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 fire_event = require "core.eventmanager".fire_event;
  31 local uuid_gen = require "util.uuid".generate;
  32
  33 local logger_init = require "util.logger".init;
  34
  35 local log = logger_init("s2smanager");
  36
  37 local sha256_hash = require "util.hashes".sha256;
  38
  39 local dialback_secret = uuid_gen();
  40
  41 local adns, dns = require "net.adns", require "net.dns";
  42
  43 local connect_timeout = config.get("*", "core", "s2s_timeout") or 60;
  44 local dns_timeout = config.get("*", "core", "dns_timeout") or 60;
  45 local max_dns_depth = config.get("*", "core", "dns_max_depth") or 3;
  46
  47 incoming_s2s = {};
  48 local incoming_s2s = incoming_s2s;
  49
  50 module "s2smanager"
  51
  52 local function compare_srv_priorities(a,b) return a.priority < b.priority or a.weight < b.weight; end
  53
  54 local function bounce_sendq(session)
  55         local sendq = session.sendq;
  56         if sendq then
  57                 session.log("info", "sending error replies for "..#sendq.." queued stanzas because of failed outgoing connection to "..tostring(session.to_host));
  58                 local dummy = {
  59                         type = "s2sin";
  60                         send = function(s)
  61                                 (session.log or log)("error", "Replying to to an s2s error reply, please report this! Traceback: %s", get_traceback());
  62                         end;
  63                         dummy = true;
  64                 };
  65                 for i, data in ipairs(sendq) do
  66                         local reply = data[2];
  67                         local xmlns = reply.attr.xmlns;
  68                         if not xmlns or xmlns == "jabber:client" or xmlns == "jabber:server" then
  69                                 reply.attr.type = "error";
  70                                 reply:tag("error", {type = "cancel"})
  71                                         :tag("remote-server-not-found", {xmlns = "urn:ietf:params:xml:ns:xmpp-stanzas"}):up();
  72                                 core_process_stanza(dummy, reply);
  73                         end
  74                         sendq[i] = nil;
  75                 end
  76                 session.sendq = nil;
  77         end
  78 end
  79
  80 function send_to_host(from_host, to_host, data)
  81         local host = hosts[from_host].s2sout[to_host];
  82         if host then
  83                 -- We have a connection to this host already
  84                 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
  85                         (host.log or log)("debug", "trying to send over unauthed s2sout to "..to_host);
  86                         if not host.notopen and not host.dialback_key and host.sends2s then
  87                                 host.log("debug", "dialback had not been initiated");
  88                                 initiate_dialback(host);
  89                         end
  90
  91                         -- Queue stanza until we are able to send it
  92                         if host.sendq then t_insert(host.sendq, {tostring(data), st.reply(data)});
  93                         else host.sendq = { {tostring(data), st.reply(data)} }; end
  94                         host.log("debug", "stanza [%s] queued ", data.name);
  95                 elseif host.type == "local" or host.type == "component" then
  96                         log("error", "Trying to send a stanza to ourselves??")
  97                         log("error", "Traceback: %s", get_traceback());
  98                         log("error", "Stanza: %s", tostring(data));
  99                 else
 100                         (host.log or log)("debug", "going to send stanza to "..to_host.." from "..from_host);
 101                         -- FIXME
 102                         if host.from_host ~= from_host then
 103                                 log("error", "WARNING! This might, possibly, be a bug, but it might not...");
 104                                 log("error", "We are going to send from %s instead of %s", tostring(host.from_host), tostring(from_host));
 105                         end
 106                         host.sends2s(data);
 107                         host.log("debug", "stanza sent over "..host.type);
 108                 end
 109         else
 110                 log("debug", "opening a new outgoing connection for this stanza");
 111                 local host_session = new_outgoing(from_host, to_host);
 112                 -- Store in buffer
 113                 host_session.sendq = { {tostring(data), st.reply(data)} };
 114                 log("debug", "stanza [%s] queued until connection complete", tostring(data.name));
 115                 if (not host_session.connecting) and (not host_session.conn) then
 116                         log("warn", "Connection to %s failed already, destroying session...", to_host);
 117                         destroy_session(host_session);
 118                 end
 119         end
 120 end
 121
 122 local open_sessions = 0;
 123
 124 function new_incoming(conn)
 125         local session = { conn = conn, type = "s2sin_unauthed", direction = "incoming", hosts = {} };
 126         if true then
 127                 session.trace = newproxy(true);
 128                 getmetatable(session.trace).__gc = function () open_sessions = open_sessions - 1; end;
 129         end
 130         open_sessions = open_sessions + 1;
 131         local w, log = conn.write, logger_init("s2sin"..tostring(conn):match("[a-f0-9]+$"));
 132         session.log = log;
 133         session.sends2s = function (t) log("debug", "sending: %s", tostring(t)); w(tostring(t)); end
 134         incoming_s2s[session] = true;
 135         add_task(connect_timeout, function ()
 136                 if session.conn ~= conn or
 137                    session.type == "s2sin" then
 138                         return; -- Ok, we're connect[ed|ing]
 139                 end
 140                 -- Not connected, need to close session and clean up
 141                 (session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 142                     session.from_host or "(unknown)", session.to_host or "(unknown)");
 143                 session:close("connection-timeout");
 144         end);
 145         return session;
 146 end
 147
 148 function new_outgoing(from_host, to_host)
 149                 local host_session = { to_host = to_host, from_host = from_host, host = from_host,
 150                                        notopen = true, type = "s2sout_unauthed", direction = "outgoing" };
 151
 152                 hosts[from_host].s2sout[to_host] = host_session;
 153
 154                 local log;
 155                 do
 156                         local conn_name = "s2sout"..tostring(host_session):match("[a-f0-9]*$");
 157                         log = logger_init(conn_name);
 158                         host_session.log = log;
 159                 end
 160
 161                 -- This is the first call, can't fail (the first step is DNS lookup)
 162                 attempt_connection(host_session);
 163
 164                 if not host_session.sends2s then
 165                         -- A sends2s which buffers data (until the stream is opened)
 166                         -- note that data in this buffer will be sent before the stream is authed
 167                         -- and will not be ack'd in any way, successful or otherwise
 168                         local buffer;
 169                         function host_session.sends2s(data)
 170                                 if not buffer then
 171                                         buffer = {};
 172                                         host_session.send_buffer = buffer;
 173                                 end
 174                                 log("debug", "Buffering data on unconnected s2sout to %s", to_host);
 175                                 buffer[#buffer+1] = data;
 176                                 log("debug", "Buffered item %d: %s", #buffer, tostring(data));
 177                         end
 178
 179                 end
 180
 181                 return host_session;
 182 end
 183
 184
 185 function attempt_connection(host_session, err)
 186         local from_host, to_host = host_session.from_host, host_session.to_host;
 187         local connect_host, connect_port = idna_to_ascii(to_host), 5269;
 188
 189         if not err then -- This is our first attempt
 190                 log("debug", "First attempt to connect to %s, starting with SRV lookup...", to_host);
 191                 host_session.connecting = true;
 192                 local handle;
 193                 handle = adns.lookup(function (answer)
 194                         handle = nil;
 195                         host_session.connecting = nil;
 196                         if answer then
 197                                 log("debug", to_host.." has SRV records, handling...");
 198                                 local srv_hosts = {};
 199                                 host_session.srv_hosts = srv_hosts;
 200                                 for _, record in ipairs(answer) do
 201                                         t_insert(srv_hosts, record.srv);
 202                                 end
 203                                 t_sort(srv_hosts, compare_srv_priorities);
 204
 205                                 local srv_choice = srv_hosts[1];
 206                                 host_session.srv_choice = 1;
 207                                 if srv_choice then
 208                                         connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 209                                         log("debug", "Best record found, will connect to %s:%d", connect_host, connect_port);
 210                                 end
 211                         else
 212                                 log("debug", to_host.." has no SRV records, falling back to A");
 213                         end
 214                         -- Try with SRV, or just the plain hostname if no SRV
 215                         local ok, err = try_connect(host_session, connect_host, connect_port);
 216                         if not ok then
 217                                 if not attempt_connection(host_session, err) then
 218                                         -- No more attempts will be made
 219                                         destroy_session(host_session);
 220                                 end
 221                         end
 222                 end, "_xmpp-server._tcp."..connect_host..".", "SRV");
 223
 224                 -- Set handler for DNS timeout
 225                 add_task(dns_timeout, function ()
 226                         if handle then
 227                                 adns.cancel(handle, true);
 228                         end
 229                 end);
 230
 231                 log("debug", "DNS lookup for %s sent, waiting for response before we can connect", to_host);
 232                 return true; -- Attempt in progress
 233         elseif host_session.srv_hosts and #host_session.srv_hosts > host_session.srv_choice then -- Not our first attempt, and we also have SRV
 234                 host_session.srv_choice = host_session.srv_choice + 1;
 235                 local srv_choice = host_session.srv_hosts[host_session.srv_choice];
 236                 connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 237                 host_session.log("info", "Connection failed (%s). Attempt #%d: This time to %s:%d", tostring(err), host_session.srv_choice, connect_host, connect_port);
 238         else
 239                 host_session.log("info", "Out of connection options, can't connect to %s", tostring(host_session.to_host));
 240                 -- We're out of options
 241                 return false;
 242         end
 243
 244         if not (connect_host and connect_port) then
 245                 -- Likely we couldn't resolve DNS
 246                 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));
 247                 return false;
 248         end
 249
 250         return try_connect(host_session, connect_host, connect_port);
 251 end
 252
 253 function try_connect(host_session, connect_host, connect_port)
 254         host_session.connecting = true;
 255         local handle;
 256         handle = adns.lookup(function (reply)
 257                 handle = nil;
 258                 host_session.connecting = nil;
 259
 260                 -- COMPAT: This is a compromise for all you CNAME-(ab)users :)
 261                 if not (reply and reply[#reply] and reply[#reply].a) then
 262                         local count = max_dns_depth;
 263                         reply = dns.peek(connect_host, "CNAME", "IN");
 264                         while count > 0 and reply and reply[#reply] and not reply[#reply].a and reply[#reply].cname do
 265                                 log("debug", "Looking up %s (DNS depth is %d)", tostring(reply[#reply].cname), count);
 266                                 reply = dns.peek(reply[#reply].cname, "A", "IN") or dns.peek(reply[#reply].cname, "CNAME", "IN");
 267                                 count = count - 1;
 268                         end
 269                 end
 270                 -- end of CNAME resolving
 271
 272                 if reply and reply[#reply] and reply[#reply].a then
 273                         log("debug", "DNS reply for %s gives us %s", connect_host, reply[#reply].a);
 274                         return make_connect(host_session, reply[#reply].a, connect_port);
 275                 else
 276                         log("debug", "DNS lookup failed to get a response for %s", connect_host);
 277                         if not attempt_connection(host_session, "name resolution failed") then -- Retry if we can
 278                                 log("debug", "No other records to try for %s - destroying", host_session.to_host);
 279                                 destroy_session(host_session); -- End of the line, we can't
 280                         end
 281                 end
 282         end, connect_host, "A", "IN");
 283
 284         -- Set handler for DNS timeout
 285         add_task(dns_timeout, function ()
 286                 if handle then
 287                         adns.cancel(handle, true);
 288                 end
 289         end);
 290
 291         return true;
 292 end
 293
 294 function make_connect(host_session, connect_host, connect_port)
 295         host_session.log("info", "Beginning new connection attempt to %s (%s:%d)", host_session.to_host, connect_host, connect_port);
 296         -- Ok, we're going to try to connect
 297
 298         local from_host, to_host = host_session.from_host, host_session.to_host;
 299
 300         local conn, handler = socket.tcp()
 301
 302         conn:settimeout(0);
 303         local success, err = conn:connect(connect_host, connect_port);
 304         if not success and err ~= "timeout" then
 305                 log("warn", "s2s connect() to %s (%s:%d) failed: %s", host_session.to_host, connect_host, connect_port, err);
 306                 return false, err;
 307         end
 308
 309         local cl = connlisteners_get("xmppserver");
 310         conn = wrapclient(conn, connect_host, connect_port, cl, cl.default_mode or 1, hosts[from_host].ssl_ctx, false );
 311         host_session.conn = conn;
 312
 313         -- Register this outgoing connection so that xmppserver_listener knows about it
 314         -- otherwise it will assume it is a new incoming connection
 315         cl.register_outgoing(conn, host_session);
 316
 317         local w = conn.write;
 318         host_session.sends2s = function (t) log("debug", "sending: %s", tostring(t)); w(tostring(t)); end
 319
 320         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));
 321         log("debug", "Connection attempt in progress...");
 322         add_task(connect_timeout, function ()
 323                 if host_session.conn ~= conn or
 324                    host_session.type == "s2sout" or
 325                    host_session.connecting then
 326                         return; -- Ok, we're connect[ed|ing]
 327                 end
 328                 -- Not connected, need to close session and clean up
 329                 (host_session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 330                     host_session.from_host or "(unknown)", host_session.to_host or "(unknown)");
 331                 host_session:close("connection-timeout");
 332         end);
 333         return true;
 334 end
 335
 336 function streamopened(session, attr)
 337         local send = session.sends2s;
 338
 339         -- TODO: #29: SASL/TLS on s2s streams
 340         session.version = tonumber(attr.version) or 0;
 341
 342         if session.secure == false then
 343                 session.secure = true;
 344         end
 345
 346         if session.version >= 1.0 and not (attr.to and attr.from) then
 347
 348                 (session.log or log)("warn", "Remote of stream "..(session.from_host or "(unknown)").."->"..(session.to_host or "(unknown)")
 349                         .." failed to specify to (%s) and/or from (%s) hostname as per RFC", tostring(attr.to), tostring(attr.from));
 350         end
 351
 352         if session.direction == "incoming" then
 353                 -- Send a reply stream header
 354                 session.to_host = attr.to and nameprep(attr.to);
 355                 session.from_host = attr.from and nameprep(attr.from);
 356
 357                 session.streamid = uuid_gen();
 358                 (session.log or log)("debug", "incoming s2s received <stream:stream>");
 359                 send("<?xml version='1.0'?>");
 360                 send(stanza("stream:stream", { xmlns='jabber:server', ["xmlns:db"]='jabber:server:dialback',
 361                                 ["xmlns:stream"]='http://etherx.jabber.org/streams', id=session.streamid, from=session.to_host, version=(session.version > 0 and "1.0" or nil) }):top_tag());
 362                 if session.to_host and not hosts[session.to_host] then
 363                         -- Attempting to connect to a host we don't serve
 364                         session:close({ condition = "host-unknown"; text = "This host does not serve "..session.to_host });
 365                         return;
 366                 end
 367                 if session.version >= 1.0 then
 368                         local features = st.stanza("stream:features");
 369
 370                         if session.to_host then
 371                                 hosts[session.to_host].events.fire_event("s2s-stream-features", { session = session, features = features });
 372                         else
 373                                 (session.log or log)("warn", "No 'to' on stream header from %s means we can't offer any features", session.from_host or "unknown host");
 374                         end
 375
 376                         log("debug", "Sending stream features: %s", tostring(features));
 377                         send(features);
 378                 end
 379         elseif session.direction == "outgoing" then
 380                 -- If we are just using the connection for verifying dialback keys, we won't try and auth it
 381                 if not attr.id then error("stream response did not give us a streamid!!!"); end
 382                 session.streamid = attr.id;
 383
 384                 -- Send unauthed buffer
 385                 -- (stanzas which are fine to send before dialback)
 386                 -- Note that this is *not* the stanza queue (which
 387                 -- we can only send if auth succeeds) :)
 388                 local send_buffer = session.send_buffer;
 389                 if send_buffer and #send_buffer > 0 then
 390                         log("debug", "Sending s2s send_buffer now...");
 391                         for i, data in ipairs(send_buffer) do
 392                                 session.sends2s(tostring(data));
 393                                 send_buffer[i] = nil;
 394                         end
 395                 end
 396                 session.send_buffer = nil;
 397
 398                 -- If server is pre-1.0, don't wait for features, just do dialback
 399                 if session.version < 1.0 then
 400                         if not session.dialback_verifying then
 401                                 log("debug", "Initiating dialback...");
 402                                 initiate_dialback(session);
 403                         else
 404                                 mark_connected(session);
 405                         end
 406                 end
 407         end
 408
 409         session.notopen = nil;
 410 end
 411
 412 function streamclosed(session)
 413         (session.log or log)("debug", "</stream:stream>");
 414         if session.sends2s then
 415                 session.sends2s("</stream:stream>");
 416         end
 417         session.notopen = true;
 418 end
 419
 420 function initiate_dialback(session)
 421         -- generate dialback key
 422         session.dialback_key = generate_dialback(session.streamid, session.to_host, session.from_host);
 423         session.sends2s(format("<db:result from='%s' to='%s'>%s</db:result>", session.from_host, session.to_host, session.dialback_key));
 424         session.log("info", "sent dialback key on outgoing s2s stream");
 425 end
 426
 427 function generate_dialback(id, to, from)
 428         return sha256_hash(id..to..from..dialback_secret, true);
 429 end
 430
 431 function verify_dialback(id, to, from, key)
 432         return key == generate_dialback(id, to, from);
 433 end
 434
 435 function make_authenticated(session, host)
 436         if session.type == "s2sout_unauthed" then
 437                 session.type = "s2sout";
 438         elseif session.type == "s2sin_unauthed" then
 439                 session.type = "s2sin";
 440                 if host then
 441                         session.hosts[host].authed = true;
 442                 end
 443         elseif session.type == "s2sin" and host then
 444                 session.hosts[host].authed = true;
 445         else
 446                 return false;
 447         end
 448         session.log("debug", "connection %s->%s is now authenticated", session.from_host or "(unknown)", session.to_host or "(unknown)");
 449
 450         mark_connected(session);
 451
 452         return true;
 453 end
 454
 455 -- Stream is authorised, and ready for normal stanzas
 456 function mark_connected(session)
 457         local sendq, send = session.sendq, session.sends2s;
 458
 459         local from, to = session.from_host, session.to_host;
 460
 461         session.log("info", session.direction.." s2s connection "..from.."->"..to.." complete");
 462
 463         local send_to_host = send_to_host;
 464         function session.send(data) send_to_host(to, from, data); end
 465
 466
 467         if session.direction == "outgoing" then
 468                 if sendq then
 469                         session.log("debug", "sending "..#sendq.." queued stanzas across new outgoing connection to "..session.to_host);
 470                         for i, data in ipairs(sendq) do
 471                                 send(data[1]);
 472                                 sendq[i] = nil;
 473                         end
 474                         session.sendq = nil;
 475                 end
 476
 477                 session.srv_hosts = nil;
 478         end
 479 end
 480
 481 function destroy_session(session)
 482         (session.log or log)("info", "Destroying "..tostring(session.direction).." session "..tostring(session.from_host).."->"..tostring(session.to_host));
 483
 484         if session.direction == "outgoing" then
 485                 hosts[session.from_host].s2sout[session.to_host] = nil;
 486                 bounce_sendq(session);
 487         elseif session.direction == "incoming" then
 488                 incoming_s2s[session] = nil;
 489         end
 490
 491         for k in pairs(session) do
 492                 if k ~= "trace" then
 493                         session[k] = nil;
 494                 end
 495         end
 496 end
 497
 498 return _M;