path: root/doc/man7/ossl-guide-tls-introduction.pod

=pod

=head1 NAME

ossl-guide-tls-introduction
- OpenSSL Guide: An introduction to SSL/TLS in OpenSSL

=head1 INTRODUCTION

This page will provide an introduction to some basic SSL/TLS concepts and
background and how it is used within OpenSSL. It assumes that you have a basic
understanding of TCP/IP and sockets.

=head1 WHAT IS TLS?

TLS stands for Transport Layer Security. TLS allows applications to securely
communicate with each other across a network such that the confidentiality of
the information exchanged is protected (i.e. it prevents eavesdroppers from
listening in to the communication). Additionally it protects the integrity of
the information exchanged to prevent an attacker from changing it. Finally it
provides authentication so that one or both parties can be sure that they are
talking to who they think they are talking to and not some imposter.

Sometimes TLS is referred to by its predecessor's name SSL (Secure Sockets
Layer). OpenSSL dates from a time when the SSL name was still in common use and
hence many of the functions and names used by OpenSSL contain the "SSL"
abbreviation. Nonetheless OpenSSL contains a fully fledged TLS implementation.

TLS is based on a client/server model. The application that initiates a
communication is known as the client. The application that responds to a
remotely initiated communication is the server. The term "endpoint" refers to
either of the client or the server in a communication. The term "peer" refers to
the endpoint at the other side of the communication that we are currently
referring to. So if we are currently talking about the client then the peer
would be the server.

TLS is a standardised protocol and there are numerous different implementations
of it. Due to the standards an OpenSSL client or server is able to communicate
seamlessly with an application using some different implementation of TLS. TLS
(and its predecessor SSL) have been around for a significant period of time and
the protocol has undergone various changes over the years. Consequently there
are different versions of the protocol available. TLS includes the ability to
perform version negotiation so that the highest protocol version that the client
and server share in common is used.

TLS acts as a security layer over some lower level transport protocol. Typically
the transport layer will be TCP.

=head1 SSL AND TLS VERSIONS

SSL was initially developed by Netscape Communications and its first publicly
released version was SSLv2 in 1995. Note that SSLv1 was never publicly released.
SSLv3 came along quickly afterwards in 1996. Subsequently development of the
protocol moved to the IETF which released the first version of TLS (TLSv1.0) in
1999 as RFC2246. TLSv1.1 was released in 2006 as RFC4346 and TLSv1.2 came along
in 2008 as RFC5246. The most recent version of the standard is TLSv1.3 which
was released in 2018 as RFC8446.

Today TLSv1.3 and TLSv1.2 are the most commonly deployed versions of the
protocol. The IETF have formally deprecated TLSv1.1 and TLSv1.0, so anything
below TLSv1.2 should be avoided since the older protocol versions are
susceptible to security problems.

OpenSSL does not support SSLv2 (it was removed in OpenSSL 1.1.0). Support for
SSLv3 is available as a compile time option - but it is not built by default.
Support for TLSv1.0, TLSv1.1, TLSv1.2 and TLSv1.3 are all available by default
in a standard build of OpenSSL. However special run-time configuration is
required in order to make TLSv1.0 and TLSv1.1 work successfully.

OpenSSL will always try to negotiate the highest protocol version that it has
been configured to support. In most cases this will mean either TLSv1.3 or
TLSv1.2 is chosen.

=head1 CERTIFICATES

In order for a client to establish a connection to a server it must authenticate
the identify of that server, i.e. it needs to confirm that the server is really
the server that it claims to be and not some imposter. In order to do this the
server will send to the client a digital certificate (also commonly referred to
as an X.509 certificate). The certificate contains various information about the
server including its full DNS hostname. Also within the certificate is the
server's public key. The server operator will have a private key which is
linked to the public key and must not be published.

Along with the certificate the server will also send to the client proof that it
knows the private key associated with the public key in the certificate. It does
this by digitally signing a message to the client using that private key. The
client can verify the signature using the public key from the certificate. If
the signature verifies successfully then the client knows that the server is in
possession of the correct private key.

The certificate that the server sends will also be signed by a Certificate
Authority. The Certificate Authority (commonly known as a CA) is a third party
organisation that is responsible for verifying the information in the server's
certificate (including its DNS hostname). The CA should only sign the
certificate if it has been able to confirm that the server operator does indeed
have control of the server associated with its DNS hostname and that the server
operator has control of the private key.

In this way, if the client trusts the CA that has signed the server's
certificate and it can verify that the server has the right private key then it
can trust that the server truly does represent the DNS hostname given in the
certificate. The client must also verify that the hostname given in the
certificate matches the hostname that it originally sent the request to.

Once all of these checks have been done the client has successfully verified the
identify of the server. OpenSSL can perform all of these checks automatically
but it must be provided with certain information in order to do so, i.e. the set
of CAs that the client trusts as well as the DNS hostname for the server that
this client is trying to connect to.

Note that it is common for certificates to be built up into a chain. For example
a server's certificate may be signed by a key owned by a an intermediate CA.
That intermediate CA also has a certificate containing its public key which is
in turn signed by a key owned by a root CA. The client may only trust the root
CA, but if the server sends both its own certificate and the certificate for the
intermediate CA then the client can still successfully verify the identity of
the server. There is a chain of trust between the root CA and the server.

By default it is only the client that authenticates the server using this
method. However it is also possible to set things up such that the server
additionally authenticates the client. This is known as "client authentication".
In this approach the client will still authenticate the server in the same way,
but the server will request a certificate from the client. The client sends the
server its certificate and the server authenticates it in the same way that the
client does.

=head1 TRUSTED CERTIFICATE STORE

The system described above only works if a chain of trust can be built between
the set of CAs that the endpoint trusts and the certificate that the peer is
using. The endpoint must therefore have a set of certificates for CAs that it
trusts before any communication can take place. OpenSSL itself does not provide
such a set of certificates. Therefore you will need to make sure you have them
before you start if you are going to be verifying certificates (i.e. always if
the endpoint is a client, and only if client authentication is in use for a
server).

Fortunately other organisations do maintain such a set of certificates. If you
have obtained your copy of OpenSSL from an Operating System (OS) vendor (e.g. a
Linux distribution) then normally the set of CA certificates will also be
distributed with that copy.

You can check this by running the OpenSSL command line application like this:

 openssl version -d

This will display a value for B<OPENSSLDIR>. Look in the B<certs> sub directory
of B<OPENSSLDIR> and check its contents. For example if B<OPENSSLDIR> is
"/usr/local/ssl", then check the contents of the "/usr/local/ssl/certs"
directory.

You are expecting to see a list of files, typically with the suffix ".pem" or
".0". If they exist then you already have a suitable trusted certificate store.

If you are running your version of OpenSSL on Windows then OpenSSL (from version
3.2 onwards) will use the default Windows set of trusted CAs.

If you have built your version of OpenSSL from source, or obtained it from some
other location and it does not have a set of trusted CA certificates then you
will have to obtain them yourself. One such source is the Curl project. See the
page L<https://curl.se/docs/caextract.html> where you can download trusted
certificates in a single file. Rename the file to "cert.pem" and store it
directly in B<OPENSSLDIR>. For example if B<OPENSSLDIR> is "/usr/local/ssl",
then save it as "/usr/local/ssl/cert.pem".

You can also use environment variables to override the default location that
OpenSSL will look for its trusted certificate store. Set the B<SSL_CERT_PATH>
environment variable to give the directory where OpenSSL should looks for its
certificates or the B<SSL_CERT_FILE> environment variable to give the name of
a single file containing all of the certificates. See L<openssl-env(7)> for
further details about OpenSSL environment variables. For example you could use
this capability to have multiple versions of OpenSSL all installed on the same
system using different values for B<OPENSSLDIR> but all using the same
trusted certificate store.

You can test that your trusted certificate store is setup correctly by using it
via the OpenSSL command line. Use the following command to connect to a TLS
server:

 openssl s_client www.openssl.org:443

Once the command has connected type the letter "Q" followed by "<enter>" to exit
the session. This will print a lot of information on the screen about the
connection. Look for a block of text like this:

 SSL handshake has read 4584 bytes and written 403 bytes
 Verification: OK

Hopefully if everything has worked then the "Verification" line will say "OK".
If its not working as expected then you might see output like this instead:

 SSL handshake has read 4584 bytes and written 403 bytes
 Verification error: unable to get local issuer certificate

The "unable to get local issuer certificate" error means that OpenSSL has been
unable to find a trusted CA for the chain of certificates provided by the server
in its trusted certificate store. Check your trusted certificate store
configuration again.

Note that s_client is a testing tool and will still allow you to connect to the
TLS server regardless of the verification error. Most applications should not do
this and should abort the connection in the event of a verification error.

=head1 IMPORTANT OBJECTS FOR AN OPENSSL TLS APPLICATION

A TLS connection is represented by the B<SSL> object in an OpenSSL based
application. Once a connection with a remote peer has been established an
endpoint can "write" data to the B<SSL> object to send data to the peer, or
"read" data from it to receive data from the server.

A new B<SSL> object is created from an B<SSL_CTX> object. Think of an B<SSL_CTX>
as a "factory" for creating B<SSL> objects. You can create a single B<SSL_CTX>
object and then create multiple connections (i.e. B<SSL> objects) from it.
Typically you can set up common configuration options on the B<SSL_CTX> so that
all the B<SSL> object created from it inherit the same configuration options.

Note that internally to OpenSSL various items that are shared between multiple
B<SSL> objects are cached in the B<SSL_CTX> for performance reasons. Therefore
it is considered best practice to create one B<SSL_CTX> for use by multiple
B<SSL> objects instead of having one B<SSL_CTX> for each B<SSL> object that you
create.

Each B<SSL> object is also associated with two B<BIO> objects. A B<BIO> object
is used for sending or receiving data from the underlying transport layer. For
example you might create a B<BIO> to represent a TCP socket. The B<SSL> object
uses one B<BIO> for reading data and one B<BIO> for writing data. In most cases
you would use the same B<BIO> for each direction but there could be some
circumstances where you want them to be different.

It is up to the application programmer to create the B<BIO> objects that are
needed and supply them to the B<SSL> object. See
L<ossl-guide-tls-client-block(7)> for further information.

Finally, an endpoint can establish a "session" with its peer. The session holds
various TLS parameters about the connection between the client and the server.
The session details can then be reused in a subsequent connection attempt to
speed up the process of connecting. This is known as "resumption". Sessions are
represented in OpenSSL by the B<SSL_SESSION> object. In TLSv1.2 there is always
exactly one session per connection. In TLSv1.3 there can be any number per
connection including none.

=head1 PHASES OF A TLS CONNECTION

A TLS connection starts with an initial "set up" phase. The endpoint creates the
B<SSL_CTX> (if one has not already been created) and configures it.

A client then creates an B<SSL> object to represent the new TLS connection. Any
connection specific configuration parameters are then applied and the underlying
socket is created and associated with the B<SSL> via B<BIO> objects.

A server will create a socket for listening for incoming connection attempts
from clients. Once a connection attempt is made the server will create an B<SSL>
object in the same way as for a client and associate it with a B<BIO> for the
newly created incoming socket.

After set up is complete the TLS "handshake" phase begins. A TLS handshake
consists of the client and server exchanging a series of TLS handshake messages
to establish the connection. The client starts by sending a "ClientHello"
handshake message and the server responds with a "ServerHello". The handshake is
complete once an endpoint has sent its last message (known as the "Finished"
message) and received a Finished message from its peer. Note that this might
occur at slightly different times for each peer. For example in TLSv1.3 the
server always sends its Finished message before the client. The client later
responds with its Finished message. At this point the client has completed the
handshake because it has both sent and received a Finished message. The server
has sent its Finished message but the Finished message from the client may still
be in-flight, so the server is still in the handshake phase. It is even possible
that the server will fail to complete the handshake (if it considers there is
some problem with the messages sent from the client), even though the client may
have already progressed to sending application data. In TLSv1.2 this can happen
the other way around, i.e. the server finishes first and the client finishes
second.

Once the handshake is complete the application data transfer phase begins.
Strictly speaking there are some situations where the client can start sending
application data even earlier (using the TLSv1.3 "early data" capability) - but
we're going to skip over that for this basic introduction.

During application data transfer the client and server can read and write data
to the connection freely. The details of this are typically left to some higher
level application protocol (for example HTTP). Not all information exchanged
during this phase is application data. Some protocol level messages may still
be exchanged - so it is not necessarily the case that, just because the
underlying socket is "readable", that application data will be available to read.

When the connection is no longer required then it should be shutdown. A shutdown
may be initiated by either the client or the server via a message known as a
"close_notify" alert. The client or server that receives a close_notify may
respond with one and then the connection is fully closed and application data
can no longer be sent or received.

Once shutdown is complete a TLS application must clean up by freeing the SSL
object.

=head1 FURTHER READING

See L<ossl-guide-tls-client-block(7)> to see an example of applying these
concepts in order to write a simple TLS client based on a blocking socket.
See L<ossl-guide-quic-introduction(7)> for an introduction to QUIC in OpenSSL.

=head1 SEE ALSO

L<ossl-guide-introduction(7)>, L<ossl-guide-libraries-introduction(7)>,
L<ossl-guide-libssl-introduction(7)>, L<ossl-guide-tls-client-block(7)>,
L<ossl-guide-quic-introduction(7)>

=head1 COPYRIGHT

Copyright 2023 The OpenSSL Project Authors. All Rights Reserved.

Licensed under the Apache License 2.0 (the "License").  You may not use
this file except in compliance with the License.  You can obtain a copy
in the file LICENSE in the source distribution or at
L<https://www.openssl.org/source/license.html>.

=cut