Auto-update anyone_hosts from the .anyone DNS service (v2) (#147)

Author: jim-toth

changes/anyone-hosts-autoupdate

@@ -0,0 +1,16 @@
+  o New features (anyone-hosts auto-update):
+    - The client now automatically fetches a fresh copy of the anyone_hosts
+      DNS mapping file from the .anyone DNS service nodes when new
+      directory information arrives and on a periodic schedule.  Fetches
+      are routed by .anyone name over hidden-service circuits.  A fetched
+      file is installed only if it carries a valid signature from a
+      trusted DNS signer, is unexpired, contains at least one mapping, and
+      is not older than the installed mapping.  New configuration options
+      AnyoneHostsUpdate, AnyoneHostsUpdateInterval, and AnyoneHostsURL
+      control the behaviour.
+
+  o Minor features (documentation):
+    - DNSMappingFileMaxSize now documents bit-based units (KBits, MBits,
+      GBits, TBits) in the man page, consistent with other MEMUNIT options.
+    - Clarified that DNSMappingFileMaxSize only limits loading an existing
+      file; a missing file is always recreated with the default mapping.

doc/man/anon.1.txt

@@ -1222,10 +1222,32 @@ The following options are useful only for clients (that is, if
     addresses/ports. See <<SocksPort,SocksPort>> for an explanation of isolation
     flags. (Default: 0)
 
-[[DNSMappingFileMaxSize]] **DNSMappingFileMaxSize** __N__ **bytes**|**KBytes**|**MBytes**|**GBytes**|**TBytes**::
+[[DNSMappingFileMaxSize]] **DNSMappingFileMaxSize** __N__ **bytes**|**KBytes**|**MBytes**|**GBytes**|**TBytes**|**KBits**|**MBits**|**GBits**|**TBits**::
     The maximum size allowed for the `anyone_hosts` DNS mapping file in the
     data directory. If the file exceeds this size, it is rejected. Set this
-    option to 0 to disable the size limit. (Default: 10 MB)
+    option to 0 to disable the size limit.  This limit applies only when
+    reading or accepting an existing file; if the file is absent, a new
+    default file is created regardless of this setting. (Default: 10 MB)
+
+[[AnyoneHostsUpdate]] **AnyoneHostsUpdate** **0**|**1**::
+    If 1, the client will automatically attempt to fetch a fresh copy of
+    the `anyone_hosts` DNS mapping file from the .anyone DNS service nodes
+    when new directory information arrives and on a periodic schedule.
+    Fetched files are installed only if they carry a valid signature from
+    a trusted DNS signer, have not expired, and are not older than the
+    currently installed mapping. (Default: 1)
+
+[[AnyoneHostsUpdateInterval]] **AnyoneHostsUpdateInterval** __N__ **seconds**|**minutes**|**hours**|**days**|**weeks**::
+    How often to attempt fetching a fresh `anyone_hosts` file.  The actual
+    schedule is randomly jittered by up to +/-10% around this value.
+    The minimum is 1 hour. (Default: 12 hours)
+
+[[AnyoneHostsURL]] **AnyoneHostsURL** __address__::
+    A .anyone onion address of a DNS service node to try when fetching the
+    `anyone_hosts` file.  This option may be specified multiple times; the
+    addresses are tried in order before the built-in defaults.  A failing
+    address is skipped on the next attempt; a working address keeps being
+    used. (Default: none)
 
 [[DownloadExtraInfo]] **DownloadExtraInfo** **0**|**1**::
     If true, Tor downloads and caches "extra-info" documents. These documents

src/app/config/config.c

@@ -327,6 +327,9 @@ static const config_var_t option_vars_[] = {
   V(AlternateBridgeAuthority,    LINELIST, NULL),
   V(AlternateDirAuthority,       LINELIST, NULL),
   OBSOLETE("AlternateHSAuthority"),
+  V(AnyoneHostsUpdate,           BOOL,     "1"),
+  V(AnyoneHostsUpdateInterval,   INTERVAL, "12 hours"),
+  V(AnyoneHostsURL,              LINELIST, NULL),
   V(AssumeReachable,             BOOL,     "0"),
   V(AssumeReachableIPv6,         AUTOBOOL, "auto"),
   OBSOLETE("AuthDirBadDir"),
@@ -4062,6 +4065,23 @@ options_validate_cb(const void *old_options_, void *options_, char **msg)
     return -1;
   }
 
+  if (options->AnyoneHostsURL) {
+    const config_line_t *cl;
+    for (cl = options->AnyoneHostsURL; cl; cl = cl->next) {
+      const char *v = cl->value;
+      if (!v || !strlen(v) || strcmpend(v, ".anyone") ||
+          strstr(v, "://") || strchr(v, '/') || strchr(v, ':') ||
+          strpbrk(v, " \t\r\n")) {
+        REJECT("AnyoneHostsURL entries must be bare .anyone hostnames "
+               "(no scheme, port, path, or whitespace) ending in "
+               "\".anyone\".");
+      }
+    }
+  }
+  if (options->AnyoneHostsUpdateInterval < 3600) {
+    REJECT("AnyoneHostsUpdateInterval must be at least 1 hour.");
+  }
+
   return 0;
 }
 

src/app/config/or_options_st.h

@@ -705,6 +705,14 @@ struct or_options_t {
   /** Maximum size in bytes allowed for the anyone_hosts DNS mapping file.
    * A value of 0 disables the size limit. */
   uint64_t DNSMappingFileMaxSize;
+  /** If true, automatically update the anyone_hosts DNS mapping file when
+   * new directory information arrives or on a periodic schedule. */
+  int AnyoneHostsUpdate;
+  /** How often (in seconds) to check for a new anyone_hosts file. */
+  int AnyoneHostsUpdateInterval;
+  /** List of .anyone service addresses to try when fetching the
+   * anyone_hosts file.  Tried in order before the built-in defaults. */
+  struct config_line_t *AnyoneHostsURL;
 
   /** If true, do not accept any requests to connect to internal addresses
    * over randomly chosen exits. */

src/core/mainloop/mainloop.c

@@ -67,6 +67,7 @@
 #include "core/or/connection_or.h"
 #include "core/or/dos.h"
 #include "core/or/status.h"
+#include "feature/anyone/anyone_hosts_update.h"
 #include "feature/client/addressmap.h"
 #include "feature/client/bridges.h"
 #include "feature/client/dnsserv.h"
@@ -1381,6 +1382,7 @@ CALLBACK(write_stats_file);
 CALLBACK(control_per_second_events);
 CALLBACK(second_elapsed);
 CALLBACK(manage_vglite);
+CALLBACK(update_anyone_hosts);
 
 #undef CALLBACK
 
@@ -1406,6 +1408,9 @@ STATIC periodic_event_item_t mainloop_periodic_events[] = {
   /* Update vanguards-lite once per hour, if we have networking */
   CALLBACK(manage_vglite, NET_PARTICIPANT, FL(NEED_NET)),
 
+  /* Periodically update the anyone_hosts DNS mapping file. */
+  CALLBACK(update_anyone_hosts, CLIENT, FL(NEED_NET)),
+
   /* XXXX Do we have a reason to do this on a callback? Does it do any good at
    * all?  For now, if we're dormant, we can let our listeners decay. */
   CALLBACK(retry_listeners, NET_PARTICIPANT, FL(NEED_NET)),
@@ -1553,6 +1558,8 @@ initialize_periodic_events(void)
   NAMED_CALLBACK(launch_descriptor_fetches);
   NAMED_CALLBACK(check_dns_honesty);
   NAMED_CALLBACK(save_state);
+
+  anyone_hosts_update_init();
 }
 
 STATIC void
@@ -1691,6 +1698,17 @@ manage_vglite_callback(time_t now, const or_options_t *options)
   return VANGUARDS_LITE_INTERVAL;
 }
 
+/** Periodic-event callback: attempt to update the anyone_hosts DNS mapping
+ * file, and return the number of seconds until the next run
+ * (AnyoneHostsUpdateInterval, jittered).  Whether an update may actually
+ * run (enabled, client-only, timing) is decided inside the anyone module.
+ */
+static int
+update_anyone_hosts_callback(time_t now, const or_options_t *options)
+{
+  return anyone_hosts_update_callback(now, options);
+}
+
 /** Perform regular maintenance tasks.  This function gets run once per
  * second.
  */

src/core/or/connection_edge.c

@@ -3640,6 +3640,43 @@ connection_ap_make_link(connection_t *partner,
   return conn;
 }
 
+/** Like connection_ap_make_link(), but for a target that is an onion-routed
+ * hostname (a .anyone name) rather than a numeric address.
+ *
+ * connection_ap_make_link() queues the stream for direct circuit attachment,
+ * which treats the address as an exit-resolvable hostname; a .anyone name
+ * would reach the exit unresolved and fail.  Instead, send the linked stream
+ * through connection_ap_handshake_rewrite_and_attach(), the same path SOCKS
+ * client streams take, so the name is translated via the anyone_hosts
+ * mapping and attached to a hidden-service circuit.
+ *
+ * Return the new entry connection, or NULL on error.  On error the stream
+ * has already been marked for close, so the partner connection will see EOF.
+ */
+entry_connection_t *
+connection_ap_make_link_onion(connection_t *partner,
+                              char *address, uint16_t port,
+                              int session_group, int isolation_flags)
+{
+  entry_connection_t *conn =
+    connection_ap_make_link(partner, address, port, NULL /* digest */,
+                            session_group, isolation_flags,
+                            0 /* use_begindir */, 0 /* want_onehop */);
+  if (!conn)
+    return NULL;
+
+  /* Undo the direct-attach queueing and take the rewrite path instead. */
+  connection_ap_mark_as_non_pending_circuit(conn);
+  /* The rewrite path refuses onion targets unless the stream's entry
+   * configuration allows them; internal fetches always may. */
+  conn->entry_cfg.onion_traffic = 1;
+  if (connection_ap_handshake_rewrite_and_attach(conn, NULL, NULL) < 0) {
+    /* The connection has already been marked unattached and closed. */
+    return NULL;
+  }
+  return conn;
+}
+
 /** Notify any interested controller connections about a new hostname resolve
  * or resolve error.  Takes the same arguments as does
  * connection_ap_handshake_socks_resolved(). */

src/core/or/connection_edge.h

@@ -114,6 +114,11 @@ entry_connection_t  *connection_ap_make_link(connection_t *partner,
                                             int session_group,
                                             int isolation_flags,
                                             int use_begindir, int want_onehop);
+entry_connection_t *connection_ap_make_link_onion(connection_t *partner,
+                                                  char *address,
+                                                  uint16_t port,
+                                                  int session_group,
+                                                  int isolation_flags);
 void connection_ap_handshake_socks_reply(entry_connection_t *conn, char *reply,
                                          size_t replylen,
                                          int endreason);