kill_conn_list_calls, OP_EQ, 6);
+
+ done:
+
+  UNMOCK(pick_oos_victims);
+  UNMOCK(kill_conn_list_for_oos);
+  UNMOCK(connection_count_moribund);
+  UNMOCK(get_options);
+
+  return;
+}
+
+static int cfe_calls = 0;
+
+static void
+close_for_error_mock(or_connection_t *orconn, int flush)
+{
+  (void)flush;
+
+  tt_assert(orconn != NULL);
+  ++cfe_calls;
+
+ done:
+  return;
+}
+
+static int mark_calls = 0;
+
+static void
+mark_for_close_oos_mock(connection_t *conn,
+                        int line, const char *file)
+{
+  (void)line;
+  (void)file;
+
+  tt_assert(conn != NULL);
+  ++mark_calls;
+
+ done:
+  return;
+}
+
+static void
+test_oos_kill_conn_list(void *arg)
+{
+  connection_t *c1, *c2;
+  or_connection_t *or_c1 = NULL;
+  dir_connection_t *dir_c2 = NULL;
+  smartlist_t *l = NULL;
+  (void)arg;
+
+  /* Set up mocks */
+  mark_calls = 0;
+  MOCK(connection_mark_for_close_internal_, mark_for_close_oos_mock);
+  cfe_calls = 0;
+  MOCK(connection_or_close_for_error, close_for_error_mock);
+
+  /* Make fake conns */
+  or_c1 = tor_malloc_zero(sizeof(*or_c1));
+  or_c1->base_.magic = OR_CONNECTION_MAGIC;
+  or_c1->base_.type = CONN_TYPE_OR;
+  c1 = TO_CONN(or_c1);
+  dir_c2 = tor_malloc_zero(sizeof(*dir_c2));
+  dir_c2->base_.magic = DIR_CONNECTION_MAGIC;
+  dir_c2->base_.type = CONN_TYPE_DIR;
+  c2 = TO_CONN(dir_c2);
+
+  tt_assert(c1 != NULL);
+  tt_assert(c2 != NULL);
+
+  /* Make list */
+  l = smartlist_new();
+  smartlist_add(l, c1);
+  smartlist_add(l, c2);
+
+  /* Run kill_conn_list_for_oos() */
+  kill_conn_list_for_oos(l);
+
+  /* Check call counters */
+  tt_int_op(mark_calls, OP_EQ, 1);
+  tt_int_op(cfe_calls, OP_EQ, 1);
+
+ done:
+
+  UNMOCK(connection_or_close_for_error);
+  UNMOCK(connection_mark_for_close_internal_);
+
+  if (l) smartlist_free(l);
+  tor_free(or_c1);
+  tor_free(dir_c2);
+
+  return;
+}
+
+static smartlist_t *conns_for_mock = NULL;
+
+static smartlist_t *
+get_conns_mock(void)
+{
+  return conns_for_mock;
+}
+
+/*
+ * For this mock, we pretend all conns have either zero or one circuits,
+ * depending on if this appears on the list of things to say have a circuit.
+ */
+
+static smartlist_t *conns_with_circs = NULL;
+
+static int
+get_num_circuits_mock(or_connection_t *conn)
+{
+  int circs = 0;
+
+  tt_assert(conn != NULL);
+
+  if (conns_with_circs &&
+      smartlist_contains(conns_with_circs, TO_CONN(conn))) {
+    circs = 1;
+  }
+
+ done:
+  return circs;
+}
+
+static void
+test_oos_pick_oos_victims(void *arg)
+{
+  (void)arg;
+  or_connection_t *ortmp;
+  dir_connection_t *dirtmp;
+  smartlist_t *picked;
+
+  /* Set up mocks */
+  conns_for_mock = smartlist_new();
+  MOCK(get_connection_array, get_conns_mock);
+  conns_with_circs = smartlist_new();
+  MOCK(connection_or_get_num_circuits, get_num_circuits_mock);
+
+  /* Make some fake connections */
+  ortmp = tor_malloc_zero(sizeof(*ortmp));
+  ortmp->base_.magic = OR_CONNECTION_MAGIC;
+  ortmp->base_.type = CONN_TYPE_OR;
+  smartlist_add(conns_for_mock, TO_CONN(ortmp));
+  /* We'll pretend this one has a circuit too */
+  smartlist_add(conns_with_circs, TO_CONN(ortmp));
+  /* Next one */
+  ortmp = tor_malloc_zero(sizeof(*ortmp));
+  ortmp->base_.magic = OR_CONNECTION_MAGIC;
+  ortmp->base_.type = CONN_TYPE_OR;
+  smartlist_add(conns_for_mock, TO_CONN(ortmp));
+  /* Next one is moribund */
+  ortmp = tor_malloc_zero(sizeof(*ortmp));
+  ortmp->base_.magic = OR_CONNECTION_MAGIC;
+  ortmp->base_.type = CONN_TYPE_OR;
+  ortmp->base_.marked_for_close = 1;
+  smartlist_add(conns_for_mock, TO_CONN(ortmp));
+  /* Last one isn't an orconn */
+  dirtmp = tor_malloc_zero(sizeof(*dirtmp));
+  dirtmp->base_.magic = DIR_CONNECTION_MAGIC;
+  dirtmp->base_.type = CONN_TYPE_DIR;
+  smartlist_add(conns_for_mock, TO_CONN(dirtmp));
+
+  /* Try picking one */
+  picked = pick_oos_victims(1);
+  /* It should be the one with circuits */
+  tt_assert(picked != NULL);
+  tt_int_op(smartlist_len(picked), OP_EQ, 1);
+  tt_assert(smartlist_contains(picked, smartlist_get(conns_for_mock, 0)));
+  smartlist_free(picked);
+
+  /* Try picking none */
+  picked = pick_oos_victims(0);
+  /* We should get an empty list */
+  tt_assert(picked != NULL);
+  tt_int_op(smartlist_len(picked), OP_EQ, 0);
+  smartlist_free(picked);
+
+  /* Try picking two */
+  picked = pick_oos_victims(2);
+  /* We should get both active orconns */
+  tt_assert(picked != NULL);
+  tt_int_op(smartlist_len(picked), OP_EQ, 2);
+  tt_assert(smartlist_contains(picked, smartlist_get(conns_for_mock, 0)));
+  tt_assert(smartlist_contains(picked, smartlist_get(conns_for_mock, 1)));
+  smartlist_free(picked);
+
+  /* Try picking three - only two are eligible */
+  picked = pick_oos_victims(3);
+  tt_int_op(smartlist_len(picked), OP_EQ, 2);
+  tt_assert(smartlist_contains(picked, smartlist_get(conns_for_mock, 0)));
+  tt_assert(smartlist_contains(picked, smartlist_get(conns_for_mock, 1)));
+  smartlist_free(picked);
+
+ done:
+
+  /* Free leftover stuff */
+  if (conns_with_circs) {
+    smartlist_free(conns_with_circs);
+    conns_with_circs = NULL;
+  }
+
+  UNMOCK(connection_or_get_num_circuits);
+
+  if (conns_for_mock) {
+    SMARTLIST_FOREACH(conns_for_mock, connection_t *, c, tor_free(c));
+    smartlist_free(conns_for_mock);
+    conns_for_mock = NULL;
+  }
+
+  UNMOCK(get_connection_array);
+
+  return;
+}
+
+struct testcase_t oos_tests[] = {
+  { "connection_check_oos", test_oos_connection_check_oos,
+    TT_FORK, NULL, NULL },
+  { "kill_conn_list", test_oos_kill_conn_list, TT_FORK, NULL, NULL },
+  { "pick_oos_victims", test_oos_pick_oos_victims, TT_FORK, NULL, NULL },
+  END_OF_TESTCASES
+};
+