multissl: curl_global_cleanup doesn't restore Curl_ssl as its default (Curl_ssl_multi)

[davidedec]

I noticed that after a curl_global_cleanup it's not possible to curl_global_sslset and curl_global_init again in order to change the ssl backend.

This can be achieved in vtls.c by setting back the Curl_ssl = &Curl_ssl_multi after the Curl_ssl_cleanup

The idea is to be able to change the backend in this sequence:

curl_global_sslset(CURLSSLBACKEND_SCHANNEL, NULL, NULL);
curl_global_init(CURL_GLOBAL_DEFAULT);
//....
curl_global_cleanup();

curl_global_sslset(CURLSSLBACKEND_OPENSSL, NULL, NULL);
curl_global_init(CURL_GLOBAL_DEFAULT);
//....
curl_global_cleanup();

Thanks

[jay]

The documentation says it can only be set once, but I would think it could be set after cleanup, in other words once each time it's uninitialized. @dscho

[dscho]

Makes sense.

I think it should be as easy as adding the lines

#ifdef CURL_SSL_MULTI
    Curl_ssl = &Curl_ssl_multi;
#endif

to

curl/lib/vtls/vtls.c

Lines 179 to 186 in 9c703ea

	void Curl_ssl_cleanup(void)
	{
	if(init_ssl) {
	/* only cleanup if we did a previous init */
	Curl_ssl->cleanup();
	init_ssl = FALSE;
	}
	}

@davidedec could you test that?

[davidedec]

@dscho thanks for your prompt reply.
I'll test that tomorrow and I'll come back to you.

Thanks again

[dscho]

Actually, I think those three lines should go into Curl_none_cleanup(). Consequently, that function should be renamed to Curl_multissl_cleanup(), and it should be moved directly after Curl_multissl_init().

[dscho]

Ah! And of course it should call Curl_ssl->cleanup(), unless Curl_ssl still points to Curl_ssl_multi.

[davidedec]

Right. Tomorrow I will tell you if it works and pass tests

[jay]

If it works can one of you turn it into a PR for review please

[dscho]

I actually pulled out the computer and am doing precisely that.

[dscho]

Phew, that took longer than I hoped for. I'm glad I tried to verify that my patch works, as my initial version (the one modifying Curl_none_cleanup()) totally did not work, like, at all.

@davidedec would you mind testing #5257?

[davidedec]

Just tested #5257 and worked as expected for me