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