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 local config = require "core.configmanager";
  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 _G.prosody.incoming_s2s = incoming_s2s;
  49 local incoming_s2s = incoming_s2s;
  50
  51 module "s2smanager"
  52
  53 function compare_srv_priorities(a,b)
  54         return a.priority < b.priority or (a.priority == b.priority and a.weight > b.weight);
  55 end
  56
  57 local function bounce_sendq(session)
  58         local sendq = session.sendq;
  59         if sendq then
  60                 session.log("info", "sending error replies for "..#sendq.." queued stanzas because of failed outgoing connection to "..tostring(session.to_host));
  61                 local dummy = {
  62                         type = "s2sin";
  63                         send = function(s)
  64                                 (session.log or log)("error", "Replying to to an s2s error reply, please report this! Traceback: %s", get_traceback());
  65                         end;
  66                         dummy = true;
  67                 };
  68                 for i, data in ipairs(sendq) do
  69                         local reply = data[2];
  70                         local xmlns = reply.attr.xmlns;
  71                         if not xmlns or xmlns == "jabber:client" or xmlns == "jabber:server" then
  72                                 reply.attr.type = "error";
  73                                 reply:tag("error", {type = "cancel"})
  74                                         :tag("remote-server-not-found", {xmlns = "urn:ietf:params:xml:ns:xmpp-stanzas"}):up();
  75                                 core_process_stanza(dummy, reply);
  76                         end
  77                         sendq[i] = nil;
  78                 end
  79                 session.sendq = nil;
  80         end
  81 end
  82
  83 function send_to_host(from_host, to_host, data)
  84         if not hosts[from_host] then
  85                 log("warn", "Attempt to send stanza from %s - a host we don't serve", from_host);
  86                 return false;
  87         end
  88         local host = hosts[from_host].s2sout[to_host];
  89         if host then
  90                 -- We have a connection to this host already
  91                 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
  92                         (host.log or log)("debug", "trying to send over unauthed s2sout to "..to_host);
  93
  94                         -- Queue stanza until we are able to send it
  95                         if host.sendq then t_insert(host.sendq, {tostring(data), st.reply(data)});
  96                         else host.sendq = { {tostring(data), st.reply(data)} }; end
  97                         host.log("debug", "stanza [%s] queued ", data.name);
  98                 elseif host.type == "local" or host.type == "component" then
  99                         log("error", "Trying to send a stanza to ourselves??")
 100                         log("error", "Traceback: %s", get_traceback());
 101                         log("error", "Stanza: %s", tostring(data));
 102                 else
 103                         (host.log or log)("debug", "going to send stanza to "..to_host.." from "..from_host);
 104                         -- FIXME
 105                         if host.from_host ~= from_host then
 106                                 log("error", "WARNING! This might, possibly, be a bug, but it might not...");
 107                                 log("error", "We are going to send from %s instead of %s", tostring(host.from_host), tostring(from_host));
 108                         end
 109                         host.sends2s(data);
 110                         host.log("debug", "stanza sent over "..host.type);
 111                 end
 112         else
 113                 log("debug", "opening a new outgoing connection for this stanza");
 114                 local host_session = new_outgoing(from_host, to_host);
 115
 116                 -- Store in buffer
 117                 host_session.sendq = { {tostring(data), st.reply(data)} };
 118                 log("debug", "stanza [%s] queued until connection complete", tostring(data.name));
 119                 if (not host_session.connecting) and (not host_session.conn) then
 120                         log("warn", "Connection to %s failed already, destroying session...", to_host);
 121                         destroy_session(host_session);
 122                 end
 123         end
 124 end
 125
 126 local open_sessions = 0;
 127
 128 function new_incoming(conn)
 129         local session = { conn = conn, type = "s2sin_unauthed", direction = "incoming", hosts = {} };
 130         if true then
 131                 session.trace = newproxy(true);
 132                 getmetatable(session.trace).__gc = function () open_sessions = open_sessions - 1; end;
 133         end
 134         open_sessions = open_sessions + 1;
 135         local w, log = conn.write, logger_init("s2sin"..tostring(conn):match("[a-f0-9]+$"));
 136         session.log = log;
 137         session.sends2s = function (t) log("debug", "sending: %s", tostring(t)); w(tostring(t)); end
 138         incoming_s2s[session] = true;
 139         add_task(connect_timeout, function ()
 140                 if session.conn ~= conn or
 141                    session.type == "s2sin" then
 142                         return; -- Ok, we're connect[ed|ing]
 143                 end
 144                 -- Not connected, need to close session and clean up
 145                 (session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 146                     session.from_host or "(unknown)", session.to_host or "(unknown)");
 147                 session:close("connection-timeout");
 148         end);
 149         return session;
 150 end
 151
 152 function new_outgoing(from_host, to_host)
 153                 local host_session = { to_host = to_host, from_host = from_host, host = from_host,
 154                                        notopen = true, type = "s2sout_unauthed", direction = "outgoing" };
 155
 156                 hosts[from_host].s2sout[to_host] = host_session;
 157
 158                 local log;
 159                 do
 160                         local conn_name = "s2sout"..tostring(host_session):match("[a-f0-9]*$");
 161                         log = logger_init(conn_name);
 162                         host_session.log = log;
 163                 end
 164
 165                 -- Kick the connection attempting machine
 166                 attempt_connection(host_session);
 167
 168                 if not host_session.sends2s then
 169                         -- A sends2s which buffers data (until the stream is opened)
 170                         -- note that data in this buffer will be sent before the stream is authed
 171                         -- and will not be ack'd in any way, successful or otherwise
 172                         local buffer;
 173                         function host_session.sends2s(data)
 174                                 if not buffer then
 175                                         buffer = {};
 176                                         host_session.send_buffer = buffer;
 177                                 end
 178                                 log("debug", "Buffering data on unconnected s2sout to %s", to_host);
 179                                 buffer[#buffer+1] = data;
 180                                 log("debug", "Buffered item %d: %s", #buffer, tostring(data));
 181                         end
 182
 183                 end
 184
 185                 return host_session;
 186 end
 187
 188
 189 function attempt_connection(host_session, err)
 190         local from_host, to_host = host_session.from_host, host_session.to_host;
 191         local connect_host, connect_port = idna_to_ascii(to_host), 5269;
 192
 193         if not connect_host then
 194                 return false;
 195         end
 196
 197         if not err then -- This is our first attempt
 198                 log("debug", "First attempt to connect to %s, starting with SRV lookup...", to_host);
 199                 host_session.connecting = true;
 200                 local handle;
 201                 handle = adns.lookup(function (answer)
 202                         handle = nil;
 203                         host_session.connecting = nil;
 204                         if answer then
 205                                 log("debug", to_host.." has SRV records, handling...");
 206                                 local srv_hosts = {};
 207                                 host_session.srv_hosts = srv_hosts;
 208                                 for _, record in ipairs(answer) do
 209                                         t_insert(srv_hosts, record.srv);
 210                                 end
 211                                 t_sort(srv_hosts, compare_srv_priorities);
 212
 213                                 local srv_choice = srv_hosts[1];
 214                                 host_session.srv_choice = 1;
 215                                 if srv_choice then
 216                                         connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 217                                         log("debug", "Best record found, will connect to %s:%d", connect_host, connect_port);
 218                                 end
 219                         else
 220                                 log("debug", to_host.." has no SRV records, falling back to A");
 221                         end
 222                         -- Try with SRV, or just the plain hostname if no SRV
 223                         local ok, err = try_connect(host_session, connect_host, connect_port);
 224                         if not ok then
 225                                 if not attempt_connection(host_session, err) then
 226                                         -- No more attempts will be made
 227                                         destroy_session(host_session);
 228                                 end
 229                         end
 230                 end, "_xmpp-server._tcp."..connect_host..".", "SRV");
 231
 232                 -- Set handler for DNS timeout
 233                 add_task(dns_timeout, function ()
 234                         if handle then
 235                                 adns.cancel(handle, true);
 236                         end
 237                 end);
 238
 239                 log("debug", "DNS lookup for %s sent, waiting for response before we can connect", to_host);
 240                 return true; -- Attempt in progress
 241         elseif host_session.srv_hosts and #host_session.srv_hosts > host_session.srv_choice then -- Not our first attempt, and we also have SRV
 242                 host_session.srv_choice = host_session.srv_choice + 1;
 243                 local srv_choice = host_session.srv_hosts[host_session.srv_choice];
 244                 connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 245                 host_session.log("info", "Connection failed (%s). Attempt #%d: This time to %s:%d", tostring(err), host_session.srv_choice, connect_host, connect_port);
 246         else
 247                 host_session.log("info", "Out of connection options, can't connect to %s", tostring(host_session.to_host));
 248                 -- We're out of options
 249                 return false;
 250         end
 251
 252         if not (connect_host and connect_port) then
 253                 -- Likely we couldn't resolve DNS
 254                 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));
 255                 return false;
 256         end
 257
 258         return try_connect(host_session, connect_host, connect_port);
 259 end
 260
 261 function try_connect(host_session, connect_host, connect_port)
 262         host_session.connecting = true;
 263         local handle;
 264         handle = adns.lookup(function (reply)
 265                 handle = nil;
 266                 host_session.connecting = nil;
 267
 268                 -- COMPAT: This is a compromise for all you CNAME-(ab)users :)
 269                 if not (reply and reply[#reply] and reply[#reply].a) then
 270                         local count = max_dns_depth;
 271                         reply = dns.peek(connect_host, "CNAME", "IN");
 272                         while count > 0 and reply and reply[#reply] and not reply[#reply].a and reply[#reply].cname do
 273                                 log("debug", "Looking up %s (DNS depth is %d)", tostring(reply[#reply].cname), count);
 274                                 reply = dns.peek(reply[#reply].cname, "A", "IN") or dns.peek(reply[#reply].cname, "CNAME", "IN");
 275                                 count = count - 1;
 276                         end
 277                 end
 278                 -- end of CNAME resolving
 279
 280                 if reply and reply[#reply] and reply[#reply].a then
 281                         log("debug", "DNS reply for %s gives us %s", connect_host, reply[#reply].a);
 282                         return make_connect(host_session, reply[#reply].a, connect_port);
 283                 else
 284                         log("debug", "DNS lookup failed to get a response for %s", connect_host);
 285                         if not attempt_connection(host_session, "name resolution failed") then -- Retry if we can
 286                                 log("debug", "No other records to try for %s - destroying", host_session.to_host);
 287                                 destroy_session(host_session); -- End of the line, we can't
 288                         end
 289                 end
 290         end, connect_host, "A", "IN");
 291
 292         -- Set handler for DNS timeout
 293         add_task(dns_timeout, function ()
 294                 if handle then
 295                         adns.cancel(handle, true);
 296                 end
 297         end);
 298
 299         return true;
 300 end
 301
 302 function make_connect(host_session, connect_host, connect_port)
 303         host_session.log("info", "Beginning new connection attempt to %s (%s:%d)", host_session.to_host, connect_host, connect_port);
 304         -- Ok, we're going to try to connect
 305
 306         local from_host, to_host = host_session.from_host, host_session.to_host;
 307
 308         local conn, handler = socket.tcp()
 309
 310         if not conn then
 311                 log("warn", "Failed to create outgoing connection, system error: %s", handler);
 312                 return false, handler;
 313         end
 314
 315         conn:settimeout(0);
 316         local success, err = conn:connect(connect_host, connect_port);
 317         if not success and err ~= "timeout" then
 318                 log("warn", "s2s connect() to %s (%s:%d) failed: %s", host_session.to_host, connect_host, connect_port, err);
 319                 return false, err;
 320         end
 321
 322         local cl = connlisteners_get("xmppserver");
 323         conn = wrapclient(conn, connect_host, connect_port, cl, cl.default_mode or 1, hosts[from_host].ssl_ctx, false );
 324         host_session.conn = conn;
 325
 326         -- Register this outgoing connection so that xmppserver_listener knows about it
 327         -- otherwise it will assume it is a new incoming connection
 328         cl.register_outgoing(conn, host_session);
 329
 330         local w, log = conn.write, host_session.log;
 331         host_session.sends2s = function (t) log("debug", "sending: %s", tostring(t)); w(tostring(t)); end
 332
 333         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));
 334         log("debug", "Connection attempt in progress...");
 335         add_task(connect_timeout, function ()
 336                 if host_session.conn ~= conn or
 337                    host_session.type == "s2sout" or
 338                    host_session.connecting then
 339                         return; -- Ok, we're connect[ed|ing]
 340                 end
 341                 -- Not connected, need to close session and clean up
 342                 (host_session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 343                     host_session.from_host or "(unknown)", host_session.to_host or "(unknown)");
 344                 host_session:close("connection-timeout");
 345         end);
 346         return true;
 347 end
 348
 349 function streamopened(session, attr)
 350         local send = session.sends2s;
 351
 352         -- TODO: #29: SASL/TLS on s2s streams
 353         session.version = tonumber(attr.version) or 0;
 354
 355         if session.secure == false then
 356                 session.secure = true;
 357         end
 358
 359         if session.version >= 1.0 and not (attr.to and attr.from) then
 360
 361                 (session.log or log)("warn", "Remote of stream "..(session.from_host or "(unknown)").."->"..(session.to_host or "(unknown)")
 362                         .." failed to specify to (%s) and/or from (%s) hostname as per RFC", tostring(attr.to), tostring(attr.from));
 363         end
 364
 365         if session.direction == "incoming" then
 366                 -- Send a reply stream header
 367                 session.to_host = attr.to and nameprep(attr.to);
 368                 session.from_host = attr.from and nameprep(attr.from);
 369
 370                 session.streamid = uuid_gen();
 371                 (session.log or log)("debug", "incoming s2s received <stream:stream>");
 372                 send("<?xml version='1.0'?>");
 373                 send(stanza("stream:stream", { xmlns='jabber:server', ["xmlns:db"]='jabber:server:dialback',
 374                                 ["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());
 375                 if session.to_host and not hosts[session.to_host] then
 376                         -- Attempting to connect to a host we don't serve
 377                         session:close({ condition = "host-unknown"; text = "This host does not serve "..session.to_host });
 378                         return;
 379                 end
 380                 if session.version >= 1.0 then
 381                         local features = st.stanza("stream:features");
 382
 383                         if session.to_host then
 384                                 hosts[session.to_host].events.fire_event("s2s-stream-features", { session = session, features = features });
 385                         else
 386                                 (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");
 387                         end
 388
 389                         log("debug", "Sending stream features: %s", tostring(features));
 390                         send(features);
 391                 end
 392         elseif session.direction == "outgoing" then
 393                 -- If we are just using the connection for verifying dialback keys, we won't try and auth it
 394                 if not attr.id then error("stream response did not give us a streamid!!!"); end
 395                 session.streamid = attr.id;
 396
 397                 -- Send unauthed buffer
 398                 -- (stanzas which are fine to send before dialback)
 399                 -- Note that this is *not* the stanza queue (which
 400                 -- we can only send if auth succeeds) :)
 401                 local send_buffer = session.send_buffer;
 402                 if send_buffer and #send_buffer > 0 then
 403                         log("debug", "Sending s2s send_buffer now...");
 404                         for i, data in ipairs(send_buffer) do
 405                                 session.sends2s(tostring(data));
 406                                 send_buffer[i] = nil;
 407                         end
 408                 end
 409                 session.send_buffer = nil;
 410
 411                 -- If server is pre-1.0, don't wait for features, just do dialback
 412                 if session.version < 1.0 then
 413                         if not session.dialback_verifying then
 414                                 log("debug", "Initiating dialback...");
 415                                 initiate_dialback(session);
 416                         else
 417                                 mark_connected(session);
 418                         end
 419                 end
 420         end
 421
 422         session.notopen = nil;
 423 end
 424
 425 function streamclosed(session)
 426         (session.log or log)("debug", "</stream:stream>");
 427         if session.sends2s then
 428                 session.sends2s("</stream:stream>");
 429         end
 430         session.notopen = true;
 431 end
 432
 433 function initiate_dialback(session)
 434         -- generate dialback key
 435         session.dialback_key = generate_dialback(session.streamid, session.to_host, session.from_host);
 436         session.sends2s(format("<db:result from='%s' to='%s'>%s</db:result>", session.from_host, session.to_host, session.dialback_key));
 437         session.log("info", "sent dialback key on outgoing s2s stream");
 438 end
 439
 440 function generate_dialback(id, to, from)
 441         return sha256_hash(id..to..from..dialback_secret, true);
 442 end
 443
 444 function verify_dialback(id, to, from, key)
 445         return key == generate_dialback(id, to, from);
 446 end
 447
 448 function make_authenticated(session, host)
 449         if session.type == "s2sout_unauthed" then
 450                 session.type = "s2sout";
 451         elseif session.type == "s2sin_unauthed" then
 452                 session.type = "s2sin";
 453                 if host then
 454                         session.hosts[host].authed = true;
 455                 end
 456         elseif session.type == "s2sin" and host then
 457                 session.hosts[host].authed = true;
 458         else
 459                 return false;
 460         end
 461         session.log("debug", "connection %s->%s is now authenticated", session.from_host or "(unknown)", session.to_host or "(unknown)");
 462
 463         mark_connected(session);
 464
 465         return true;
 466 end
 467
 468 -- Stream is authorised, and ready for normal stanzas
 469 function mark_connected(session)
 470         local sendq, send = session.sendq, session.sends2s;
 471
 472         local from, to = session.from_host, session.to_host;
 473
 474         session.log("info", session.direction.." s2s connection "..from.."->"..to.." complete");
 475
 476         local send_to_host = send_to_host;
 477         function session.send(data) send_to_host(to, from, data); end
 478
 479
 480         if session.direction == "outgoing" then
 481                 if sendq then
 482                         session.log("debug", "sending "..#sendq.." queued stanzas across new outgoing connection to "..session.to_host);
 483                         for i, data in ipairs(sendq) do
 484                                 send(data[1]);
 485                                 sendq[i] = nil;
 486                         end
 487                         session.sendq = nil;
 488                 end
 489
 490                 session.srv_hosts = nil;
 491         end
 492 end
 493
 494 function destroy_session(session)
 495         (session.log or log)("info", "Destroying "..tostring(session.direction).." session "..tostring(session.from_host).."->"..tostring(session.to_host));
 496
 497         if session.direction == "outgoing" then
 498                 hosts[session.from_host].s2sout[session.to_host] = nil;
 499                 bounce_sendq(session);
 500         elseif session.direction == "incoming" then
 501                 incoming_s2s[session] = nil;
 502         end
 503
 504         for k in pairs(session) do
 505                 if k ~= "trace" then
 506                         session[k] = nil;
 507                 end
 508         end
 509 end
 510
 511 return _M;