crypto: add ticketKeys option in createSecureContext

[ryzokuken]

There's a method to initialize a TLS Server using tls.createSever by
specifying a ticketKeys option, but none in the underlying constructor,
tls.createSecureContext.

This PR adds the ticketKeys option to tls.createSecureContext.

Fixes: #20908

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

/cc @nodejs/crypto @bnoordhuis@silverwind@DiegoTUI

I guess this qualifies as a semver-minor?

[ryzokuken]

Running tests locally.

[silverwind]

Change seems good to me. Couly you update the docs and add a test that verifies that both createServer and createSecureContext honor the option (compare the given buffer with the output of .getTicketKeys())?

[ryzokuken]

@silverwind exactly what I had in mind once the tests passed. I guess I'd also have to add a "Changes" entry to the docs.

[ryzokuken]

Tests pass. Let me add the docs for this.

[DiegoTUI]

Other than the docs, I think it also makes sense to add the methods getTicketKeys() and setTicketKeys(keys) to SecureContext. tls.Server has them and it ends up getting the keys from the context. Both methods are pretty convenient for sharing keys among contexts.

[ryzokuken]

@DiegoTUI that does sound like a logical thing to do, thanks. I will add those today.

If I had any doubts regarding the change, this makes it certain that it'll indeed be a semver-minor.

[bnoordhuis]

Needs documentation and tests. It's not useful to expose this without a documented way to:

1. get the ticket keys
2. configure the session timeout

See tls.Server#getTicketKeys() and the sessionTimeout option to tls.createServer().

[bnoordhuis]

Superfluous comment.

[ryzokuken]

Will remove it, in that case.

[bnoordhuis]

Can you add a trailing comma?

[ryzokuken]

I certainly will! I didn't add one because there wasn't one to begin with 😅

[silverwind]

Noticed that some documentation links to the ticketKeys option of createServe in tls.md, those references have to be updated.

[ryzokuken]

Helpful checklist:

• Remove superfluous comment
• Add a trailing comma
• getTicketKeys
  • function
  • docs
  • tests
• setTicketKeys
  • function
  • docs
  • tests
• sessionTimeout
  • function
  • docs
  • tests
• overall docs
• overall tests
• changes block

[ryzokuken]

@bnoordhuis@silverwind No docs or tests still, but added the required functions and option, so PTAL

[bnoordhuis]

I realize you copied this from elsewhere but this method doesn't take any arguments, it returns a new buffer.

[ryzokuken]

@bnoordhuis quite right, I actually took my cues from https://github.com/ryzokuken/node/blob/6269853a9a444a3fbed351ddeb12a6e97114543d/lib/_tls_wrap.js#L928

[ryzokuken]

I'd look into the original context's getTicketKeys() method and try working with that one.

[ryzokuken]

Umm, it can take arguments? Idk, the code was:

voidSecureContext::GetTicketKeys(const FunctionCallbackInfo<Value>& args) {
#if !defined(OPENSSL_NO_TLSEXT) && defined(SSL_CTX_get_tlsext_ticket_keys)
SecureContext* wrap;
ASSIGN_OR_RETURN_UNWRAP(&wrap, args.Holder());
Local<Object> buff = Buffer::New(wrap->env(), 48).ToLocalChecked();
memcpy(Buffer::Data(buff), wrap->ticket_key_name_, 16);
memcpy(Buffer::Data(buff) + 16, wrap->ticket_key_hmac_, 16);
memcpy(Buffer::Data(buff) + 32, wrap->ticket_key_aes_, 16);
args.GetReturnValue().Set(buff);
#endif// !def(OPENSSL_NO_TLSEXT) && def(SSL_CTX_get_tlsext_ticket_keys)
}

[ryzokuken]

Anyway, I think most if not all people call it without any args plus no arguments mentioned in the docs, so I'll make the change.

[bnoordhuis]

You mean args? Any C++ API function takes that as its argument.

If you're referring to args.Holder(), that's the this object. Positional arguments are accessed using array notation, i.e., args[0].

[ryzokuken]

@bnoordhuis TIL, I guess 😅 Not super good at the C++ part of things.

[ryzokuken]

In that case, getTicketKeys indeed doesn't use any arguments while setTicketKeys takes one, perfect.

[ryzokuken]

@bnoordhuis added a few tests, they're running locally. PTAL and let me know if I missed anything.

[ryzokuken]

I needed deepStrictEqual instead! Made the change, will push alongside others, please ignore that part.

[ryzokuken]

@bnoordhuis I have added the sessionTimeout option, but cannot decide how to test it. I could either add a getter for it and assert the value from the getter with the one I enter at the constructor or do something closer to the real world but a tad hacky like what we do at test-tls-session-timeout.js.

test-tls-session-timeout.js

'use strict';constcommon=require('../common');if(!common.opensslCli)common.skip('node compiled without OpenSSL CLI.');if(!common.hasCrypto)common.skip('missing crypto');consttmpdir=require('../common/tmpdir');doTest();// This test consists of three TLS requests --// * The first one should result in a new connection because we don't have// a valid session ticket.// * The second one should result in connection resumption because we used// the session ticket we saved from the first connection.// * The third one should result in a new connection because the ticket// that we used has expired by now.functiondoTest(){constassert=require('assert');consttls=require('tls');constfs=require('fs');constjoin=require('path').join;constfixtures=require('../common/fixtures');constspawn=require('child_process').spawn;constSESSION_TIMEOUT=1;constkey=fixtures.path('agent.key');constcert=fixtures.path('agent.crt');constoptions={key: key,cert: cert,ca: [cert],sessionTimeout: SESSION_TIMEOUT};// We need to store a sample session ticket in the fixtures directory because// `s_client` behaves incorrectly if we do not pass in both the `-sess_in`// and the `-sess_out` flags, and the `-sess_in` argument must point to a// file containing a proper serialization of a session ticket.// To avoid a source control diff, we copy the ticket to a temporary file.constsessionFileName=(function(){constticketFileName='tls-session-ticket.txt';consttmpPath=join(tmpdir.path,ticketFileName);fs.writeFileSync(tmpPath,fixtures.readSync(ticketFileName));returntmpPath;}());// Expects a callback -- cb(connectionType : enum ['New'|'Reused'])functionClient(cb){constflags=['s_client','-connect',`localhost:${common.PORT}`,'-sess_in',sessionFileName,'-sess_out',sessionFileName];constclient=spawn(common.opensslCli,flags,{stdio: ['ignore','pipe','ignore']});letclientOutput='';client.stdout.on('data',function(data){clientOutput+=data.toString();});client.on('exit',function(code){letconnectionType;constgrepConnectionType=(line)=>{constmatches=line.match(/(New|Reused),/);if(matches){connectionType=matches[1];returntrue;}};constlines=clientOutput.split('\n');if(!lines.some(grepConnectionType)){thrownewError('unexpected output from openssl client');}cb(connectionType);});}constserver=tls.createServer(options,function(cleartext){cleartext.on('error',function(er){if(er.code!=='ECONNRESET')thrower;});cleartext.end();});server.listen(common.PORT,function(){Client(function(connectionType){assert.strictEqual(connectionType,'New');Client(function(connectionType){assert.strictEqual(connectionType,'Reused');setTimeout(function(){Client(function(connectionType){assert.strictEqual(connectionType,'New');server.close();});},(SESSION_TIMEOUT+1)*1000);});});});}

[bnoordhuis]

You could test sessionTimeout by setting it to 1s, creating a ticket, then checking that it's gone after one second.

[ryzokuken]

@bnoordhuis I assigned the sessionTimeout property in a way similar to how we did it in tls.Server, but it probably doesn't work. Pushing the failing test (or passing, in this case, but it shouldn't) for inspection.

[ryzokuken]

Built Node in debug mode and checked it out myself, SSL_CTX_set_timeout is infact being called. I wonder what else is wrong?

Breakpoint and Backtrace

~/Code/nodejs/node crypto-20908*
❯ lldb -- ./out/Debug/node test/parallel/test-tls-securecontext-ticketkeys.js(lldb) target create "./out/Debug/node"Current executable set to './out/Debug/node' (x86_64).(lldb) settings set -- target.run-args "test/parallel/test-tls-securecontext-ticketkeys.js"(lldb) breakpoint set --name SSL_CTX_set_timeoutBreakpoint 1: where = node`SSL_CTX_set_timeout + 12 at ssl_sess.c:916, address = 0x0000000101a910bc(lldb) rProcess 56423 launched: './out/Debug/node' (x86_64)Process 56423 stopped* thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.1 frame #0: 0x0000000101a910bc node`SSL_CTX_set_timeout(s=0x00000001053064b0, t=1) at ssl_sess.c:916 913 long SSL_CTX_set_timeout(SSL_CTX *s, long t) 914 { 915 long l;-> 916 if (s == NULL) 917 return (0); 918 l = s->session_timeout; 919 s->session_timeout = t;Target 0: (node) stopped.(lldb) bt* thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.1 * frame #0: 0x0000000101a910bc node`SSL_CTX_set_timeout(s=0x00000001053064b0, t=1) at ssl_sess.c:916 frame #1: 0x00000001002b5d44 node`node::crypto::SecureContext::SetSessionTimeout(args=0x00007ffeefbfcf70) at node_crypto.cc:1025 frame #2: 0x0000000100607cb3 node`v8::internal::FunctionCallbackArguments::Call(this=0x00007ffeefbfd0d0, handler=0x00003969690bdd41) at api-arguments.cc:29 frame #3: 0x00000001007283ee node`v8::internal::MaybeHandle<v8::internal::Object> v8::internal::(anonymous namespace)::HandleApiCallHelper<false>(isolate=0x0000000105804c00, function=Handle<v8::internal::HeapObject> @ 0x00007ffeefbfd1e0, new_target=Handle<v8::internal::HeapObject> @ 0x00007ffeefbfd1d8, fun_data=Handle<v8::internal::FunctionTemplateInfo> @ 0x00007ffeefbfd1d0, receiver=Handle<v8::internal::Object> @ 0x00007ffeefbfd1c8, args=BuiltinArguments @ 0x00007ffeefbfd200) at builtins-api.cc:108 frame #4: 0x0000000100726d0a node`v8::internal::Builtin_Impl_HandleApiCall(args=BuiltinArguments @ 0x00007ffeefbfd330, isolate=0x0000000105804c00) at builtins-api.cc:138 frame #5: 0x000000010072683d node`v8::internal::Builtin_HandleApiCall(args_length=6, args_object=0x00007ffeefbfd410, isolate=0x0000000105804c00) at builtins-api.cc:126 frame #6: 0x00001f0f67a042c4 frame #7: 0x00001f0f67a1cf78 frame #8: 0x00001f0f67a108e3 frame #9: 0x00001f0f67a1cf78 frame #10: 0x00001f0f67a1cf78 frame #11: 0x00001f0f67a1cf78 frame #12: 0x00001f0f67a1cf78 frame #13: 0x00001f0f67a1cf78 frame #14: 0x00001f0f67a1cf78 frame #15: 0x00001f0f67a1cf78 frame #16: 0x00001f0f67a1cf78 frame #17: 0x00001f0f67a1cf78 frame #18: 0x00001f0f67a16e15 frame #19: 0x00001f0f67a0b581 frame #20: 0x0000000100fd0efd node`v8::internal::GeneratedCode<v8::internal::Object*, v8::internal::Object*, v8::internal::Object*, v8::internal::Object*, int, v8::internal::Object***>::Call(this=0x00007ffeefbfdd50, args=0x00003969a12822e1, args=0x000039699cc0e4d9, args=0x00003969a1282201, args=2, args=0x00007ffeefbfe3e0) at simulator.h:110 frame #21: 0x0000000100fcf182 node`v8::internal::(anonymous namespace)::Invoke(isolate=0x0000000105804c00, is_construct=false, target=Handle<v8::internal::Object> @ 0x00007ffeefbfde70, receiver=Handle<v8::internal::Object> @ 0x00007ffeefbfde68, argc=2, args=0x00007ffeefbfe3e0, new_target=Handle<v8::internal::Object> @ 0x00007ffeefbfde60, message_handling=kReport, execution_target=kCallable) at execution.cc:153 frame #22: 0x0000000100fcea03 node`v8::internal::(anonymous namespace)::CallInternal(isolate=0x0000000105804c00, callable=Handle<v8::internal::Object> @ 0x00007ffeefbfdf40, receiver=Handle<v8::internal::Object> @ 0x00007ffeefbfdf38, argc=2, argv=0x00007ffeefbfe3e0, message_handling=kReport, target=kCallable) at execution.cc:189 frame #23: 0x0000000100fce89d node`v8::internal::Execution::Call(isolate=0x0000000105804c00, callable=Handle<v8::internal::Object> @ 0x00007ffeefbfdfa0, receiver=Handle<v8::internal::Object> @ 0x00007ffeefbfdf98, argc=2, argv=0x00007ffeefbfe3e0) at execution.cc:200 frame #24: 0x00000001005b0f48 node`v8::Function::Call(this=0x000000010703ea68, context=(val_ = 0x00000001058216e0), recv=(val_ = 0x0000000105804c78), argc=2, argv=0x00007ffeefbfe3e0) at api.cc:5077 frame #25: 0x000000010008a872 node`node::ExecuteBootstrapper(env=0x00007ffeefbfe4f0, bootstrapper=(val_ = 0x000000010703ea68), argc=2, argv=0x00007ffeefbfe3e0, out=0x00007ffeefbfe258) at node.cc:3067 frame #26: 0x0000000100089a0c node`node::LoadEnvironment(env=0x00007ffeefbfe4f0) at node.cc:3167 frame #27: 0x000000010009460e node`node::Start(isolate=0x0000000105804c00, isolate_data=0x0000000107040400, argc=2, argv=0x0000000105200050, exec_argc=0, exec_argv=0x00000001052001f0) at node.cc:4256 frame #28: 0x000000010008f19b node`node::Start(event_loop=0x0000000102da7a40, argc=2, argv=0x0000000105200050, exec_argc=0, exec_argv=0x00000001052001f0) at node.cc:4363 frame #29: 0x000000010008e7c5 node`node::Start(argc=2, argv=0x0000000105200050) at node.cc:4414 frame #30: 0x0000000101cb912e node`main(argc=2, argv=0x00007ffeefbff170) at node_main.cc:124 frame #31: 0x0000000100001034 node`start + 52(lldb) cProcess 56423 resuming<Buffer 38 67 52 0b 46 dc 16 f1 df df cf 62 81 f8 a0 3b ed 45 27 75 d4 f9 9e 00 2d 5b fc 1d 8a 3e 60 20 eb ae ea 7f 31 45 0d f8 24 f6 4a 48 6d e1 ca 04> <Buffer 38 67 52 0b 46 dc 16 f1 df df cf 62 81 f8 a0 3b ed 45 27 75 d4 f9 9e 00 2d 5b fc 1d 8a 3e 60 20 eb ae ea 7f 31 45 0d f8 24 f6 4a 48 6d e1 ca 04>Process 56423 exited with status = 0 (0x00000000)(lldb) exit

[ryzokuken]

@nodejs/crypto @bnoordhuis PTAL.

[ryzokuken]

@nodejs/crypto anyone willing to give this a shot with me during the summit?

P.S. giving this a shot == helping me debug this, I'm clueless.

[ryzokuken]

ping @nodejs/crypto please help?

[tniessen]

Sorry, didn't have time to look at this yet, and I am not that confident with the TLS APIs of OpenSSL. Are you sure that the test case is correct?

Edit: Also, I believe the subsystem should be tls, not crypto.

[ryzokuken]

@tniessen that's what I'm afraid of. I'm so bad with the specifics of this that I don't know if the test case is correct. Let me try looking deeper too.

[jasnell]

Ping @ryzokuken

[Trott]

@ryzokuken Still working on this?

[BridgeAR]

Closing due to long inactivity. @ryzokuken please reopen / open a new PR in case you want to continue working on this.

[ryzokuken]

I'm sorry I put this in the backseat, @jasnell@Trott@BridgeAR. Hopefully will publish a fix later.