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