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