Merge branch 'maint-0.2.9' into release-0.2.9

37 files changed, 2372 insertions, 46 deletions

diff --git a/changes/bug24952 b/changes/bug24952
new file mode 100644
index 0000000000..93174c04f5
--- /dev/null
+++ b/changes/bug24952
@@ -0,0 +1,5 @@
+  o Minor bugfix (channel connection):
+    - The accurate address of a connection is real_addr, not the addr member.
+      TLS Channel remote address is now real_addr content instead of addr
+      member. Fixes bug 24952; bugfix on 707c1e2e26 in 0.2.4.11-alpha.
+      Patch by "ffmancera".
diff --git a/changes/bug25223 b/changes/bug25223
new file mode 100644
index 0000000000..fdd5563500
--- /dev/null
+++ b/changes/bug25223
@@ -0,0 +1,4 @@
+  o Minor bugfixes (DoS mitigation):
+    - Make sure we don't modify consensus parameters if we aren't a public
+      relay when a new consensus arrives. Fixes bug 25223; bugfix on
+      0.3.3.2-alpha.
diff --git a/changes/ticket24902 b/changes/ticket24902
new file mode 100644
index 0000000000..1a2ef95cc9
--- /dev/null
+++ b/changes/ticket24902
@@ -0,0 +1,13 @@
+  o Major features (denial of service mitigation):
+    - Give relays some defenses against the recent network overload. We start
+      with three defenses (default parameters in parentheses). First: if a
+      single client address makes too many concurrent connections (>100), hang
+      up on further connections. Second: if a single client address makes
+      circuits too quickly (more than 3 per second, with an allowed burst of
+      90) while also having too many connections open (3), refuse new create
+      cells for the next while (1-2 hours). Third: if a client asks to
+      establish a rendezvous point to you directly, ignore the request. These
+      defenses can be manually controlled by new torrc options, but relays
+      will also take guidance from consensus parameters, so there's no need to
+      configure anything manually. Implements ticket 24902.
+
diff --git a/changes/ticket25122 b/changes/ticket25122
new file mode 100644
index 0000000000..2921811b22
--- /dev/null
+++ b/changes/ticket25122
@@ -0,0 +1,4 @@
+  o Minor feature (geoip cache):
+    - Make our OOM handler aware of the geoip client history cache so it
+      doesn't fill up the memory which is especially important for IPv6 and
+      our DoS mitigation subsystem. Closes ticket 25122.
diff --git a/changes/ticket25202 b/changes/ticket25202
new file mode 100644
index 0000000000..ba64abad7b
--- /dev/null
+++ b/changes/ticket25202
@@ -0,0 +1,4 @@
+  o Minor bugfixes (DoS mitigation):
+    - Add extra safety checks when refilling the circuit creation bucket to
+      ensure we never set a value that is above the allowed burst. Fixes
+      bug 25202; bugfix on 0.3.3.2-alpha.
diff --git a/doc/tor.1.txt b/doc/tor.1.txt
index 4c5d5359af..a7ee7d11ca 100644
--- a/doc/tor.1.txt
+++ b/doc/tor.1.txt
@@ -2441,6 +2441,101 @@ The following options are used to configure a hidden service.
     including setting SOCKSPort to "0".
     (Default: 0)
 
+DENIAL OF SERVICE MITIGATION OPTIONS
+------------------------------------
+
+The following options are useful only for a public relay. They control the
+Denial of Service mitigation subsystem.
+
+[[DoSCircuitCreationEnabled]] **DoSCircuitCreationEnabled** **0**|**1**|**auto**::
+
+    Enable circuit creation DoS mitigation. If enabled, tor will cache client
+    IPs along with statistics in order to detect circuit DoS attacks. If an
+    address is positively identified, tor will activate defenses against the
+    address. See the DoSCircuitCreationDefenseType option for more details.
+    This is a client to relay detection only. "auto" means use the consensus
+    parameter. If not defined in the consensus, the value is 0.
+    (Default: auto)
+
+[[DoSCircuitCreationMinConnections]] **DoSCircuitCreationMinConnections** __NUM__::
+
+    Minimum threshold of concurrent connections before a client address can be
+    flagged as executing a circuit creation DoS. In other words, once a client
+    address reaches the circuit rate and has a minimum of NUM concurrent
+    connections, a detection is positive. "0" means use the consensus
+    parameter. If not defined in the consensus, the value is 3.
+    (Default: 0)
+
+[[DoSCircuitCreationRate]] **DoSCircuitCreationRate** __NUM__::
+
+    The allowed circuit creation rate per second applied per client IP
+    address. If this option is 0, it obeys a consensus parameter. If not
+    defined in the consensus, the value is 3.
+    (Default: 0)
+
+[[DoSCircuitCreationBurst]] **DoSCircuitCreationBurst** __NUM__::
+
+    The allowed circuit creation burst per client IP address. If the circuit
+    rate and the burst are reached, a client is marked as executing a circuit
+    creation DoS. "0" means use the consensus parameter. If not defined in the
+    consensus, the value is 90.
+    (Default: 0)
+
+[[DoSCircuitCreationDefenseType]] **DoSCircuitCreationDefenseType** __NUM__::
+
+    This is the type of defense applied to a detected client address. The
+    possible values are:
+
+      1: No defense.
+      2: Refuse circuit creation for the DoSCircuitCreationDefenseTimePeriod period of time.
++
+    "0" means use the consensus parameter. If not defined in the consensus,
+    the value is 2.
+    (Default: 0)
+
+[[DoSCircuitCreationDefenseTimePeriod]] **DoSCircuitCreationDefenseTimePeriod** __N__ **seconds**|**minutes**|**hours**::
+
+    The base time period in seconds that the DoS defense is activated for. The
+    actual value is selected randomly for each activation from N+1 to 3/2 * N.
+    "0" means use the consensus parameter. If not defined in the consensus,
+    the value is 3600 seconds (1 hour).  (Default: 0)
+
+[[DoSConnectionEnabled]] **DoSConnectionEnabled** **0**|**1**|**auto**::
+
+    Enable the connection DoS mitigation. For client address only, this allows
+    tor to mitigate against large number of concurrent connections made by a
+    single IP address. "auto" means use the consensus parameter. If not
+    defined in the consensus, the value is 0.
+    (Default: auto)
+
+[[DoSConnectionMaxConcurrentCount]] **DoSConnectionMaxConcurrentCount** __NUM__::
+
+    The maximum threshold of concurrent connection from a client IP address.
+    Above this limit, a defense selected by DoSConnectionDefenseType is
+    applied. "0" means use the consensus parameter. If not defined in the
+    consensus, the value is 100.
+    (Default: 0)
+
+[[DoSConnectionDefenseType]] **DoSConnectionDefenseType** __NUM__::
+
+    This is the type of defense applied to a detected client address for the
+    connection mitigation. The possible values are:
+
+      1: No defense.
+      2: Immediately close new connections.
++
+    "0" means use the consensus parameter. If not defined in the consensus,
+    the value is 2.
+    (Default: 0)
+
+[[DoSRefuseSingleHopClientRendezvous]] **DoSRefuseSingleHopClientRendezvous** **0**|**1**|**auto**::
+
+    Refuse establishment of rendezvous points for single hop clients. In other
+    words, if a client directly connects to the relay and sends an
+    ESTABLISH_RENDEZVOUS cell, it is silently dropped. "auto" means use the
+    consensus parameter. If not defined in the consensus, the value is 0.
+    (Default: auto)
+
 TESTING NETWORK OPTIONS
 -----------------------
 
diff --git a/src/common/address.c b/src/common/address.c
index 773e688554..68ad639411 100644
--- a/src/common/address.c
+++ b/src/common/address.c
@@ -1181,6 +1181,9 @@ tor_addr_compare_masked(const tor_addr_t *addr1, const tor_addr_t *addr2,
   }
 }
 
+/** Input for siphash, to produce some output for an unspec value. */
+static const uint32_t unspec_hash_input[] = { 0x4e4df09f, 0x92985342 };
+
 /** Return a hash code based on the address addr. DOCDOC extra */
 uint64_t
 tor_addr_hash(const tor_addr_t *addr)
@@ -1189,7 +1192,7 @@ tor_addr_hash(const tor_addr_t *addr)
   case AF_INET:
     return siphash24g(&addr->addr.in_addr.s_addr, 4);
   case AF_UNSPEC:
-    return 0x4e4d5342;
+    return siphash24g(unspec_hash_input, sizeof(unspec_hash_input));
   case AF_INET6:
     return siphash24g(&addr->addr.in6_addr.s6_addr, 16);
   default:
@@ -1200,6 +1203,28 @@ tor_addr_hash(const tor_addr_t *addr)
   }
 }
 
+/** As tor_addr_hash, but use a particular siphash key. */
+uint64_t
+tor_addr_keyed_hash(const struct sipkey *key, const tor_addr_t *addr)
+{
+  /* This is duplicate code with tor_addr_hash, since this function needs to
+   * be backportable all the way to 0.2.9. */
+
+  switch (tor_addr_family(addr)) {
+  case AF_INET:
+    return siphash24(&addr->addr.in_addr.s_addr, 4, key);
+  case AF_UNSPEC:
+    return siphash24(unspec_hash_input, sizeof(unspec_hash_input), key);
+  case AF_INET6:
+    return siphash24(&addr->addr.in6_addr.s6_addr, 16, key);
+  default:
+    /* LCOV_EXCL_START */
+    tor_fragile_assert();
+    return 0;
+    /* LCOV_EXCL_END */
+  }
+}
+
 /** Return a newly allocated string with a representation of <b>addr</b>. */
 char *
 tor_addr_to_str_dup(const tor_addr_t *addr)
diff --git a/src/common/address.h b/src/common/address.h
index 51db42c315..d57abd0d9e 100644
--- a/src/common/address.h
+++ b/src/common/address.h
@@ -228,6 +228,8 @@ int tor_addr_compare_masked(const tor_addr_t *addr1, const tor_addr_t *addr2,
 #define tor_addr_eq(a,b) (0==tor_addr_compare((a),(b),CMP_EXACT))
 
 uint64_t tor_addr_hash(const tor_addr_t *addr);
+struct sipkey;
+uint64_t tor_addr_keyed_hash(const struct sipkey *key, const tor_addr_t *addr);
 int tor_addr_is_v4(const tor_addr_t *addr);
 int tor_addr_is_internal_(const tor_addr_t *ip, int for_listening,
                           const char *filename, int lineno);
diff --git a/src/common/address_set.c b/src/common/address_set.c
new file mode 100644
index 0000000000..4924cb65c2
--- /dev/null
+++ b/src/common/address_set.c
@@ -0,0 +1,129 @@
+/* Copyright (c) 2018, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+/**
+ * \file address_set.c
+ * \brief Implementation for a set of addresses.
+ *
+ * This module was first written on a semi-emergency basis to improve the
+ * robustness of the anti-DoS module.  As such, it's written in a pretty
+ * conservative way, and should be susceptible to improvement later on.
+ **/
+
+#include "orconfig.h"
+#include "address_set.h"
+#include "address.h"
+#include "compat.h"
+#include "container.h"
+#include "crypto.h"
+#include "util.h"
+#include "siphash.h"
+
+/** How many 64-bit siphash values to extract per address */
+#define N_HASHES 2
+/** How many bloom-filter bits we set per address. This is twice the N_HASHES
+ * value, since we split the siphash output into two 32-bit values. */
+#define N_BITS_PER_ITEM (N_HASHES * 2)
+
+/* XXXX This code is largely duplicated with digestset_t.  We should merge
+ * them together into a common bloom-filter implementation.  I'm keeping
+ * them separate for now, though, since this module needs to be backported
+ * all the way to 0.2.9.
+ *
+ * The main difference between digestset_t and this code is that we use
+ * independent siphashes rather than messing around with bit-shifts.  The
+ * approach here is probably more sound, and we should prefer it if&when we
+ * unify the implementations.
+ **/
+
+struct address_set_t {
+  /** siphash keys to make N_HASHES independent hashes for each address. */
+  struct sipkey key[N_HASHES];
+  int mask; /**< One less than the number of bits in <b>ba</b>; always one less
+             * than a power of two. */
+  bitarray_t *ba; /**< A bit array to implement the Bloom filter. */
+};
+
+/**
+ * Allocate and return an address_set, suitable for holding up to
+ * <b>max_address_guess</b> distinct values.
+ */
+address_set_t *
+address_set_new(int max_addresses_guess)
+{
+  /* See digestset_new() for rationale on this equation. */
+  int n_bits = 1u << (tor_log2(max_addresses_guess)+5);
+
+  address_set_t *set = tor_malloc_zero(sizeof(address_set_t));
+  set->mask = n_bits - 1;
+  set->ba = bitarray_init_zero(n_bits);
+  crypto_rand((char*) set->key, sizeof(set->key));
+
+  return set;
+}
+
+/**
+ * Release all storage associated with <b>set</b>
+ */
+void
+address_set_free(address_set_t *set)
+{
+  if (! set)
+    return;
+
+  bitarray_free(set->ba);
+  tor_free(set);
+}
+
+/** Yield the bit index corresponding to 'val' for set. */
+#define BIT(set, val) ((val) & (set)->mask)
+
+/**
+ * Add <b>addr</b> to <b>set</b>.
+ *
+ * All future queries for <b>addr</b> in set will return true. Removing
+ * items is not possible.
+ */
+void
+address_set_add(address_set_t *set, const struct tor_addr_t *addr)
+{
+  int i;
+  for (i = 0; i < N_HASHES; ++i) {
+    uint64_t h = tor_addr_keyed_hash(&set->key[i], addr);
+    uint32_t high_bits = (uint32_t)(h >> 32);
+    uint32_t low_bits = (uint32_t)(h);
+    bitarray_set(set->ba, BIT(set, high_bits));
+    bitarray_set(set->ba, BIT(set, low_bits));
+  }
+}
+
+/** As address_set_add(), but take an ipv4 address in host order. */
+void
+address_set_add_ipv4h(address_set_t *set, uint32_t addr)
+{
+  tor_addr_t a;
+  tor_addr_from_ipv4h(&a, addr);
+  address_set_add(set, &a);
+}
+
+/**
+ * Return true if <b>addr</b> if a member of <b>set</b>.  (And probably,
+ * return false if <b>addr</b> is not a member of set.)
+ */
+int
+address_set_probably_contains(address_set_t *set,
+                              const struct tor_addr_t *addr)
+{
+  int i, matches = 0;
+  for (i = 0; i < N_HASHES; ++i) {
+    uint64_t h = tor_addr_keyed_hash(&set->key[i], addr);
+    uint32_t high_bits = (uint32_t)(h >> 32);
+    uint32_t low_bits = (uint32_t)(h);
+    // Note that !! is necessary here, since bitarray_is_set does not
+    // necessarily return 1 on true.
+    matches += !! bitarray_is_set(set->ba, BIT(set, high_bits));
+    matches += !! bitarray_is_set(set->ba, BIT(set, low_bits));
+  }
+  return matches == N_BITS_PER_ITEM;
+}
+
diff --git a/src/common/address_set.h b/src/common/address_set.h
new file mode 100644
index 0000000000..aedf17fc66
--- /dev/null
+++ b/src/common/address_set.h
@@ -0,0 +1,35 @@
+/* Copyright (c) 2018, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+/**
+ * \file addressset.h
+ * \brief Types to handle sets of addresses.
+ *
+ * This module was first written on a semi-emergency basis to improve the
+ * robustness of the anti-DoS module.  As such, it's written in a pretty
+ * conservative way, and should be susceptible to improvement later on.
+ **/
+
+#ifndef TOR_ADDRESS_SET_H
+#define TOR_ADDRESS_SET_H
+
+#include "orconfig.h"
+#include "torint.h"
+
+/**
+ * An address_set_t represents a set of tor_addr_t values. The implementation
+ * is probabilistic: false negatives cannot occur but false positives are
+ * possible.
+ */
+typedef struct address_set_t address_set_t;
+struct tor_addr_t;
+
+address_set_t *address_set_new(int max_addresses_guess);
+void address_set_free(address_set_t *set);
+void address_set_add(address_set_t *set, const struct tor_addr_t *addr);
+void address_set_add_ipv4h(address_set_t *set, uint32_t addr);
+int address_set_probably_contains(address_set_t *set,
+                                  const struct tor_addr_t *addr);
+
+#endif
+
diff --git a/src/common/include.am b/src/common/include.am
index 40c463c9d9..cb307e9d5f 100644
--- a/src/common/include.am
+++ b/src/common/include.am
@@ -80,6 +80,7 @@ src_common_libor_ctime_testing_a_CFLAGS = @CFLAGS_CONSTTIME@ $(TEST_CFLAGS)
 
 LIBOR_A_SRC = \
   src/common/address.c					\
+  src/common/address_set.c				\
   src/common/backtrace.c				\
   src/common/compat.c					\
   src/common/compat_threads.c				\
@@ -135,6 +136,7 @@ src_common_libor_event_testing_a_CFLAGS = $(AM_CFLAGS) $(TEST_CFLAGS)
 
 COMMONHEADERS = \
   src/common/address.h				\
+  src/common/address_set.h			\
   src/common/backtrace.h			\
   src/common/aes.h				\
   src/common/ciphers.inc			\
diff --git a/src/common/log.c b/src/common/log.c
index 56adc77f84..4db1c9f0d0 100644
--- a/src/common/log.c
+++ b/src/common/log.c
@@ -1177,7 +1177,7 @@ static const char *domain_list[] = {
   "GENERAL", "CRYPTO", "NET", "CONFIG", "FS", "PROTOCOL", "MM",
   "HTTP", "APP", "CONTROL", "CIRC", "REND", "BUG", "DIR", "DIRSERV",
   "OR", "EDGE", "ACCT", "HIST", "HANDSHAKE", "HEARTBEAT", "CHANNEL",
-  "SCHED", NULL
+  "SCHED", "DOS", NULL
 };
 
 /** Return a bitmask for the log domain for which <b>domain</b> is the name,
diff --git a/src/common/torlog.h b/src/common/torlog.h
index 6732a42741..20b7d938f0 100644
--- a/src/common/torlog.h
+++ b/src/common/torlog.h
@@ -99,8 +99,10 @@
 #define LD_CHANNEL   (1u<<21)
 /** Scheduler */
 #define LD_SCHED     (1u<<22)
+/** Denial of Service mitigation. */
+#define LD_DOS       (1u<<23)
 /** Number of logging domains in the code. */
-#define N_LOGGING_DOMAINS 23
+#define N_LOGGING_DOMAINS 24
 
 /** This log message is not safe to send to a callback-based logger
  * immediately.  Used as a flag, not a log domain. */
diff --git a/src/or/channel.c b/src/or/channel.c
index f547aea1b3..54e10666d2 100644
--- a/src/or/channel.c
+++ b/src/or/channel.c
@@ -2583,6 +2583,7 @@ channel_do_open_actions(channel_t *chan)
     if (!router_get_by_id_digest(chan->identity_digest)) {
       if (channel_get_addr_if_possible(chan, &remote_addr)) {
         char *transport_name = NULL;
+        channel_tls_t *tlschan = BASE_CHAN_TO_TLS(chan);
         if (chan->get_transport_name(chan, &transport_name) < 0)
           transport_name = NULL;
 
@@ -2590,6 +2591,10 @@ channel_do_open_actions(channel_t *chan)
                                &remote_addr, transport_name,
                                now);
         tor_free(transport_name);
+        /* Notify the DoS subsystem of a new client. */
+        if (tlschan && tlschan->conn) {
+          dos_new_client_conn(tlschan->conn);
+        }
       }
       /* Otherwise the underlying transport can't tell us this, so skip it */
     }
@@ -3840,8 +3845,8 @@ channel_get_canonical_remote_descr(channel_t *chan)
  * supports this operation, and return 1.  Return 0 if the underlying transport
  * doesn't let us do this.
  */
-int
-channel_get_addr_if_possible(channel_t *chan, tor_addr_t *addr_out)
+MOCK_IMPL(int,
+channel_get_addr_if_possible,(channel_t *chan, tor_addr_t *addr_out))
 {
   tor_assert(chan);
   tor_assert(addr_out);
diff --git a/src/or/channel.h b/src/or/channel.h
index a711b56d44..bcd345e8d2 100644
--- a/src/or/channel.h
+++ b/src/or/channel.h
@@ -550,7 +550,8 @@ MOCK_DECL(void, channel_dump_statistics, (channel_t *chan, int severity));
 void channel_dump_transport_statistics(channel_t *chan, int severity);
 const char * channel_get_actual_remote_descr(channel_t *chan);
 const char * channel_get_actual_remote_address(channel_t *chan);
-int channel_get_addr_if_possible(channel_t *chan, tor_addr_t *addr_out);
+MOCK_DECL(int, channel_get_addr_if_possible, (channel_t *chan,
+                                              tor_addr_t *addr_out));
 const char * channel_get_canonical_remote_descr(channel_t *chan);
 int channel_has_queued_writes(channel_t *chan);
 int channel_is_bad_for_new_circs(channel_t *chan);
diff --git a/src/or/channeltls.c b/src/or/channeltls.c
index 31641c3db0..3a352d47fe 100644
--- a/src/or/channeltls.c
+++ b/src/or/channeltls.c
@@ -514,7 +514,7 @@ channel_tls_get_remote_addr_method(channel_t *chan, tor_addr_t *addr_out)
   tor_assert(addr_out);
 
   if (tlschan->conn) {
-    tor_addr_copy(addr_out, &(TO_CONN(tlschan->conn)->addr));
+    tor_addr_copy(addr_out, &(tlschan->conn->real_addr));
     rv = 1;
   } else tor_addr_make_unspec(addr_out);
 
diff --git a/src/or/command.c b/src/or/command.c
index 8831446f0b..3f5386c5a3 100644
--- a/src/or/command.c
+++ b/src/or/command.c
@@ -46,6 +46,7 @@
 #include "config.h"
 #include "control.h"
 #include "cpuworker.h"
+#include "dos.h"
 #include "hibernate.h"
 #include "nodelist.h"
 #include "onion.h"
@@ -247,6 +248,11 @@ command_process_create_cell(cell_t *cell, channel_t *chan)
             (unsigned)cell->circ_id,
             U64_PRINTF_ARG(chan->global_identifier), chan);
 
+  /* First thing we do, even though the cell might be invalid, is inform the
+   * DoS mitigation subsystem layer of this event. Validation is done by this
+   * function. */
+  dos_cc_new_create_cell(chan);
+
   /* We check for the conditions that would make us drop the cell before
    * we check for the conditions that would make us send a DESTROY back,
    * since those conditions would make a DESTROY nonsensical. */
@@ -284,6 +290,13 @@ command_process_create_cell(cell_t *cell, channel_t *chan)
     return;
   }
 
+  /* Check if we should apply a defense for this channel. */
+  if (dos_cc_get_defense_type(chan) == DOS_CC_DEFENSE_REFUSE_CELL) {
+    channel_send_destroy(cell->circ_id, chan,
+                         END_CIRC_REASON_RESOURCELIMIT);
+    return;
+  }
+
   if (!server_mode(options) ||
       (!public_server_mode(options) && channel_is_outgoing(chan))) {
     log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL,
diff --git a/src/or/config.c b/src/or/config.c
index 42ff25877e..3b40274339 100644
--- a/src/or/config.c
+++ b/src/or/config.c
@@ -29,6 +29,7 @@
 #include "dirserv.h"
 #include "dirvote.h"
 #include "dns.h"
+#include "dos.h"
 #include "entrynodes.h"
 #include "geoip.h"
 #include "hibernate.h"
@@ -241,6 +242,19 @@ static config_var_t option_vars_[] = {
   OBSOLETE("DynamicDHGroups"),
   VPORT(DNSPort,                     LINELIST, NULL),
   V(DNSListenAddress,            LINELIST, NULL),
+  /* DoS circuit creation options. */
+  V(DoSCircuitCreationEnabled,   AUTOBOOL, "auto"),
+  V(DoSCircuitCreationMinConnections,      UINT, "0"),
+  V(DoSCircuitCreationRate,      UINT,     "0"),
+  V(DoSCircuitCreationBurst,     UINT,     "0"),
+  V(DoSCircuitCreationDefenseType,         INT,  "0"),
+  V(DoSCircuitCreationDefenseTimePeriod,   INTERVAL, "0"),
+  /* DoS connection options. */
+  V(DoSConnectionEnabled,        AUTOBOOL, "auto"),
+  V(DoSConnectionMaxConcurrentCount,       UINT, "0"),
+  V(DoSConnectionDefenseType,    INT,      "0"),
+  /* DoS single hop client options. */
+  V(DoSRefuseSingleHopClientRendezvous,    AUTOBOOL, "auto"),
   V(DownloadExtraInfo,           BOOL,     "0"),
   V(TestingEnableConnBwEvent,    BOOL,     "0"),
   V(TestingEnableCellStatsEvent, BOOL,     "0"),
@@ -2039,6 +2053,17 @@ options_act(const or_options_t *old_options)
     }
   }
 
+  /* DoS mitigation subsystem only applies to public relay. */
+  if (public_server_mode(options)) {
+    /* If we are configured as a relay, initialize the subsystem. Even on HUP,
+     * this is safe to call as it will load data from the current options
+     * or/and the consensus. */
+    dos_init();
+  } else if (old_options && public_server_mode(old_options)) {
+    /* Going from relay to non relay, clean it up. */
+    dos_free_all();
+  }
+
   /* Load the webpage we're going to serve every time someone asks for '/' on
      our DirPort. */
   tor_free(global_dirfrontpagecontents);
diff --git a/src/or/connection.c b/src/or/connection.c
index 8b00d637f6..791fd95c27 100644
--- a/src/or/connection.c
+++ b/src/or/connection.c
@@ -78,6 +78,7 @@
 #include "dirserv.h"
 #include "dns.h"
 #include "dnsserv.h"
+#include "dos.h"
 #include "entrynodes.h"
 #include "ext_orport.h"
 #include "geoip.h"
@@ -687,6 +688,13 @@ connection_free,(connection_t *conn))
                                                   "connection_free");
   }
 #endif
+
+  /* Notify the circuit creation DoS mitigation subsystem that an OR client
+   * connection has been closed. And only do that if we track it. */
+  if (conn->type == CONN_TYPE_OR) {
+    dos_close_client_conn(TO_OR_CONN(conn));
+  }
+
   connection_unregister_events(conn);
   connection_free_(conn);
 }
@@ -1592,6 +1600,14 @@ connection_handle_listener_read(connection_t *conn, int new_type)
         return 0;
       }
     }
+    if (new_type == CONN_TYPE_OR) {
+      /* Assess with the connection DoS mitigation subsystem if this address
+       * can open a new connection. */
+      if (dos_conn_addr_get_defense_type(&addr) == DOS_CONN_DEFENSE_CLOSE) {
+        tor_close_socket(news);
+        return 0;
+      }
+    }
 
     newconn = connection_new(new_type, conn->socket_family);
     newconn->s = news;
diff --git a/src/or/dos.c b/src/or/dos.c
new file mode 100644
index 0000000000..4d1797eece
--- /dev/null
+++ b/src/or/dos.c
@@ -0,0 +1,794 @@
+/* Copyright (c) 2018, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+/*
+ * \file dos.c
+ * \brief Implement Denial of Service mitigation subsystem.
+ */
+
+#define DOS_PRIVATE
+
+#include "or.h"
+#include "channel.h"
+#include "config.h"
+#include "geoip.h"
+#include "main.h"
+#include "networkstatus.h"
+#include "nodelist.h"
+#include "router.h"
+
+#include "dos.h"
+
+/*
+ * Circuit creation denial of service mitigation.
+ *
+ * Namespace used for this mitigation framework is "dos_cc_" where "cc" is for
+ * Circuit Creation.
+ */
+
+/* Is the circuit creation DoS mitigation enabled? */
+static unsigned int dos_cc_enabled = 0;
+
+/* Consensus parameters. They can be changed when a new consensus arrives.
+ * They are initialized with the hardcoded default values. */
+static uint32_t dos_cc_min_concurrent_conn;
+static uint32_t dos_cc_circuit_rate;
+static uint32_t dos_cc_circuit_burst;
+static dos_cc_defense_type_t dos_cc_defense_type;
+static int32_t dos_cc_defense_time_period;
+
+/* Keep some stats for the heartbeat so we can report out. */
+static uint64_t cc_num_rejected_cells;
+static uint32_t cc_num_marked_addrs;
+
+/*
+ * Concurrent connection denial of service mitigation.
+ *
+ * Namespace used for this mitigation framework is "dos_conn_".
+ */
+
+/* Is the connection DoS mitigation enabled? */
+static unsigned int dos_conn_enabled = 0;
+
+/* Consensus parameters. They can be changed when a new consensus arrives.
+ * They are initialized with the hardcoded default values. */
+static uint32_t dos_conn_max_concurrent_count;
+static dos_conn_defense_type_t dos_conn_defense_type;
+
+/* Keep some stats for the heartbeat so we can report out. */
+static uint64_t conn_num_addr_rejected;
+
+/*
+ * General interface of the denial of service mitigation subsystem.
+ */
+
+/* Keep stats for the heartbeat. */
+static uint64_t num_single_hop_client_refused;
+
+/* Return true iff the circuit creation mitigation is enabled. We look at the
+ * consensus for this else a default value is returned. */
+MOCK_IMPL(STATIC unsigned int,
+get_param_cc_enabled, (const networkstatus_t *ns))
+{
+  if (get_options()->DoSCircuitCreationEnabled != -1) {
+    return get_options()->DoSCircuitCreationEnabled;
+  }
+
+  return !!networkstatus_get_param(ns, "DoSCircuitCreationEnabled",
+                                   DOS_CC_ENABLED_DEFAULT, 0, 1);
+}
+
+/* Return the parameter for the minimum concurrent connection at which we'll
+ * start counting circuit for a specific client address. */
+STATIC uint32_t
+get_param_cc_min_concurrent_connection(const networkstatus_t *ns)
+{
+  if (get_options()->DoSCircuitCreationMinConnections) {
+    return get_options()->DoSCircuitCreationMinConnections;
+  }
+  return networkstatus_get_param(ns, "DoSCircuitCreationMinConnections",
+                                 DOS_CC_MIN_CONCURRENT_CONN_DEFAULT,
+                                 1, INT32_MAX);
+}
+
+/* Return the parameter for the time rate that is how many circuits over this
+ * time span. */
+static uint32_t
+get_param_cc_circuit_rate(const networkstatus_t *ns)
+{
+  /* This is in seconds. */
+  if (get_options()->DoSCircuitCreationRate) {
+    return get_options()->DoSCircuitCreationRate;
+  }
+  return networkstatus_get_param(ns, "DoSCircuitCreationRate",
+                                 DOS_CC_CIRCUIT_RATE_DEFAULT,
+                                 1, INT32_MAX);
+}
+
+/* Return the parameter for the maximum circuit count for the circuit time
+ * rate. */
+STATIC uint32_t
+get_param_cc_circuit_burst(const networkstatus_t *ns)
+{
+  if (get_options()->DoSCircuitCreationBurst) {
+    return get_options()->DoSCircuitCreationBurst;
+  }
+  return networkstatus_get_param(ns, "DoSCircuitCreationBurst",
+                                 DOS_CC_CIRCUIT_BURST_DEFAULT,
+                                 1, INT32_MAX);
+}
+
+/* Return the consensus parameter of the circuit creation defense type. */
+static uint32_t
+get_param_cc_defense_type(const networkstatus_t *ns)
+{
+  if (get_options()->DoSCircuitCreationDefenseType) {
+    return get_options()->DoSCircuitCreationDefenseType;
+  }
+  return networkstatus_get_param(ns, "DoSCircuitCreationDefenseType",
+                                 DOS_CC_DEFENSE_TYPE_DEFAULT,
+                                 DOS_CC_DEFENSE_NONE, DOS_CC_DEFENSE_MAX);
+}
+
+/* Return the consensus parameter of the defense time period which is how much
+ * time should we defend against a malicious client address. */
+static int32_t
+get_param_cc_defense_time_period(const networkstatus_t *ns)
+{
+  /* Time in seconds. */
+  if (get_options()->DoSCircuitCreationDefenseTimePeriod) {
+    return get_options()->DoSCircuitCreationDefenseTimePeriod;
+  }
+  return networkstatus_get_param(ns, "DoSCircuitCreationDefenseTimePeriod",
+                                 DOS_CC_DEFENSE_TIME_PERIOD_DEFAULT,
+                                 0, INT32_MAX);
+}
+
+/* Return true iff connection mitigation is enabled. We look at the consensus
+ * for this else a default value is returned. */
+MOCK_IMPL(STATIC unsigned int,
+get_param_conn_enabled, (const networkstatus_t *ns))
+{
+  if (get_options()->DoSConnectionEnabled != -1) {
+    return get_options()->DoSConnectionEnabled;
+  }
+  return !!networkstatus_get_param(ns, "DoSConnectionEnabled",
+                                   DOS_CONN_ENABLED_DEFAULT, 0, 1);
+}
+
+/* Return the consensus parameter for the maximum concurrent connection
+ * allowed. */
+STATIC uint32_t
+get_param_conn_max_concurrent_count(const networkstatus_t *ns)
+{
+  if (get_options()->DoSConnectionMaxConcurrentCount) {
+    return get_options()->DoSConnectionMaxConcurrentCount;
+  }
+  return networkstatus_get_param(ns, "DoSConnectionMaxConcurrentCount",
+                                 DOS_CONN_MAX_CONCURRENT_COUNT_DEFAULT,
+                                 1, INT32_MAX);
+}
+
+/* Return the consensus parameter of the connection defense type. */
+static uint32_t
+get_param_conn_defense_type(const networkstatus_t *ns)
+{
+  if (get_options()->DoSConnectionDefenseType) {
+    return get_options()->DoSConnectionDefenseType;
+  }
+  return networkstatus_get_param(ns, "DoSConnectionDefenseType",
+                                 DOS_CONN_DEFENSE_TYPE_DEFAULT,
+                                 DOS_CONN_DEFENSE_NONE, DOS_CONN_DEFENSE_MAX);
+}
+
+/* Set circuit creation parameters located in the consensus or their default
+ * if none are present. Called at initialization or when the consensus
+ * changes. */
+static void
+set_dos_parameters(const networkstatus_t *ns)
+{
+  /* Get the default consensus param values. */
+  dos_cc_enabled = get_param_cc_enabled(ns);
+  dos_cc_min_concurrent_conn = get_param_cc_min_concurrent_connection(ns);
+  dos_cc_circuit_rate = get_param_cc_circuit_rate(ns);
+  dos_cc_circuit_burst = get_param_cc_circuit_burst(ns);
+  dos_cc_defense_time_period = get_param_cc_defense_time_period(ns);
+  dos_cc_defense_type = get_param_cc_defense_type(ns);
+
+  /* Connection detection. */
+  dos_conn_enabled = get_param_conn_enabled(ns);
+  dos_conn_max_concurrent_count = get_param_conn_max_concurrent_count(ns);
+  dos_conn_defense_type = get_param_conn_defense_type(ns);
+}
+
+/* Free everything for the circuit creation DoS mitigation subsystem. */
+static void
+cc_free_all(void)
+{
+  /* If everything is freed, the circuit creation subsystem is not enabled. */
+  dos_cc_enabled = 0;
+}
+
+/* Called when the consensus has changed. Do appropriate actions for the
+ * circuit creation subsystem. */
+static void
+cc_consensus_has_changed(const networkstatus_t *ns)
+{
+  /* Looking at the consensus, is the circuit creation subsystem enabled? If
+   * not and it was enabled before, clean it up. */
+  if (dos_cc_enabled && !get_param_cc_enabled(ns)) {
+    cc_free_all();
+  }
+}
+
+/** Return the number of circuits we allow per second under the current
+ *  configuration. */
+STATIC uint64_t
+get_circuit_rate_per_second(void)
+{
+  return dos_cc_circuit_rate;
+}
+
+/* Given the circuit creation client statistics object, refill the circuit
+ * bucket if needed. This also works if the bucket was never filled in the
+ * first place. The addr is only used for logging purposes. */
+STATIC void
+cc_stats_refill_bucket(cc_client_stats_t *stats, const tor_addr_t *addr)
+{
+  uint32_t new_circuit_bucket_count;
+  uint64_t num_token, elapsed_time_last_refill = 0, circuit_rate = 0;
+  time_t now;
+  int64_t last_refill_ts;
+
+  tor_assert(stats);
+  tor_assert(addr);
+
+  now = approx_time();
+  last_refill_ts = (int64_t)stats->last_circ_bucket_refill_ts;
+
+  /* If less than a second has elapsed, don't add any tokens.
+   * Note: If a relay's clock is ever 0, any new clients won't get a refill
+   * until the next second. But a relay that thinks it is 1970 will never
+   * validate the public consensus. */
+  if ((int64_t)now == last_refill_ts) {
+    goto done;
+  }
+
+  /* At this point, we know we might need to add token to the bucket. We'll
+   * first get the circuit rate that is how many circuit are we allowed to do
+   * per second. */
+  circuit_rate = get_circuit_rate_per_second();
+
+  /* We've never filled the bucket so fill it with the maximum being the burst
+   * and we are done.
+   * Note: If a relay's clock is ever 0, all clients that were last refilled
+   * in that zero second will get a full refill here. */
+  if (last_refill_ts == 0) {
+    num_token = dos_cc_circuit_burst;
+    goto end;
+  }
+
+  /* Our clock jumped backward so fill it up to the maximum. Not filling it
+   * could trigger a detection for a valid client. Also, if the clock jumped
+   * negative but we didn't notice until the elapsed time became positive
+   * again, then we potentially spent many seconds not refilling the bucket
+   * when we should have been refilling it. But the fact that we didn't notice
+   * until now means that no circuit creation requests came in during that
+   * time, so the client doesn't end up punished that much from this hopefully
+   * rare situation.*/
+  if ((int64_t)now < last_refill_ts) {
+    /* Use the maximum allowed value of token. */
+    num_token = dos_cc_circuit_burst;
+    goto end;
+  }
+
+  /* How many seconds have elapsed between now and the last refill?
+   * This subtraction can't underflow, because now >= last_refill_ts.
+   * And it can't overflow, because INT64_MAX - (-INT64_MIN) == UINT64_MAX. */
+  elapsed_time_last_refill = (uint64_t)now - last_refill_ts;
+
+  /* If the elapsed time is very large, it means our clock jumped forward.
+   * If the multiplication would overflow, use the maximum allowed value. */
+  if (elapsed_time_last_refill > UINT32_MAX) {
+    num_token = dos_cc_circuit_burst;
+    goto end;
+  }
+
+  /* Compute how many circuits we are allowed in that time frame which we'll
+   * add to the bucket. This can't overflow, because both multiplicands
+   * are less than or equal to UINT32_MAX, and num_token is uint64_t. */
+  num_token = elapsed_time_last_refill * circuit_rate;
+
+ end:
+  /* If the sum would overflow, use the maximum allowed value. */
+  if (num_token > UINT32_MAX - stats->circuit_bucket) {
+    new_circuit_bucket_count = dos_cc_circuit_burst;
+  } else {
+    /* We cap the bucket to the burst value else this could overflow uint32_t
+     * over time. */
+    new_circuit_bucket_count = MIN(stats->circuit_bucket + (uint32_t)num_token,
+                                   dos_cc_circuit_burst);
+  }
+
+  /* This function is not allowed to make the bucket count larger than the
+   * burst value */
+  tor_assert_nonfatal(new_circuit_bucket_count <= dos_cc_circuit_burst);
+  /* This function is not allowed to make the bucket count smaller, unless it
+   * is decreasing it to a newly configured, lower burst value. We allow the
+   * bucket to stay the same size, in case the circuit rate is zero. */
+  tor_assert_nonfatal(new_circuit_bucket_count >= stats->circuit_bucket ||
+                      new_circuit_bucket_count == dos_cc_circuit_burst);
+
+  log_debug(LD_DOS, "DoS address %s has its circuit bucket value: %" PRIu32
+                    ". Filling it to %" PRIu32 ". Circuit rate is %" PRIu64
+                    ". Elapsed time is %" PRIi64,
+            fmt_addr(addr), stats->circuit_bucket, new_circuit_bucket_count,
+            circuit_rate, (int64_t)elapsed_time_last_refill);
+
+  stats->circuit_bucket = new_circuit_bucket_count;
+  stats->last_circ_bucket_refill_ts = now;
+
+ done:
+  return;
+}
+
+/* Return true iff the circuit bucket is down to 0 and the number of
+ * concurrent connections is greater or equal the minimum threshold set the
+ * consensus parameter. */
+static int
+cc_has_exhausted_circuits(const dos_client_stats_t *stats)
+{
+  tor_assert(stats);
+  return stats->cc_stats.circuit_bucket == 0 &&
+         stats->concurrent_count >= dos_cc_min_concurrent_conn;
+}
+
+/* Mark client address by setting a timestamp in the stats object which tells
+ * us until when it is marked as positively detected. */
+static void
+cc_mark_client(cc_client_stats_t *stats)
+{
+  tor_assert(stats);
+  /* We add a random offset of a maximum of half the defense time so it is
+   * less predictable. */
+  stats->marked_until_ts =
+    approx_time() + dos_cc_defense_time_period +
+    crypto_rand_int_range(1, dos_cc_defense_time_period / 2);
+}
+
+/* Return true iff the given channel address is marked as malicious. This is
+ * called a lot and part of the fast path of handling cells. It has to remain
+ * as fast as we can. */
+static int
+cc_channel_addr_is_marked(channel_t *chan)
+{
+  time_t now;
+  tor_addr_t addr;
+  clientmap_entry_t *entry;
+  cc_client_stats_t *stats = NULL;
+
+  if (chan == NULL) {
+    goto end;
+  }
+  /* Must be a client connection else we ignore. */
+  if (!channel_is_client(chan)) {
+    goto end;
+  }
+  /* Without an IP address, nothing can work. */
+  if (!channel_get_addr_if_possible(chan, &addr)) {
+    goto end;
+  }
+
+  /* We are only interested in client connection from the geoip cache. */
+  entry = geoip_lookup_client(&addr, NULL, GEOIP_CLIENT_CONNECT);
+  if (entry == NULL) {
+    /* We can have a connection creating circuits but not tracked by the geoip
+     * cache. Once this DoS subsystem is enabled, we can end up here with no
+     * entry for the channel. */
+    goto end;
+  }
+  now = approx_time();
+  stats = &entry->dos_stats.cc_stats;
+
+ end:
+  return stats && stats->marked_until_ts >= now;
+}
+
+/* Concurrent connection private API. */
+
+/* Free everything for the connection DoS mitigation subsystem. */
+static void
+conn_free_all(void)
+{
+  dos_conn_enabled = 0;
+}
+
+/* Called when the consensus has changed. Do appropriate actions for the
+ * connection mitigation subsystem. */
+static void
+conn_consensus_has_changed(const networkstatus_t *ns)
+{
+  /* Looking at the consensus, is the connection mitigation subsystem enabled?
+   * If not and it was enabled before, clean it up. */
+  if (dos_conn_enabled && !get_param_conn_enabled(ns)) {
+    conn_free_all();
+  }
+}
+
+/* General private API */
+
+/* Return true iff we have at least one DoS detection enabled. This is used to
+ * decide if we need to allocate any kind of high level DoS object. */
+static inline int
+dos_is_enabled(void)
+{
+  return (dos_cc_enabled || dos_conn_enabled);
+}
+
+/* Circuit creation public API. */
+
+/* Called when a CREATE cell is received from the given channel. */
+void
+dos_cc_new_create_cell(channel_t *chan)
+{
+  tor_addr_t addr;
+  clientmap_entry_t *entry;
+
+  tor_assert(chan);
+
+  /* Skip everything if not enabled. */
+  if (!dos_cc_enabled) {
+    goto end;
+  }
+
+  /* Must be a client connection else we ignore. */
+  if (!channel_is_client(chan)) {
+    goto end;
+  }
+  /* Without an IP address, nothing can work. */
+  if (!channel_get_addr_if_possible(chan, &addr)) {
+    goto end;
+  }
+
+  /* We are only interested in client connection from the geoip cache. */
+  entry = geoip_lookup_client(&addr, NULL, GEOIP_CLIENT_CONNECT);
+  if (entry == NULL) {
+    /* We can have a connection creating circuits but not tracked by the geoip
+     * cache. Once this DoS subsystem is enabled, we can end up here with no
+     * entry for the channel. */
+    goto end;
+  }
+
+  /* General comment. Even though the client can already be marked as
+   * malicious, we continue to track statistics. If it keeps going above
+   * threshold while marked, the defense period time will grow longer. There
+   * is really no point at unmarking a client that keeps DoSing us. */
+
+  /* First of all, we'll try to refill the circuit bucket opportunistically
+   * before we assess. */
+  cc_stats_refill_bucket(&entry->dos_stats.cc_stats, &addr);
+
+  /* Take a token out of the circuit bucket if we are above 0 so we don't
+   * underflow the bucket. */
+  if (entry->dos_stats.cc_stats.circuit_bucket > 0) {
+    entry->dos_stats.cc_stats.circuit_bucket--;
+  }
+
+  /* This is the detection. Assess at every CREATE cell if the client should
+   * get marked as malicious. This should be kept as fast as possible. */
+  if (cc_has_exhausted_circuits(&entry->dos_stats)) {
+    /* If this is the first time we mark this entry, log it a info level.
+     * Under heavy DDoS, logging each time we mark would results in lots and
+     * lots of logs. */
+    if (entry->dos_stats.cc_stats.marked_until_ts == 0) {
+      log_debug(LD_DOS, "Detected circuit creation DoS by address: %s",
+                fmt_addr(&addr));
+      cc_num_marked_addrs++;
+    }
+    cc_mark_client(&entry->dos_stats.cc_stats);
+  }
+
+ end:
+  return;
+}
+
+/* Return the defense type that should be used for this circuit.
+ *
+ * This is part of the fast path and called a lot. */
+dos_cc_defense_type_t
+dos_cc_get_defense_type(channel_t *chan)
+{
+  tor_assert(chan);
+
+  /* Skip everything if not enabled. */
+  if (!dos_cc_enabled) {
+    goto end;
+  }
+
+  /* On an OR circuit, we'll check if the previous channel is a marked client
+   * connection detected by our DoS circuit creation mitigation subsystem. */
+  if (cc_channel_addr_is_marked(chan)) {
+    /* We've just assess that this circuit should trigger a defense for the
+     * cell it just seen. Note it down. */
+    cc_num_rejected_cells++;
+    return dos_cc_defense_type;
+  }
+
+ end:
+  return DOS_CC_DEFENSE_NONE;
+}
+
+/* Concurrent connection detection public API. */
+
+/* Return true iff the given address is permitted to open another connection.
+ * A defense value is returned for the caller to take appropriate actions. */
+dos_conn_defense_type_t
+dos_conn_addr_get_defense_type(const tor_addr_t *addr)
+{
+  clientmap_entry_t *entry;
+
+  tor_assert(addr);
+
+  /* Skip everything if not enabled. */
+  if (!dos_conn_enabled) {
+    goto end;
+  }
+
+  /* We are only interested in client connection from the geoip cache. */
+  entry = geoip_lookup_client(addr, NULL, GEOIP_CLIENT_CONNECT);
+  if (entry == NULL) {
+    goto end;
+  }
+
+  /* Need to be above the maximum concurrent connection count to trigger a
+   * defense. */
+  if (entry->dos_stats.concurrent_count > dos_conn_max_concurrent_count) {
+    conn_num_addr_rejected++;
+    return dos_conn_defense_type;
+  }
+
+ end:
+  return DOS_CONN_DEFENSE_NONE;
+}
+
+/* General API */
+
+/* Take any appropriate actions for the given geoip entry that is about to get
+ * freed. This is called for every entry that is being freed.
+ *
+ * This function will clear out the connection tracked flag if the concurrent
+ * count of the entry is above 0 so if those connections end up being seen by
+ * this subsystem, we won't try to decrement the counter for a new geoip entry
+ * that might have been added after this call for the same address. */
+void
+dos_geoip_entry_about_to_free(const clientmap_entry_t *geoip_ent)
+{
+  tor_assert(geoip_ent);
+
+  /* The count is down to 0 meaning no connections right now, we can safely
+   * clear the geoip entry from the cache. */
+  if (geoip_ent->dos_stats.concurrent_count == 0) {
+    goto end;
+  }
+
+  /* For each connection matching the geoip entry address, we'll clear the
+   * tracked flag because the entry is about to get removed from the geoip
+   * cache. We do not try to decrement if the flag is not set. */
+  SMARTLIST_FOREACH_BEGIN(get_connection_array(), connection_t *, conn) {
+    if (conn->type == CONN_TYPE_OR) {
+      or_connection_t *or_conn = TO_OR_CONN(conn);
+      if (!tor_addr_compare(&geoip_ent->addr, &or_conn->real_addr,
+                            CMP_EXACT)) {
+        or_conn->tracked_for_dos_mitigation = 0;
+      }
+    }
+  } SMARTLIST_FOREACH_END(conn);
+
+ end:
+  return;
+}
+
+/* Note down that we've just refused a single hop client. This increments a
+ * counter later used for the heartbeat. */
+void
+dos_note_refuse_single_hop_client(void)
+{
+  num_single_hop_client_refused++;
+}
+
+/* Return true iff single hop client connection (ESTABLISH_RENDEZVOUS) should
+ * be refused. */
+int
+dos_should_refuse_single_hop_client(void)
+{
+  /* If we aren't a public relay, this shouldn't apply to anything. */
+  if (!public_server_mode(get_options())) {
+    return 0;
+  }
+
+  if (get_options()->DoSRefuseSingleHopClientRendezvous != -1) {
+    return get_options()->DoSRefuseSingleHopClientRendezvous;
+  }
+
+  return (int) networkstatus_get_param(NULL,
+                                       "DoSRefuseSingleHopClientRendezvous",
+                                       0 /* default */, 0, 1);
+}
+
+/* Log a heartbeat message with some statistics. */
+void
+dos_log_heartbeat(void)
+{
+  char *conn_msg = NULL;
+  char *cc_msg = NULL;
+  char *single_hop_client_msg = NULL;
+
+  if (!dos_is_enabled()) {
+    goto end;
+  }
+
+  if (dos_cc_enabled) {
+    tor_asprintf(&cc_msg,
+                 " %" PRIu64 " circuits rejected,"
+                 " %" PRIu32 " marked addresses.",
+                 cc_num_rejected_cells, cc_num_marked_addrs);
+  }
+
+  if (dos_conn_enabled) {
+    tor_asprintf(&conn_msg,
+                 " %" PRIu64 " connections closed.",
+                 conn_num_addr_rejected);
+  }
+
+  if (dos_should_refuse_single_hop_client()) {
+    tor_asprintf(&single_hop_client_msg,
+                 " %" PRIu64 " single hop clients refused.",
+                 num_single_hop_client_refused);
+  }
+
+  log_notice(LD_HEARTBEAT,
+             "DoS mitigation since startup:%s%s%s",
+             (cc_msg != NULL) ? cc_msg : " [cc not enabled]",
+             (conn_msg != NULL) ? conn_msg : " [conn not enabled]",
+             (single_hop_client_msg != NULL) ? single_hop_client_msg : "");
+
+  tor_free(conn_msg);
+  tor_free(cc_msg);
+  tor_free(single_hop_client_msg);
+
+ end:
+  return;
+}
+
+/* Called when a new client connection has been established on the given
+ * address. */
+void
+dos_new_client_conn(or_connection_t *or_conn)
+{
+  clientmap_entry_t *entry;
+
+  tor_assert(or_conn);
+
+  /* Past that point, we know we have at least one DoS detection subsystem
+   * enabled so we'll start allocating stuff. */
+  if (!dos_is_enabled()) {
+    goto end;
+  }
+
+  /* We ignore any known address meaning an address of a known relay. The
+   * reason to do so is because network reentry is possible where a client
+   * connection comes from an Exit node. Even when we'll fix reentry, this is
+   * a robust defense to keep in place. */
+  if (nodelist_probably_contains_address(&or_conn->real_addr)) {
+    goto end;
+  }
+
+  /* We are only interested in client connection from the geoip cache. */
+  entry = geoip_lookup_client(&or_conn->real_addr, NULL,
+                              GEOIP_CLIENT_CONNECT);
+  if (BUG(entry == NULL)) {
+    /* Should never happen because we note down the address in the geoip
+     * cache before this is called. */
+    goto end;
+  }
+
+  entry->dos_stats.concurrent_count++;
+  or_conn->tracked_for_dos_mitigation = 1;
+  log_debug(LD_DOS, "Client address %s has now %u concurrent connections.",
+            fmt_addr(&or_conn->real_addr),
+            entry->dos_stats.concurrent_count);
+
+ end:
+  return;
+}
+
+/* Called when a client connection for the given IP address has been closed. */
+void
+dos_close_client_conn(const or_connection_t *or_conn)
+{
+  clientmap_entry_t *entry;
+
+  tor_assert(or_conn);
+
+  /* We have to decrement the count on tracked connection only even if the
+   * subsystem has been disabled at runtime because it might be re-enabled
+   * after and we need to keep a synchronized counter at all time. */
+  if (!or_conn->tracked_for_dos_mitigation) {
+    goto end;
+  }
+
+  /* We are only interested in client connection from the geoip cache. */
+  entry = geoip_lookup_client(&or_conn->real_addr, NULL,
+                              GEOIP_CLIENT_CONNECT);
+  if (entry == NULL) {
+    /* This can happen because we can close a connection before the channel
+     * got to be noted down in the geoip cache. */
+    goto end;
+  }
+
+  /* Extra super duper safety. Going below 0 means an underflow which could
+   * lead to most likely a false positive. In theory, this should never happen
+   * but lets be extra safe. */
+  if (BUG(entry->dos_stats.concurrent_count == 0)) {
+    goto end;
+  }
+
+  entry->dos_stats.concurrent_count--;
+  log_debug(LD_DOS, "Client address %s has lost a connection. Concurrent "
+                    "connections are now at %u",
+            fmt_addr(&or_conn->real_addr),
+            entry->dos_stats.concurrent_count);
+
+ end:
+  return;
+}
+
+/* Called when the consensus has changed. We might have new consensus
+ * parameters to look at. */
+void
+dos_consensus_has_changed(const networkstatus_t *ns)
+{
+  /* There are two ways to configure this subsystem, one at startup through
+   * dos_init() which is called when the options are parsed. And this one
+   * through the consensus. We don't want to enable any DoS mitigation if we
+   * aren't a public relay. */
+  if (!public_server_mode(get_options())) {
+    return;
+  }
+
+  cc_consensus_has_changed(ns);
+  conn_consensus_has_changed(ns);
+
+  /* We were already enabled or we just became enabled but either way, set the
+   * consensus parameters for all subsystems. */
+  set_dos_parameters(ns);
+}
+
+/* Return true iff the DoS mitigation subsystem is enabled. */
+int
+dos_enabled(void)
+{
+  return dos_is_enabled();
+}
+
+/* Free everything from the Denial of Service subsystem. */
+void
+dos_free_all(void)
+{
+  /* Free the circuit creation mitigation subsystem. It is safe to do this
+   * even if it wasn't initialized. */
+  cc_free_all();
+
+  /* Free the connection mitigation subsystem. It is safe to do this even if
+   * it wasn't initialized. */
+  conn_free_all();
+}
+
+/* Initialize the Denial of Service subsystem. */
+void
+dos_init(void)
+{
+  /* To initialize, we only need to get the parameters. */
+  set_dos_parameters(NULL);
+}
+
diff --git a/src/or/dos.h b/src/or/dos.h
new file mode 100644
index 0000000000..5d35a2b12e
--- /dev/null
+++ b/src/or/dos.h
@@ -0,0 +1,140 @@
+/* Copyright (c) 2018, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+/*
+ * \file dos.h
+ * \brief Header file for dos.c
+ */
+
+#ifndef TOR_DOS_H
+#define TOR_DOS_H
+
+/* Structure that keeps stats of client connection per-IP. */
+typedef struct cc_client_stats_t {
+  /* Number of allocated circuits remaining for this address.  It is
+   * decremented every time a new circuit is seen for this client address and
+   * if the count goes to 0, we have a positive detection. */
+  uint32_t circuit_bucket;
+
+  /* When was the last time we've refilled the circuit bucket? This is used to
+   * know if we need to refill the bucket when a new circuit is seen. It is
+   * synchronized using approx_time(). */
+  time_t last_circ_bucket_refill_ts;
+
+  /* This client address was detected to be above the circuit creation rate
+   * and this timestamp indicates until when it should remain marked as
+   * detected so we can apply a defense for the address. It is synchronized
+   * using the approx_time(). */
+  time_t marked_until_ts;
+} cc_client_stats_t;
+
+/* This object is a top level object that contains everything related to the
+ * per-IP client DoS mitigation. Because it is per-IP, it is used in the geoip
+ * clientmap_entry_t object. */
+typedef struct dos_client_stats_t {
+  /* Concurrent connection count from the specific address. 2^32 is most
+   * likely way too big for the amount of allowed file descriptors. */
+  uint32_t concurrent_count;
+
+  /* Circuit creation statistics. This is only used if the circuit creation
+   * subsystem has been enabled (dos_cc_enabled). */
+  cc_client_stats_t cc_stats;
+} dos_client_stats_t;
+
+/* General API. */
+
+/* Stub. */
+struct clientmap_entry_t;
+
+void dos_init(void);
+void dos_free_all(void);
+void dos_consensus_has_changed(const networkstatus_t *ns);
+int dos_enabled(void);
+void dos_log_heartbeat(void);
+void dos_geoip_entry_about_to_free(const struct clientmap_entry_t *geoip_ent);
+
+void dos_new_client_conn(or_connection_t *or_conn);
+void dos_close_client_conn(const or_connection_t *or_conn);
+
+int dos_should_refuse_single_hop_client(void);
+void dos_note_refuse_single_hop_client(void);
+
+/*
+ * Circuit creation DoS mitigation subsystemn interface.
+ */
+
+/* DoSCircuitCreationEnabled default. Disabled by default. */
+#define DOS_CC_ENABLED_DEFAULT 0
+/* DoSCircuitCreationDefenseType maps to the dos_cc_defense_type_t enum. */
+#define DOS_CC_DEFENSE_TYPE_DEFAULT DOS_CC_DEFENSE_REFUSE_CELL
+/* DoSCircuitCreationMinConnections default */
+#define DOS_CC_MIN_CONCURRENT_CONN_DEFAULT 3
+/* DoSCircuitCreationRateTenths is 3 per seconds. */
+#define DOS_CC_CIRCUIT_RATE_DEFAULT 3
+/* DoSCircuitCreationBurst default. */
+#define DOS_CC_CIRCUIT_BURST_DEFAULT 90
+/* DoSCircuitCreationDefenseTimePeriod in seconds. */
+#define DOS_CC_DEFENSE_TIME_PERIOD_DEFAULT (60 * 60)
+
+/* Type of defense that we can use for the circuit creation DoS mitigation. */
+typedef enum dos_cc_defense_type_t {
+  /* No defense used. */
+  DOS_CC_DEFENSE_NONE             = 1,
+  /* Refuse any cells which means a DESTROY cell will be sent back. */
+  DOS_CC_DEFENSE_REFUSE_CELL      = 2,
+
+  /* Maximum value that can be used. Useful for the boundaries of the
+   * consensus parameter. */
+  DOS_CC_DEFENSE_MAX              = 2,
+} dos_cc_defense_type_t;
+
+void dos_cc_new_create_cell(channel_t *channel);
+dos_cc_defense_type_t dos_cc_get_defense_type(channel_t *chan);
+
+/*
+ * Concurrent connection DoS mitigation interface.
+ */
+
+/* DoSConnectionEnabled default. Disabled by default. */
+#define DOS_CONN_ENABLED_DEFAULT 0
+/* DoSConnectionMaxConcurrentCount default. */
+#define DOS_CONN_MAX_CONCURRENT_COUNT_DEFAULT 100
+/* DoSConnectionDefenseType maps to the dos_conn_defense_type_t enum. */
+#define DOS_CONN_DEFENSE_TYPE_DEFAULT DOS_CONN_DEFENSE_CLOSE
+
+/* Type of defense that we can use for the concurrent connection DoS
+ * mitigation. */
+typedef enum dos_conn_defense_type_t {
+  /* No defense used. */
+  DOS_CONN_DEFENSE_NONE             = 1,
+  /* Close immediately the connection meaning refuse it. */
+  DOS_CONN_DEFENSE_CLOSE            = 2,
+
+  /* Maximum value that can be used. Useful for the boundaries of the
+   * consensus parameter. */
+  DOS_CONN_DEFENSE_MAX              = 2,
+} dos_conn_defense_type_t;
+
+dos_conn_defense_type_t dos_conn_addr_get_defense_type(const tor_addr_t *addr);
+
+#ifdef DOS_PRIVATE
+
+STATIC uint32_t get_param_conn_max_concurrent_count(
+                                              const networkstatus_t *ns);
+STATIC uint32_t get_param_cc_circuit_burst(const networkstatus_t *ns);
+STATIC uint32_t get_param_cc_min_concurrent_connection(
+                                            const networkstatus_t *ns);
+
+STATIC uint64_t get_circuit_rate_per_second(void);
+STATIC void cc_stats_refill_bucket(cc_client_stats_t *stats,
+                                   const tor_addr_t *addr);
+
+MOCK_DECL(STATIC unsigned int, get_param_cc_enabled,
+          (const networkstatus_t *ns));
+MOCK_DECL(STATIC unsigned int, get_param_conn_enabled,
+          (const networkstatus_t *ns));
+
+#endif /* TOR_DOS_PRIVATE */
+
+#endif /* TOR_DOS_H */
+
diff --git a/src/or/geoip.c b/src/or/geoip.c
index 00c055bbe7..a39366ed13 100644
--- a/src/or/geoip.c
+++ b/src/or/geoip.c
@@ -33,6 +33,7 @@
 #include "config.h"
 #include "control.h"
 #include "dnsserv.h"
+#include "dos.h"
 #include "geoip.h"
 #include "routerlist.h"
 
@@ -72,6 +73,38 @@ static smartlist_t *geoip_ipv4_entries = NULL, *geoip_ipv6_entries = NULL;
 static char geoip_digest[DIGEST_LEN];
 static char geoip6_digest[DIGEST_LEN];
 
+/* Total size in bytes of the geoip client history cache. Used by the OOM
+ * handler. */
+static size_t geoip_client_history_cache_size;
+
+/* Increment the geoip client history cache size counter with the given bytes.
+ * This prevents an overflow and set it to its maximum in that case. */
+static inline void
+geoip_increment_client_history_cache_size(size_t bytes)
+{
+  /* This is shockingly high, lets log it so it can be reported. */
+  IF_BUG_ONCE(geoip_client_history_cache_size > (SIZE_MAX - bytes)) {
+    geoip_client_history_cache_size = SIZE_MAX;
+    return;
+  }
+  geoip_client_history_cache_size += bytes;
+}
+
+/* Decrement the geoip client history cache size counter with the given bytes.
+ * This prevents an underflow and set it to 0 in that case. */
+static inline void
+geoip_decrement_client_history_cache_size(size_t bytes)
+{
+  /* Going below 0 means that we either allocated an entry without
+   * incrementing the counter or we have different sizes when allocating and
+   * freeing. It shouldn't happened so log it. */
+  IF_BUG_ONCE(geoip_client_history_cache_size < bytes) {
+    geoip_client_history_cache_size = 0;
+    return;
+  }
+  geoip_client_history_cache_size -= bytes;
+}
+
 /** Return the index of the <b>country</b>'s entry in the GeoIP
  * country list if it is a valid 2-letter country code, otherwise
  * return -1. */
@@ -472,24 +505,6 @@ geoip_db_digest(sa_family_t family)
     return hex_str(geoip6_digest, DIGEST_LEN);
 }
 
-/** Entry in a map from IP address to the last time we've seen an incoming
- * connection from that IP address. Used by bridges only, to track which
- * countries have them blocked. */
-typedef struct clientmap_entry_t {
-  HT_ENTRY(clientmap_entry_t) node;
-  tor_addr_t addr;
- /* Name of pluggable transport used by this client. NULL if no
-    pluggable transport was used. */
-  char *transport_name;
-
-  /** Time when we last saw this IP address, in MINUTES since the epoch.
-   *
-   * (This will run out of space around 4011 CE.  If Tor is still in use around
-   * 4000 CE, please remember to add more bits to last_seen_in_minutes.) */
-  unsigned int last_seen_in_minutes:30;
-  unsigned int action:2;
-} clientmap_entry_t;
-
 /** Largest allowable value for last_seen_in_minutes.  (It's a 30-bit field,
  * so it can hold up to (1u<<30)-1, or 0x3fffffffu.
  */
@@ -526,6 +541,15 @@ HT_PROTOTYPE(clientmap, clientmap_entry_t, node, clientmap_entry_hash,
 HT_GENERATE2(clientmap, clientmap_entry_t, node, clientmap_entry_hash,
              clientmap_entries_eq, 0.6, tor_reallocarray_, tor_free_)
 
+/** Return the size of a client map entry. */
+static inline size_t
+clientmap_entry_size(const clientmap_entry_t *ent)
+{
+  tor_assert(ent);
+  return (sizeof(clientmap_entry_t) +
+          (ent->transport_name ? strlen(ent->transport_name) : 0));
+}
+
 /** Free all storage held by <b>ent</b>. */
 static void
 clientmap_entry_free(clientmap_entry_t *ent)
@@ -533,10 +557,40 @@ clientmap_entry_free(clientmap_entry_t *ent)
   if (!ent)
     return;
 
+  /* This entry is about to be freed so pass it to the DoS subsystem to see if
+   * any actions can be taken about it. */
+  dos_geoip_entry_about_to_free(ent);
+  geoip_decrement_client_history_cache_size(clientmap_entry_size(ent));
+
   tor_free(ent->transport_name);
   tor_free(ent);
 }
 
+/* Return a newly allocated clientmap entry with the given action and address
+ * that are mandatory. The transport_name can be optional. This can't fail. */
+static clientmap_entry_t *
+clientmap_entry_new(geoip_client_action_t action, const tor_addr_t *addr,
+                    const char *transport_name)
+{
+  clientmap_entry_t *entry;
+
+  tor_assert(action == GEOIP_CLIENT_CONNECT ||
+             action == GEOIP_CLIENT_NETWORKSTATUS);
+  tor_assert(addr);
+
+  entry = tor_malloc_zero(sizeof(clientmap_entry_t));
+  entry->action = action;
+  tor_addr_copy(&entry->addr, addr);
+  if (transport_name) {
+    entry->transport_name = tor_strdup(transport_name);
+  }
+
+  /* Allocated and initialized, note down its size for the OOM handler. */
+  geoip_increment_client_history_cache_size(clientmap_entry_size(entry));
+
+  return entry;
+}
+
 /** Clear history of connecting clients used by entry and bridge stats. */
 static void
 client_history_clear(void)
@@ -564,14 +618,17 @@ geoip_note_client_seen(geoip_client_action_t action,
                        time_t now)
 {
   const or_options_t *options = get_options();
-  clientmap_entry_t lookup, *ent;
-  memset(&lookup, 0, sizeof(clientmap_entry_t));
+  clientmap_entry_t *ent;
 
   if (action == GEOIP_CLIENT_CONNECT) {
-    /* Only remember statistics as entry guard or as bridge. */
-    if (!options->EntryStatistics &&
-        (!(options->BridgeRelay && options->BridgeRecordUsageByCountry)))
-      return;
+    /* Only remember statistics if the DoS mitigation subsystem is enabled. If
+     * not, only if as entry guard or as bridge. */
+    if (!dos_enabled()) {
+      if (!options->EntryStatistics &&
+          (!(options->BridgeRelay && options->BridgeRecordUsageByCountry))) {
+        return;
+      }
+    }
   } else {
     /* Only gather directory-request statistics if configured, and
      * forcibly disable them on bridge authorities. */
@@ -583,17 +640,9 @@ geoip_note_client_seen(geoip_client_action_t action,
             safe_str_client(fmt_addr((addr))),
             transport_name ? transport_name : "<no transport>");
 
-  tor_addr_copy(&lookup.addr, addr);
-  lookup.action = (int)action;
-  lookup.transport_name = (char*) transport_name;
-  ent = HT_FIND(clientmap, &client_history, &lookup);
-
+  ent = geoip_lookup_client(addr, transport_name, action);
   if (! ent) {
-    ent = tor_malloc_zero(sizeof(clientmap_entry_t));
-    tor_addr_copy(&ent->addr, addr);
-    if (transport_name)
-      ent->transport_name = tor_strdup(transport_name);
-    ent->action = (int)action;
+    ent = clientmap_entry_new(action, addr, transport_name);
     HT_INSERT(clientmap, &client_history, ent);
   }
   if (now / 60 <= (int)MAX_LAST_SEEN_IN_MINUTES && now >= 0)
@@ -635,6 +684,94 @@ geoip_remove_old_clients(time_t cutoff)
                           &cutoff);
 }
 
+/* Return a client entry object matching the given address, transport name and
+ * geoip action from the clientmap. NULL if not found. The transport_name can
+ * be NULL. */
+clientmap_entry_t *
+geoip_lookup_client(const tor_addr_t *addr, const char *transport_name,
+                    geoip_client_action_t action)
+{
+  clientmap_entry_t lookup;
+
+  tor_assert(addr);
+
+  /* We always look for a client connection with no transport. */
+  tor_addr_copy(&lookup.addr, addr);
+  lookup.action = action;
+  lookup.transport_name = (char *) transport_name;
+
+  return HT_FIND(clientmap, &client_history, &lookup);
+}
+
+/* Cleanup client entries older than the cutoff. Used for the OOM. Return the
+ * number of bytes freed. If 0 is returned, nothing was freed. */
+static size_t
+oom_clean_client_entries(time_t cutoff)
+{
+  size_t bytes = 0;
+  clientmap_entry_t **ent, **ent_next;
+
+  for (ent = HT_START(clientmap, &client_history); ent; ent = ent_next) {
+    clientmap_entry_t *entry = *ent;
+    if (entry->last_seen_in_minutes < (cutoff / 60)) {
+      ent_next = HT_NEXT_RMV(clientmap, &client_history, ent);
+      bytes += clientmap_entry_size(entry);
+      clientmap_entry_free(entry);
+    } else {
+      ent_next = HT_NEXT(clientmap, &client_history, ent);
+    }
+  }
+  return bytes;
+}
+
+/* Below this minimum lifetime, the OOM won't cleanup any entries. */
+#define GEOIP_CLIENT_CACHE_OOM_MIN_CUTOFF (4 * 60 * 60)
+/* The OOM moves the cutoff by that much every run. */
+#define GEOIP_CLIENT_CACHE_OOM_STEP (15 * 50)
+
+/* Cleanup the geoip client history cache called from the OOM handler. Return
+ * the amount of bytes removed. This can return a value below or above
+ * min_remove_bytes but will stop as oon as the min_remove_bytes has been
+ * reached. */
+size_t
+geoip_client_cache_handle_oom(time_t now, size_t min_remove_bytes)
+{
+  time_t k;
+  size_t bytes_removed = 0;
+
+  /* Our OOM handler called with 0 bytes to remove is a code flow error. */
+  tor_assert(min_remove_bytes != 0);
+
+  /* Set k to the initial cutoff of an entry. We then going to move it by step
+   * to try to remove as much as we can. */
+  k = WRITE_STATS_INTERVAL;
+
+  do {
+    time_t cutoff;
+
+    /* If k has reached the minimum lifetime, we have to stop else we might
+     * remove every single entries which would be pretty bad for the DoS
+     * mitigation subsystem if by just filling the geoip cache, it was enough
+     * to trigger the OOM and clean every single entries. */
+    if (k <= GEOIP_CLIENT_CACHE_OOM_MIN_CUTOFF) {
+      break;
+    }
+
+    cutoff = now - k;
+    bytes_removed += oom_clean_client_entries(cutoff);
+    k -= GEOIP_CLIENT_CACHE_OOM_STEP;
+  } while (bytes_removed < min_remove_bytes);
+
+  return bytes_removed;
+}
+
+/* Return the total size in bytes of the client history cache. */
+size_t
+geoip_client_cache_total_allocation(void)
+{
+  return geoip_client_history_cache_size;
+}
+
 /** How many responses are we giving to clients requesting v3 network
  * statuses? */
 static uint32_t ns_v3_responses[GEOIP_NS_RESPONSE_NUM];
diff --git a/src/or/geoip.h b/src/or/geoip.h
index 070296dd07..c8ea9f85ea 100644
--- a/src/or/geoip.h
+++ b/src/or/geoip.h
@@ -13,6 +13,7 @@
 #define TOR_GEOIP_H
 
 #include "testsupport.h"
+#include "dos.h"
 
 #ifdef GEOIP_PRIVATE
 STATIC int geoip_parse_entry(const char *line, sa_family_t family);
@@ -20,6 +21,29 @@ STATIC int geoip_get_country_by_ipv4(uint32_t ipaddr);
 STATIC int geoip_get_country_by_ipv6(const struct in6_addr *addr);
 STATIC void clear_geoip_db(void);
 #endif
+
+/** Entry in a map from IP address to the last time we've seen an incoming
+ * connection from that IP address. Used by bridges only to track which
+ * countries have them blocked, or the DoS mitigation subsystem if enabled. */
+typedef struct clientmap_entry_t {
+  HT_ENTRY(clientmap_entry_t) node;
+  tor_addr_t addr;
+  /* Name of pluggable transport used by this client. NULL if no
+     pluggable transport was used. */
+  char *transport_name;
+
+  /** Time when we last saw this IP address, in MINUTES since the epoch.
+   *
+   * (This will run out of space around 4011 CE.  If Tor is still in use around
+   * 4000 CE, please remember to add more bits to last_seen_in_minutes.) */
+  unsigned int last_seen_in_minutes:30;
+  unsigned int action:2;
+
+  /* This object is used to keep some statistics per client address for the
+   * DoS mitigation subsystem. */
+  dos_client_stats_t dos_stats;
+} clientmap_entry_t;
+
 int should_record_bridge_info(const or_options_t *options);
 int geoip_load_file(sa_family_t family, const char *filename);
 MOCK_DECL(int, geoip_get_country_by_addr, (const tor_addr_t *addr));
@@ -33,6 +57,11 @@ void geoip_note_client_seen(geoip_client_action_t action,
                             const tor_addr_t *addr, const char *transport_name,
                             time_t now);
 void geoip_remove_old_clients(time_t cutoff);
+clientmap_entry_t *geoip_lookup_client(const tor_addr_t *addr,
+                                       const char *transport_name,
+                                       geoip_client_action_t action);
+size_t geoip_client_cache_total_allocation(void);
+size_t geoip_client_cache_handle_oom(time_t now, size_t min_remove_bytes);
 
 void geoip_note_ns_response(geoip_ns_response_t response);
 char *geoip_get_transport_history(void);
diff --git a/src/or/include.am b/src/or/include.am
index ae493b7225..5108a08e53 100644
--- a/src/or/include.am
+++ b/src/or/include.am
@@ -43,6 +43,7 @@ LIBTOR_A_SOURCES = \
 	src/or/dirvote.c				\
 	src/or/dns.c					\
 	src/or/dnsserv.c				\
+	src/or/dos.c					\
 	src/or/fp_pair.c				\
 	src/or/geoip.c					\
 	src/or/entrynodes.c				\
@@ -151,6 +152,7 @@ ORHEADERS = \
 	src/or/dns.h					\
 	src/or/dns_structs.h				\
 	src/or/dnsserv.h				\
+	src/or/dos.h					\
 	src/or/ext_orport.h				\
 	src/or/fallback_dirs.inc			\
 	src/or/fp_pair.h				\
diff --git a/src/or/main.c b/src/or/main.c
index 187b255bfb..fcd8dc9024 100644
--- a/src/or/main.c
+++ b/src/or/main.c
@@ -34,6 +34,7 @@
 #include "dirvote.h"
 #include "dns.h"
 #include "dnsserv.h"
+#include "dos.h"
 #include "entrynodes.h"
 #include "geoip.h"
 #include "hibernate.h"
@@ -2989,6 +2990,7 @@ tor_free_all(int postfork)
   control_free_all();
   sandbox_free_getaddrinfo_cache();
   protover_free_all();
+  dos_free_all();
   if (!postfork) {
     config_free_all();
     or_state_free_all();
diff --git a/src/or/networkstatus.c b/src/or/networkstatus.c
index 991cf80121..d9ae32560e 100644
--- a/src/or/networkstatus.c
+++ b/src/or/networkstatus.c
@@ -23,6 +23,7 @@
 #include "directory.h"
 #include "dirserv.h"
 #include "dirvote.h"
+#include "dos.h"
 #include "entrynodes.h"
 #include "main.h"
 #include "microdesc.h"
@@ -1502,6 +1503,15 @@ notify_control_networkstatus_changed(const networkstatus_t *old_c,
   smartlist_free(changed);
 }
 
+/* Called when the consensus has changed from old_c to new_c. */
+static void
+notify_networkstatus_changed(const networkstatus_t *old_c,
+                             const networkstatus_t *new_c)
+{
+  notify_control_networkstatus_changed(old_c, new_c);
+  dos_consensus_has_changed(new_c);
+}
+
 /** Copy all the ancillary information (like router download status and so on)
  * from <b>old_c</b> to <b>new_c</b>. */
 static void
@@ -1826,8 +1836,7 @@ networkstatus_set_current_consensus(const char *consensus,
   const int is_usable_flavor = flav == usable_consensus_flavor();
 
   if (is_usable_flavor) {
-    notify_control_networkstatus_changed(
-                         networkstatus_get_latest_consensus(), c);
+    notify_networkstatus_changed(networkstatus_get_latest_consensus(), c);
   }
   if (flav == FLAV_NS) {
     if (current_ns_consensus) {
diff --git a/src/or/nodelist.c b/src/or/nodelist.c
index 0e9a651818..5a02648c5c 100644
--- a/src/or/nodelist.c
+++ b/src/or/nodelist.c
@@ -14,6 +14,7 @@
 
 #include "or.h"
 #include "address.h"
+#include "address_set.h"
 #include "config.h"
 #include "control.h"
 #include "dirserv.h"
@@ -52,6 +53,7 @@ static void count_usable_descriptors(int *num_present,
 static void update_router_have_minimum_dir_info(void);
 static double get_frac_paths_needed_for_circs(const or_options_t *options,
                                               const networkstatus_t *ns);
+static void node_add_to_address_set(const node_t *node);
 
 /** A nodelist_t holds a node_t object for every router we're "willing to use
  * for something".  Specifically, it should hold a node_t for every node that
@@ -63,6 +65,8 @@ typedef struct nodelist_t {
   /* Hash table to map from node ID digest to node. */
   HT_HEAD(nodelist_map, node_t) nodes_by_id;
 
+  /* Set of addresses that belong to nodes we believe in. */
+  address_set_t *node_addrs;
 } nodelist_t;
 
 static inline unsigned int
@@ -150,6 +154,50 @@ node_addrs_changed(node_t *node)
   node->country = -1;
 }
 
+/** Add all address information about <b>node</b> to the current address
+ * set (if there is one).
+ */
+static void
+node_add_to_address_set(const node_t *node)
+{
+  if (!the_nodelist || !the_nodelist->node_addrs)
+    return;
+
+  /* These various address sources can be redundant, but it's likely faster
+   * to add them all than to compare them all for equality. */
+
+  if (node->rs) {
+    if (node->rs->addr)
+      address_set_add_ipv4h(the_nodelist->node_addrs, node->rs->addr);
+    if (!tor_addr_is_null(&node->rs->ipv6_addr))
+      address_set_add(the_nodelist->node_addrs, &node->rs->ipv6_addr);
+  }
+  if (node->ri) {
+    if (node->ri->addr)
+      address_set_add_ipv4h(the_nodelist->node_addrs, node->ri->addr);
+    if (!tor_addr_is_null(&node->ri->ipv6_addr))
+      address_set_add(the_nodelist->node_addrs, &node->ri->ipv6_addr);
+  }
+  if (node->md) {
+    if (!tor_addr_is_null(&node->md->ipv6_addr))
+      address_set_add(the_nodelist->node_addrs, &node->md->ipv6_addr);
+  }
+}
+
+/** Return true if <b>addr</b> is the address of some node in the nodelist.
+ * If not, probably return false. */
+int
+nodelist_probably_contains_address(const tor_addr_t *addr)
+{
+  if (BUG(!addr))
+    return 0;
+
+  if (!the_nodelist || !the_nodelist->node_addrs)
+    return 0;
+
+  return address_set_probably_contains(the_nodelist->node_addrs, addr);
+}
+
 /** Add <b>ri</b> to an appropriate node in the nodelist.  If we replace an
  * old routerinfo, and <b>ri_old_out</b> is not NULL, set *<b>ri_old_out</b>
  * to the previous routerinfo.
@@ -188,6 +236,8 @@ nodelist_set_routerinfo(routerinfo_t *ri, routerinfo_t **ri_old_out)
     dirserv_set_node_flags_from_authoritative_status(node, status);
   }
 
+  node_add_to_address_set(node);
+
   return node;
 }
 
@@ -219,9 +269,23 @@ nodelist_add_microdesc(microdesc_t *md)
     node->md = md;
     md->held_by_nodes++;
   }
+
+  node_add_to_address_set(node);
+
   return node;
 }
 
+/* Default value. */
+#define ESTIMATED_ADDRESS_PER_NODE 2
+
+/* Return the estimated number of address per node_t. This is used for the
+ * size of the bloom filter in the nodelist (node_addrs). */
+MOCK_IMPL(int,
+get_estimated_address_per_node, (void))
+{
+  return ESTIMATED_ADDRESS_PER_NODE;
+}
+
 /** Tell the nodelist that the current usable consensus is <b>ns</b>.
  * This makes the nodelist change all of the routerstatus entries for
  * the nodes, drop nodes that no longer have enough info to get used,
@@ -240,6 +304,12 @@ nodelist_set_consensus(networkstatus_t *ns)
   SMARTLIST_FOREACH(the_nodelist->nodes, node_t *, node,
                     node->rs = NULL);
 
+  /* Conservatively estimate that every node will have 2 addresses. */
+  const int estimated_addresses = smartlist_len(ns->routerstatus_list) *
+                                  get_estimated_address_per_node();
+  address_set_free(the_nodelist->node_addrs);
+  the_nodelist->node_addrs = address_set_new(estimated_addresses);
+
   SMARTLIST_FOREACH_BEGIN(ns->routerstatus_list, routerstatus_t *, rs) {
     node_t *node = node_get_or_create(rs->identity_digest);
     node->rs = rs;
@@ -278,6 +348,11 @@ nodelist_set_consensus(networkstatus_t *ns)
 
   nodelist_purge();
 
+  /* Now add all the nodes we have to the address set. */
+  SMARTLIST_FOREACH_BEGIN(the_nodelist->nodes, node_t *, node) {
+    node_add_to_address_set(node);
+  } SMARTLIST_FOREACH_END(node);
+
   if (! authdir) {
     SMARTLIST_FOREACH_BEGIN(the_nodelist->nodes, node_t *, node) {
       /* We have no routerstatus for this router. Clear flags so we can skip
@@ -430,6 +505,9 @@ nodelist_free_all(void)
 
   smartlist_free(the_nodelist->nodes);
 
+  address_set_free(the_nodelist->node_addrs);
+  the_nodelist->node_addrs = NULL;
+
   tor_free(the_nodelist);
 }
 
diff --git a/src/or/nodelist.h b/src/or/nodelist.h
index 71a91e107f..098f1d1555 100644
--- a/src/or/nodelist.h
+++ b/src/or/nodelist.h
@@ -22,6 +22,7 @@ const node_t *node_get_by_hex_id(const char *identity_digest);
 node_t *nodelist_set_routerinfo(routerinfo_t *ri, routerinfo_t **ri_old_out);
 node_t *nodelist_add_microdesc(microdesc_t *md);
 void nodelist_set_consensus(networkstatus_t *ns);
+int nodelist_probably_contains_address(const tor_addr_t *addr);
 
 void nodelist_remove_microdesc(const char *identity_digest, microdesc_t *md);
 void nodelist_remove_routerinfo(routerinfo_t *ri);
@@ -124,5 +125,7 @@ void router_dir_info_changed(void);
 const char *get_dir_info_status_string(void);
 int count_loading_descriptors_progress(void);
 
+MOCK_DECL(int, get_estimated_address_per_node, (void));
+
 #endif
 
diff --git a/src/or/or.h b/src/or/or.h
index 75a02a531e..024a9cff0f 100644
--- a/src/or/or.h
+++ b/src/or/or.h
@@ -1500,6 +1500,10 @@ typedef struct or_connection_t {
   /** True iff this connection has had its bootstrap failure logged with
    * control_event_bootstrap_problem. */
   unsigned int have_noted_bootstrap_problem:1;
+  /** True iff this is a client connection and its address has been put in the
+   * geoip cache and handled by the DoS mitigation subsystem. We use this to
+   * insure we have a coherent count of concurrent connection. */
+  unsigned int tracked_for_dos_mitigation : 1;
 
   uint16_t link_proto; /**< What protocol version are we using? 0 for
                         * "none negotiated yet." */
@@ -4510,6 +4514,35 @@ typedef struct {
 
   /** If 1, we skip all OOS checks. */
   int DisableOOSCheck;
+
+  /** Autobool: Is the circuit creation DoS mitigation subsystem enabled? */
+  int DoSCircuitCreationEnabled;
+  /** Minimum concurrent connection needed from one single address before any
+   * defense is used. */
+  int DoSCircuitCreationMinConnections;
+  /** Circuit rate used to refill the token bucket. */
+  int DoSCircuitCreationRate;
+  /** Maximum allowed burst of circuits. Reaching that value, the address is
+   * detected as malicious and a defense might be used. */
+  int DoSCircuitCreationBurst;
+  /** When an address is marked as malicous, what defense should be used
+   * against it. See the dos_cc_defense_type_t enum. */
+  int DoSCircuitCreationDefenseType;
+  /** For how much time (in seconds) the defense is applicable for a malicious
+   * address. A random time delta is added to the defense time of an address
+   * which will be between 1 second and half of this value. */
+  int DoSCircuitCreationDefenseTimePeriod;
+
+  /** Autobool: Is the DoS connection mitigation subsystem enabled? */
+  int DoSConnectionEnabled;
+  /** Maximum concurrent connection allowed per address. */
+  int DoSConnectionMaxConcurrentCount;
+  /** When an address is reaches the maximum count, what defense should be
+   * used against it. See the dos_conn_defense_type_t enum. */
+  int DoSConnectionDefenseType;
+
+  /** Autobool: Do we refuse single hop client rendezvous? */
+  int DoSRefuseSingleHopClientRendezvous;
 } or_options_t;
 
 /** Persistent state for an onion router, as saved to disk. */
diff --git a/src/or/relay.c b/src/or/relay.c
index 29f34ca033..22ce767523 100644
--- a/src/or/relay.c
+++ b/src/or/relay.c
@@ -2469,24 +2469,34 @@ static time_t last_time_under_memory_pressure = 0;
 STATIC int
 cell_queues_check_size(void)
 {
+  time_t now = time(NULL);
   size_t alloc = cell_queues_get_total_allocation();
   alloc += buf_get_total_allocation();
   alloc += tor_zlib_get_total_allocation();
   const size_t rend_cache_total = rend_cache_get_total_allocation();
   alloc += rend_cache_total;
+  const size_t geoip_client_cache_total =
+    geoip_client_cache_total_allocation();
+  alloc += geoip_client_cache_total;
   if (alloc >= get_options()->MaxMemInQueues_low_threshold) {
     last_time_under_memory_pressure = approx_time();
     if (alloc >= get_options()->MaxMemInQueues) {
       /* If we're spending over 20% of the memory limit on hidden service
-       * descriptors, free them until we're down to 10%.
-       */
+       * descriptors, free them until we're down to 10%. Do the same for geoip
+       * client cache. */
       if (rend_cache_total > get_options()->MaxMemInQueues / 5) {
         const size_t bytes_to_remove =
           rend_cache_total - (size_t)(get_options()->MaxMemInQueues / 10);
-        rend_cache_clean_v2_descs_as_dir(time(NULL), bytes_to_remove);
+        rend_cache_clean_v2_descs_as_dir(now, bytes_to_remove);
         alloc -= rend_cache_total;
         alloc += rend_cache_get_total_allocation();
       }
+      if (geoip_client_cache_total > get_options()->MaxMemInQueues / 5) {
+        const size_t bytes_to_remove =
+          geoip_client_cache_total -
+          (size_t)(get_options()->MaxMemInQueues / 10);
+        alloc -= geoip_client_cache_handle_oom(now, bytes_to_remove);
+      }
       circuits_handle_oom(alloc);
       return 1;
     }
diff --git a/src/or/rendmid.c b/src/or/rendmid.c
index ca0ad7b0d4..441d5043ce 100644
--- a/src/or/rendmid.c
+++ b/src/or/rendmid.c
@@ -8,9 +8,11 @@
  **/
 
 #include "or.h"
+#include "channel.h"
 #include "circuitlist.h"
 #include "circuituse.h"
 #include "config.h"
+#include "dos.h"
 #include "relay.h"
 #include "rendmid.h"
 #include "rephist.h"
@@ -246,6 +248,16 @@ rend_mid_establish_rendezvous(or_circuit_t *circ, const uint8_t *request,
     goto err;
   }
 
+  /* Check if we are configured to accept established rendezvous cells from
+   * client or in other words tor2web clients. */
+  if (channel_is_client(circ->p_chan) &&
+      dos_should_refuse_single_hop_client()) {
+    /* Note it down for the heartbeat log purposes. */
+    dos_note_refuse_single_hop_client();
+    /* Silent drop so the client has to time out before moving on. */
+    return 0;
+  }
+
   if (circ->base_.n_chan) {
     log_warn(LD_PROTOCOL,
              "Tried to establish rendezvous on non-edge circuit");
diff --git a/src/or/status.c b/src/or/status.c
index fce6a10157..fa2238b9f9 100644
--- a/src/or/status.c
+++ b/src/or/status.c
@@ -27,6 +27,7 @@
 #include "hibernate.h"
 #include "rephist.h"
 #include "statefile.h"
+#include "dos.h"
 
 static void log_accounting(const time_t now, const or_options_t *options);
 #include "geoip.h"
@@ -145,6 +146,7 @@ log_heartbeat(time_t now)
   if (public_server_mode(options)) {
     rep_hist_log_circuit_handshake_stats(now);
     rep_hist_log_link_protocol_counts();
+    dos_log_heartbeat();
   }
 
   circuit_log_ancient_one_hop_circuits(1800);
diff --git a/src/test/include.am b/src/test/include.am
index 8ecfaf10c6..0ee3d1169f 100644
--- a/src/test/include.am
+++ b/src/test/include.am
@@ -72,6 +72,7 @@ src_test_test_SOURCES = \
 	src/test/test_accounting.c \
 	src/test/test_addr.c \
 	src/test/test_address.c \
+	src/test/test_address_set.c \
 	src/test/test_buffers.c \
 	src/test/test_cell_formats.c \
 	src/test/test_cell_queue.c \
@@ -87,6 +88,7 @@ src_test_test_SOURCES = \
 	src/test/test_controller.c \
 	src/test/test_controller_events.c \
 	src/test/test_crypto.c \
+	src/test/test_dos.c \
 	src/test/test_data.c \
 	src/test/test_dir.c \
 	src/test/test_dir_common.c \
diff --git a/src/test/test.c b/src/test/test.c
index 9a41b976b8..0fef697909 100644
--- a/src/test/test.c
+++ b/src/test/test.c
@@ -912,6 +912,24 @@ test_geoip(void *arg)
   tt_str_op(entry_stats_2,OP_EQ, s);
   tor_free(s);
 
+  /* Test the OOM handler. Add a client, run the OOM. */
+  geoip_entry_stats_init(now);
+  SET_TEST_ADDRESS(100);
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, &addr, NULL,
+                         now - (12 * 60 * 60));
+  /* We've seen this 12 hours ago. Run the OOM, it should clean the entry
+   * because it is above the minimum cutoff of 4 hours. */
+  size_t bytes_removed = geoip_client_cache_handle_oom(now, 1000);
+  tt_size_op(bytes_removed, OP_GT, 0);
+
+  /* Do it again but this time with an entry with a lower cutoff. */
+  geoip_entry_stats_init(now);
+  SET_TEST_ADDRESS(100);
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, &addr, NULL,
+                         now - (3 * 60 * 60));
+  bytes_removed = geoip_client_cache_handle_oom(now, 1000);
+  tt_size_op(bytes_removed, OP_EQ, 0);
+
   /* Stop collecting entry statistics. */
   geoip_entry_stats_term();
   get_options_mutable()->EntryStatistics = 0;
@@ -1182,6 +1200,7 @@ struct testgroup_t testgroups[] = {
   { "accounting/", accounting_tests },
   { "addr/", addr_tests },
   { "address/", address_tests },
+  { "address_set/", address_set_tests },
   { "buffer/", buffer_tests },
   { "cellfmt/", cell_format_tests },
   { "cellqueue/", cell_queue_tests },
@@ -1197,6 +1216,7 @@ struct testgroup_t testgroups[] = {
   { "control/", controller_tests },
   { "control/event/", controller_event_tests },
   { "crypto/", crypto_tests },
+  { "dos/", dos_tests },
   { "dir/", dir_tests },
   { "dir_handle_get/", dir_handle_get_tests },
   { "dir/md/", microdesc_tests },
diff --git a/src/test/test.h b/src/test/test.h
index 25336ac83e..028082386e 100644
--- a/src/test/test.h
+++ b/src/test/test.h
@@ -175,6 +175,7 @@ extern const struct testcase_setup_t ed25519_test_setup;
 extern struct testcase_t accounting_tests[];
 extern struct testcase_t addr_tests[];
 extern struct testcase_t address_tests[];
+extern struct testcase_t address_set_tests[];
 extern struct testcase_t buffer_tests[];
 extern struct testcase_t cell_format_tests[];
 extern struct testcase_t cell_queue_tests[];
@@ -190,6 +191,7 @@ extern struct testcase_t container_tests[];
 extern struct testcase_t controller_tests[];
 extern struct testcase_t controller_event_tests[];
 extern struct testcase_t crypto_tests[];
+extern struct testcase_t dos_tests[];
 extern struct testcase_t dir_tests[];
 extern struct testcase_t dir_handle_get_tests[];
 extern struct testcase_t entryconn_tests[];
diff --git a/src/test/test_address_set.c b/src/test/test_address_set.c
new file mode 100644
index 0000000000..df022f539a
--- /dev/null
+++ b/src/test/test_address_set.c
@@ -0,0 +1,174 @@
+/* Copyright (c) 2017, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+#include "or.h"
+#include "address_set.h"
+#include "microdesc.h"
+#include "networkstatus.h"
+#include "nodelist.h"
+#include "routerlist.h"
+#include "torcert.h"
+
+#include "test.h"
+
+static networkstatus_t *dummy_ns = NULL;
+static networkstatus_t *
+mock_networkstatus_get_latest_consensus(void)
+{
+  return dummy_ns;
+}
+
+static networkstatus_t *
+mock_networkstatus_get_latest_consensus_by_flavor(consensus_flavor_t f)
+{
+  tor_assert(f == FLAV_MICRODESC);
+  return dummy_ns;
+}
+
+/* Number of address a single node_t can have. Default to the production
+ * value. This is to control the size of the bloom filter. */
+static int addr_per_node = 2;
+static int
+mock_get_estimated_address_per_node(void)
+{
+  return addr_per_node;
+}
+
+static void
+test_contains(void *arg)
+{
+  int ret;
+  address_set_t *set = NULL;
+
+  (void) arg;
+
+  /* Setup an IPv4 and IPv6 addresses. */
+  tor_addr_t addr_v6;
+  tor_addr_parse(&addr_v6, "1:2:3:4::");
+  tor_addr_t addr_v4;
+  tor_addr_parse(&addr_v4, "42.42.42.42");
+  uint32_t ipv4h = tor_addr_to_ipv4h(&addr_v4);
+
+  /* Make it very big so the chance of failing the contain test will be
+   * extremely rare. */
+  set = address_set_new(1024);
+  tt_assert(set);
+
+  /* Add and lookup IPv6. */
+  address_set_add(set, &addr_v6);
+  ret = address_set_probably_contains(set, &addr_v6);
+  tt_int_op(ret, OP_EQ, 1);
+
+  /* Add and lookup IPv4. */
+  address_set_add_ipv4h(set, ipv4h);
+  ret = address_set_probably_contains(set, &addr_v4);
+  tt_int_op(ret, OP_EQ, 1);
+
+  /* Try a lookup of rubbish. */
+  tor_addr_t dummy_addr;
+  memset(&dummy_addr, 'A', sizeof(dummy_addr));
+  dummy_addr.family = AF_INET;
+  ret = address_set_probably_contains(set, &dummy_addr);
+  tt_int_op(ret, OP_EQ, 0);
+  dummy_addr.family = AF_INET6;
+  ret = address_set_probably_contains(set, &dummy_addr);
+  tt_int_op(ret, OP_EQ, 0);
+
+ done:
+  address_set_free(set);
+}
+
+static void
+test_nodelist(void *arg)
+{
+  int ret;
+  routerstatus_t *rs = NULL; microdesc_t *md = NULL; routerinfo_t *ri = NULL;
+
+  (void) arg;
+
+  MOCK(networkstatus_get_latest_consensus,
+       mock_networkstatus_get_latest_consensus);
+  MOCK(networkstatus_get_latest_consensus_by_flavor,
+       mock_networkstatus_get_latest_consensus_by_flavor);
+  MOCK(get_estimated_address_per_node,
+       mock_get_estimated_address_per_node);
+
+  dummy_ns = tor_malloc_zero(sizeof(*dummy_ns));
+  dummy_ns->flavor = FLAV_MICRODESC;
+  dummy_ns->routerstatus_list = smartlist_new();
+
+  tor_addr_t addr_v4, addr_v6, dummy_addr;
+  tor_addr_parse(&addr_v4, "42.42.42.42");
+  uint32_t ipv4h = tor_addr_to_ipv4h(&addr_v4);
+  tor_addr_parse(&addr_v6, "1:2:3:4::");
+  memset(&dummy_addr, 'A', sizeof(dummy_addr));
+
+  /* This will make the nodelist bloom filter very large
+   * (the_nodelist->node_addrs) so we will fail the contain test rarely. */
+  addr_per_node = 1024;
+
+  /* No node no nothing. The lookups should be empty. */
+  nodelist_set_consensus(dummy_ns);
+
+  /* The address set should be empty. */
+  ret = nodelist_probably_contains_address(&addr_v4);
+  tt_int_op(ret, OP_EQ, 0);
+  ret = nodelist_probably_contains_address(&addr_v6);
+  tt_int_op(ret, OP_EQ, 0);
+  dummy_addr.family = AF_INET;
+  ret = nodelist_probably_contains_address(&dummy_addr);
+  tt_int_op(ret, OP_EQ, 0);
+  dummy_addr.family = AF_INET6;
+  ret = nodelist_probably_contains_address(&dummy_addr);
+  tt_int_op(ret, OP_EQ, 0);
+
+  md = tor_malloc_zero(sizeof(*md));
+  ri = tor_malloc_zero(sizeof(*ri));
+  rs = tor_malloc_zero(sizeof(*rs));
+  crypto_rand(rs->identity_digest, sizeof(rs->identity_digest));
+  crypto_rand(md->digest, sizeof(md->digest));
+  memcpy(rs->descriptor_digest, md->digest, DIGEST256_LEN);
+
+  /* Setup the rs, ri and md addresses. */
+  rs->addr = ipv4h;
+  tor_addr_parse(&rs->ipv6_addr, "1:2:3:4::");
+  ri->addr = ipv4h;
+  tor_addr_parse(&ri->ipv6_addr, "1:2:3:4::");
+  tor_addr_parse(&md->ipv6_addr, "1:2:3:4::");
+
+  /* Add the rs to the consensus becoming a node_t. */
+  smartlist_add(dummy_ns->routerstatus_list, rs);
+  nodelist_set_consensus(dummy_ns);
+
+  /* At this point, the address set should be initialized in the nodelist and
+   * we should be able to lookup. */
+  ret = nodelist_probably_contains_address(&addr_v4);
+  tt_int_op(ret, OP_EQ, 1);
+  ret = nodelist_probably_contains_address(&addr_v6);
+  tt_int_op(ret, OP_EQ, 1);
+  /* Lookup unknown address. */
+  dummy_addr.family = AF_INET;
+  ret = nodelist_probably_contains_address(&dummy_addr);
+  tt_int_op(ret, OP_EQ, 0);
+  dummy_addr.family = AF_INET6;
+  ret = nodelist_probably_contains_address(&dummy_addr);
+  tt_int_op(ret, OP_EQ, 0);
+
+ done:
+  routerstatus_free(rs); routerinfo_free(ri); microdesc_free(md);
+  smartlist_clear(dummy_ns->routerstatus_list);
+  networkstatus_vote_free(dummy_ns);
+  UNMOCK(networkstatus_get_latest_consensus);
+  UNMOCK(networkstatus_get_latest_consensus_by_flavor);
+  UNMOCK(get_estimated_address_per_node);
+}
+
+struct testcase_t address_set_tests[] = {
+  { "contains", test_contains, TT_FORK,
+    NULL, NULL },
+  { "nodelist", test_nodelist, TT_FORK,
+    NULL, NULL },
+
+  END_OF_TESTCASES
+};
+
diff --git a/src/test/test_dos.c b/src/test/test_dos.c
new file mode 100644
index 0000000000..cb9d9e559c
--- /dev/null
+++ b/src/test/test_dos.c
@@ -0,0 +1,497 @@
+/* Copyright (c) 2018, The Tor Project, Inc. */
+/* See LICENSE for licensing information */
+
+#define DOS_PRIVATE
+#define TOR_CHANNEL_INTERNAL_
+#define CIRCUITLIST_PRIVATE
+
+#include "or.h"
+#include "dos.h"
+#include "circuitlist.h"
+#include "geoip.h"
+#include "channel.h"
+#include "microdesc.h"
+#include "networkstatus.h"
+#include "nodelist.h"
+#include "routerlist.h"
+#include "test.h"
+#include "log_test_helpers.h"
+
+static networkstatus_t *dummy_ns = NULL;
+static networkstatus_t *
+mock_networkstatus_get_latest_consensus(void)
+{
+  return dummy_ns;
+}
+
+static networkstatus_t *
+mock_networkstatus_get_latest_consensus_by_flavor(consensus_flavor_t f)
+{
+  tor_assert(f == FLAV_MICRODESC);
+  return dummy_ns;
+}
+
+/* Number of address a single node_t can have. Default to the production
+ * value. This is to control the size of the bloom filter. */
+static int addr_per_node = 2;
+static int
+mock_get_estimated_address_per_node(void)
+{
+  return addr_per_node;
+}
+
+static unsigned int
+mock_enable_dos_protection(const networkstatus_t *ns)
+{
+  (void) ns;
+  return 1;
+}
+
+/** Test that the connection tracker of the DoS subsystem will block clients
+ *  who try to establish too many connections */
+static void
+test_dos_conn_creation(void *arg)
+{
+  (void) arg;
+
+  MOCK(get_param_cc_enabled, mock_enable_dos_protection);
+  MOCK(get_param_conn_enabled, mock_enable_dos_protection);
+
+  /* Initialize test data */
+  or_connection_t or_conn;
+  time_t now = 1281533250; /* 2010-08-11 13:27:30 UTC */
+  tt_int_op(AF_INET,OP_EQ, tor_addr_parse(&or_conn.real_addr,
+                                          "18.0.0.1"));
+  tor_addr_t *addr = &or_conn.real_addr;
+
+  /* Get DoS subsystem limits */
+  dos_init();
+  uint32_t max_concurrent_conns = get_param_conn_max_concurrent_count(NULL);
+
+  /* Introduce new client */
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, addr, NULL, now);
+  { /* Register many conns from this client but not enough to get it blocked */
+    unsigned int i;
+    for (i = 0; i < max_concurrent_conns; i++) {
+      dos_new_client_conn(&or_conn);
+    }
+  }
+
+  /* Check that new conns are still permitted */
+  tt_int_op(DOS_CONN_DEFENSE_NONE, OP_EQ,
+            dos_conn_addr_get_defense_type(addr));
+
+  /* Register another conn and check that new conns are not allowed anymore */
+  dos_new_client_conn(&or_conn);
+  tt_int_op(DOS_CONN_DEFENSE_CLOSE, OP_EQ,
+            dos_conn_addr_get_defense_type(addr));
+
+  /* Close a client conn and see that a new conn will be permitted again */
+  dos_close_client_conn(&or_conn);
+  tt_int_op(DOS_CONN_DEFENSE_NONE, OP_EQ,
+            dos_conn_addr_get_defense_type(addr));
+
+  /* Register another conn and see that defense measures get reactivated */
+  dos_new_client_conn(&or_conn);
+  tt_int_op(DOS_CONN_DEFENSE_CLOSE, OP_EQ,
+            dos_conn_addr_get_defense_type(addr));
+
+ done:
+  dos_free_all();
+}
+
+/** Helper mock: Place a fake IP addr for this channel in <b>addr_out</b> */
+static int
+mock_channel_get_addr_if_possible(channel_t *chan, tor_addr_t *addr_out)
+{
+  (void)chan;
+  tt_int_op(AF_INET,OP_EQ, tor_addr_parse(addr_out, "18.0.0.1"));
+  return 1;
+
+ done:
+  return 0;
+}
+
+/** Test that the circuit tracker of the DoS subsystem will block clients who
+ *  try to establish too many circuits. */
+static void
+test_dos_circuit_creation(void *arg)
+{
+  (void) arg;
+  unsigned int i;
+
+  MOCK(get_param_cc_enabled, mock_enable_dos_protection);
+  MOCK(get_param_conn_enabled, mock_enable_dos_protection);
+  MOCK(channel_get_addr_if_possible,
+       mock_channel_get_addr_if_possible);
+
+  /* Initialize channels/conns/circs that will be used */
+  channel_t *chan = tor_malloc_zero(sizeof(channel_t));
+  channel_init(chan);
+  chan->is_client = 1;
+
+  /* Initialize test data */
+  or_connection_t or_conn;
+  time_t now = 1281533250; /* 2010-08-11 13:27:30 UTC */
+  tt_int_op(AF_INET,OP_EQ, tor_addr_parse(&or_conn.real_addr,
+                                          "18.0.0.1"));
+  tor_addr_t *addr = &or_conn.real_addr;
+
+  /* Get DoS subsystem limits */
+  dos_init();
+  uint32_t max_circuit_count = get_param_cc_circuit_burst(NULL);
+  uint32_t min_conc_conns_for_cc =
+    get_param_cc_min_concurrent_connection(NULL);
+
+  /* Introduce new client and establish enough connections to activate the
+   * circuit counting subsystem */
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, addr, NULL, now);
+  for (i = 0; i < min_conc_conns_for_cc ; i++) {
+    dos_new_client_conn(&or_conn);
+  }
+
+  /* Register new circuits for this client and conn, but not enough to get
+   * detected as dos */
+  for (i=0; i < max_circuit_count-1; i++) {
+    dos_cc_new_create_cell(chan);
+  }
+  /* see that we didn't get detected for dosing */
+  tt_int_op(DOS_CC_DEFENSE_NONE, OP_EQ, dos_cc_get_defense_type(chan));
+
+  /* Register another CREATE cell that will push us over the limit. Check that
+   * the cell gets refused. */
+  dos_cc_new_create_cell(chan);
+  tt_int_op(DOS_CC_DEFENSE_REFUSE_CELL, OP_EQ, dos_cc_get_defense_type(chan));
+
+  /* TODO: Wait a few seconds before sending the cell, and check that the
+     buckets got refilled properly. */
+  /* TODO: Actually send a Tor cell (instead of calling the DoS function) and
+   * check that it will get refused */
+
+ done:
+  tor_free(chan);
+  dos_free_all();
+}
+
+/** Test that the DoS subsystem properly refills the circuit token buckets. */
+static void
+test_dos_bucket_refill(void *arg)
+{
+  (void) arg;
+  int i;
+  /* For this test, this variable is set to the current circ count of the token
+   * bucket. */
+  uint32_t current_circ_count;
+
+  MOCK(get_param_cc_enabled, mock_enable_dos_protection);
+  MOCK(get_param_conn_enabled, mock_enable_dos_protection);
+  MOCK(channel_get_addr_if_possible,
+       mock_channel_get_addr_if_possible);
+
+  time_t now = 1281533250; /* 2010-08-11 13:27:30 UTC */
+  update_approx_time(now);
+
+  /* Initialize channels/conns/circs that will be used */
+  channel_t *chan = tor_malloc_zero(sizeof(channel_t));
+  channel_init(chan);
+  chan->is_client = 1;
+  or_connection_t or_conn;
+  tt_int_op(AF_INET,OP_EQ, tor_addr_parse(&or_conn.real_addr,
+                                          "18.0.0.1"));
+  tor_addr_t *addr = &or_conn.real_addr;
+
+  /* Initialize DoS subsystem and get relevant limits */
+  dos_init();
+  uint32_t max_circuit_count = get_param_cc_circuit_burst(NULL);
+  uint64_t circ_rate = get_circuit_rate_per_second();
+  /* Check that the circuit rate is a positive number and smaller than the max
+   * circuit count */
+  tt_u64_op(circ_rate, OP_GT, 1);
+  tt_u64_op(circ_rate, OP_LT, max_circuit_count);
+
+  /* Register this client */
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, addr, NULL, now);
+  dos_new_client_conn(&or_conn);
+
+  /* Fetch this client from the geoip cache and get its DoS structs */
+  clientmap_entry_t *entry = geoip_lookup_client(addr, NULL,
+                                                 GEOIP_CLIENT_CONNECT);
+  tt_assert(entry);
+  dos_client_stats_t* dos_stats = &entry->dos_stats;
+  /* Check that the circuit bucket is still uninitialized */
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, 0);
+
+  /* Send a create cell: then check that the circ token bucket got initialized
+   * and one circ was subtracted. */
+  dos_cc_new_create_cell(chan);
+  current_circ_count = max_circuit_count - 1;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send 29 more CREATEs and ensure that the bucket is missing 30
+   * tokens */
+  for (i=0; i < 29; i++) {
+   dos_cc_new_create_cell(chan);
+   current_circ_count--;
+  }
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* OK! Progress time forward one sec, refill the bucket and check that the
+   * refill happened correctly. */
+  now += 1;
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  /* check refill */
+  current_circ_count += circ_rate;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+   dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now progress time a week forward, and check that the token bucket does not
+   * have more than max_circs allowance, even tho we let it simmer for so
+   * long. */
+  now += 604800; /* a week */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now use a very large time, and check that the token bucket does not have
+   * more than max_circs allowance, even tho we let it simmer for so long. */
+  now = INT32_MAX; /* 2038? */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now use a very small time, and check that the token bucket has exactly
+   * the max_circs allowance, because backward clock jumps are rare. */
+  now = INT32_MIN; /* 19?? */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Progress time forward one sec again, refill the bucket and check that the
+   * refill happened correctly. */
+  now += 1;
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  /* check refill */
+  current_circ_count += circ_rate;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now use a very large time (again), and check that the token bucket does
+   * not have more than max_circs allowance, even tho we let it simmer for so
+   * long. */
+  now = INT32_MAX; /* 2038? */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* This code resets the time to zero with 32-bit time_t, which triggers the
+   * code that initialises the bucket. */
+#if SIZEOF_TIME_T == 8
+  /* Now use a very very small time, and check that the token bucket has
+   * exactly the max_circs allowance, because backward clock jumps are rare.
+   */
+  now = (time_t)INT64_MIN; /* ???? */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Progress time forward one sec again, refill the bucket and check that the
+   * refill happened correctly. */
+  now += 1;
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  /* check refill */
+  current_circ_count += circ_rate;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now use a very very small time, and check that the token bucket has
+   * exactly the max_circs allowance, because backward clock jumps are rare.
+   */
+  now = (time_t)INT64_MIN; /* ???? */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now use a very very large time, and check that the token bucket does not
+   * have more than max_circs allowance, even tho we let it simmer for so
+   * long. */
+  now = (time_t)INT64_MAX; /* ???? */
+  update_approx_time(now);
+  cc_stats_refill_bucket(&dos_stats->cc_stats, addr);
+  current_circ_count += max_circuit_count;
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+
+  /* Now send as many CREATE cells as needed to deplete our token bucket
+   * completely */
+  for (; current_circ_count != 0; current_circ_count--) {
+    dos_cc_new_create_cell(chan);
+  }
+  tt_uint_op(current_circ_count, OP_EQ, 0);
+  tt_uint_op(dos_stats->cc_stats.circuit_bucket, OP_EQ, current_circ_count);
+#endif
+
+ done:
+  tor_free(chan);
+  dos_free_all();
+}
+
+/* Test if we avoid counting a known relay. */
+static void
+test_known_relay(void *arg)
+{
+  clientmap_entry_t *entry = NULL;
+  routerstatus_t *rs = NULL; microdesc_t *md = NULL; routerinfo_t *ri = NULL;
+
+  (void) arg;
+
+  MOCK(networkstatus_get_latest_consensus,
+       mock_networkstatus_get_latest_consensus);
+  MOCK(networkstatus_get_latest_consensus_by_flavor,
+       mock_networkstatus_get_latest_consensus_by_flavor);
+  MOCK(get_estimated_address_per_node,
+       mock_get_estimated_address_per_node);
+  MOCK(get_param_cc_enabled, mock_enable_dos_protection);
+
+  dos_init();
+
+  dummy_ns = tor_malloc_zero(sizeof(*dummy_ns));
+  dummy_ns->flavor = FLAV_MICRODESC;
+  dummy_ns->routerstatus_list = smartlist_new();
+
+  /* Setup an OR conn so we can pass it to the DoS subsystem. */
+  or_connection_t or_conn;
+  tor_addr_parse(&or_conn.real_addr, "42.42.42.42");
+
+  rs = tor_malloc_zero(sizeof(*rs));
+  rs->addr = tor_addr_to_ipv4h(&or_conn.real_addr);
+  crypto_rand(rs->identity_digest, sizeof(rs->identity_digest));
+  smartlist_add(dummy_ns->routerstatus_list, rs);
+
+  /* This will make the nodelist bloom filter very large
+   * (the_nodelist->node_addrs) so we will fail the contain test rarely. */
+  addr_per_node = 1024;
+  nodelist_set_consensus(dummy_ns);
+
+  /* We have now a node in our list so we'll make sure we don't count it as a
+   * client connection. */
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, &or_conn.real_addr, NULL, 0);
+  /* Suppose we have 5 connections in rapid succession, the counter should
+   * always be 0 because we should ignore this. */
+  dos_new_client_conn(&or_conn);
+  dos_new_client_conn(&or_conn);
+  dos_new_client_conn(&or_conn);
+  dos_new_client_conn(&or_conn);
+  dos_new_client_conn(&or_conn);
+  entry = geoip_lookup_client(&or_conn.real_addr, NULL, GEOIP_CLIENT_CONNECT);
+  tt_assert(entry);
+  /* We should have a count of 0. */
+  tt_uint_op(entry->dos_stats.concurrent_count, OP_EQ, 0);
+
+  /* To make sure that his is working properly, make a unknown client
+   * connection and see if we do get it. */
+  tor_addr_parse(&or_conn.real_addr, "42.42.42.43");
+  geoip_note_client_seen(GEOIP_CLIENT_CONNECT, &or_conn.real_addr, NULL, 0);
+  dos_new_client_conn(&or_conn);
+  dos_new_client_conn(&or_conn);
+  entry = geoip_lookup_client(&or_conn.real_addr, NULL, GEOIP_CLIENT_CONNECT);
+  tt_assert(entry);
+  /* We should have a count of 2. */
+  tt_uint_op(entry->dos_stats.concurrent_count, OP_EQ, 2);
+
+ done:
+  routerstatus_free(rs); routerinfo_free(ri); microdesc_free(md);
+  smartlist_clear(dummy_ns->routerstatus_list);
+  networkstatus_vote_free(dummy_ns);
+  dos_free_all();
+  UNMOCK(networkstatus_get_latest_consensus);
+  UNMOCK(networkstatus_get_latest_consensus_by_flavor);
+  UNMOCK(get_estimated_address_per_node);
+  UNMOCK(get_param_cc_enabled);
+}
+
+struct testcase_t dos_tests[] = {
+  { "conn_creation", test_dos_conn_creation, TT_FORK, NULL, NULL },
+  { "circuit_creation", test_dos_circuit_creation, TT_FORK, NULL, NULL },
+  { "bucket_refill", test_dos_bucket_refill, TT_FORK, NULL, NULL },
+  { "known_relay" , test_known_relay, TT_FORK,
+    NULL, NULL },
+  END_OF_TESTCASES
+};
+