diff options
Diffstat (limited to 'lib/diameter/src/base/diameter_peer_fsm.erl')
| -rw-r--r-- | lib/diameter/src/base/diameter_peer_fsm.erl | 311 |
1 files changed, 164 insertions, 147 deletions
diff --git a/lib/diameter/src/base/diameter_peer_fsm.erl b/lib/diameter/src/base/diameter_peer_fsm.erl index 5dab6214b1..66342f7b62 100644 --- a/lib/diameter/src/base/diameter_peer_fsm.erl +++ b/lib/diameter/src/base/diameter_peer_fsm.erl @@ -18,10 +18,10 @@ %% %% -%% This module implements (as a process) the RFC 3588 Peer State +%% This module implements (as a process) the RFC 3588/6733 Peer State %% Machine modulo the necessity of adapting the peer election to the -%% fact that we don't know the identity of a peer until we've -%% received a CER/CEA from it. +%% fact that we don't know the identity of a peer until we've received +%% a CER/CEA from it. %% -module(diameter_peer_fsm). @@ -46,16 +46,19 @@ -include_lib("diameter/include/diameter.hrl"). -include("diameter_internal.hrl"). --include("diameter_gen_base_rfc3588.hrl"). %% Values of Disconnect-Cause in DPR. --define(GOAWAY, ?'DIAMETER_BASE_DISCONNECT-CAUSE_DO_NOT_WANT_TO_TALK_TO_YOU'). --define(REBOOT, ?'DIAMETER_BASE_DISCONNECT-CAUSE_REBOOTING'). --define(BUSY, ?'DIAMETER_BASE_DISCONNECT-CAUSE_BUSY'). +-define(GOAWAY, 2). %% DO_NOT_WANT_TO_TALK_TO_YOU +-define(BUSY, 1). %% BUSY +-define(REBOOT, 0). %% REBOOTING +%% Values of Inband-Security-Id. -define(NO_INBAND_SECURITY, 0). -define(TLS, 1). +%% Note that the a common dictionary hrl is purposely not included +%% since the common dictionary is an argument to start/3. + %% Keys in process dictionary. -define(CB_KEY, cb). %% capabilities callback -define(DPR_KEY, dpr). %% disconnect callback @@ -100,11 +103,13 @@ | {'Wait-CEA', uint32(), uint32()} | 'Open', mode :: accept | connect | {connect, reference()}, - parent :: pid(), %% watchdog process - transport :: pid(), %% transport process + parent :: pid(), %% watchdog process + transport :: pid(), %% transport process + dictionary :: module(), %% common dictionary service :: #diameter_service{}, - dpr = false :: false | {uint32(), uint32()}}). + dpr = false :: false | {uint32(), uint32()}, %% | hop by hop and end to end identifiers + length_errors :: exit | handle | discard}). %% There are non-3588 states possible as a consequence of 5.6.1 of the %% standard and the corresponding problem for incoming CEA's: we don't @@ -126,24 +131,18 @@ %% State Machine rather than closer to the transport. This is what we %% now do below: connect/accept call diameter_watchdog and return the %% pid of the watchdog process, and the watchdog in turn calls start/3 -%% below to start the process implementing the Peer State Machine. The -%% former is a "peer" in diameter_service while the latter is a -%% "conn". In a sense, diameter_service sees the watchdog as -%% implementing the Peer State Machine and the process implemented -%% here as being the transport, not being aware of the watchdog at -%% all. +%% below to start the process implementing the Peer State Machine. %% -%%% --------------------------------------------------------------------------- -%%% # start({connect|accept, Ref}, Opts, Service) -%%% -%%% Output: Pid -%%% --------------------------------------------------------------------------- +%% --------------------------------------------------------------------------- +%% # start/3 +%% --------------------------------------------------------------------------- -spec start(T, [Opt], {diameter:sequence(), - diameter:restriction(), + [node()], + module(), #diameter_service{}}) - -> pid() + -> {reference(), pid()} when T :: {connect|accept, diameter:transport_ref()}, Opt :: diameter:transport_opt(). @@ -152,9 +151,15 @@ %% specified on the transport in question. Check here that the list is %% still non-empty. -start({_,_} = Type, Opts, MS) -> - {ok, Pid} = diameter_peer_fsm_sup:start_child({self(), Type, Opts, MS}), - Pid. +start({_,_} = Type, Opts, S) -> + Ack = make_ref(), + T = {Ack, self(), Type, Opts, S}, + {ok, Pid} = diameter_peer_fsm_sup:start_child(T), + try + {erlang:monitor(process, Pid), Pid} + after + Pid ! Ack + end. start_link(T) -> {ok, _} = proc_lib:start_link(?MODULE, @@ -163,8 +168,8 @@ start_link(T) -> infinity, diameter_lib:spawn_opts(server, [])). -%%% --------------------------------------------------------------------------- -%%% --------------------------------------------------------------------------- +%% --------------------------------------------------------------------------- +%% --------------------------------------------------------------------------- %% init/1 @@ -172,12 +177,14 @@ init(T) -> proc_lib:init_ack({ok, self()}), gen_server:enter_loop(?MODULE, [], i(T)). -i({WPid, T, Opts, {Mask, Nodes, #diameter_service{applications = Apps, - capabilities = LCaps} - = Svc}}) -> - [] /= Apps orelse ?ERROR({no_apps, T, Opts}), +i({Ack, WPid, {M, Ref} = T, Opts, {Mask, + Nodes, + Dict0, + #diameter_service{capabilities = LCaps} + = Svc}}) -> + erlang:monitor(process, WPid), + wait(Ack, WPid), putr(?DWA_KEY, dwa(LCaps)), - {M, Ref} = T, diameter_stats:reg(Ref), {[Cs,Ds], Rest} = proplists:split(Opts, [capabilities_cb, disconnect_cb]), putr(?CB_KEY, {Ref, [F || {_,F} <- Cs]}), @@ -185,23 +192,39 @@ i({WPid, T, Opts, {Mask, Nodes, #diameter_service{applications = Apps, putr(?REF_KEY, Ref), putr(?SEQUENCE_KEY, Mask), putr(?RESTRICT_KEY, Nodes), - erlang:monitor(process, WPid), - {TPid, Addrs} = start_transport(T, Rest, Svc), + Tmo = proplists:get_value(capx_timeout, Opts, ?EVENT_TIMEOUT), ?IS_TIMEOUT(Tmo) orelse ?ERROR({invalid, {capx_timeout, Tmo}}), + OnLengthErr = proplists:get_value(length_errors, Opts, exit), + lists:member(OnLengthErr, [exit, handle, discard]) + orelse ?ERROR({invalid, {length_errors, OnLengthErr}}), + + {TPid, Addrs} = start_transport(T, Rest, Svc), + #state{state = {'Wait-Conn-Ack', Tmo}, parent = WPid, transport = TPid, + dictionary = Dict0, mode = M, - service = svc(Svc, Addrs)}. + service = svc(Svc, Addrs), + length_errors = OnLengthErr}. %% The transport returns its local ip addresses so that different %% transports on the same service can use different local addresses. %% The local addresses are put into Host-IP-Address avps here when %% sending capabilities exchange messages. %% %% Invalid transport config may cause us to crash but note that the -%% watchdog start (start/2) succeeds regardless so as not to crash the -%% service. +%% watchdog start (start/2) succeeds regardless. + +%% Wait for the caller to have a monitor to avoid a race with our +%% death. (Since the exit reason is used in diameter_service.) +wait(Ref, Pid) -> + receive + Ref -> + ok; + {'DOWN', _, process, Pid, _} = D -> + exit({shutdown, D}) + end. start_transport(T, Opts, #diameter_service{capabilities = LCaps} = Svc) -> Addrs0 = LCaps#diameter_caps.host_ip_address, @@ -274,13 +297,12 @@ handle_info(T, #state{} = State) -> {noreply, S}; {stop, Reason} -> ?LOG(stop, Reason), - x(Reason, State); + {stop, {shutdown, Reason}, State}; stop -> ?LOG(stop, T), - x(T, State) + {stop, {shutdown, T}, State} catch exit: {diameter_codec, encode, _} = Reason -> - close_wd(Reason, State#state.parent), ?LOG(stop, Reason), %% diameter_codec:encode/2 emits an error report. Only %% indicate the probable reason here. @@ -300,10 +322,6 @@ handle_info(T, #state{} = State) -> %% succesfully encoded. It's not checked at diameter:add_transport/2 %% since this can be called before creating the service. -x(Reason, #state{} = S) -> - close_wd(Reason, S), - {stop, {shutdown, Reason}, S}. - %% terminate/2 terminate(_, _) -> @@ -314,8 +332,8 @@ terminate(_, _) -> code_change(_, State, _) -> {ok, State}. -%%% --------------------------------------------------------------------------- -%%% --------------------------------------------------------------------------- +%% --------------------------------------------------------------------------- +%% --------------------------------------------------------------------------- putr(Key, Val) -> put({?MODULE, Key}, Val). @@ -378,9 +396,8 @@ transition({diameter, {recv, Pkt}}, S) -> recv(Pkt, S); %% Timeout when still in the same state ... -transition({timeout = T, PS}, #state{state = PS} = S) -> - close({capx(PS), T}, S), - stop; +transition({timeout = T, PS}, #state{state = PS}) -> + {stop, {capx(PS), T}}; %% ... or not. transition({timeout, _}, _) -> @@ -393,8 +410,6 @@ transition({send, Msg}, #state{transport = TPid}) -> %% Request for graceful shutdown at remove_transport, stop_service of %% application shutdown. -transition({shutdown = T, Pid}, S) -> - transition({T, Pid, transport}, S); transition({shutdown, Pid, Reason}, #state{parent = Pid, dpr = false} = S) -> dpr(Reason, S); transition({shutdown, Pid, _}, #state{parent = Pid}) -> @@ -454,18 +469,19 @@ start_next(#state{service = Svc0} = S) -> send_CER(#state{state = {'Wait-Conn-Ack', Tmo}, mode = {connect, Remote}, service = #diameter_service{capabilities = LCaps}, - transport = TPid} + transport = TPid, + dictionary = Dict} = S) -> OH = LCaps#diameter_caps.origin_host, req_send_CER(OH, Remote) orelse - close({already_connected, Remote, LCaps}, S), + close({already_connected, Remote, LCaps}), CER = build_CER(S), ?LOG(send, 'CER'), #diameter_packet{header = #diameter_header{end_to_end_id = Eid, hop_by_hop_id = Hid}} = Pkt - = encode(CER), + = encode(CER, Dict), send(TPid, Pkt), start_timer(Tmo, S#state{state = {'Wait-CEA', Hid, Eid}}). @@ -487,42 +503,29 @@ start_timer(Tmo, #state{state = PS} = S) -> %% build_CER/1 -build_CER(#state{service = #diameter_service{capabilities = LCaps}}) -> - {ok, CER} = diameter_capx:build_CER(LCaps), +build_CER(#state{service = #diameter_service{capabilities = LCaps}, + dictionary = Dict}) -> + {ok, CER} = diameter_capx:build_CER(LCaps, Dict), CER. -%% encode/1 +%% encode/2 -encode(Rec) -> +encode(Rec, Dict) -> Seq = diameter_session:sequence({_,_} = getr(?SEQUENCE_KEY)), Hdr = #diameter_header{version = ?DIAMETER_VERSION, end_to_end_id = Seq, hop_by_hop_id = Seq}, - diameter_codec:encode(?BASE, #diameter_packet{header = Hdr, - msg = Rec}). + diameter_codec:encode(Dict, #diameter_packet{header = Hdr, + msg = Rec}). %% recv/2 -%% RFC 3588 has result code 5015 for an invalid length but if a -%% transport is detecting message boundaries using the length header -%% then a length error will likely lead to further errors. - -recv(#diameter_packet{header = #diameter_header{length = Len} - = Hdr, - bin = Bin}, - S) - when Len < 20; - (0 /= Len rem 4 orelse bit_size(Bin) /= 8*Len) -> - discard(invalid_message_length, recv, [size(Bin), - bit_size(Bin) rem 8, - Hdr, - S]); - recv(#diameter_packet{header = #diameter_header{} = Hdr} = Pkt, - #state{parent = Pid} + #state{parent = Pid, + dictionary = Dict0} = S) -> - Name = diameter_codec:msg_name(Hdr), + Name = diameter_codec:msg_name(Dict0, Hdr), Pid ! {recv, self(), Name, Pkt}, diameter_stats:incr({msg_id(Name, Hdr), recv}), %% count received rcv(Name, Pkt, S); @@ -531,29 +534,52 @@ recv(#diameter_packet{header = undefined, bin = Bin} = Pkt, S) -> - recv(Pkt#diameter_packet{header = diameter_codec:decode_header(Bin)}, S); + recv(diameter_codec:decode_header(Bin), Pkt, S); -recv(Bin, S) - when is_binary(Bin) -> - recv(#diameter_packet{bin = Bin}, S); +recv(Bin, S) -> + recv(#diameter_packet{bin = Bin}, S). -recv(#diameter_packet{header = false} = Pkt, S) -> - discard(truncated_header, recv, [Pkt, S]). +%% recv/3 -msg_id({_,_,_} = T, _) -> - T; -msg_id(_, Hdr) -> - diameter_codec:msg_id(Hdr). +recv(#diameter_header{length = Len} + = H, + #diameter_packet{bin = Bin} + = Pkt, + #state{length_errors = E} + = S) + when E == handle; + 0 == Len rem 4, bit_size(Bin) == 8*Len -> + recv(Pkt#diameter_packet{header = H}, S); + +recv(#diameter_header{} + = H, + #diameter_packet{bin = Bin}, + #state{length_errors = E} + = S) -> + invalid(E, + invalid_message_length, + recv, + [size(Bin), bit_size(Bin) rem 8, H, S]); -%% Treat invalid length as a transport error and die. Especially in -%% the TCP case, in which there's no telling where the next message -%% begins in the incoming byte stream, keeping a crippled connection -%% alive may just make things worse. +recv(false, Pkt, #state{length_errors = E} = S) -> + invalid(E, truncated_header, recv, [Pkt, S]). -discard(Reason, F, A) -> +%% Note that counters here only count discarded messages. +invalid(E, Reason, F, A) -> diameter_stats:incr(Reason), + abort(E, Reason, F, A). + +abort(exit, Reason, F, A) -> diameter_lib:warning_report(Reason, {?MODULE, F, A}), - throw({?MODULE, abort, Reason}). + throw({?MODULE, abort, Reason}); + +abort(_, _, _, _) -> + ok. + +msg_id({_,_,_} = T, _) -> + T; +msg_id(_, Hdr) -> + {_,_,_} = diameter_codec:msg_id(Hdr). %% rcv/3 @@ -607,13 +633,13 @@ send(Pid, Msg) -> %% handle_request/3 -handle_request(Type, #diameter_packet{} = Pkt, S) -> +handle_request(Type, #diameter_packet{} = Pkt, #state{dictionary = D} = S) -> ?LOG(recv, Type), - send_answer(Type, diameter_codec:decode(?BASE, Pkt), S). + send_answer(Type, diameter_codec:decode(D, Pkt), S). %% send_answer/3 -send_answer(Type, ReqPkt, #state{transport = TPid} = S) -> +send_answer(Type, ReqPkt, #state{transport = TPid, dictionary = Dict} = S) -> #diameter_packet{header = H, transport_data = TD} = ReqPkt, @@ -630,13 +656,15 @@ send_answer(Type, ReqPkt, #state{transport = TPid} = S) -> msg = Msg, transport_data = TD}, - send(TPid, diameter_codec:encode(?BASE, Pkt)), + send(TPid, diameter_codec:encode(Dict, Pkt)), eval(PostF, S). eval([F|A], S) -> apply(F, A ++ [S]); eval(ok, S) -> - S. + S; +eval(T, _) -> + close(T). %% build_answer/3 @@ -647,11 +675,11 @@ build_answer('CER', is_error = false}, errors = []} = Pkt, - S) -> - {SupportedApps, RCaps, #diameter_base_CEA{'Result-Code' = RC, - 'Inband-Security-Id' = IS} - = CEA} - = recv_CER(CER, S), + #state{dictionary = Dict0} + = S) -> + {SupportedApps, RCaps, CEA} = recv_CER(CER, S), + + [RC, IS] = Dict0:'#get-'(['Result-Code', 'Inband-Security-Id'], CEA), #diameter_caps{origin_host = {OH, DH}} = Caps @@ -664,10 +692,10 @@ build_answer('CER', orelse ?THROW(4003), %% DIAMETER_ELECTION_LOST caps_cb(Caps) of - N -> {cea(CEA, N), [fun open/5, Pkt, - SupportedApps, - Caps, - {accept, hd([_] = IS)}]} + N -> {cea(CEA, N, Dict0), [fun open/5, Pkt, + SupportedApps, + Caps, + {accept, hd([_] = IS)}]} catch ?FAILURE(Reason) -> rejected(Reason, {'CER', Reason, Caps, Pkt}, S) @@ -684,25 +712,25 @@ build_answer(Type, RC = rc(H, Es), {answer(Type, RC, Es, S), post(Type, RC, Pkt, S)}. -cea(CEA, ok) -> +cea(CEA, ok, _) -> CEA; -cea(CEA, 2001) -> +cea(CEA, 2001, _) -> CEA; -cea(CEA, RC) -> - CEA#diameter_base_CEA{'Result-Code' = RC}. +cea(CEA, RC, Dict0) -> + Dict0:'#set-'({'Result-Code', RC}, CEA). post('CER' = T, RC, Pkt, S) -> - [fun close/2, {T, caps(S), {RC, Pkt}}]; + {T, caps(S), {RC, Pkt}}; post(_, _, _, _) -> ok. rejected({capabilities_cb, _F, Reason}, T, S) -> rejected(Reason, T, S); -rejected(discard, T, S) -> - close(T, S); +rejected(discard, T, _) -> + close(T); rejected({N, Es}, T, S) -> - {answer('CER', N, Es, S), [fun close/2, T]}; + {answer('CER', N, Es, S), T}; rejected(N, T, S) -> rejected({N, []}, T, S). @@ -728,7 +756,7 @@ is_origin({N, _}) -> orelse N == 'Origin-State-Id'. %% failed_avp/1 - + failed_avp([] = No) -> No; failed_avp(Avps) -> @@ -817,22 +845,23 @@ a('DPR', #diameter_caps{origin_host = {Host, _}, %% recv_CER/2 -recv_CER(CER, #state{service = Svc}) -> - {ok, T} = diameter_capx:recv_CER(CER, Svc), +recv_CER(CER, #state{service = Svc, dictionary = Dict}) -> + {ok, T} = diameter_capx:recv_CER(CER, Svc, Dict), T. %% handle_CEA/1 handle_CEA(#diameter_packet{bin = Bin} = Pkt, - #state{service = #diameter_service{capabilities = LCaps}} + #state{dictionary = Dict0, + service = #diameter_service{capabilities = LCaps}} = S) when is_binary(Bin) -> ?LOG(recv, 'CEA'), #diameter_packet{msg = CEA} = DPkt - = diameter_codec:decode(?BASE, Pkt), + = diameter_codec:decode(Dict0, Pkt), {SApps, IS, RCaps} = recv_CEA(DPkt, S), @@ -840,8 +869,7 @@ handle_CEA(#diameter_packet{bin = Bin} = Caps = capz(LCaps, RCaps), - #diameter_base_CEA{'Result-Code' = RC} - = CEA, + RC = Dict0:'#get-'('Result-Code', CEA), %% Ensure that we don't already have a connection to the peer in %% question. This isn't the peer election of 3588 except in the @@ -862,7 +890,7 @@ handle_CEA(#diameter_packet{bin = Bin} of _ -> open(DPkt, SApps, Caps, {connect, hd([_] = IS)}, S) catch - ?FAILURE(Reason) -> close({'CEA', Reason, Caps, DPkt}, S) + ?FAILURE(Reason) -> close({'CEA', Reason, Caps, DPkt}) end. %% Check more than the result code since the peer could send success %% regardless. If not 2001 then a peer_up callback could do anything @@ -877,12 +905,13 @@ recv_CEA(#diameter_packet{header = #diameter_header{version is_error = false}, msg = CEA, errors = []}, - #state{service = Svc}) -> - {ok, T} = diameter_capx:recv_CEA(CEA, Svc), + #state{service = Svc, + dictionary = Dict}) -> + {ok, T} = diameter_capx:recv_CEA(CEA, Svc, Dict), T; recv_CEA(Pkt, S) -> - close({'CEA', caps(S), Pkt}, S). + close({'CEA', caps(S), Pkt}). caps(#diameter_service{capabilities = Caps}) -> Caps; @@ -935,14 +964,14 @@ open(Pkt, SupportedApps, Caps, {Type, IS}, #state{parent = Pid, %% We've advertised TLS support: tell the transport the result %% and expect a reply when the handshake is complete. -tls_ack(true, Caps, Type, IS, #state{transport = TPid} = S) -> +tls_ack(true, Caps, Type, IS, #state{transport = TPid}) -> Ref = make_ref(), TPid ! {diameter, {tls, Ref, Type, IS == ?TLS}}, receive {diameter, {tls, Ref}} -> ok; {'DOWN', _, process, TPid, Reason} -> - close({tls_ack, Reason, Caps}, S) + close({tls_ack, Reason, Caps}) end; %% Or not. Don't send anything to the transport so that transports @@ -955,25 +984,11 @@ capz(#diameter_caps{} = L, #diameter_caps{} = R) -> = list_to_tuple([diameter_caps | lists:zip(tl(tuple_to_list(L)), tl(tuple_to_list(R)))]). -%% close/2 +%% close/1 -%% Tell the watchdog that our death isn't due to transport failure. -close(Reason, #state{parent = Pid}) -> - close_wd(Reason, Pid), +close(Reason) -> throw({?MODULE, close, Reason}). -%% close_wd/2 - -%% Ensure the watchdog dies if DPR has been sent ... -close_wd(_, #state{dpr = false}) -> - ok; -close_wd(Reason, #state{parent = Pid}) -> - close_wd(Reason, Pid); - -%% ... or otherwise -close_wd(Reason, Pid) -> - Pid ! {close, self(), Reason}. - %% dwa/1 dwa(#diameter_caps{origin_host = OH, @@ -1035,13 +1050,14 @@ dpr([CB|Rest], [Reason | _] = Args, S) -> diameter_lib:error_report(failure, No), {stop, No} end; - + dpr([], [Reason | _], S) -> send_dpr(Reason, [], S). -record(opts, {cause, timeout = ?DPA_TIMEOUT}). send_dpr(Reason, Opts, #state{transport = TPid, + dictionary = Dict, service = #diameter_service{capabilities = Caps}} = S) -> #opts{cause = Cause, timeout = Tmo} @@ -1061,7 +1077,8 @@ send_dpr(Reason, Opts, #state{transport = TPid, = Pkt = encode(['DPR', {'Origin-Host', OH}, {'Origin-Realm', OR}, - {'Disconnect-Cause', Cause}]), + {'Disconnect-Cause', Cause}], + Dict), send(TPid, Pkt), dpa_timer(Tmo), ?LOG(send, 'DPR'), |