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 adns, dns = require "net.adns", require "net.dns";
  40 local config = require "core.configmanager";
  41 local connect_timeout = config.get("*", "core", "s2s_timeout") or 60;
  42 local dns_timeout = config.get("*", "core", "dns_timeout") or 60;
  43 local max_dns_depth = config.get("*", "core", "dns_max_depth") or 3;
  44 local dialback_secret = config.get("*", "core", "dialback_secret") or uuid_gen();
  45
  46 incoming_s2s = {};
  47 _G.prosody.incoming_s2s = 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         if not hosts[from_host] then
  82                 log("warn", "Attempt to send stanza from %s - a host we don't serve", from_host);
  83                 return false;
  84         end
  85         local host = hosts[from_host].s2sout[to_host];
  86         if host then
  87                 -- We have a connection to this host already
  88                 if host.type == "s2sout_unauthed" and (data.name ~= "db:verify" or not host.dialback_key) and ((not data.xmlns) or data.xmlns == "jabber:client" or data.xmlns == "jabber:server") then
  89                         (host.log or log)("debug", "trying to send over unauthed s2sout to "..to_host);
  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
 113                 -- Store in buffer
 114                 host_session.sendq = { {tostring(data), st.reply(data)} };
 115                 log("debug", "stanza [%s] queued until connection complete", tostring(data.name));
 116                 if (not host_session.connecting) and (not host_session.conn) then
 117                         log("warn", "Connection to %s failed already, destroying session...", to_host);
 118                         destroy_session(host_session);
 119                 end
 120         end
 121 end
 122
 123 local open_sessions = 0;
 124
 125 function new_incoming(conn)
 126         local session = { conn = conn, type = "s2sin_unauthed", direction = "incoming", hosts = {} };
 127         if true then
 128                 session.trace = newproxy(true);
 129                 getmetatable(session.trace).__gc = function () open_sessions = open_sessions - 1; end;
 130         end
 131         open_sessions = open_sessions + 1;
 132         local w, log = conn.write, logger_init("s2sin"..tostring(conn):match("[a-f0-9]+$"));
 133         session.log = log;
 134         session.sends2s = function (t) log("debug", "sending: %s", t.top_tag and t:top_tag() or t:match("^([^>]*>?)")); w(conn, tostring(t)); end
 135         incoming_s2s[session] = true;
 136         add_task(connect_timeout, function ()
 137                 if session.conn ~= conn or
 138                    session.type == "s2sin" then
 139                         return; -- Ok, we're connect[ed|ing]
 140                 end
 141                 -- Not connected, need to close session and clean up
 142                 (session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 143                     session.from_host or "(unknown)", session.to_host or "(unknown)");
 144                 session:close("connection-timeout");
 145         end);
 146         return session;
 147 end
 148
 149 function new_outgoing(from_host, to_host)
 150                 local host_session = { to_host = to_host, from_host = from_host, host = from_host,
 151                                        notopen = true, type = "s2sout_unauthed", direction = "outgoing" };
 152
 153                 hosts[from_host].s2sout[to_host] = host_session;
 154
 155                 local log;
 156                 do
 157                         local conn_name = "s2sout"..tostring(host_session):match("[a-f0-9]*$");
 158                         log = logger_init(conn_name);
 159                         host_session.log = log;
 160                 end
 161
 162                 -- Kick the connection attempting machine
 163                 attempt_connection(host_session);
 164
 165                 if not host_session.sends2s then
 166                         -- A sends2s which buffers data (until the stream is opened)
 167                         -- note that data in this buffer will be sent before the stream is authed
 168                         -- and will not be ack'd in any way, successful or otherwise
 169                         local buffer;
 170                         function host_session.sends2s(data)
 171                                 if not buffer then
 172                                         buffer = {};
 173                                         host_session.send_buffer = buffer;
 174                                 end
 175                                 log("debug", "Buffering data on unconnected s2sout to %s", to_host);
 176                                 buffer[#buffer+1] = data;
 177                                 log("debug", "Buffered item %d: %s", #buffer, tostring(data));
 178                         end
 179
 180                 end
 181
 182                 return host_session;
 183 end
 184
 185
 186 function attempt_connection(host_session, err)
 187         local from_host, to_host = host_session.from_host, host_session.to_host;
 188         local connect_host, connect_port = idna_to_ascii(to_host), 5269;
 189
 190         if not connect_host then
 191                 return false;
 192         end
 193
 194         if not err then -- This is our first attempt
 195                 log("debug", "First attempt to connect to %s, starting with SRV lookup...", to_host);
 196                 host_session.connecting = true;
 197                 local handle;
 198                 handle = adns.lookup(function (answer)
 199                         handle = nil;
 200                         host_session.connecting = nil;
 201                         if answer then
 202                                 log("debug", to_host.." has SRV records, handling...");
 203                                 local srv_hosts = {};
 204                                 host_session.srv_hosts = srv_hosts;
 205                                 for _, record in ipairs(answer) do
 206                                         t_insert(srv_hosts, record.srv);
 207                                 end
 208                                 t_sort(srv_hosts, compare_srv_priorities);
 209
 210                                 local srv_choice = srv_hosts[1];
 211                                 host_session.srv_choice = 1;
 212                                 if srv_choice then
 213                                         connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 214                                         log("debug", "Best record found, will connect to %s:%d", connect_host, connect_port);
 215                                 end
 216                         else
 217                                 log("debug", to_host.." has no SRV records, falling back to A");
 218                         end
 219                         -- Try with SRV, or just the plain hostname if no SRV
 220                         local ok, err = try_connect(host_session, connect_host, connect_port);
 221                         if not ok then
 222                                 if not attempt_connection(host_session, err) then
 223                                         -- No more attempts will be made
 224                                         destroy_session(host_session);
 225                                 end
 226                         end
 227                 end, "_xmpp-server._tcp."..connect_host..".", "SRV");
 228
 229                 -- Set handler for DNS timeout
 230                 add_task(dns_timeout, function ()
 231                         if handle then
 232                                 adns.cancel(handle, true);
 233                         end
 234                 end);
 235
 236                 log("debug", "DNS lookup for %s sent, waiting for response before we can connect", to_host);
 237                 return true; -- Attempt in progress
 238         elseif host_session.srv_hosts and #host_session.srv_hosts > host_session.srv_choice then -- Not our first attempt, and we also have SRV
 239                 host_session.srv_choice = host_session.srv_choice + 1;
 240                 local srv_choice = host_session.srv_hosts[host_session.srv_choice];
 241                 connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 242                 host_session.log("info", "Connection failed (%s). Attempt #%d: This time to %s:%d", tostring(err), host_session.srv_choice, connect_host, connect_port);
 243         else
 244                 host_session.log("info", "Out of connection options, can't connect to %s", tostring(host_session.to_host));
 245                 -- We're out of options
 246                 return false;
 247         end
 248
 249         if not (connect_host and connect_port) then
 250                 -- Likely we couldn't resolve DNS
 251                 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));
 252                 return false;
 253         end
 254
 255         return try_connect(host_session, connect_host, connect_port);
 256 end
 257
 258 function try_connect(host_session, connect_host, connect_port)
 259         host_session.connecting = true;
 260         local handle;
 261         handle = adns.lookup(function (reply)
 262                 handle = nil;
 263                 host_session.connecting = nil;
 264
 265                 -- COMPAT: This is a compromise for all you CNAME-(ab)users :)
 266                 if not (reply and reply[#reply] and reply[#reply].a) then
 267                         local count = max_dns_depth;
 268                         reply = dns.peek(connect_host, "CNAME", "IN");
 269                         while count > 0 and reply and reply[#reply] and not reply[#reply].a and reply[#reply].cname do
 270                                 log("debug", "Looking up %s (DNS depth is %d)", tostring(reply[#reply].cname), count);
 271                                 reply = dns.peek(reply[#reply].cname, "A", "IN") or dns.peek(reply[#reply].cname, "CNAME", "IN");
 272                                 count = count - 1;
 273                         end
 274                 end
 275                 -- end of CNAME resolving
 276
 277                 if reply and reply[#reply] and reply[#reply].a then
 278                         log("debug", "DNS reply for %s gives us %s", connect_host, reply[#reply].a);
 279                         return make_connect(host_session, reply[#reply].a, connect_port);
 280                 else
 281                         log("debug", "DNS lookup failed to get a response for %s", connect_host);
 282                         if not attempt_connection(host_session, "name resolution failed") then -- Retry if we can
 283                                 log("debug", "No other records to try for %s - destroying", host_session.to_host);
 284                                 destroy_session(host_session); -- End of the line, we can't
 285                         end
 286                 end
 287         end, connect_host, "A", "IN");
 288
 289         -- Set handler for DNS timeout
 290         add_task(dns_timeout, function ()
 291                 if handle then
 292                         adns.cancel(handle, true);
 293                 end
 294         end);
 295
 296         return true;
 297 end
 298
 299 function make_connect(host_session, connect_host, connect_port)
 300         host_session.log("info", "Beginning new connection attempt to %s (%s:%d)", host_session.to_host, connect_host, connect_port);
 301         -- Ok, we're going to try to connect
 302
 303         local from_host, to_host = host_session.from_host, host_session.to_host;
 304
 305         local conn, handler = socket.tcp()
 306
 307         conn:settimeout(0);
 308         local success, err = conn:connect(connect_host, connect_port);
 309         if not success and err ~= "timeout" then
 310                 log("warn", "s2s connect() to %s (%s:%d) failed: %s", host_session.to_host, connect_host, connect_port, err);
 311                 return false, err;
 312         end
 313
 314         local cl = connlisteners_get("xmppserver");
 315         conn = wrapclient(conn, connect_host, connect_port, cl, cl.default_mode or 1, hosts[from_host].ssl_ctx, false );
 316         host_session.conn = conn;
 317
 318         -- Register this outgoing connection so that xmppserver_listener knows about it
 319         -- otherwise it will assume it is a new incoming connection
 320         cl.register_outgoing(conn, host_session);
 321
 322         local w, log = conn.write, host_session.log;
 323         host_session.sends2s = function (t) log("debug", "sending: %s", (t.top_tag and t:top_tag()) or t:match("^[^>]*>?")); w(conn, tostring(t)); end
 324
 325         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));
 326         log("debug", "Connection attempt in progress...");
 327         add_task(connect_timeout, function ()
 328                 if host_session.conn ~= conn or
 329                    host_session.type == "s2sout" or
 330                    host_session.connecting then
 331                         return; -- Ok, we're connect[ed|ing]
 332                 end
 333                 -- Not connected, need to close session and clean up
 334                 (host_session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 335                     host_session.from_host or "(unknown)", host_session.to_host or "(unknown)");
 336                 host_session:close("connection-timeout");
 337         end);
 338         return true;
 339 end
 340
 341 function streamopened(session, attr)
 342         local send = session.sends2s;
 343
 344         -- TODO: #29: SASL/TLS on s2s streams
 345         session.version = tonumber(attr.version) or 0;
 346
 347         if session.secure == false then
 348                 session.secure = true;
 349         end
 350
 351         if session.version >= 1.0 and not (attr.to and attr.from) then
 352
 353                 (session.log or log)("warn", "Remote of stream "..(session.from_host or "(unknown)").."->"..(session.to_host or "(unknown)")
 354                         .." failed to specify to (%s) and/or from (%s) hostname as per RFC", tostring(attr.to), tostring(attr.from));
 355         end
 356
 357         if session.direction == "incoming" then
 358                 -- Send a reply stream header
 359                 session.to_host = attr.to and nameprep(attr.to);
 360                 session.from_host = attr.from and nameprep(attr.from);
 361
 362                 session.streamid = uuid_gen();
 363                 (session.log or log)("debug", "incoming s2s received <stream:stream>");
 364                 send("<?xml version='1.0'?>");
 365                 send(stanza("stream:stream", { xmlns='jabber:server', ["xmlns:db"]='jabber:server:dialback',
 366                                 ["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());
 367                 if session.to_host and not hosts[session.to_host] then
 368                         -- Attempting to connect to a host we don't serve
 369                         session:close({ condition = "host-unknown"; text = "This host does not serve "..session.to_host });
 370                         return;
 371                 end
 372                 if session.version >= 1.0 then
 373                         local features = st.stanza("stream:features");
 374
 375                         if session.to_host then
 376                                 hosts[session.to_host].events.fire_event("s2s-stream-features", { session = session, features = features });
 377                         else
 378                                 (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");
 379                         end
 380
 381                         log("debug", "Sending stream features: %s", tostring(features));
 382                         send(features);
 383                 end
 384         elseif session.direction == "outgoing" then
 385                 -- If we are just using the connection for verifying dialback keys, we won't try and auth it
 386                 if not attr.id then error("stream response did not give us a streamid!!!"); end
 387                 session.streamid = attr.id;
 388
 389                 -- Send unauthed buffer
 390                 -- (stanzas which are fine to send before dialback)
 391                 -- Note that this is *not* the stanza queue (which
 392                 -- we can only send if auth succeeds) :)
 393                 local send_buffer = session.send_buffer;
 394                 if send_buffer and #send_buffer > 0 then
 395                         log("debug", "Sending s2s send_buffer now...");
 396                         for i, data in ipairs(send_buffer) do
 397                                 session.sends2s(tostring(data));
 398                                 send_buffer[i] = nil;
 399                         end
 400                 end
 401                 session.send_buffer = nil;
 402
 403                 -- If server is pre-1.0, don't wait for features, just do dialback
 404                 if session.version < 1.0 then
 405                         if not session.dialback_verifying then
 406                                 log("debug", "Initiating dialback...");
 407                                 initiate_dialback(session);
 408                         else
 409                                 mark_connected(session);
 410                         end
 411                 end
 412         end
 413
 414         session.notopen = nil;
 415 end
 416
 417 function streamclosed(session)
 418         (session.log or log)("debug", "</stream:stream>");
 419         if session.sends2s then
 420                 session.sends2s("</stream:stream>");
 421         end
 422         session.notopen = true;
 423 end
 424
 425 function initiate_dialback(session)
 426         -- generate dialback key
 427         session.dialback_key = generate_dialback(session.streamid, session.to_host, session.from_host);
 428         session.sends2s(format("<db:result from='%s' to='%s'>%s</db:result>", session.from_host, session.to_host, session.dialback_key));
 429         session.log("info", "sent dialback key on outgoing s2s stream");
 430 end
 431
 432 function generate_dialback(id, to, from)
 433         return sha256_hash(id..to..from..dialback_secret, true);
 434 end
 435
 436 function verify_dialback(id, to, from, key)
 437         return key == generate_dialback(id, to, from);
 438 end
 439
 440 function make_authenticated(session, host)
 441         if session.type == "s2sout_unauthed" then
 442                 session.type = "s2sout";
 443         elseif session.type == "s2sin_unauthed" then
 444                 session.type = "s2sin";
 445                 if host then
 446                         session.hosts[host].authed = true;
 447                 end
 448         elseif session.type == "s2sin" and host then
 449                 session.hosts[host].authed = true;
 450         else
 451                 return false;
 452         end
 453         session.log("debug", "connection %s->%s is now authenticated", session.from_host or "(unknown)", session.to_host or "(unknown)");
 454
 455         mark_connected(session);
 456
 457         return true;
 458 end
 459
 460 -- Stream is authorised, and ready for normal stanzas
 461 function mark_connected(session)
 462         local sendq, send = session.sendq, session.sends2s;
 463
 464         local from, to = session.from_host, session.to_host;
 465
 466         session.log("info", session.direction.." s2s connection "..from.."->"..to.." complete");
 467
 468         local send_to_host = send_to_host;
 469         function session.send(data) send_to_host(to, from, data); end
 470
 471
 472         if session.direction == "outgoing" then
 473                 if sendq then
 474                         session.log("debug", "sending "..#sendq.." queued stanzas across new outgoing connection to "..session.to_host);
 475                         for i, data in ipairs(sendq) do
 476                                 send(data[1]);
 477                                 sendq[i] = nil;
 478                         end
 479                         session.sendq = nil;
 480                 end
 481
 482                 session.srv_hosts = nil;
 483         end
 484 end
 485
 486 function destroy_session(session)
 487         (session.log or log)("info", "Destroying "..tostring(session.direction).." session "..tostring(session.from_host).."->"..tostring(session.to_host));
 488
 489         if session.direction == "outgoing" then
 490                 hosts[session.from_host].s2sout[session.to_host] = nil;
 491                 bounce_sendq(session);
 492         elseif session.direction == "incoming" then
 493                 incoming_s2s[session] = nil;
 494         end
 495
 496         for k in pairs(session) do
 497                 if k ~= "trace" then
 498                         session[k] = nil;
 499                 end
 500         end
 501 end
 502
 503 return _M;