Repository navigation
Replies: 3 comments
|
it's good that encription only exist on pro server. In normal server People like me who doesn't like and doesn't need this encription feature can entirely not have it. |
|
You're right that the standard server doesn't build with encryption. The clients are a different matter though. That's the group this affects. Not people like you who don't use encryption, but people who specifically chose it, by running or connecting to a Pro server. They get their traffic encrypted and reasonably assume that means they're talking to the server they think they are. At the moment they aren't, because So the request isn't to add encryption anywhere it isn't already, or to turn it on for anyone. It's that where it does exist, it should authenticate the server rather than just encrypt to whoever answers. |
|
In client and server you can configure whether to verify peers. If you want something other than default then change the configuration. The default configuration is designed to make it easy for new users. If you want verification in both directions then it's not easy for users to set up. Quite the contrary. You can find more information on how to configure client and server here: https://www.bearware.dk/teamtalksdk/v5.22a/docs/C-API/serversetup.html#ttsrvencrypt |
Uh oh!
There was an error while loading. Please reload this page.
Description
Encrypted connections currently give confidentiality but not authentication. A client has no way to tell whether it is talking to the intended server, so an attacker positioned to intercept the connection can terminate the TLS session, present any certificate, and relay traffic while reading logins, private messages and voice.
This is not an argument for turning encryption off. Unauthenticated TLS still defeats passive capture on shared wifi or at an ISP, which is the common case, and removing it would expose everything to anyone running
tcpdump. The point is that the protection stops short of what the UI leads people to expect.Two separate gaps, both verified in the current tree.
1. Peer verification is off by default, in every client
Client/qtTeamTalk/common.h:80,bool verifypeer = false;Client/TeamTalkAndroid/src/main/java/dk/bearware/data/ServerEntry.java:65,public boolean verifypeer = false;Library/TeamTalkLib/bin/ttsrv/ServerConfig.cpp:896,GetCertificateVerify(false)So a default client accepts whatever certificate is offered.
Turning it on is also deliberately awkward.
Client/qtTeamTalk/encryptionsetupdlg.cpp:223only enables the checkbox once a CA certificate has been pasted into that specific server entry, and nothing consults the operating system trust store. An operator who does the right thing and puts a real certificate from a public CA on a real hostname gets no benefit from it, because there is no configuration that says "trust the normal CAs".2. Hostnames are never verified, even with
bVerifyPeerenabledLibrary/TeamTalkLib/bin/dll/Convert.cpp:1804passes the flag to ACE:That is chain validation. OpenSSL does not check the name the client asked for unless the application sets it, through
SSL_set1_hostorX509_VERIFY_PARAM_set1_host.Searching
Library/TeamTalkLibforX509,SSL_set1_host,X509_check_host,SSL_get_verify_result,get_peer_certificateandpeer_certreturns nothing. The library never inspects the peer certificate at all.The consequence is that a certificate issued by the configured CA for some other host is accepted. Enabling verification therefore delivers less than the option name suggests.
Suggestion
Ordered by what seems like the best return for the effort. Happy to work on whichever of these is wanted, or none, if the direction is wrong.
Trust on first use. Record the server's certificate fingerprint on first connect and warn clearly if it changes later. This fits TeamTalk better than anything CA based, because most servers are self signed and always will be. It is the model SSH uses for the same problem, it needs no infrastructure from operators, and it breaks nothing that works today. It turns a silent interception into a visible one.
Hostname verification when
bVerifyPeeris set. Contained change inSetupEncryptionContext, and it makes the existing option mean what users already assume it means.Optional use of the system trust store. Gives operators with a real certificate and a real domain a path that simply works, instead of asking every user to paste a CA certificate by hand.
Say what the connection actually is. Showing the fingerprint, or stating that a connection is encrypted but unauthenticated, would stop the UI implying a guarantee that is not there.
Application
Platform
All reactions