[jira] [Commented] (MESOS-5320) SSL related error messages can be misguiding or incomplete

["Till Toenshoff (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-5320?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15268645#comment-15268645 ] 

Till Toenshoff commented on MESOS-5320:
---------------------------------------

For re-creating my faulty setup and hence the error mentioned, the following might be useful:

{noformat}
openssl req -new -x509 -nodes -newkey rsa:4096 -days 3650 -subj "/C=US/ST=CA/L=San Francisco/O=Acme Inc./CN=Acme Secure Certificate Authority" -keyout ca.key -out ca.crt
openssl genrsa -out snakeoil.key 2048
openssl req -new -key snakeoil.key -out snakeoil.csr -subj /CN=*/
openssl x509 -req -in snakeoil.csr -out snakeoil.crt -signkey ca.key -days 1095
chmod 600 ca.key snakeoil.key
{noformat}

Then running the Mesos master:
{noformat}
$ export SSL_ENABLED=true
$ export SSL_SUPPORT_DOWNGRADE=true
$ export SSL_VERIFY_CERT=false
$ export SSL_REQUIRE_CERT=false
$ export SSL_CA_FILE=/home/till/scratchpad/mesos/build/ca.crt
$ export SSL_CERT_FILE=/home/till/scratchpad/mesos/build/snakeoil.crt
$ export SSL_KEY_FILE=/home/till/scratchpad/mesos/build/snakeoil.key
$ ./bin/mesos-master.sh
{noformat}

> SSL related error messages can be misguiding or incomplete
> ----------------------------------------------------------
>
>                 Key: MESOS-5320
>                 URL: https://issues.apache.org/jira/browse/MESOS-5320
>             Project: Mesos
>          Issue Type: Bug
>    Affects Versions: 0.29.0
>            Reporter: Till Toenshoff
>              Labels: ssl
>
> I was trying to activate SSL within Mesos but had rendered an invalid certificate, it was signed with a mismatching key. Once I started the master, the error message I received was rather confusing to me:
> {noformat}
> W0503 10:15:58.027343  6696 openssl.cpp:363] Failed SSL connections will be downgraded to a non-SSL socket
> Could not load key file
> {noformat} 
> To me, this error message hinted that the key file was not existing or had rights issues. However, a quick {{strace}} revealed  that the key-file was properly accessed, no sign of a file-not-found or alike.
> The problem here is the hardcoded error-message, not taking OpenSSL's human readable error strings into account.
> The code that misguided me is located at  https://github.com/apache/mesos/blob/master/3rdparty/libprocess/src/openssl.cpp#L471
> We might want to change
> {noformat}
>   // Set private key.
>   if (SSL_CTX_use_PrivateKey_file(
>           ctx,
>           ssl_flags->key_file.get().c_str(),
>           SSL_FILETYPE_PEM) != 1) {
>     EXIT(EXIT_FAILURE) << "Could not load key file";
>   }
> {noformat}
> Towards something like this
> {noformat}
>   // Set private key.
>   if (SSL_CTX_use_PrivateKey_file(
>           ctx,
>           ssl_flags->key_file.get().c_str(),
>           SSL_FILETYPE_PEM) != 1) {
>     EXIT(EXIT_FAILURE) << "Could not use key file: " << ERR_error_string(ERR_get_error(), NULL);
>   }
> {noformat}
> To receive a much more helpful message like this
> {noformat}
> W0503 13:18:12.551364 11572 openssl.cpp:363] Failed SSL connections will be downgraded to a non-SSL socket
> Could not use key file: error:0B080074:x509 certificate routines:X509_check_private_key:key values mismatch
> {noformat}
> A quick scan of the implementation within {{openssl.cpp}} to me suggests that there are more places that we might want to update with more deterministic error messages. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)