core/s2smanager.lua

   1 -- Prosody IM
   2 -- Copyright (C) 2008-2009 Matthew Wild
   3 -- Copyright (C) 2008-2009 Waqas Hussain
   4 --
   5 -- This project is MIT/X11 licensed. Please see the
   6 -- COPYING file in the source package for more information.
   7 --
   8
   9
  10
  11 local hosts = hosts;
  12 local sessions = sessions;
  13 local core_process_stanza = function(a, b) core_process_stanza(a, b); end
  14 local add_task = require "util.timer".add_task;
  15 local socket = require "socket";
  16 local format = string.format;
  17 local t_insert, t_sort = table.insert, table.sort;
  18 local get_traceback = debug.traceback;
  19 local tostring, pairs, ipairs, getmetatable, newproxy, error, tonumber
  20     = tostring, pairs, ipairs, getmetatable, newproxy, error, tonumber;
  21
  22 local idna_to_ascii = require "util.encodings".idna.to_ascii;
  23 local connlisteners_get = require "net.connlisteners".get;
  24 local wrapclient = require "net.server".wrapclient;
  25 local modulemanager = require "core.modulemanager";
  26 local st = require "stanza";
  27 local stanza = st.stanza;
  28 local nameprep = require "util.encodings".stringprep.nameprep;
  29
  30 local uuid_gen = require "util.uuid".generate;
  31
  32 local logger_init = require "util.logger".init;
  33
  34 local log = logger_init("s2smanager");
  35
  36 local sha256_hash = require "util.hashes".sha256;
  37
  38 local dialback_secret = uuid_gen();
  39
  40 local adns, dns = require "net.adns", require "net.dns";
  41
  42 local connect_timeout = config.get("*", "core", "s2s_timeout") or 60;
  43 local dns_timeout = config.get("*", "core", "dns_timeout") or 60;
  44
  45 incoming_s2s = {};
  46 local incoming_s2s = incoming_s2s;
  47
  48 module "s2smanager"
  49
  50 local function compare_srv_priorities(a,b) return a.priority < b.priority or a.weight < b.weight; end
  51
  52 local function bounce_sendq(session)
  53         local sendq = session.sendq;
  54         if sendq then
  55                 session.log("info", "sending error replies for "..#sendq.." queued stanzas because of failed outgoing connection to "..tostring(session.to_host));
  56                 local dummy = {
  57                         type = "s2sin";
  58                         send = function(s)
  59                                 (session.log or log)("error", "Replying to to an s2s error reply, please report this! Traceback: %s", get_traceback());
  60                         end;
  61                         dummy = true;
  62                 };
  63                 for i, data in ipairs(sendq) do
  64                         local reply = data[2];
  65                         local xmlns = reply.attr.xmlns;
  66                         if not xmlns or xmlns == "jabber:client" or xmlns == "jabber:server" then
  67                                 reply.attr.type = "error";
  68                                 reply:tag("error", {type = "cancel"})
  69                                         :tag("remote-server-not-found", {xmlns = "urn:ietf:params:xml:ns:xmpp-stanzas"}):up();
  70                                 core_process_stanza(dummy, reply);
  71                         end
  72                         sendq[i] = nil;
  73                 end
  74                 session.sendq = nil;
  75         end
  76 end
  77
  78 function send_to_host(from_host, to_host, data)
  79         local host = hosts[from_host].s2sout[to_host];
  80         if host then
  81                 -- We have a connection to this host already
  82                 if host.type == "s2sout_unauthed" and data.name ~= "db:verify" and ((not data.xmlns) or data.xmlns == "jabber:client" or data.xmlns == "jabber:server") then
  83                         (host.log or log)("debug", "trying to send over unauthed s2sout to "..to_host);
  84                         if not host.notopen and not host.dialback_key and host.sends2s then
  85                                 host.log("debug", "dialback had not been initiated");
  86                                 initiate_dialback(host);
  87                         end
  88
  89                         -- Queue stanza until we are able to send it
  90                         if host.sendq then t_insert(host.sendq, {tostring(data), st.reply(data)});
  91                         else host.sendq = { {tostring(data), st.reply(data)} }; end
  92                         host.log("debug", "stanza [%s] queued ", data.name);
  93                 elseif host.type == "local" or host.type == "component" then
  94                         log("error", "Trying to send a stanza to ourselves??")
  95                         log("error", "Traceback: %s", get_traceback());
  96                         log("error", "Stanza: %s", tostring(data));
  97                 else
  98                         (host.log or log)("debug", "going to send stanza to "..to_host.." from "..from_host);
  99                         -- FIXME
 100                         if host.from_host ~= from_host then
 101                                 log("error", "WARNING! This might, possibly, be a bug, but it might not...");
 102                                 log("error", "We are going to send from %s instead of %s", tostring(host.from_host), tostring(from_host));
 103                         end
 104                         host.sends2s(data);
 105                         host.log("debug", "stanza sent over "..host.type);
 106                 end
 107         else
 108                 log("debug", "opening a new outgoing connection for this stanza");
 109                 local host_session = new_outgoing(from_host, to_host);
 110                 -- Store in buffer
 111                 host_session.sendq = { {tostring(data), st.reply(data)} };
 112                 log("debug", "stanza [%s] queued until connection complete", tostring(data.name));
 113                 if (not host_session.connecting) and (not host_session.conn) then
 114                         log("warn", "Connection to %s failed already, destroying session...", to_host);
 115                         destroy_session(host_session);
 116                 end
 117         end
 118 end
 119
 120 local open_sessions = 0;
 121
 122 function new_incoming(conn)
 123         local session = { conn = conn, type = "s2sin_unauthed", direction = "incoming", hosts = {} };
 124         if true then
 125                 session.trace = newproxy(true);
 126                 getmetatable(session.trace).__gc = function () open_sessions = open_sessions - 1; end;
 127         end
 128         open_sessions = open_sessions + 1;
 129         local w, log = conn.write, logger_init("s2sin"..tostring(conn):match("[a-f0-9]+$"));
 130         session.sends2s = function (t) log("debug", "sending: %s", tostring(t)); w(tostring(t)); end
 131         incoming_s2s[session] = true;
 132         add_task(connect_timeout, function ()
 133                 if session.conn ~= conn or
 134                    session.type == "s2sin" then
 135                         return; -- Ok, we're connect[ed|ing]
 136                 end
 137                 -- Not connected, need to close session and clean up
 138                 (session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 139                     session.from_host or "(unknown)", session.to_host or "(unknown)");
 140                 session:close("connection-timeout");
 141         end);
 142         return session;
 143 end
 144
 145 function new_outgoing(from_host, to_host)
 146                 local host_session = { to_host = to_host, from_host = from_host, notopen = true, type = "s2sout_unauthed", direction = "outgoing" };
 147                 hosts[from_host].s2sout[to_host] = host_session;
 148
 149                 local log;
 150                 do
 151                         local conn_name = "s2sout"..tostring(host_session):match("[a-f0-9]*$");
 152                         log = logger_init(conn_name);
 153                         host_session.log = log;
 154                 end
 155
 156                 -- This is the first call, can't fail (the first step is DNS lookup)
 157                 attempt_connection(host_session);
 158
 159                 if not host_session.sends2s then
 160                         -- A sends2s which buffers data (until the stream is opened)
 161                         -- note that data in this buffer will be sent before the stream is authed
 162                         -- and will not be ack'd in any way, successful or otherwise
 163                         local buffer;
 164                         function host_session.sends2s(data)
 165                                 if not buffer then
 166                                         buffer = {};
 167                                         host_session.send_buffer = buffer;
 168                                 end
 169                                 log("debug", "Buffering data on unconnected s2sout to %s", to_host);
 170                                 buffer[#buffer+1] = data;
 171                                 log("debug", "Buffered item %d: %s", #buffer, tostring(data));
 172                         end
 173
 174                 end
 175
 176                 return host_session;
 177 end
 178
 179
 180 function attempt_connection(host_session, err)
 181         local from_host, to_host = host_session.from_host, host_session.to_host;
 182         local connect_host, connect_port = idna_to_ascii(to_host), 5269;
 183
 184         if not err then -- This is our first attempt
 185                 log("debug", "First attempt to connect to %s, starting with SRV lookup...", to_host);
 186                 host_session.connecting = true;
 187                 local handle;
 188                 handle = adns.lookup(function (answer)
 189                         handle = nil;
 190                         host_session.connecting = nil;
 191                         if answer then
 192                                 log("debug", to_host.." has SRV records, handling...");
 193                                 local srv_hosts = {};
 194                                 host_session.srv_hosts = srv_hosts;
 195                                 for _, record in ipairs(answer) do
 196                                         t_insert(srv_hosts, record.srv);
 197                                 end
 198                                 t_sort(srv_hosts, compare_srv_priorities);
 199
 200                                 local srv_choice = srv_hosts[1];
 201                                 host_session.srv_choice = 1;
 202                                 if srv_choice then
 203                                         connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 204                                         log("debug", "Best record found, will connect to %s:%d", connect_host, connect_port);
 205                                 end
 206                         else
 207                                 log("debug", to_host.." has no SRV records, falling back to A");
 208                         end
 209                         -- Try with SRV, or just the plain hostname if no SRV
 210                         local ok, err = try_connect(host_session, connect_host, connect_port);
 211                         if not ok then
 212                                 if not attempt_connection(host_session, err) then
 213                                         -- No more attempts will be made
 214                                         destroy_session(host_session);
 215                                 end
 216                         end
 217                 end, "_xmpp-server._tcp."..connect_host..".", "SRV");
 218
 219                 -- Set handler for DNS timeout
 220                 add_task(dns_timeout, function ()
 221                         if handle then
 222                                 adns.cancel(handle, true);
 223                         end
 224                 end);
 225
 226                 log("debug", "DNS lookup for %s sent, waiting for response before we can connect", to_host);
 227                 return true; -- Attempt in progress
 228         elseif host_session.srv_hosts and #host_session.srv_hosts > host_session.srv_choice then -- Not our first attempt, and we also have SRV
 229                 host_session.srv_choice = host_session.srv_choice + 1;
 230                 local srv_choice = host_session.srv_hosts[host_session.srv_choice];
 231                 connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 232                 host_session.log("info", "Connection failed (%s). Attempt #%d: This time to %s:%d", tostring(err), host_session.srv_choice, connect_host, connect_port);
 233         else
 234                 host_session.log("info", "Out of connection options, can't connect to %s", tostring(host_session.to_host));
 235                 -- We're out of options
 236                 return false;
 237         end
 238
 239         if not (connect_host and connect_port) then
 240                 -- Likely we couldn't resolve DNS
 241                 log("warn", "Hmm, we're without a host (%s) and port (%s) to connect to for %s, giving up :(", tostring(connect_host), tostring(connect_port), tostring(to_host));
 242                 return false;
 243         end
 244
 245         return try_connect(host_session, connect_host, connect_port);
 246 end
 247
 248 function try_connect(host_session, connect_host, connect_port)
 249         host_session.connecting = true;
 250         local handle;
 251         handle = adns.lookup(function (reply)
 252                 handle = nil;
 253                 host_session.connecting = nil;
 254
 255                 -- COMPAT: This is a compromise for all you CNAME-(ab)users :)
 256                 if not (reply and reply[1] and reply[1].a) then
 257                         reply = dns.peek(connect_host, "CNAME", "IN");
 258                         while reply and reply[1] and not reply[1].a and reply[1].cname do
 259                                 reply = dns.peek(reply[1].cname, "A", "IN") or dns.peek(reply[1].cname, "CNAME", "IN");
 260                         end
 261                 end
 262                 -- end of CNAME resolving
 263
 264                 if reply and reply[1] and reply[1].a then
 265                         log("debug", "DNS reply for %s gives us %s", connect_host, reply[1].a);
 266                         return make_connect(host_session, reply[1].a, connect_port);
 267                 else
 268                         log("debug", "DNS lookup failed to get a response for %s", connect_host);
 269                         if not attempt_connection(host_session, "name resolution failed") then -- Retry if we can
 270                                 log("debug", "No other records to try for %s - destroying", host_session.to_host);
 271                                 destroy_session(host_session); -- End of the line, we can't
 272                         end
 273                 end
 274         end, connect_host, "A", "IN");
 275
 276         -- Set handler for DNS timeout
 277         add_task(dns_timeout, function ()
 278                 if handle then
 279                         adns.cancel(handle, true);
 280                 end
 281         end);
 282
 283         return true;
 284 end
 285
 286 function make_connect(host_session, connect_host, connect_port)
 287         host_session.log("info", "Beginning new connection attempt to %s (%s:%d)", host_session.to_host, connect_host, connect_port);
 288         -- Ok, we're going to try to connect
 289
 290         local from_host, to_host = host_session.from_host, host_session.to_host;
 291
 292         local conn, handler = socket.tcp()
 293
 294         conn:settimeout(0);
 295         local success, err = conn:connect(connect_host, connect_port);
 296         if not success and err ~= "timeout" then
 297                 log("warn", "s2s connect() to %s (%s:%d) failed: %s", host_session.to_host, connect_host, connect_port, err);
 298                 return false, err;
 299         end
 300
 301         local cl = connlisteners_get("xmppserver");
 302         conn = wrapclient(conn, connect_host, connect_port, cl, cl.default_mode or 1, hosts[from_host].ssl_ctx, false );
 303         host_session.conn = conn;
 304
 305         -- Register this outgoing connection so that xmppserver_listener knows about it
 306         -- otherwise it will assume it is a new incoming connection
 307         cl.register_outgoing(conn, host_session);
 308
 309         local w = conn.write;
 310         host_session.sends2s = function (t) log("debug", "sending: %s", tostring(t)); w(tostring(t)); end
 311
 312         conn.write(format([[<stream:stream xmlns='jabber:server' xmlns:db='jabber:server:dialback' xmlns:stream='http://etherx.jabber.org/streams' from='%s' to='%s' version='1.0' xml:lang='en'>]], from_host, to_host));
 313         log("debug", "Connection attempt in progress...");
 314         add_task(connect_timeout, function ()
 315                 if host_session.conn ~= conn or
 316                    host_session.type == "s2sout" or
 317                    host_session.connecting then
 318                         return; -- Ok, we're connect[ed|ing]
 319                 end
 320                 -- Not connected, need to close session and clean up
 321                 (host_session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 322                     host_session.from_host or "(unknown)", host_session.to_host or "(unknown)");
 323                 host_session:close("connection-timeout");
 324         end);
 325         return true;
 326 end
 327
 328 function streamopened(session, attr)
 329         local send = session.sends2s;
 330
 331         -- TODO: #29: SASL/TLS on s2s streams
 332         session.version = 0; --tonumber(attr.version) or 0;
 333
 334         if session.version >= 1.0 and not (attr.to and attr.from) then
 335                 log("warn", (session.to_host or "(unknown)").." failed to specify 'to' or 'from' hostname as per RFC");
 336         end
 337
 338         if session.direction == "incoming" then
 339                 -- Send a reply stream header
 340                 session.to_host = attr.to and nameprep(attr.to);
 341                 session.from_host = attr.from and nameprep(attr.from);
 342
 343                 session.streamid = uuid_gen();
 344                 (session.log or log)("debug", "incoming s2s received <stream:stream>");
 345                 send("<?xml version='1.0'?>");
 346                 send(stanza("stream:stream", { xmlns='jabber:server', ["xmlns:db"]='jabber:server:dialback',
 347                                 ["xmlns:stream"]='http://etherx.jabber.org/streams', id=session.streamid, from=session.to_host }):top_tag());
 348                 if session.to_host and not hosts[session.to_host] then
 349                         -- Attempting to connect to a host we don't serve
 350                         session:close({ condition = "host-unknown"; text = "This host does not serve "..session.to_host });
 351                         return;
 352                 end
 353                 if session.version >= 1.0 then
 354                         send(st.stanza("stream:features")
 355                                         :tag("dialback", { xmlns='urn:xmpp:features:dialback' }):tag("optional"):up():up());
 356                 end
 357         elseif session.direction == "outgoing" then
 358                 -- If we are just using the connection for verifying dialback keys, we won't try and auth it
 359                 if not attr.id then error("stream response did not give us a streamid!!!"); end
 360                 session.streamid = attr.id;
 361
 362                 -- Send unauthed buffer
 363                 -- (stanzas which are fine to send before dialback)
 364                 -- Note that this is *not* the stanza queue (which
 365                 -- we can only send if auth succeeds) :)
 366                 local send_buffer = session.send_buffer;
 367                 if send_buffer and #send_buffer > 0 then
 368                         log("debug", "Sending s2s send_buffer now...");
 369                         for i, data in ipairs(send_buffer) do
 370                                 session.sends2s(tostring(data));
 371                                 send_buffer[i] = nil;
 372                         end
 373                 end
 374                 session.send_buffer = nil;
 375
 376                 if not session.dialback_verifying then
 377                         initiate_dialback(session);
 378                 else
 379                         mark_connected(session);
 380                 end
 381         end
 382
 383         session.notopen = nil;
 384 end
 385
 386 function streamclosed(session)
 387         (session.log or log)("debug", "</stream:stream>");
 388         if session.sends2s then
 389                 session.sends2s("</stream:stream>");
 390         end
 391         session.notopen = true;
 392 end
 393
 394 function initiate_dialback(session)
 395         -- generate dialback key
 396         session.dialback_key = generate_dialback(session.streamid, session.to_host, session.from_host);
 397         session.sends2s(format("<db:result from='%s' to='%s'>%s</db:result>", session.from_host, session.to_host, session.dialback_key));
 398         session.log("info", "sent dialback key on outgoing s2s stream");
 399 end
 400
 401 function generate_dialback(id, to, from)
 402         return sha256_hash(id..to..from..dialback_secret, true);
 403 end
 404
 405 function verify_dialback(id, to, from, key)
 406         return key == generate_dialback(id, to, from);
 407 end
 408
 409 function make_authenticated(session, host)
 410         if session.type == "s2sout_unauthed" then
 411                 session.type = "s2sout";
 412         elseif session.type == "s2sin_unauthed" then
 413                 session.type = "s2sin";
 414                 if host then
 415                         session.hosts[host].authed = true;
 416                 end
 417         elseif session.type == "s2sin" and host then
 418                 session.hosts[host].authed = true;
 419         else
 420                 return false;
 421         end
 422         session.log("debug", "connection %s->%s is now authenticated", session.from_host or "(unknown)", session.to_host or "(unknown)");
 423
 424         mark_connected(session);
 425
 426         return true;
 427 end
 428
 429 function mark_connected(session)
 430         local sendq, send = session.sendq, session.sends2s;
 431
 432         local from, to = session.from_host, session.to_host;
 433
 434         session.log("info", session.direction.." s2s connection "..from.."->"..to.." complete");
 435
 436         local send_to_host = send_to_host;
 437         function session.send(data) send_to_host(to, from, data); end
 438
 439
 440         if session.direction == "outgoing" then
 441                 if sendq then
 442                         session.log("debug", "sending "..#sendq.." queued stanzas across new outgoing connection to "..session.to_host);
 443                         for i, data in ipairs(sendq) do
 444                                 send(data[1]);
 445                                 sendq[i] = nil;
 446                         end
 447                         session.sendq = nil;
 448                 end
 449
 450                 session.srv_hosts = nil;
 451         end
 452 end
 453
 454 function destroy_session(session)
 455         (session.log or log)("info", "Destroying "..tostring(session.direction).." session "..tostring(session.from_host).."->"..tostring(session.to_host));
 456
 457         if session.direction == "outgoing" then
 458                 hosts[session.from_host].s2sout[session.to_host] = nil;
 459                 bounce_sendq(session);
 460         elseif session.direction == "incoming" then
 461                 incoming_s2s[session] = nil;
 462         end
 463
 464         for k in pairs(session) do
 465                 if k ~= "trace" then
 466                         session[k] = nil;
 467                 end
 468         end
 469 end
 470
 471 return _M;