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