core/s2smanager.lua

   1 -- Prosody IM
   2 -- Copyright (C) 2008-2010 Matthew Wild
   3 -- Copyright (C) 2008-2010 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       setmetatable
  21     = tostring, pairs, ipairs, getmetatable, newproxy, error, tonumber,
  22       setmetatable;
  23
  24 local idna_to_ascii = require "util.encodings".idna.to_ascii;
  25 local connlisteners_get = require "net.connlisteners".get;
  26 local wrapclient = require "net.server".wrapclient;
  27 local modulemanager = require "core.modulemanager";
  28 local st = require "stanza";
  29 local stanza = st.stanza;
  30 local nameprep = require "util.encodings".stringprep.nameprep;
  31
  32 local fire_event = require "core.eventmanager".fire_event;
  33 local uuid_gen = require "util.uuid".generate;
  34
  35 local logger_init = require "util.logger".init;
  36
  37 local log = logger_init("s2smanager");
  38
  39 local sha256_hash = require "util.hashes".sha256;
  40
  41 local adns, dns = require "net.adns", require "net.dns";
  42 local config = require "core.configmanager";
  43 local connect_timeout = config.get("*", "core", "s2s_timeout") or 60;
  44 local dns_timeout = config.get("*", "core", "dns_timeout") or 60;
  45 local max_dns_depth = config.get("*", "core", "dns_max_depth") or 3;
  46
  47 incoming_s2s = {};
  48 _G.prosody.incoming_s2s = incoming_s2s;
  49 local incoming_s2s = incoming_s2s;
  50
  51 module "s2smanager"
  52
  53 function compare_srv_priorities(a,b)
  54         return a.priority < b.priority or (a.priority == b.priority and a.weight > b.weight);
  55 end
  56
  57 local function bounce_sendq(session, reason)
  58         local sendq = session.sendq;
  59         if sendq then
  60                 session.log("info", "sending error replies for "..#sendq.." queued stanzas because of failed outgoing connection to "..tostring(session.to_host));
  61                 local dummy = {
  62                         type = "s2sin";
  63                         send = function(s)
  64                                 (session.log or log)("error", "Replying to to an s2s error reply, please report this! Traceback: %s", get_traceback());
  65                         end;
  66                         dummy = true;
  67                 };
  68                 for i, data in ipairs(sendq) do
  69                         local reply = data[2];
  70                         local xmlns = reply.attr.xmlns;
  71                         if not xmlns then
  72                                 reply.attr.type = "error";
  73                                 reply:tag("error", {type = "cancel"})
  74                                         :tag("remote-server-not-found", {xmlns = "urn:ietf:params:xml:ns:xmpp-stanzas"}):up();
  75                                 if reason then
  76                                         reply:tag("text", {xmlns = "urn:ietf:params:xml:ns:xmpp-stanzas"}):text("Connection failed: "..reason):up();
  77                                 end
  78                                 core_process_stanza(dummy, reply);
  79                         end
  80                         sendq[i] = nil;
  81                 end
  82                 session.sendq = nil;
  83         end
  84 end
  85
  86 function send_to_host(from_host, to_host, data)
  87         if not hosts[from_host] then
  88                 log("warn", "Attempt to send stanza from %s - a host we don't serve", from_host);
  89                 return false;
  90         end
  91         local host = hosts[from_host].s2sout[to_host];
  92         if host then
  93                 -- We have a connection to this host already
  94                 if host.type == "s2sout_unauthed" and (data.name ~= "db:verify" or not host.dialback_key) then
  95                         (host.log or log)("debug", "trying to send over unauthed s2sout to "..to_host);
  96
  97                         -- Queue stanza until we are able to send it
  98                         if host.sendq then t_insert(host.sendq, {tostring(data), st.reply(data)});
  99                         else host.sendq = { {tostring(data), st.reply(data)} }; end
 100                         host.log("debug", "stanza [%s] queued ", data.name);
 101                 elseif host.type == "local" or host.type == "component" then
 102                         log("error", "Trying to send a stanza to ourselves??")
 103                         log("error", "Traceback: %s", get_traceback());
 104                         log("error", "Stanza: %s", tostring(data));
 105                 else
 106                         (host.log or log)("debug", "going to send stanza to "..to_host.." from "..from_host);
 107                         -- FIXME
 108                         if host.from_host ~= from_host then
 109                                 log("error", "WARNING! This might, possibly, be a bug, but it might not...");
 110                                 log("error", "We are going to send from %s instead of %s", tostring(host.from_host), tostring(from_host));
 111                         end
 112                         host.sends2s(data);
 113                         host.log("debug", "stanza sent over "..host.type);
 114                 end
 115         else
 116                 log("debug", "opening a new outgoing connection for this stanza");
 117                 local host_session = new_outgoing(from_host, to_host);
 118
 119                 -- Store in buffer
 120                 host_session.sendq = { {tostring(data), st.reply(data)} };
 121                 log("debug", "stanza [%s] queued until connection complete", tostring(data.name));
 122                 if (not host_session.connecting) and (not host_session.conn) then
 123                         log("warn", "Connection to %s failed already, destroying session...", to_host);
 124                         destroy_session(host_session);
 125                 end
 126         end
 127 end
 128
 129 local open_sessions = 0;
 130
 131 function new_incoming(conn)
 132         local session = { conn = conn, type = "s2sin_unauthed", direction = "incoming", hosts = {} };
 133         if true then
 134                 session.trace = newproxy(true);
 135                 getmetatable(session.trace).__gc = function () open_sessions = open_sessions - 1; end;
 136         end
 137         open_sessions = open_sessions + 1;
 138         local w, log = conn.write, logger_init("s2sin"..tostring(conn):match("[a-f0-9]+$"));
 139         session.log = log;
 140         session.sends2s = function (t) log("debug", "sending: %s", t.top_tag and t:top_tag() or t:match("^([^>]*>?)")); w(conn, tostring(t)); end
 141         incoming_s2s[session] = true;
 142         add_task(connect_timeout, function ()
 143                 if session.conn ~= conn or
 144                    session.type == "s2sin" then
 145                         return; -- Ok, we're connect[ed|ing]
 146                 end
 147                 -- Not connected, need to close session and clean up
 148                 (session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 149                     session.from_host or "(unknown)", session.to_host or "(unknown)");
 150                 session:close("connection-timeout");
 151         end);
 152         return session;
 153 end
 154
 155 function new_outgoing(from_host, to_host, connect)
 156                 local host_session = { to_host = to_host, from_host = from_host, host = from_host,
 157                                        notopen = true, type = "s2sout_unauthed", direction = "outgoing",
 158                                        open_stream = session_open_stream };
 159
 160                 hosts[from_host].s2sout[to_host] = host_session;
 161
 162                 local log;
 163                 do
 164                         local conn_name = "s2sout"..tostring(host_session):match("[a-f0-9]*$");
 165                         log = logger_init(conn_name);
 166                         host_session.log = log;
 167                 end
 168
 169                 if connect ~= false then
 170                         -- Kick the connection attempting machine into life
 171                         attempt_connection(host_session);
 172                 end
 173
 174                 if not host_session.sends2s then
 175                         -- A sends2s which buffers data (until the stream is opened)
 176                         -- note that data in this buffer will be sent before the stream is authed
 177                         -- and will not be ack'd in any way, successful or otherwise
 178                         local buffer;
 179                         function host_session.sends2s(data)
 180                                 if not buffer then
 181                                         buffer = {};
 182                                         host_session.send_buffer = buffer;
 183                                 end
 184                                 log("debug", "Buffering data on unconnected s2sout to %s", to_host);
 185                                 buffer[#buffer+1] = data;
 186                                 log("debug", "Buffered item %d: %s", #buffer, tostring(data));
 187                         end
 188                 end
 189
 190                 return host_session;
 191 end
 192
 193
 194 function attempt_connection(host_session, err)
 195         local from_host, to_host = host_session.from_host, host_session.to_host;
 196         local connect_host, connect_port = idna_to_ascii(to_host), 5269;
 197
 198         if not connect_host then
 199                 return false;
 200         end
 201
 202         if not err then -- This is our first attempt
 203                 log("debug", "First attempt to connect to %s, starting with SRV lookup...", to_host);
 204                 host_session.connecting = true;
 205                 local handle;
 206                 handle = adns.lookup(function (answer)
 207                         handle = nil;
 208                         host_session.connecting = nil;
 209                         if answer then
 210                                 log("debug", to_host.." has SRV records, handling...");
 211                                 local srv_hosts = {};
 212                                 host_session.srv_hosts = srv_hosts;
 213                                 for _, record in ipairs(answer) do
 214                                         t_insert(srv_hosts, record.srv);
 215                                 end
 216                                 t_sort(srv_hosts, compare_srv_priorities);
 217
 218                                 local srv_choice = srv_hosts[1];
 219                                 host_session.srv_choice = 1;
 220                                 if srv_choice then
 221                                         connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 222                                         log("debug", "Best record found, will connect to %s:%d", connect_host, connect_port);
 223                                 end
 224                         else
 225                                 log("debug", to_host.." has no SRV records, falling back to A");
 226                         end
 227                         -- Try with SRV, or just the plain hostname if no SRV
 228                         local ok, err = try_connect(host_session, connect_host, connect_port);
 229                         if not ok then
 230                                 if not attempt_connection(host_session, err) then
 231                                         -- No more attempts will be made
 232                                         destroy_session(host_session, err);
 233                                 end
 234                         end
 235                 end, "_xmpp-server._tcp."..connect_host..".", "SRV");
 236
 237                 -- Set handler for DNS timeout
 238                 add_task(dns_timeout, function ()
 239                         if handle then
 240                                 adns.cancel(handle, true);
 241                         end
 242                 end);
 243
 244                 log("debug", "DNS lookup for %s sent, waiting for response before we can connect", to_host);
 245                 return true; -- Attempt in progress
 246         elseif host_session.srv_hosts and #host_session.srv_hosts > host_session.srv_choice then -- Not our first attempt, and we also have SRV
 247                 host_session.srv_choice = host_session.srv_choice + 1;
 248                 local srv_choice = host_session.srv_hosts[host_session.srv_choice];
 249                 connect_host, connect_port = srv_choice.target or to_host, srv_choice.port or connect_port;
 250                 host_session.log("info", "Connection failed (%s). Attempt #%d: This time to %s:%d", tostring(err), host_session.srv_choice, connect_host, connect_port);
 251         else
 252                 host_session.log("info", "Out of connection options, can't connect to %s", tostring(host_session.to_host));
 253                 -- We're out of options
 254                 return false;
 255         end
 256
 257         if not (connect_host and connect_port) then
 258                 -- Likely we couldn't resolve DNS
 259                 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));
 260                 return false;
 261         end
 262
 263         return try_connect(host_session, connect_host, connect_port);
 264 end
 265
 266 function try_connect(host_session, connect_host, connect_port)
 267         host_session.connecting = true;
 268         local handle;
 269         handle = adns.lookup(function (reply)
 270                 handle = nil;
 271                 host_session.connecting = nil;
 272
 273                 -- COMPAT: This is a compromise for all you CNAME-(ab)users :)
 274                 if not (reply and reply[#reply] and reply[#reply].a) then
 275                         local count = max_dns_depth;
 276                         reply = dns.peek(connect_host, "CNAME", "IN");
 277                         while count > 0 and reply and reply[#reply] and not reply[#reply].a and reply[#reply].cname do
 278                                 log("debug", "Looking up %s (DNS depth is %d)", tostring(reply[#reply].cname), count);
 279                                 reply = dns.peek(reply[#reply].cname, "A", "IN") or dns.peek(reply[#reply].cname, "CNAME", "IN");
 280                                 count = count - 1;
 281                         end
 282                 end
 283                 -- end of CNAME resolving
 284
 285                 if reply and reply[#reply] and reply[#reply].a then
 286                         log("debug", "DNS reply for %s gives us %s", connect_host, reply[#reply].a);
 287                         return make_connect(host_session, reply[#reply].a, connect_port);
 288                 else
 289                         log("debug", "DNS lookup failed to get a response for %s", connect_host);
 290                         if not attempt_connection(host_session, "name resolution failed") then -- Retry if we can
 291                                 log("debug", "No other records to try for %s - destroying", host_session.to_host);
 292                                 destroy_session(host_session, "DNS resolution failed"); -- End of the line, we can't
 293                         end
 294                 end
 295         end, connect_host, "A", "IN");
 296
 297         -- Set handler for DNS timeout
 298         add_task(dns_timeout, function ()
 299                 if handle then
 300                         adns.cancel(handle, true);
 301                 end
 302         end);
 303
 304         return true;
 305 end
 306
 307 function make_connect(host_session, connect_host, connect_port)
 308         (host_session.log or log)("info", "Beginning new connection attempt to %s (%s:%d)", host_session.to_host, connect_host, connect_port);
 309         -- Ok, we're going to try to connect
 310
 311         local from_host, to_host = host_session.from_host, host_session.to_host;
 312
 313         local conn, handler = socket.tcp()
 314
 315         if not conn then
 316                 log("warn", "Failed to create outgoing connection, system error: %s", handler);
 317                 return false, handler;
 318         end
 319
 320         conn:settimeout(0);
 321         local success, err = conn:connect(connect_host, connect_port);
 322         if not success and err ~= "timeout" then
 323                 log("warn", "s2s connect() to %s (%s:%d) failed: %s", host_session.to_host, connect_host, connect_port, err);
 324                 return false, err;
 325         end
 326
 327         local cl = connlisteners_get("xmppserver");
 328         conn = wrapclient(conn, connect_host, connect_port, cl, cl.default_mode or 1 );
 329         host_session.conn = conn;
 330
 331         -- Register this outgoing connection so that xmppserver_listener knows about it
 332         -- otherwise it will assume it is a new incoming connection
 333         cl.register_outgoing(conn, host_session);
 334
 335         local w, log = conn.write, host_session.log;
 336         host_session.sends2s = function (t) log("debug", "sending: %s", (t.top_tag and t:top_tag()) or t:match("^[^>]*>?")); w(conn, tostring(t)); end
 337
 338         host_session:open_stream(from_host, to_host);
 339
 340         log("debug", "Connection attempt in progress...");
 341         add_task(connect_timeout, function ()
 342                 if host_session.conn ~= conn or
 343                    host_session.type == "s2sout" or
 344                    host_session.connecting then
 345                         return; -- Ok, we're connect[ed|ing]
 346                 end
 347                 -- Not connected, need to close session and clean up
 348                 (host_session.log or log)("warn", "Destroying incomplete session %s->%s due to inactivity",
 349                     host_session.from_host or "(unknown)", host_session.to_host or "(unknown)");
 350                 host_session:close("connection-timeout");
 351         end);
 352         return true;
 353 end
 354
 355 function session_open_stream(session, from, to)
 356         session.sends2s(st.stanza("stream:stream", {
 357                 xmlns='jabber:server', ["xmlns:db"]='jabber:server:dialback',
 358                 ["xmlns:stream"]='http://etherx.jabber.org/streams',
 359                 from=from, to=to, version='1.0', ["xml:lang"]='en'}):top_tag());
 360 end
 361
 362 function streamopened(session, attr)
 363         local send = session.sends2s;
 364
 365         -- TODO: #29: SASL/TLS on s2s streams
 366         session.version = tonumber(attr.version) or 0;
 367
 368         if session.secure == false then
 369                 session.secure = true;
 370         end
 371
 372         if session.direction == "incoming" then
 373                 -- Send a reply stream header
 374                 session.to_host = attr.to and nameprep(attr.to);
 375                 session.from_host = attr.from and nameprep(attr.from);
 376
 377                 session.streamid = uuid_gen();
 378                 (session.log or log)("debug", "incoming s2s received <stream:stream>");
 379                 if session.to_host and not hosts[session.to_host] then
 380                         -- Attempting to connect to a host we don't serve
 381                         session:close({ condition = "host-unknown"; text = "This host does not serve "..session.to_host });
 382                         return;
 383                 end
 384                 send("<?xml version='1.0'?>");
 385                 send(stanza("stream:stream", { xmlns='jabber:server', ["xmlns:db"]='jabber:server:dialback',
 386                                 ["xmlns:stream"]='http://etherx.jabber.org/streams', id=session.streamid, from=session.to_host, to=session.from_host, version=(session.version > 0 and "1.0" or nil) }):top_tag());
 387                 if session.version >= 1.0 then
 388                         local features = st.stanza("stream:features");
 389
 390                         if session.to_host then
 391                                 hosts[session.to_host].events.fire_event("s2s-stream-features", { origin = session, features = features });
 392                         else
 393                                 (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");
 394                         end
 395
 396                         log("debug", "Sending stream features: %s", tostring(features));
 397                         send(features);
 398                 end
 399         elseif session.direction == "outgoing" then
 400                 -- If we are just using the connection for verifying dialback keys, we won't try and auth it
 401                 if not attr.id then error("stream response did not give us a streamid!!!"); end
 402                 session.streamid = attr.id;
 403
 404                 -- Send unauthed buffer
 405                 -- (stanzas which are fine to send before dialback)
 406                 -- Note that this is *not* the stanza queue (which
 407                 -- we can only send if auth succeeds) :)
 408                 local send_buffer = session.send_buffer;
 409                 if send_buffer and #send_buffer > 0 then
 410                         log("debug", "Sending s2s send_buffer now...");
 411                         for i, data in ipairs(send_buffer) do
 412                                 session.sends2s(tostring(data));
 413                                 send_buffer[i] = nil;
 414                         end
 415                 end
 416                 session.send_buffer = nil;
 417
 418                 -- If server is pre-1.0, don't wait for features, just do dialback
 419                 if session.version < 1.0 then
 420                         if not session.dialback_verifying then
 421                                 log("debug", "Initiating dialback...");
 422                                 initiate_dialback(session);
 423                         else
 424                                 mark_connected(session);
 425                         end
 426                 end
 427         end
 428         session.notopen = nil;
 429 end
 430
 431 function streamclosed(session)
 432         (session.log or log)("debug", "Received </stream:stream>");
 433         session:close();
 434 end
 435
 436 function initiate_dialback(session)
 437         -- generate dialback key
 438         session.dialback_key = generate_dialback(session.streamid, session.to_host, session.from_host);
 439         session.sends2s(format("<db:result from='%s' to='%s'>%s</db:result>", session.from_host, session.to_host, session.dialback_key));
 440         session.log("info", "sent dialback key on outgoing s2s stream");
 441 end
 442
 443 function generate_dialback(id, to, from)
 444         return sha256_hash(id..to..from..hosts[from].dialback_secret, true);
 445 end
 446
 447 function verify_dialback(id, to, from, key)
 448         return key == generate_dialback(id, to, from);
 449 end
 450
 451 function make_authenticated(session, host)
 452         if not session.secure then
 453                 local local_host = session.direction == "incoming" and session.to_host or session.from_host;
 454                 if config.get(local_host, "core", "s2s_require_encryption") then
 455                         session:close({
 456                                 condition = "policy-violation",
 457                                 text = "Encrypted server-to-server communication is required but was not "
 458                                        ..((session.direction == "outgoing" and "offered") or "used")
 459                         });
 460                 end
 461         end
 462         if session.type == "s2sout_unauthed" then
 463                 session.type = "s2sout";
 464         elseif session.type == "s2sin_unauthed" then
 465                 session.type = "s2sin";
 466                 if host then
 467                         session.hosts[host].authed = true;
 468                 end
 469         elseif session.type == "s2sin" and host then
 470                 session.hosts[host].authed = true;
 471         else
 472                 return false;
 473         end
 474         session.log("debug", "connection %s->%s is now authenticated", session.from_host or "(unknown)", session.to_host or "(unknown)");
 475
 476         mark_connected(session);
 477
 478         return true;
 479 end
 480
 481 -- Stream is authorised, and ready for normal stanzas
 482 function mark_connected(session)
 483         local sendq, send = session.sendq, session.sends2s;
 484
 485         local from, to = session.from_host, session.to_host;
 486
 487         session.log("info", session.direction.." s2s connection "..from.."->"..to.." complete");
 488
 489         local send_to_host = send_to_host;
 490         function session.send(data) send_to_host(to, from, data); end
 491
 492
 493         if session.direction == "outgoing" then
 494                 if sendq then
 495                         session.log("debug", "sending "..#sendq.." queued stanzas across new outgoing connection to "..session.to_host);
 496                         for i, data in ipairs(sendq) do
 497                                 send(data[1]);
 498                                 sendq[i] = nil;
 499                         end
 500                         session.sendq = nil;
 501                 end
 502
 503                 session.srv_hosts = nil;
 504         end
 505 end
 506
 507 local resting_session = { -- Resting, not dead
 508                 destroyed = true;
 509                 type = "s2s_destroyed";
 510                 open_stream = function (session)
 511                         session.log("debug", "Attempt to open stream on resting session");
 512                 end;
 513                 close = function (session)
 514                         session.log("debug", "Attempt to close already-closed session");
 515                 end;
 516         }; resting_session.__index = resting_session;
 517
 518 function retire_session(session)
 519         local log = session.log or log;
 520         for k in pairs(session) do
 521                 if k ~= "trace" and k ~= "log" and k ~= "id" then
 522                         session[k] = nil;
 523                 end
 524         end
 525
 526         function session.send(data) log("debug", "Discarding data sent to resting session: %s", tostring(data)); end
 527         function session.data(data) log("debug", "Discarding data received from resting session: %s", tostring(data)); end
 528         return setmetatable(session, resting_session);
 529 end
 530
 531 function destroy_session(session, reason)
 532         if session.destroyed then return; end
 533         (session.log or log)("info", "Destroying "..tostring(session.direction).." session "..tostring(session.from_host).."->"..tostring(session.to_host));
 534
 535         if session.direction == "outgoing" then
 536                 hosts[session.from_host].s2sout[session.to_host] = nil;
 537                 bounce_sendq(session, reason);
 538         elseif session.direction == "incoming" then
 539                 incoming_s2s[session] = nil;
 540         end
 541
 542         retire_session(session); -- Clean session until it is GC'd
 543 end
 544
 545 return _M;