My evoluion mail (opensuse tumbleweed) complain about the mail server's certificate. It looks like gnutls reject the certificate. Can anyone give me some pointers to the problem? Thanks
Information for package gnutls:
Repository : openSUSE-20181224-0
Name : gnutls
Version : 3.7.1-1.1 Arch :
x86_64 Vendor : openSUSE
While trying to setup the exchange email from scratch with gnome-shell online-account, the server could not be contacted with Error Connecting to Microsoft Exchange server:
Code:6 -- Unexpected response from server
Searching for the cause, I found this solution 1, basically lower the
system-wide security level or accepting shorter key or old algorithm:
update-crypto-policies --set LEGACY
1 [The Diffie-Hellman prime sent by the server is not acceptable]
Conclusion from my observation, the certificate may contain legacy encription algorithms or short keys that are no longer secure for newer TLS version.
Related
The website is on Windows server 2019 with the AWS Load Balancer with ELB SecurityPolicy-2016-08. This policy definitely has the ECDHE_ECDSA cipher enabled. I have checked their docs. SSL certificate is installed on LB.
Running TLS Cipher Suites in PowerShell Windows server 2019 also shows these suits enabled but when running the website domain with SSLLabs or Zenmap. These suites are not appearing
TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384
TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256
or even these:
TLS_DHE_RSA_WITH_AES_256_GCM_SHA384
TLS_DHE_RSA_WITH_AES_128_GCM_SHA256
any ideas? the website is ASP.NetFramework 4.7. but I hardly think it has anything to do with the ciphers. Any help will be appreciated. Thanks
Zenmap Snapshot
AWS load balancer Snapshot
PowerShell Snapshot
Meta: this isn't about programming, and I'm not sure 'how to operate cloud' counts as development, so I authorize deletion if this is voted offtopic.
Your server is irrelevant and nothing you set or change on it will affect client(s).
You don't tell us which AWS load balancer you use but to be at HTTPS level it must be Application or Classic, and in either case to do HTTPS it must terminate the SSL/TLS protocol -- in other words, the LB establishes one SSL/TLS connection with the client and decrypts the incoming request, parses it, and then optionally uses a separate SSL/TLS connection to the backend to re-encrypt, and reverses the process on the response: decrypt from the backend if necessary and re-encrypt to the client. See the line "SSL Offloading" well down in the table on that page; that's a jargon way of saying "LB does the SSL/TLS for the client, your server does not".
Thus the settings in the LB, only, control the SSL/TLS seen by the client(s). ELBSecurityPolicy-2016-08 which is the default (and I'm guessing that might be why you used it) excludes all DHE-RSA ciphersuites. (To avoid confusion, note the AWS webpage uses the OpenSSL names for ciphersuites, where RSA-only keyexchange is omitted from the name, whereas Zenmap/nmap uses the RFC names TLS_RSA_with_whatever.) It does allow ECDHE_ECDSA suites, but those will actually be negotiated, and thus seen by a scanner like Zenmap/nmap, only if you configure an ECDSA certificate and key -- which I bet you didn't.
The TL;DR version
I'd like to know:
Where does the specification for the use of ECDHE get defined (in a cert parameter or a server configuration of SSL contexts, or elsewhere)?
In a non-home-rolled certificate setup, who's responsibility is it to define the ECDHE public and private information (the end user or cert provider)?
Can an existing Certificate which does not appear to use ECDHE be made to without causing issues with the Certificate?
Are there any examples of someone using SSL in Boost::ASIO with an ECDHE setup?
The Longer Version
We've been building an application which is using a proper-paid-for certificate from an external Cert Authority. The application uses a home-rolled server setup based off of Boost ASIO and Boost Beast, and we only recently noticed it doesn't play nice with iOS - ASIO says there is no shared cipher.
Reading into how TLS works has led me to the fact that some part of our server was preventing us from serving TLS using the ECDHE-* suite of ciphers (which iOS seems to want) - but I'm having difficulty in figuring out how to wrangle ASIO and our current cert/key into serving ECDHE.
What I've tried:
Using the same cert and key, adding in the results of openssl dhparam into ASIO using set_tmp_dh, then specifying ciphers. Curl reports that this allows a connection using DHE but not ECDHE. Specifying ciphers that only use ECDHE causes errors when connecting.
Trying to pass the output of openssl ecparam to ASIO using a similar method to the above. I've not been able to format something that ASIO accepts.
Trying to see if there is a way you can use the output of openssl ecparam with another combining function to modify the original cert into one that uses ECDHE. I clued onto this one from the OpenSSL wiki suggesting that if the cert does not contain the line ASN1 OID: prime256v1 (or a similar named curve), then it is not suitable for ECDHE usage.
At this point I'm unsure as to where the issue truly lies (in ASIO, in the certificates or in how I'm putting it all together) and most of the information on the internet I can find relates to home-rolling everything from scratch, rather than working with existing certs.
Update 11/05/19
https://github.com/chriskohlhoff/asio/pull/117 pulled in changes for ASIO with ECDHE. Will need to wait a while to see which Boost lib version it makes it into.
Original Answer
I seem to have found an answer for any googlers - ASIO does not appear to support ECDHE natively at the time of writing. This issue from the main repo suggests that ECDHE is on the cards for support but is not yet implemented.
Here is a link to the ECDHE implementation that's been waiting to be merged since 2016: https://github.com/chriskohlhoff/asio/pull/117.
+1 to get the attention of the Boost ASIO maintainer; he's been pretty slow with it.
I'm developing application in C++ (cross-platform; Windows, Mac and Linux) that needs to communicate securely with servers using https protocol with libcurl (built with winssl/darwinssl/openssl on Windows/Mac/Linux respectively). I've changed a curl option, CURLOPT_SSL_VERIFYPEER from 0 to 1 which should help prevent MitM issues.
This has caused issues that an initial search points to turning that option off, but after digging deeper I found:
Get a CA certificate that can verify the remote server and use the proper option to point out this CA cert for verification when connecting. For libcurl hackers: curl_easy_setopt(curl, CURLOPT_CAPATH, capath); from curl docs
and
Get a better/different/newer CA cert bundle! One option is to extract the one a recent Firefox browser uses by running 'make ca-bundle' in the curl build tree root, or possibly download a version that was generated this way for you.
from curl docs
I actually use CURLOPT_CAINFO to the bundle as I had seen some word of issues using CURLOPT_CAPATH on Windows; curl docs. I have downloaded and installed this bundle along with the application on Windows and Mac and I'd like to know if this is the correct way to do it or if there is a better practice.
Initially this caused issues for users of the application running behind some corporate networks or proxy which seemed to get fixed by building libcurl against winssl instead of openssl on Windows; though potentially disguising itself as a firewall issue, still unclear although it seems likely.
Sorry for the length.
Is anything silly about installing the ca-cert-bundle.crt along with the application, and is there anything that should be done differently to communicate securely with the server from this installed application?
A slightly separate, but still very related, issue I have is CURLOPT_CAINFO on Linux giving the error:
error setting certificate verify locations:
CAfile: ../share/my_application/curl-ca-bundle.crt
CApath: none
Though attempting to open the file for reading from within the application does work successfully. Edit: This issue I solved by NOT setting the CURLOPT_CAINFO field on Linux (leaving it blank) and adding the dependency package ca-certificates to the application package. The default path is correctly /etc/ssl/certs/ca-certificates.crt and seems to be working. To me this feels a bit better than installing the bundle with the application.
Edit2: Although solved it appears the ca-certificates package sometimes doesn't install ca-certificates.crt and instead ca-bundle.crt and the locations vary on different distros as this source, happyassassin.net shows that different Linux systems store the CA bundles in different locations. It did not seem to have a clear answer as to HOW to handle this. Should I be using a value in the configuration file that the user can then modify, or any other thoughts on the subject?
Edit3: Some users have pointed out that my name exists in one of the paths curl looks for, I'm not entirely sure how that is possible as the only thing I've specified for curl is where I built openssl/cares libraries...
I realize this is a loaded/multipart question but it is all on the same subject as the title states, I'd appreciate any help.
Thanks.
In my opinion, it is better to use system certificate then package certificates with application (if you are not using some special certs). For the linux it should be easy according to https://serverfault.com/questions/394815/how-to-update-curl-ca-bundle-on-redhat And for windows you can either use winssl or create the file from system https://superuser.com/questions/442793/why-cant-curl-properly-verify-a-certificate-on-windows Configure cURL to use default system cert store
A default libcurl build is setup to attempt to use the "right" CA bundle.
Linux
A libcurl built on Linux will scan and check where the CA store is located on your system and use that. If you install libcurl on a regular Linux distro, it should've been built to use the distro's "typical" CA store.
macOS
If you build libcurl for mac and tell it to use the Secure Transport backend, it will automatically use the macOS CA store. So will the default-installed curl and libcurls that come shipped bundled with macOS from Apple.
Windows
If you build libcurl for Windows to use Schannel (the windows TLS system) it will by default use the Windows CA store.
Other setups
If you deviate from these setups, you basically opt to not use the CA store that comes bundled in the operating system you're using. Then you need to handle and update the CA store yourself.
I developed a SSLProxy as a man in the middle between client and server. Handshaking between client-proxy and proxy-server doing well. I receive a message from client and decrypt it with client_side SSL. Then encrypt it with server_side SSL. All thing is good except on thing: OpenSSL received all the message data in one SSL record but it sent them in 2 SSL records.
Question: How can I force OpenSSL to send data in 1 SSL record, because server configured only to use 1 SSL record?
Wireshark Screen:
192.168.2.127 is client.
192.168.0.230 is server.
Update: I need something like this. I tried to use them but I faced with this error:
error: ‘SSL_CTX_set_split_send_fragment’ was not declared in this scope
With SQL Server 2008 R2 SP3 and update KB3144114 to support TLSv1.2 my problem solved. But In SQL Server 2008 R2 (with TLSv1) problem still remain same as before and I couldn't find any solution for it.
You must use OpenSSL 1.1.X series for using "SSL_CTX_set_split_send_fragment function", that function is not available in older versions.
I am searching for a client TLS connection example in C++. Best for Visual Studio, but honestly it can be any compiler. I found several C samples. But no one worked. I started with this sample in C:
https://wiki.openssl.org/index.php/SSL/TLS_Client
But it failes on
res = BIO_do_connect(web);
with "system library" if I want to connect to my own node.js server (using the direct ip address) or with "bad hostname lookup" using encrypted.google.com as url.
Both with libressl and Visual Studio 2013.
Next stop: http://fm4dd.com/openssl/sslconnect.htm
Here the program runs successful. But any attempt to write to the SSL connection at the end with:
std::string json = "{'test':'huhu'}";
char buff[1024];
sprintf(buff, "POST /test.de HTTP/1.1 \nHost: test.de\nContent-Type: application/json\nContent-Length: %d\n\n", json.length());
std::string post = buff;
int snd = SSL_write(ssl, post.data(), post.length());
snd = SSL_write(ssl, json.data(), json.length());
forces the server to close the connection (I do not see exactly what happend as I do not now how I can tell node.js to tell me more).
So I search for a working sample or how to get a TLS connection with own certificate running in C++
I am searching for a client TLS connection example in C++.
I think there are a couple of ports of OpenSSL to C++. They try to do the full class wrapper thing. See openssl++ class on Google.
When I use it in C++, I use unique pointers for cleanup. See, for example, How to properly print RSA* as string in C++?. I use it primarily to ensure cleanup. I think its similar to Resource Acquisition Is Initialization pattern.
OpenSSL also provides a page for similar libraries and frameworks. See the Related Links page on the OpenSSL wiki.
But it fails on
res = BIO_do_connect(web);
with "system library" if I want to connect to my own node.js server (using the > direct ip address) or with "bad hostname lookup"
My guess here would be the name in the certificate does not match the name used in the URL to connect.
You can make the names work by adding an entry in your host file. Effectively, this is your local DNS override. See Microsoft TCP/IP Host Name Resolution Order.
Or, you can generate a certificate with all the required names. For that, see How to create a self-signed certificate with openssl?
forces the server to close the connection (I do not see exactly what happend as I do not now how I can tell node.js to tell me more).
"POST /test.de HTTP/1.1 \nHost: test.de\nContent-Type:
application/json\nContent-Length: %d\n\n"
Since you lack the Connection: close request header, the server is probably following RFC 7230, HTTP/1.1 Message Syntax and Routing, Section 6.1:
A server that does not support persistent connections MUST send the
"close" connection option in every response message that does not
have a 1xx (Informational) status code.
Also, that should probably be:
"POST /test.de HTTP/1.1\r\nHost: test.de\r\nContent-Type:
application/json\r\nContent-Length:%d\r\n\r\n"
\r\n is used as new line, not \r and not \n. A double \r\n is used to terminate the header. You can quickly verify be searching for "CRLF" in the standard. You will land in a discussion of the ABNF grammar.
So I search for a working sample or how to get a TLS connection with own certificate running in C++
The trick here is creating a well-formed certificate. For that, see How to create a self-signed certificate with openssl?
Here's an updated example for LibreSSL using pinned cert bundle: C++ libtls example on github