logger: fix file open issue if crypto algorithm is disabled

[jnippula]

The init_logfile_encryption() function is called after the buffer's start_log() is called and file is already opened.

[jlaitine]

This looks a bit weird, is it so that the file might be created in two places?

[jlaitine]

That is, should we really just push the key to the _buffers and let the file creation be only there?

[jnippula]

Right, it was a bit confusing where and when the file is opened and how. Original idea was to create file already in init_logfile_encryption() even there is no key data (no_crypto) and then start_log would any case open file as append.

Anyway, I made it now a bit different way: init_logfile_encryption() now gets keydata/keydatasize as param, allocate memory and store keydata there. Later the LogFileBuffer::start_log() gets the keydata buffer and if any data exists, it stores the keydata into the beginning of the recently opened file. How that looks?

[jpaali]

LGTM!

[jnippula]

The proposal to store keydata into _buffers was not working as such, because it caused a "paradox" where the key header along the wrapper key was encrypted with the key itself before written to the file..

Implemented it now so that the init_logfile_encryption() function is called after the buffer's start_log is called and file is opened. Then it can write the header+key to file without needs to open the file first, because it was already opened in start_log. And it is opened in one place only.

Some tunings needed to properly close logging in case the init_logfile_encryption() fails (close the file and set _should_run to false).

[jlaitine]

Thanks, lgtm!