Chapter 8. Securing the Mail Transport

Introduction

Chapter 7 contains recipes that use the AUTH SMTP extension to provide strong authentication of the end points in a mail exchange. In this chapter, the recipes use the SMTP STARTTLS extension for both strong authentication and encryption.

While encryption is one of the primary benefits of the SMTP STARTTLS extension, it is important to remember that this is not end-to-end encryption. Mail can take multiple hops before it is delivered. Some intervening hops may not use STARTTLS. Additionally, the mail message is still stored as clear text by both the sender and the recipient. STARTTLS only provides encryption for mail passing over a single hop between two sendmail systems configured for STARTTLS.

Chapter 7 explained how sendmail relied on the Simple Authentication and Security Layer (SASL) to provide the security tools necessary for AUTH authentication. STARTTLS relies on the Transport Layer Security (TLS) protocol for both authentication and encryption.

Transport Layer Security

TLS is an Internet standard protocol, defined in RFC 2246, The TLS Protocol Version 1.0. TLS is based on the Secure Sockets Layer (SSL) protocol that was originally developed by Netscape for web security. The manner in which TLS is used to secure mail transport is defined in RFC 2487, SMTP Service Extension for Secure SMTP over TLS. The SMTP extension that supports TLS is called STARTTLS. The receiving system advertises support for STARTTLS in its response to the client’s EHLO command. The connecting system requests TLS security by issuing the STARTTLS command in response to the advertisement.

TLS uses public key encryption, which is also called asymmetric encryption. It is asymmetric because it uses two different keys. A public key that is made available to the world and a private key that is kept secret. Anything encrypted with the public key can only be decrypted with the private key, and anything encrypted with the private key can only be decrypted with the public key. Four keys are needed to authenticate both the TLS client and the TLS server: the client’s public and private keys and the server’s public and private keys.[1]

The certificate

A TLS public key is distributed in a file called a certificate. It is called a certificate because it is certified to contain a valid public key by a digital signature. There are a few different techniques used to sign the certificates used by sendmail.

First, there are commercial certificate authorities (CAs). For a fee, a commercial CA will sign your certificate request creating a signed certificate. A CA is trusted to provide reasonable assurance that the certificate is correct and represents the organization it claims to represent. Commercial CAs are well established because they have been signing certificates for web sites for a number of years. Using a commercial CA to sign sendmail certificates is easy, and it has the added benefit of supporting a global customer base—certificates signed by a commercial CA are accepted around the world. Because many web sites want access to a global customer base, they use commercial CAs. In fact, a list of trusted commercial CAs can be obtained from any web browser. Figure 8-1 shows the list of CAs displayed by Netscape.

An alternative to using a commercial CA is to create a private CA. Creating a private CA is more complex than using a commercial CA, and it limits the customer base to those sites that are willing to accept certificates from the private CA. While a private CA is inadequate for a web site that must support a global customer base, it is a viable solution for sendmail security because mail security is often used between a limited number of partners. Recipe 8.1 shows you how to build a private CA, and Recipe 8.3 uses that CA to create the certificates used in the other recipes.

Finally, it is possible to create a self-signed certificate, which is not signed by any type of CA. sendmail does not accept self-signed certificates for authentication because, without the certification of a recognized CA, there is no way to verify the certificate’s information. sendmail does, however, accept self-signed certificates for encryption. In fact, if you only want encryption, the TLS client doesn’t need a certificate at all. However, using a CA increases the utility of the STARTTLS protocol because it adds strong authentication to basic encryption.

If the sendmail system is also a web server, it may already have a commercially signed certificate that can be used for sendmail. If not, use openssl to create certificates. To have a sendmail certificate signed by a commercial CA, create a certificate signing request (CSR)—also called a certificate request—and send it to the commercial CA for signing. (See the web site of the CA you select for the exact details.) If you run a private CA, create the CSR and use the private CA to sign it. Recipe 8.2 and Recipe 8.3 show examples of creating a CSR and of signing it to produce a signed certificate.

Certificates contain much more than just a public key. Use openssl to examine the contents of a certificate, as in this example:

# openssl x509 -noout -text -in rodent.pem
Certificate:
    Data:
        Version: 3 (0x2)
        Serial Number: 2 (0x2)
        Signature Algorithm: md5WithRSAEncryption
        Issuer: C=US, ST=Maryland, L=Gaithersburg, O=WroteTheBook, 
               CN=chef.wrotethebook.com/Email=craig@chef.wrotethebook.com
        Validity
            Not Before: Jan 28 16:25:58 2003 GMT
            Not After : Jan 28 16:25:58 2004 GMT
        Subject: C=US, ST=Maryland, L=Gaithersburg, O=WroteTheBook, 
                CN=rodent.wrotethebook.com/Email=alana@rodent.wrotethebook.com
        Subject Public Key Info:
            Public Key Algorithm: rsaEncryption
            RSA Public Key: (1024 bit)
                Modulus (1024 bit):
                    00:b0:91:de:51:14:19:91:61:39:ed:23:61:ef:9d:
                    e4:44:9f:ff:6a:fd:5c:cc:e8:bd:79:1f:2d:bc:eb:
                    ca:66:da:86:fb:99:17:79:07:94:dd:94:90:aa:6a:
                    44:55:ff:ec:65:e9:4e:41:b1:e9:e9:c4:09:89:81:
                    47:2b:a1:bd:a0:0b:bf:9c:c5:5e:39:de:c5:49:a3:
                    e6:5c:af:9e:32:eb:ae:63:80:7f:c6:ad:12:d0:87:
                    01:db:06:a3:93:bf:41:9f:51:81:4f:09:fa:da:e3:
                    ca:69:dc:35:4f:cd:43:fd:ab:6c:51:3d:24:97:60:
                    bd:bf:e8:91:9c:5a:f4:b2:df
                Exponent: 65537 (0x10001)
        X509v3 extensions:
            X509v3 Basic Constraints:
                CA:FALSE
            Netscape Comment:
                OpenSSL Generated Certificate
            X509v3 Subject Key Identifier:
                76:6E:60:D0:EE:B4:62:49:84:46:04:7C:44:E6:F8:9E:CF:84:B6:DC
            X509v3 Authority Key Identifier:
                keyid:3C:D9:71:02:66:8A:D4:AD:3C:07:FD:57:2A:C5:03:C2:B8:7D:FF:D5
                DirName:/C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/
                       CN=chef.wrotethebook.com/Email=craig@chef.wrotethebook.com
                serial:00

    Signature Algorithm: md5WithRSAEncryption
        c4:45:4a:e4:01:eb:0b:83:c4:26:c5:de:9e:06:8f:7a:b6:2d:
        f0:f3:63:dd:ba:1c:22:2d:f9:ac:3d:92:04:c4:4f:a8:9c:d1:
        7c:de:a7:6d:94:10:62:cb:69:c6:ba:8f:09:e7:a9:49:18:97:
        4b:aa:50:b9:0a:c9:49:59:0d:65:b6:44:1a:d6:88:6c:bd:d9:
        3e:89:e6:1c:76:c8:54:69:7f:b4:9d:7a:a7:de:92:12:cc:f5:
        43:e5:e5:da:b0:34:9d:cf:ee:b9:65:a1:a8:d7:cc:aa:96:12:
        10:7f:2b:99:94:79:f3:b4:bf:0a:58:54:65:7f:98:36:be:a5:
        22:de

Several important pieces of information can be seen in this display. Foremost, of course, is the public key. It is the 1024-bit RSA public key that appears about halfway through the display. This is the key used to encrypt the session information sent to the system holding the private key—the session information includes the session key that will be used to encrypt the mail.

The system that owns this certificate is identified by the Subject field. The Subject field contains the distinguished name (DN) of the remote host. The DN is a formal name that can contain several fields. In this example, the DN contains a country code (C), a state code (ST), a locality (L), an organizational name (O), and a common name (CN). Inside the sendmail configuration, the DN of the subject is found in the ${cert_subject} macro, and the CN of the subject is found in the ${cn_subject} macro. In the example, the certificate belongs to the host rodent.wrotethebook.com.

The CA that signed the certificate is identified by the Issuer field, which contains the DN of the CA. Inside the sendmail configuration, the DN of the CA is found in the ${cert_issuer} macro, and the CN of the CA is found in the ${cn_issuer} macro. In the example, chef.wrotethebook.com is the private CA that is created in Recipe 8.1.

At the bottom of the display is the digital signature of the certificate created by the CA. It is an MD5 digest of the certificate, which is stored in the sendmail configuration in the macro ${cert_md5}. The digest is encrypted by the CA’s private key. sendmail uses the CA’s public key to decrypt the digest. If sendmail successfully decrypts and verifies the digest, it stores OK in the macro ${verify} and considers the certificate valid for authentication.

The ${verify} macro can contain any one of several different values depending on the result of the verification process. The possible values for the ${verify} macro are:

Once again, refer back to the certificate displayed by the openssl x509 command. Near the top of the output, the TLS version number is identified in the Version field of the certificate. Inside the sendmail configuration, the TLS version number is stored in the ${tls_version} macro.

As we have seen, several sendmail macros can be traced directly back to the TLS certificate. Two other important TLS related macros get their values from the session encryption used for the email connection. ${cipher} identifies the type of encryption used for the connection, and ${cipher_bits} contains the number of bits of encryption used for the connection. All of the important TLS encryption and authentication data is available for use inside the sendmail configuration.

The best way to control how TLS security information is applied on your system is through the access database. These are the access database records that provide control over a TLS connection:

CERTISSUER:

Authorizes relaying based on the DN of the CA that signed the certificate. Recipe 8.5 and Recipe 8.6 provide example of using the CERTISSUER: record.

CERTSUBJECT:

Authorizes relaying based on the DN of the host that presented the certificate. Recipe 8.6 covers the use of this record.

TLS_Srv:

Defines the minimum security required for an outbound TLS connection. Recipe 8.7 and Recipe 8.9 show how to use the TLS_SRV: record.

TLS_Clt:

Defines the minimum security required for an inbound TLS connection. Recipe 8.8 and Recipe 8.9 contain examples of the TLS_Clt: record.

TLS_Rcpt:

Defines the minimum security required for mail sent to the specified recipient. Recipe 8.10 uses this record.

Try_TLS:

Controls whether the STARTTLS command is issued to the specified remote TLS server. See Recipe 8.11 for an example of how this record is used.

Srv_Features:

Controls whether STARTTLS is advertised to the specified remote TLS client, and whether the specified remote TLS client is required to present a certificate. See Recipe 8.12 and Recipe 8.13 for examples of this record.

sendmail provides hooks into the rulesets that process the access database entries related to STARTTLS:

These hooks provide access to the full range of sendmail’s capabilities, including all of the information that sendmail has at the time that the access database records are being processed. Anything that cannot be done within the confines of the standard database record formats can probably be done through a custom ruleset. The power and simplicity of using the access database to control delivery over a TLS connection make custom TLS rulesets largely unnecessary. However, if you have an application for a custom ruleset, sendmail provides all of the tools you need to roll your own.

The sendmail implementation of STARTTLS depends on OpenSSL for authentication, encryption, and support tools. Before any of the recipes in this chapter can be used, OpenSSL must be installed.

Modern systems often include OpenSSL in order to provide web security. Our sample Red Hat Linux system includes an OpenSSL RPM file, as this simple command shows:

$ rpm -q openssl
openssl-0.9.6b-18

If your system includes OpenSSL, you can proceed. If it is not installed, install the version provided by your Unix vendor using the vendor’s package management tools. If your Unix vendor does not provide OpenSSL, download the source from http://www.openssl.org/source/or ftp://ftp.openssl.org/source/. The source comes as a gzipped tarball. Unzip and restore the tarball. Go to the newly created directory and run config and then make to compile the source code. Then run make test and make install to install OpenSSL. If you have any problems compiling the source, see the INSTALL file delivered with the source code.

Additionally, sendmail must be specially compiled to support STARTTLS. If your sendmail binary was not compiled with STARTTLS support, recompile sendmail as described in Recipe 1.6 and Recipe 1.7.

8.1. Building a Private Certificate Authority

Select a directory in which to place the CA directory structure (/etc/mail/certs is a common choice). Change to that directory and run the CA script provided with the OpenSSL distribution. Use the -newca command-line option of the CA script.[2] The script allows you to enter a certificate filename to work with an existing CA certificate. In this case, we are not working with an existing CA certificate, so just press the Return key to create a new CA. When prompted for a PEM passphrase, enter the password that will be required whenever the certificate authority is used to sign a certificate request. Finally, enter the distinguished name of the system that is acting as the CA. Here is an example:

# cd /etc/mail
# mkdir certs
# cd certs
# /usr/share/ssl/misc/CA -newca
CA certificate filename (or enter to create)

Making CA certificate ...
Using configuration from /usr/share/ssl/openssl.cnf
Generating a 1024 bit RSA private key
...............++++++
.............++++++
writing new private key to './demoCA/private/./cakey.pem'
Enter PEM pass phrase: SaytheSECRETword
               !
Verifying password - Enter PEM pass phrase: SaytheSECRETword
               !
-----
You are about to be asked to enter information that will be incorporated
into your certificate request.
What you are about to enter is what is called a Distinguished Name or a DN.
There are quite a few fields but you can leave some blank
For some fields there will be a default value,
If you enter '.', the field will be left blank.
-----
Country Name (2 letter code) [GB]:US
State or Province Name (full name) [Berkshire]:Maryland
Locality Name (eg, city) [Newbury]:Gaithersburg
Organization Name (eg, company) [My Company Ltd]:WroteTheBook
Organizational Unit Name (eg, section) []:
Common Name (eg, your name or your server's hostname) []:chef.wrotethebook.com
Email Address []:craig@chef.wrotethebook.com

The CA script creates a directory structure it calls demoCA, which contains the files needed for a private CA. Change the name demoCA to something that sounds less temporary, for example, to the directory name CA:

# mv demoCA CA

Change to the new directory structure. Tighten the security on the private directory, which holds the certificate authority’s private key:

# cd CA
# chmod 0700 private

Copy the openssl.cnf file to the CA directory. The copy will be used by the private CA when signing sendmail certificates. To distinguish it from the original openssl.cnf file, the copy is renamed sendmailssl.cnf when it is copied to the CA directory:

# cp /usr/share/ssl/openssl.cnf sendmailssl.cnf

Finally, edit the sendmailssl.cnf file to point to the newly created CA directory structure. Change the line:

dir             = ./demoCA                # Where everything is kept

to read:

dir             = /etc/mail/certs/CA      # Where everything is kept

The private CA can now be used to sign certificates as described in Recipe 8.3.

Discussion

It is not absolutely necessary to build a private CA just to create the certificates needed by STARTTLS. Introduction to this chapter describes alternative ways to obtain signed certificates. This recipe should not be taken as a recommendation for creating a private CA. It is incorporated as a recipe in this chapter because building a private CA is the most complex method of signing certificates, and thus it is the method most in need of a recipe to explain how it is done. Before you decide to build a private CA, evaluate the alternatives to make sure you choose the approach that is most suitable for your situation.

The complexity of creating a private CA is substantially reduced by the CA script provided with the OpenSSL distribution. The CA script accepts several different command-line arguments, but the one that is most useful for a sendmail system is the -newca option. -newca causes the script to create the directory structure and files needed by a CA. It also causes the script to build the private key and the certificate that will be used by the CA. An ls of the directory created by the CA script shows the following:

# ls
cacert.pem  certs  crl  index.txt  newcerts  private  serial
# ls private
cakey.pem

The directory created by the CA script contains three files:

The private key associated with the CA certificate is stored in the private subdirectory and is named cakey.pem, as the second ls command in the example shows. The private key is kept safe on the CA. It is never distributed to, or used on, any other system. The cakey.pem file is encrypted and can only be used by someone who knows the PEM passphrase provided to the CA script when the cakey.pem file was created. In the example in the Solution section, the PEM passphrase is SaytheSECRETword!. The passphrase is required in Recipe 8.3 when the CA is used to sign certificates.

In addition to the files just described, the directory created by the CA script contains four subdirectories:

private

This directory is used to hold private keys. The CA script places the certificate authority’s private key here. In the Solution section, the permissions on this directory are changed to 0700.

newcerts

This directory holds copies of all the certificates signed by this CA. The certificates in this directory are identified by serial numbers. This directory is empty when created by the CA script. Files are added to the directory by the signature process.

crl

This directory holds certificate revocation lists. The CA script creates this directory to be empty. See Network Security with OpenSSL, by Viega, Messier, and Chandra (O’Reilly), for information on CRLs.

certs

This directory can be used to hold certificates. However, the CA script does not place the CA’s certificate in this directory.

This recipe recommends adding a copy of openssl.cnf to the files and directories created by the CA script. The openssl.cnf configuration file is read by openssl every time it is executed. openssl is used to create and sign certificates, and it will be used in subsequent recipes in this chapter. To simplify customizing the OpenSSL configuration for sendmail certificate management, and to avoid any possibility of interfering with the OpenSSL configuration used by the web site administrator, we copy openssl.cnf to sendmailssl.cnf. The new name is intended to make it clear that this configuration is used only by the sendmail CA. We then edit the sendmailssl.cnf file to point to /etc/mail/certs/CA as the directory used by the default CA. This recipe assumes that the private CA is being used exclusively for sendmail certificate management. Because this CA is used exclusively for sendmail, it makes sense to create a sendmail-specific OpenSSL configuration. Recipe 8.3 uses the newly created sendmailssl.cnf file.

See Also

Recipe 8.2 and Recipe 8.3 are both directly related to this recipe. Introduction to this chapter contains important information about certificates and certificate authorities. The sendmail book discusses certificates in Section 10.10.2. The config manpage provides information on the openssl.cnf file. Network Security with OpenSSL, by Viega, Messier, and Chandra (O’Reilly), provides information about certificates, certificate authorities, and certificate revocation lists.

8.2. Creating a Certificate Request

On the sendmail host, create a directory to hold the certificate and private key. Here is an example:

# cd /etc/mail
# mkdir certs

Change to the new directory. Use the openssl req command to create an unsigned X.509 certificate and a private key. When prompted for the distinguished name, enter the DN of the sendmail host for which the certificate is being created. In this example we create a certificate for crab.wrotethebook.com:

# cd /etc/mail/certs
# umask 0066
# openssl req -nodes -new -x509 -keyout key.pem -out newcert.pem
Using configuration from /usr/share/ssl/openssl.cnf
Generating a 1024 bit RSA private key
.++++++
...................++++++
writing new private key to 'key.pem'
-----
You are about to be asked to enter information that will be incorporated
into your certificate request.
What you are about to enter is a Distinguished Name or a DN.
There are quite a few fields but you can leave some blank
For some fields there will be a default value,
If you enter '.', the field will be left blank.
-----
Country Name (2 letter code) [GB]:US
State or Province Name (full name) [Berkshire]:Maryland
Locality Name (eg, city) [Newbury]:Gaithersburg
Organization Name (eg, company) [My Company Ltd]:WroteTheBook
Organizational Unit Name (eg, section) []:
Common Name (eg, your name or the server's name) []:crab.wrotethebook.com
Email Address []:admin@wrotethebook.com

Next, use openssl x509 to generate a certificate signing request from the certificate and key pair created above. The newcert.pem file created above is a temporary file that can be removed after the CSR is created:

# openssl x509 -x509toreq -in newcert.pem -signkey key.pem -out csr.pem
Getting request Private Key
Generating certificate request
# rm -f newcert.pem

Send the CSR, csr.pem in this example, to the CA. The CA signs the sendmail host’s certificate request. Recipe 8.3 shows how this is done using the private CA created in Recipe 8.1.

The CA returns a signed certificate to the sendmail host, along with a copy of the CA’s certificate. On the sendmail host, place the sendmail host’s signed certificate and the CA’s certificate in the certificate directory created in the first step. Create a symbolic link to a hash of the CA’s certificate using the following command:

ln -s ca_filename `openssl x509 -noout -hash < ca_filename`.0

In the command above, replace ca_filename with the filename of the CA’s certificate. sendmail uses the hash created by this command for certificate verification. Run this command, replacing ca_filename with the name of the new CA certificate file, every time a CA certificate is added to the directory identified by the confCACERT_PATH define. Recipe 8.4 discusses the confCACERT_PATH define.

Discussion

Most of the work in this recipe is done on the sendmail host. However, it doesn’t have to be. All of the files necessary to request a signed certificate for crab.wrotethebook.com can be generated on any system that has OpenSSL installed. The advantage of creating the CSR on the sendmail host is that the private key is created there and never has to leave the system, making it easier to maintain private key security. The disadvantage is that the administrator of crab.wrotethebook.com needs to use the complex openssl commands to build the CSR. Many sites prefer to have one CA administrator who creates the certificate and key for each sendmail host on the CA server and then distributes the signed files to the sendmail systems. The approach you use is mostly a matter of organizational style.

The only thing that must be created on the sendmail host is the directory that will hold the certificate and the private key. The name of this directory, the name of the certificate file, the name of the key file, and the name of the CA certificate file, are all important values used to configure sendmail for STARTTLS, as can be seen in Recipe 8.4.

The openssl req command in the Solution section creates a new, self-signed X.509 certificate (-new -x509). The certificate is written to a file named newcert.pem (-out newcert.pem). The private key associated with the certificate is stored in a file named key.pem (-keyout key.pem). The key.pem file is not encrypted with DES (-nodes). This is important. If the private key is accidentally placed in an encrypted file, the sendmail administrator is prompted for the password to decrypt the file every time sendmail needs to access the private key. This could cause the sendmail system to hang during the boot process until the password is entered.

The openssl x509 command processes an X.509 certificate. In this case, the command converts an X.509 certificate to a certificate request (-x509toreq). The certificate that is being converted is the newcert.pem file created in the first step (-in newcert.pem -signkey key.pem). The CSR is written to a file named csr.pem (-out csr.pem).

The CSR file is sent to the CA. The CA uses that file as input to the signature process and returns another file, which is the signed certificate. Often, one file is sent to the CA and one is returned. The Solution section, however, also mentions a second file—the CA’s certificate. This certificate may not be sent from the CA as part of the signature process. There may be some other means for obtaining that file. Regardless of how the CA certificate is obtained, a hash should be made from the CA certificate as shown in the last step of the Solution. sendmail uses the hash during certificate verification.

Recipe 8.3 covers signing the certificate request created in this recipe.

8.3. Signing a Certificate Request

On the CA, use the openssl ca command to sign the certificate signing request provided by the sendmail host. When prompted for the PEM passphrase, enter the password created for the private CA. Answer “yes” to the two questions, and the certificate is signed. Here is an example in which the CSR created in Recipe 8.2 is signed on the CA created in Recipe 8.1:

# openssl ca -config ./sendmailssl.cnf -policy policy_anything -out cert.pem -infiles csr.pem
Using configuration from ./sendmailssl.cnf
Enter PEM pass phrase: SaytheSECRETword
               !
Check that the request matches the signature
Signature ok
The Subjects Distinguished Name is as follows
countryName           :PRINTABLE:'US'
stateOrProvinceName   :PRINTABLE:'Maryland'
localityName          :PRINTABLE:'Gaithersburg'
organizationName      :PRINTABLE:'WroteTheBook'
commonName            :PRINTABLE:'crab.wrotethebook.com'
emailAddress          :IA5STRING:'admin@wrotethebook.com'
Certificate is to be certified until Jan 27 19:50:16 2004 GMT (365 days)
Sign the certificate? [y/n]:y

1 out of 1 certificate requests certified, commit? [y/n]y
Write out database with 1 new entries
Data Base Updated

The CSR sent from the sendmail system is an intermediate file that is no longer needed. Delete the unneeded file:

# rm -f csr.pem

Send the signed certificate to the sendmail host. In this example, we send the file cert.pem to the sendmail host. We named the certificate file cert.pem because it is a neutral name well suited to textbook examples. However, the certificate filename is whatever you choose to make it, and it should be something descriptive. For example, the certificate was signed for the host crab.wrotethebook.com so we could have given it the name crab.wrotethebook.com.cert.pem.

Also, send the sendmail host a copy of the CA’s certificate. Assuming that this is the CA created in Recipe 8.1, the CA certificate file sent to the sendmail host would be cacert.pem.

Discussion

The openssl ca command is used to sign certificates. In the Solution section, the openssl command signs the csr.pem file from Recipe 8.2 (-infiles csr.pem), producing a signed certificate in a file named cert.pem (-out cert.pem).

This openssl command references the special OpenSSL configuration file (-config ./sendmailssl.cnf) that was created in Recipe 8.1 for the private sendmail CA. If the configuration file is not identified on the command line using the -config option, the openssl command uses the openssl.cnf file. The OpenSSL file provides default values used when creating and signing certificates; it defines the pathnames of the files and directories required by a CA, and it defines the certificate signing policies. By default, the openssl.cnf file, and the sendmailssl.cnf file that Recipe 8.1 created from it, contain two signature policies:

This recipe uses -policy policy_anything, which is commonly used when signing sendmail certificates. After all, the administrator is shown the DN and asked whether or not the certificate should be signed. If the administrator deems that the DN does not contain enough information for email security, the administrator can refuse to sign the certificate. However, in this specific example, policy_match could have been used and it would have worked. Remember, this is a private CA. We know that the country, state, and organization names used in the DN of the CA match the values in the DN of the server. Yet, policy_anything makes a better example because it is more flexible and deals with a larger variety of configurations. Most sendmail administrators use policy_anything and that is why it is included in this recipe.

After signing the certificate request, the CSR (csr.pem) is deleted. The certificate cert.pem is sent to crab.wrotethebook.com along with the CA’s certificate cacert.pem. The sendmail host will reference both of these certificates in the STARTTLS configuration.

Recipe 8.1 covers creating the private CA used to sign the certificate in this recipe. Recipe 8.2 creates the certificate request signed in this example. Introduction to this chapter contains important information about certificates and certificate authorities. The sendmail book discusses certificates in Section 10.10.2. The config manpage provides information on the openssl.cnf file. The req manpage documents the openssl req command. The x509 manpage documents the openssl x509 command, and the ca manpage covers the openssl ca command. TCP/IP Network Administration, Third Edition, by Craig Hunt (O’Reilly), provides examples of using openssl commands in Chapter 11. Network Security with OpenSSL, by Viega, Messier, and Chandra (O’Reilly), provides information about certificates, certificate authorities, openssl.cnf, and the openssl command.

Discussion

This recipe provides a full STARTTLS configuration. A system configured with all of these defines can act as a TLS server or client and can perform both encryption and authentication. The six defines point sendmail to the files necessary for this full range of service:

If OpenSSL is installed, sendmail is compiled with STARTTLS support, and these defines point to valid certificates, sendmail will advertise STARTTLS in response to the SMTP EHLO command. A simple telnet test shows that STARTTLS is ready for use:

# telnet localhost smtp
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
220 chef.wrotethebook.com ESMTP Sendmail 8.12.9/8.12.9; Fri, 22 Aug 2003 12:01:37 -
0400
ehlo localhost
250-chef.wrotethebook.com Hello IDENT:6l4ZhaGP3Qczqknqm/KdTFGsrBe2SCYC@localhost 
[127.0.0.1], pleased to meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-STARTTLS
250-DELIVERBY
250 HELP
QUIT
221 2.0.0 chef.wrotethebook.com closing connection
Connection closed by foreign host.

If your system doesn’t advertise STARTTLS in response to the EHLO command, check the logfile to make sure that all file permissions are correct. If they are, try increasing the LogLevel to 14 to log additional STARTTLS debugging information. See Recipe 1.10 for information on setting the LogLevel.

This recipe shows a system configured with both client and server certificates and keys. Many systems are configured as both clients and servers for STARTTLS because mail is often forwarded by one system to another. When the system receives inbound mail, it acts as a STARTTLS server. When it forwards that mail on to another system, it acts as a STARTTLS client. Therefore, the system needs to act as both a STARTTLS server and a STARTTLS client. Normally such a system uses the same certificate and key for both its client and server roles.

A STARTTLS client will attempt to use STARTTLS whenever the server offers it. Running the sendmail command with the -v option shows the STARTTLS client side of this sample configuration. An example of this is seen in the following code:

# sendmail -Am -v -t
               To: craig@chef.wrotethebook.com
               From: craig@rodent.wrotethebook.com
               Subject: First STARTTLS test

               Ctrl-D
craig@chef.wrotethebook.com... Connecting to chef.wrotethebook.com. via esmtp...
220 chef.wrotethebook.com ESMTP Sendmail 8.12.9/8.12.9; Fri, 22 Aug 2003 12:01:37 -
0400
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-STARTTLS
250-DELIVERBY
250 HELP
>>> STARTTLS
220 2.0.0 Ready to start TLS
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH EXTERNAL DIGEST-MD5 CRAM-MD5
250-DELIVERBY
250 HELP
>>> MAIL From:<craig@rodent.wrotethebook.com> SIZE=97 AUTH=craig@rodent.wrotethebook.
com
250 2.1.0 <craig@rodent.wrotethebook.com>... Sender ok
>>> RCPT To:<craig@chef.wrotethebook.com>
>>> DATA
250 2.1.5 <craig@chef.wrotethebook.com>... Recipient ok
354 Enter mail, end with "." on a line by itself
>>> .
250 2.0.0 h0UGn9P7001230 Message accepted for delivery
craig@chef.wrotethebook.com... Sent (h0UGn9P7001230 Message accepted for delivery)
Closing connection to chef.wrotethebook.com.
>>> QUIT
221 2.0.0 chef.wrotethebook.com closing connection

The client (rodent) connects to the server (chef) and issues an EHLO command. In response to that command, the server advertises that STARTTLS is available. The client then issues a STARTTLS command to open the TLS connection. The server responds with the message:

220 2.0.0 Ready to start TLS

Immediately after the response, the test shows another EHLO message. Many system administrators are confused by this. Don’t be. The STARTTLS negotiation takes place “outside” of the email exchange. STARTTLS creates the encrypted tunnel through which the mail travels. The first EHLO, the server’s STARTTLS advertisement, the client’s STARTTLS command, and the server’s TLS ready response are all used to create the tunnel. The second EHLO command and all that follows it are part of the standard SMTP protocol exchange, which is sent inside the encrypted tunnel. Note that this SMTP protocol exchange is displayed by the sendmail -v command as clear text. The mail stream is only encrypted on the network. Within the end systems, the mail stream is clear text.

The effect of the STARTTLS protocol can be indirectly observed in the headers of the mail message that passed through the encrypted tunnel. The following headers are from the test message sent by the sendmail -Am -v -t command used above:

Return-Path: <craig@rodent.wrotethebook.com>
Received: from rodent.wrotethebook.com (rodent.wrotethebook.com
    [192.168.0.3])
        by chef.wrotethebook.com (8.12.9/8.12.9) with ESMTP id h0UGn9P7001230
        (version=TLSv1/SSLv3 cipher=EDH-RSA-DES-CBC3-SHA bits=168 verify=OK)
        for <craig@chef.wrotethebook.com>; Thu, 30 Jan 2003 11:49:10 -0500
Received: (from root@localhost)
        by rodent.wrotethebook.com (8.12.9/8.12.9) id h0UGvpSD010157;
        Thu, 30 Jan 2003 11:57:51 -0500
Date: Thu, 30 Jan 2003 11:57:51 -0500
Message-Id: <200301301657.h0UGvpSD010157@rodent.wrotethebook.com>
To: craig@chef.wrotethebook.com
From: craig@rodent.wrotethebook.com
Subject: First STARTTLS test

The Received: header generated by chef.wrotethebook.com, which is the STARTTLS server in this exchange, shows the TLS characteristics of the link over which the mail was received. The header displays the version of the TLS protocol that was used for the connection, the type of encryption used for the connection, the number of bits used for the encryption, and whether the client’s certificate was verified. Because this information is only placed in the header when TLS is used for the connection, it tells us that the client is properly configured to use TLS when communicating with this server.

Recipe 8.2 and Recipe 8.3 cover how certificates and keys are created. Chapter 1 provides additional information on compiling sendmail. The sendmail book covers STARTTLS in Section 10.10.

8.5. Relaying Based on the CA

Make sure that OpenSSL is installed as described in Introduction and that sendmail is compiled with STARTTLS support as shown in Recipe 1.6 and Recipe 1.7.

Identify each trusted certificate authority in the access database using CERTISSUER: entries. The key field of each entry begins with the CERTISSUER: tag followed by the distinguished name of the CA. The return value of the entry is the keyword RELAY, indicating that any host presenting a certificate signed by the trusted CA is granted relaying privileges.

Create a STARTTLS sendmail configuration and add the access_db feature. Here are samples of the defines that configure STARTTLS and the FEATURE macro that enables the access database:

dnl Point to the CA certificate directory
define(`confCACERT_PATH', `/etc/mail/certs')
dnl Point to the root CA's certificate
define(`confCACERT', `/etc/mail/certs/cacert.pem')
dnl Point to the certificate used for inbound connections
define(`confSERVER_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for inbound connections
define(`confSERVER_KEY', `/etc/mail/certs/key.pem')
dnl Point to the certificate used for outbound connections
define(`confCLIENT_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for outbound connections
define(`confCLIENT_KEY', `/etc/mail/certs/key.pem')
dnl Enable the access database
FEATURE(`access_db')

Rebuild the sendmail.cf file, copy the new sendmail.cf file to /etc/mail, and restart sendmail. Recipe 1.8 provides an example of these steps.

Discussion

TLS can provide strong authentication of the end points and encryption of the data stream. Despite these important security features, establishing a STARTTLS connection is not, by itself, enough to grant relaying privileges. A test shows this:

# sendmail -Am -v -t
               To: craig@crab.wrotethebook.com
               From: craig@rodent.wrotethebook.com
               Subject: Attempt to relay after successful TLS connection

               Ctrl-D
craig@crab.wrotethebook.com... Connecting to chef.wrotethebook.com. via relay...
220 chef.wrotethebook.com ESMTP Sendmail 8.12.9/8.12.9; Fri, 22 Aug 2003 12:01:37 -
0400
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-STARTTLS
250-DELIVERBY
250 HELP
>>> STARTTLS
220 2.0.0 Ready to start TLS
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH EXTERNAL DIGEST-MD5 CRAM-MD5
250-DELIVERBY
250 HELP
>>> MAIL From:<craig@rodent.wrotethebook.com> SIZE=102 AUTH=craig@rodent.
wrotethebook.com
250 2.1.0 <craig@rodent.wrotethebook.com>... Sender ok
>>> RCPT To:<craig@crab.wrotethebook.com>
>>> DATA
550 5.7.1 <craig@crab.wrotethebook.com>... Relaying denied
503 5.0.0 Need RCPT (recipient)
>>> RSET
250 2.0.0 Reset state
craig... Connecting to local...
craig... Sent
Closing connection to chef.wrotethebook.com.
>>> QUIT
221 2.0.0 chef.wrotethebook.com closing connection

In response to the first EHLO command, the server advertises STARTTLS. The client then issues a STARTTLS command, and the server responds that it is ready. This indicates that an encrypted session will start. The second EHLO command is the first SMTP command in the encrypted session. It is followed by a MAIL From: and a RCPT To: command. Notice the response to the RCPT To: command.[3] The message is rejected with a “Relaying denied” error, even though it comes in the midst of an encrypted session. Clearly, TLS does not, by itself, authorize relaying. It is up to the sendmail administrator to decide when and how TLS authentication data should be used. One way to use TLS data is through the access database.

The ${verify} and ${cert_issuer} macros are used in the Local_Relay_Auth ruleset to authorize relaying. ${verify} contains OK when sendmail is able to verify the CA signature on the certificate received from a remote site. sendmail takes the distinguished name of the CA that signed the certificate from the Issuer field of the certificate and stores it in the macro ${cert_issuer}. If ${verify} equals OK, sendmail searches the access database for a CERTISSUER: entry that matches the ${cert_issuer} value. If a match is found, the action specified by the CERTISSUER: entry is taken. If no match is found, processing continues normally. Here is an example of adding a CERTISSUER: entry to the access database:

# cd /etc/mail
# cat >> access CERTISSUER:/C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/CN=chef.wrotethebook.com/
Email=craig@chef.wrotethebook.com    RELAY
               Ctrl-D
# makemap hash access < access

Given the access database entry just shown, a site that presents a certificate signed by /C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/CN=chef.wrotethebook.com/Email=craig@chef.wrotethebook.com is granted relaying privileges. /C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/CN=chef.wrotethebook.com/Email= craig@chef.wrotethebook.com is the complete DN of the private CA created in Recipe 8.1. Therefore, any host with a certificate signed by our private CA is allowed to relay mail.

A DN can contain almost any character. Several characters, however, have special meaning to sendmail and others are difficult to enter into the access database. For these reasons, the DN stored in the access database uses a special format for storing nonprintable characters, the space, the tab, <, >, (, ), “, and +. These characters are stored as their hexadecimal values preceded by a plus. For example, a space becomes +20, a ( becomes +25.

After adding the CERTISSUER: entry to the access database, rerun the test shown earlier in this section. The RCPT To: command now produces the following result:

>>> RCPT To:<craig@crab.wrotethebook.com>
250 2.1.5 <craig@crab.wrotethebook.com>... Recipient ok

The client is now granted relaying privileges because a CA that is trusted to authorize relaying signed the client’s certificate.

Recipe 8.4 covers the STARTTLS sendmail configuration used as the basis for this recipe. The private CA mentioned in the discussion of this recipe is covered in Recipe 8.1. Chapter 3 discusses the traditional techniques used to grant relaying privileges, including the access database. The content of a certificate, including the Issuer field, is covered in this chapter’s Introduction. The sendmail book discusses using the access database for STARTTLS in Section 10.10.8.

8.6. Relaying Based on the Certificate Subject

sendmail must be compiled with STARTTLS support, and OpenSSL must be installed. See Recipe 1.6 and Recipe 1.7 and Introduction to this chapter for more information.

Create an access database entry using a CERTISSUER: tag followed by the DN of the trusted certificate authority to identify the CA that signed the client’s certificate. The return value of the CERTISSUER: entry must be the keyword SUBJECT. Next, define each client that is to be granted relaying privileges using a CERTSUBJECT: entry in the access database. The key field of these entries begins with the CERTSUBJECT: tag followed by the DN of the client. The return value is the keyword RELAY.

Add the access_db feature to the STARTTLS sendmail configuration. Here are sample configuration lines:

dnl Point to the CA certificate directory
define(`confCACERT_PATH', `/etc/mail/certs')
dnl Point to the root CA's certificate
define(`confCACERT', `/etc/mail/certs/cacert.pem')
dnl Point to the certificate used for inbound connections
define(`confSERVER_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for inbound connections
define(`confSERVER_KEY', `/etc/mail/certs/key.pem')
dnl Point to the certificate used for outbound connections
define(`confCLIENT_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for outbound connections
define(`confCLIENT_KEY', `/etc/mail/certs/key.pem')
dnl Enable the access database
FEATURE(`access_db')

Following the example from Recipe 1.8, rebuild the sendmail.cf file. Install the new configuration, and restart sendmail.

Discussion

Successfully establishing a STARTTLS session does not, by itself, authorize relaying. (See Recipe 8.5 for a test that illustrates this.) However, the authentication data provided by STARTTLS can be used to control relaying via the access database. Here are access database entries that authorize relaying based on the value in the subject field of the client’s certificate:

CERTISSUER:/C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/CN=chef.wrotethebook.com/
SUBJECT
CERTSUBJECT:/C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/CN=rodent.wrotethebook.
com/     RELAY
CERTSUBJECT:/C=US/ST=Maryland/L=Gaithersburg/O=WroteTheBook/CN=crab.wrotethebook.com/
RELAY

This example checks both the Issuer field and the Subject field from the certificate presented by the client. These values are matched against the access database by the Local_Relay_Auth ruleset, but only if the ${verify} macro contains the string OK, which indicates that the certificate was successfully verified by sendmail. Both fields are looked up because, before the subject of a certificate can be trusted, the CA that signs and verifies the certificate must be trusted.

If ${verify} returns OK, the Local_Relay_Auth ruleset looks for a CERTISSUER: entry containing the distinguished name from the Issuer field of the certificate. If a match is found and the action is RELAY, relaying is authorized for any system with a certificate signed by the specified CA, as described in Recipe 8.5. If a match is found and the action is SUBJECT, a second access database lookup is performed. This time, the ruleset looks for a CERTSUBJECT: entry containing the DN from the Subject field of the certificate. If it is found, and it contains the action RELAY, relaying is permitted for the specified client. If it is not found, normal processing continues. Relaying can still be granted to the client through any of the basic methods described in Chapter 3. STARTTLS provides an additional method of granting relaying privileges; it does not invalidate the other traditional methods.

A distinguished name may contain characters that require special formatting before the DN is entered in the access database. See the Discussion of Recipe Recipe 8.5 for information on this topic.

See Also

The m4 configuration lines in this recipe are identical to those used in Recipe 8.5. Both use the access database and both are full STARTTLS configurations. See Recipe 8.5 for additional information. Chapter 3 discusses the traditional techniques used to grant relaying privileges, including the access database. The content of a certificate, including the Issuer and Subject fields, is covered in this chapter’s Introduction. The sendmail book discusses using the access database for STARTTLS in Section 10.10.8.

Discussion

Most mail is sent as clear text because the vast majority of systems do not use TLS. Even when STARTTLS is configured, a host sends outbound mail without regard to the level of encryption used. If you require encryption for mail sent to a specific site, you must tell sendmail about that requirement. The access database provides the means for informing sendmail of this requirement.

The TLS_Srv: entry defines the STARTTLS requirements for outbound connections. The format of the entry is:

TLS_Srv:name requirement

TLS_Srv: is the required tag. name is the name of the remote TLS server, which can be defined by domain name or IP address. If the name field is blank, the entry applies to all outbound mail connections. Because most sendmail systems send mail to a wide variety of remote systems, some of which have STARTTLS and some of which do not, the name field is almost never blank. requirement is the keyword that tells sendmail what is required of this TLS connection.

The ENCR: keyword in the requirement field tells sendmail that encryption is required. The keyword is followed by the minimum number of encryption bits that must be used. Here is a sample access database entry:

TLS_Srv:chef.wrotethebook.com     ENCR:168

This entry tells sendmail that when a connection is made to chef.wrotethebook.com, the connection must be encrypted with at least 168-bit encryption.

168-bits is the default level of encryption used on our sample STARTTLS connection. You can check the level of encryption used on your link by sending a test message and looking at the Received: header in the message. For example:

Received: from rodent.wrotethebook.com (rodent.wrotethebook.com
    [192.168.0.3])
        by chef.wrotethebook.com (8.12.9/8.12.9) with ESMTP id
    h13GqDl5001443
        (version=TLSv1/SSLv3 cipher=EDH-RSA-DES-CBC3-SHA bits=168 verify=OK)
        for <craig@chef.wrotethebook.com>; Mon, 3 Feb 2003 11:52:14 -0500

This header tells you that 168-bits encryption is used (bits=168). If the TLS_Srv: entry shown above was installed on rodent, mail would flow to chef because the link between the two systems has a sufficient level of encryption. However, if the link between the two systems did not meet the minimum standard of encryption, the client would abort the connection before sending the mail, as this test demonstrates:

# sendmail -Am -v -t
               To: craig@chef.wrotethebook.com
               From: craig@rodent.wrotethebook.com
               Subject: Test ENCR:168 with no TLS link

               Ctrl-D
craig@chef.wrotethebook.com... Connecting to chef.wrotethebook.com. via esmtp...
220 chef.wrotethebook.com ESMTP Sendmail 8.12.9/8.12.9; Fri, 22 Aug
2003 12:01:37 -
0400
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-DELIVERBY
250 HELP
>>> QUIT
221 2.0.0 chef.wrotethebook.com closing connection
craig@chef.wrotethebook.com... Deferred: 403 4.7.0 encryption too weak 0 less than 
168
Closing connection to chef.wrotethebook.com.

In this case, the server fails to offer STARTTLS, while the client’s access database requires 168-bit encryption on this link. The client terminates the connection with an “encryption too weak” error message. The message displays the number of encryption bits used by the server (0 in this case because TLS did not start) and the number required by the client (168 in this example). The SMTP response code of this error message is 403, and the DSN code is 4.7.0. Each code starts with a 4, which indicates that this is a temporary failure. The keyword “Deferred” in the message indicates that sendmail will attempt to deliver this message again during the next queue run.

Treating a STARTTLS failure as a temporary failure, which is the default, is generally correct. Normally, a server is identified in a TLS_Srv: entry only when the server actually offers STARTTLS. If the server fails to successfully create a TLS connection with the client, the failure is probably temporary. However, it is up to you to decide whether the failure should be considered temporary. The requirement field in a TLS_Srv: entry can be preceded by the string TEMP+ to select a temporary failure, which is the default, or by the string PERM+ to select a permanent error. For example, assume the access database on rodent contained the following TLS_Srv: entry:

TLS_Srv:chef.wrotethebook.com         PERM+ENCR:168

If chef.wrotethebook.com does not offer at least 168-bit encryption, the client terminates the session and returns the mail to the sender. No more attempts are made to deliver the mail. The sender sees the following error in the returned message:

----- The following addresses had permanent fatal errors -----
craig@chef.wrotethebook.com
    (reason: 503 5.7.0 encryption too weak 0 less than 168)

Use the define TLS_PERM_ERR as an alternative to adding PERM+ to the requirement field. The following line added to the sendmail configuration changes the default from a temporary failure to a permanent failure:

define(`TLS_PERM_ERR')

With this setting, individual actions must be flagged as temporary failures using TEMP+.

The search for TLS_Srv: access database entries occurs in the tls_server ruleset, which is called by the client when the client is ready to issue the STARTTLS command. The tls_server ruleset is called with the value returned by ${verify} in its workspace. The tls_server ruleset makes three attempts to find a TLS_Srv: match. It first uses ruleset D to find a TLS_Srv: entry that matches the hostname returned by ${server_name}. If no match is found, it then uses ruleset A to find a TLS_Srv: entry that matches the value returned by ${server_addr}. If there is still no match, it looks for a TLS_Srv: entry with a blank name field. These searches process all possible matches for a given domain name and IP address. For example, for a connection to foo.bar.example.com with IP address 192.168.23.45, the following lookups would happen, in the order listed, until a match was found:

TLS_Srv:foo.bar.example.com
TLS_Srv:bar.example.com
TLS_Srv:example.com
TLS_Srv:com
TLS_Srv:192.168.23.45
TLS_Srv:192.168.23
TLS_Srv:192.168
TLS_Srv:192
TLS_Srv:

The ruleset TLS_connection is called and passed anything returned by the access database lookups. It is TLS_connection that processes the ENCR: requirement field and compares the number of bits from that field to the number of bits returned by ${cipher_bits}. If the macro ${cipher_bits} is less than the value from the requirement field, an error message is displayed and the connection ends.

The m4 macro LOCAL_TLS_SERVER allows you to add custom sendmail.cf rewrite rules to the beginning of the tls_server process. Put the LOCAL_TLS_SERVER macro at the end of the master configuration file and list the custom rewrite rules after the macro.

Recipe 8.8 and Recipe 8.9 cover related access database entries. In particular, VERIFY: is an alternative action keyword that can be used to implement this recipe. See Recipe 8.9 for details on the VERIFY: keyword. Recipe 8.5 and Recipe 8.6 also describe how the access database is used with STARTTLS. The basic STARTTLS recipe upon which this recipe is built is covered in Recipe 8.4. The sendmail book covers the use of the access database with STARTTLS in Section 10.10.8.

8.8. Requiring Inbound Encryption

Discussion

The TLS_Clt: access database entry uses the same format as the TLS_Srv: entry covered in the Discussion of Recipe 8.7. All of the fields described for the TLS_Srv: entry—name, requirement, TEMP+, and PERM+—are used in exactly the same way for the TLS_CLT: entry. The difference is that the TLS_Clt: entry is used by the server to check for any special requirements before accepting mail from the client. The name field of the TLS_Clt: entry defines the domain name or IP address of the remote TLS client. When the name or address of a client matches a TLS_Clt: entry, the requirement field of the entry is used to define the conditions imposed on the client before mail is accepted. A sample TLS_Clt: entry is:

TLS_Clt:rodent.wrotethebook.com     ENCR:168

This entry tells sendmail that a connection from rodent.wrotethebook.com must be encrypted with at least 168-bit encryption. If it isn’t, mail from the client is rejected, as this test from the client rodent shows:

# sendmail -Am -v -t
               To: craig@chef.wrotethebook.com
               From: craig@rodent.wrotethebook.com
               Subject: Test the TLS_Clt: entry

               Ctrl-D
craig@chef.wrotethebook.com... Connecting to chef.wrotethebook.com. via esmtp...
220 chef.wrotethebook.com ESMTP Sendmail 8.12.9/8.12.9; Fri, 22 Aug 2003 12:01:37 -
0400
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-STARTTLS
250-DELIVERBY
250 HELP
>>> STARTTLS
220 2.0.0 Ready to start TLS
>>> EHLO rodent.wrotethebook.com
250-chef.wrotethebook.com Hello rodent.wrotethebook.com [192.168.0.3], pleased to 
meet you
250 ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-DELIVERBY
250 HELP
>>> MAIL From:<craig@rodent.wrotethebook.com>
403 4.7.0 encryption too weak 56 less than 168
craig@chef.wrotethebook.com... Deferred: 403 4.7.0 encryption too weak 56 less than 
168
Closing connection to chef.wrotethebook.com.
>>> QUIT
221 2.0.0 chef.wrotethebook.com closing connection

In this test, the error message “encryption too weak” is in response to the client’s MAIL From: command. The server checks the access database TLS_Clt: entries in the tls_client ruleset. That ruleset is called at two different points in the SMTP protocol exchange. First, it is called when the STARTTLS command is received, and it is called again when the MAIL From command is received. When the tls_client ruleset is called, it is passed the value returned by ${verify}, the literal $|, and either of the keywords STARTTLS or MAIL.

The tls_client ruleset is very similar to the tls_server ruleset described in Recipe 8.7. tls_client first uses ruleset D to look for a TLS_Clt: entry that matches the value found in ${client_name}. If no match is found, it uses ruleset A to look for a TLS_Clt: entry that matches the address in the ${client_addr} macro. If still no match is found, tls_client looks for a TLS_Clt: entry with a blank name field. tls_client would do the following lookups for a connection from a client named foo.bar.example.com that had the IP address 192.168.23.45, until a match was found:

TLS_Clt:foo.bar.example.com
TLS_Clt:bar.example.com
TLS_Clt:example.com
TLS_Clt:com
TLS_Clt:192.168.23.45
TLS_Clt:192.168.23
TLS_Clt:192.168
TLS_Clt:192
TLS_Clt:

tls_client then calls TLS_connection and passes that ruleset anything it has obtained from the access database. The TLS_connection ruleset checks to see if the connection meets the requirements defined by the TLS_Clt: record.

The m4 macro LOCAL_TLS_CLIENT provides a hook into the tls_client ruleset. Use the macro to add custom sendmail.cf rewrite rules to the beginning of the tls_client process.

Recipe 8.7 and Recipe 8.9 cover related access database entries. In particular, VERIFY: is an alternative action keyword that can be used to implement this recipe. See Recipe 8.9 for details on the VERIFY: keyword. Recipe 8.5 and Recipe 8.6 also describe how the access database is used with STARTTLS. The STARTTLS recipe upon which this recipe is built is covered in Recipe 8.4. The sendmail book covers the use of the access database with STARTTLS in Section 10.10.8.

8.9. Requiring a Verified Certificate

The system must have OpenSSL installed and sendmail compiled with STARTTLS support before attempting to implement this recipe. See this chapter’s Introduction and Chapter 1 for more information about OpenSSL and about recompiling sendmail.

Create an access database entry for each host that is required to provide a verified certificate. Use the TLS_Clt: tag if the remote host initiates the connection into the local host, or use the TLS_Srv: tag if the local host initiates the connection to the remote host. Use the VERIFY keyword in the return field to require a validated certificate from the remote system. Add a colon and a numeric value after the VERIFY keyword to require a level of encryption in addition to the verified certificate.

Create a STARTTLS sendmail configuration that includes the access_db feature. Here are sample lines that might be added to the sendmail configuration:

dnl Point to the CA certificate directory
define(`confCACERT_PATH', `/etc/mail/certs')
dnl Point to the root CA's certificate
define(`confCACERT', `/etc/mail/certs/cacert.pem')
dnl Point to the certificate used for inbound connections
define(`confSERVER_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for inbound connections
define(`confSERVER_KEY', `/etc/mail/certs/key.pem')
dnl Point to the certificate used for outbound connections
define(`confCLIENT_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for outbound connections
define(`confCLIENT_KEY', `/etc/mail/certs/key.pem')
dnl Enable the access database
FEATURE(`access_db')

Rebuild the sendmail.cf file, copy the new sendmail.cf file to /etc/mail, and restart sendmail, as shown in Recipe 1.8.

Discussion

When both end points are properly configured to support STARTTLS, sendmail uses TLS to encrypt the mail transport even if the end points do not present verified certificates. In this context, a verified certificate is one signed by a CA that sendmail trusts. If sendmail can verify the signature on the certificate, sendmail sets the ${verify} macro to OK. When a certificate is received that cannot be verified, ${verify} is set to one of several possible values, all of which are covered in Introduction of this chapter. For example, NO is one of the values indicating that the certificate was not verified. Yet TLS is still used to encrypt the link even when ${VERIFY} is set to NO, as can be seen in the Received: header shown below:

Return-Path: <craig@horseshoe.wrotethebook.com>
Received: from horseshoe.wrotethebook.com (horseshoe.wrotethebook.com [192.168.0.2])
        by chef.wrotethebook.com (8.12.9/8.12.9) with ESMTP id h14HeQAd001228
        (version=TLSv1/SSLv3 cipher=EDH-RSA-DES-CBC3-SHA bits=168 verify=NO)
        for <craig@chef.wrotethebook.com>; Tue, 4 Feb 2003 12:40:27 -0500
Received: (from root@localhost)
        by horseshoe.wrotethebook.com (8.12.9/8.12.9) id h14Hnn91001180;
        Tue, 4 Feb 2003 12:49:49 -0500
Date: Tue, 4 Feb 2003 12:49:49 -0500
Message-Id: <200302041749.h14Hnn91001180@horseshoe.wrotethebook.com>
To: craig@chef.wrotethebook.com
From: craig@horseshoe.wrotethebook.com
Subject: Test self-signed certificate

The first Received: header shows that the link was encrypted with 168-bit encryption, and that no certificate was verified for horseshoe.wrotethebook.com. The implications of this are two-fold:

  • First, it is possible to use STARTTLS without recourse to a certificate authority if the only thing you want to do is encrypt the link.

  • Second, optional configuration is required to limit STARTTLS connections to those system with verifiable certificates.

The VERIFY requirement is used with TLS_Clt: or TLS_Srv: to require a verified certificate. When VERIFY is specified, the current value stored in the ${verify} macro must be OK. Otherwise, the connection is rejected. Here are some example access database entries:

TLS_Clt:crab.wrotethebook.com          VERIFY
TLS_Srv:smtp.wrotethebook.com          VERIFY:168
TLS_Clt:horseshoe.wrotethebook.com     VERIFY:168

The first access database entry only accepts inbound mail from crab.wrotethebook.com if the certificate crab presents is verified. The TLS_Clt: entries are processed by the tls_client ruleset. See Recipe 8.8 for a description of that ruleset.

The second access database entry tells sendmail that an outbound connection made to the server smtp.wrotethebook.com must use 168-bit encryption and that the certificate presented by the server must be verified. If either of these conditions is not met, the connection is terminated. The TLS_Srv: entry is processed by the tls_server ruleset covered in Recipe 8.7.

The last entry requires 168-bit encryption and a verified certificate for the TLS client horseshoe.wrotethebook.com. If chef.wrotethebook.com had been configured with that entry, the mail that created the Received: header shown at the beginning of this discussion would not have been accepted. As the Received: header made clear, horseshoe used 168-bit encryption, but it did not present a verifiable certificate. A test run from horseshoe shows the impact of adding this TLS_Clt: entry to the configuration of the receiving host:

# sendmail -Am -v -t
               To: craig@chef.wrotethebook.com
               From: craig@horseshoe.wrotethebook.com
               Subject: Test VERIFY:168

               Ctrl-D
craig@chef.wrotethebook.com... Connecting to chef.wrotethebook.com. via esmtp...
220 chef.wrotethebook.com ESMTP Sendmail 8.12.9/8.12.9; Fri, 22 Aug 2003 12:01:37 -
0400
>>> EHLO horseshoe.wrotethebook.com
250-chef.wrotethebook.com Hello horseshoe.wrotethebook.com [192.168.0.2], pleased to 
meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-STARTTLS
250-DELIVERBY
250 HELP
>>> STARTTLS
220 2.0.0 Ready to start TLS
>>> EHLO horseshoe.wrotethebook.com
250-chef.wrotethebook.com Hello horseshoe.wrotethebook.com [192.168.0.2], pleased to 
meet you
250 ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH DIGEST-MD5 CRAM-MD5
250-DELIVERBY
250 HELP
>>> MAIL From:<craig@horseshoe.wrotethebook.com>
403 4.7.0 not authenticated
craig@chef.wrotethebook.com... Deferred: 403 4.7.0 not authenticated
Closing connection to chef.wrotethebook.com.
>>> QUIT
221 2.0.0 chef.wrotethebook.com closing connection

The STARTTLS command is issued and accepted. The second EHLO command indicates the successful start of the encrypted session. However, the response to the MAIL From: command is the error “not authenticated” because a TLS_Clt: entry in the access database on chef required a certificate signed by a trusted certificate authority.

Notice that the response codes 403 and 4.7.0 are used on the error message. The 4 at the start of each code indicates that this is a temporary failure. In response to these temporary failures, the client will queue the message and attempt to deliver it during subsequent queue runs. By default, errors generated by a VERIFY requirement are temporary errors. Use PERM+ to make the error a permanent failure in order to prevent the remote system from resending the message at each queue run. For example:

TLS_Clt:horseshoe.wrotethebook.com     PERM+VERIFY:168

See Recipe 8.7 for more information on PERM+, TEMP+, and the define TLS_PERM_ERR, all of which control whether TLS errors are permanent or temporary.

The test shown above points out the difference between encryption and authentication. Encryption can be used even if the end points have not been authenticated. Therefore, special privileges, such as relaying, cannot be extended on the basis of the level of encryption used. Special privileges require authentication. Happily, TLS can provide both encryption and authentication when they are needed.

See Also

Recipe 8.7 and Recipe 8.8 provide additional examples of the TLS_Clt: and TLS_Srv: access database entries. Recipe 8.5 and Recipe 8.6 describe other ways the access database is used with STARTTLS. This recipe contains m4 commands described in Recipe 8.4. The sendmail book covers the use of the access database with STARTTLS in Section 10.10.8.

8.10. Requiring TLS for a Recipient

Install OpenSSL and recompile sendmail with STARTTLS support as described in the Introduction and in Recipe 1.6 and Recipe 1.7.

Use TLS_Rcpt: access database entries to define those recipients that require TLS security and to specify the level of security required. The key field of each entry contains the tag TLS_Rcpt: and the full or partial address of the recipient. The return value defines the level of security required, and it can be either the keyword ENCR:, followed by the number of bits of encryption required, or the keyword VERIFY, optionally followed by a number of bits of encryption.

Add the access_db feature and the STARTTLS defines to the sendmail configuration. Here are sample lines that could be added to the configuration:

dnl Point to the CA certificate directory
define(`confCACERT_PATH', `/etc/mail/certs')
dnl Point to the root CA's certificate
define(`confCACERT', `/etc/mail/certs/cacert.pem')
dnl Point to the certificate used for inbound connections
define(`confSERVER_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for inbound connections
define(`confSERVER_KEY', `/etc/mail/certs/key.pem')
dnl Point to the certificate used for outbound connections
define(`confCLIENT_CERT', `/etc/mail/certs/cert.pem')
dnl Point to the private key used for outbound connections
define(`confCLIENT_KEY', `/etc/mail/certs/key.pem')
dnl Enable the access database
FEATURE(`access_db')

Rebuild the sendmail.cf file, copy the new sendmail.cf file to /etc/mail, and restart sendmail. See Recipe Recipe 1.8 for an example.

Discussion

The TLS_Clt: and TLS_Srv: access database entries used in Recipe 8.7, Recipe 8.8, and Recipe 8.9 enforce minimum TLS security requirements based on the name or address of the remote host. sendmail checks for ${client_name} and ${client_addr} in TLS_CLT: entries for inbound connections, and for ${server_name} and ${server_addr} in TLS_Srv: entries for outbound connections. These entries ensure that connections between the specified remote hosts and local host are secured, but these entries do not guarantee that mail addressed to a specific user will be secure. There is no guarantee because mail can reach recipients over multiple paths—some of which may be secured and others may not. Look at some sample MX records:

wrotethebook.com.     IN MX     10  chef.wrotethebook.com.
wrotethebook.com.     IN MX     20  rodent.wrotethebook.com.
wrotethebook.com.     IN MX     30  mail.example.com.

Given these three MX records, mail bound for pat@wrotethebook.com could go over a connection to chef, rodent, or mail.example.com. Assume that chef and rodent have STARTTLS support but that mail.example.com does not. Mail sent from our system to pat@wrotethebook.com would be encrypted if the delivery connection was made to chef or rodent, but the mail would be sent as clear text if the connection was made to mail.example.com. We don’t run mail.example.com, so we can’t change the way it is configured—what we can do is ensure that mail sent to pat@wrotethebook.com only goes via an adequately encrypted connection. The following TLS_Rcpt: entry ensures that mail is only sent to pat@wrotethebook.com if the remote host has presented a verified certificate and the link is using 168-bit encryption:

TLS_Rcpt:pat@wrotethebook.com     VERIFY:168

The basic format of the TLS_Rcpt: entry is:

TLS_Rcpt:recipient     
               requirements+suffix++suffix

The string TLS_Rcpt: is the required tag for this database record. recipient is the full or partial email address of the recipient. The recipient address can be:

  • A full address in the format user @ host.domain. This format matches a specific user on a specific host. TLS_Rcpt:craig@crab.wrotethebook.com is an example.

  • A hostname in the format host.domain; e.g., TLS_Rcpt:crab.wrotethebook.com. Any user on the specified host is matched.

  • A domain name to match every user on every host in the specified domain. For example, TLS_Rcpt:wrotethebook.com would match every recipient in the wrotethebook.com domain.

  • A username written in the form user @; e.g., TLS_Rcpt:craig@. This format matches any recipient with the specified username on any host in any domain.

  • A blank field, which matches every possible recipient address. This format is used to define a default security requirement for all outbound mail.

The requirements field defines the minimum security requirements that must be met before mail is sent to the recipient. The keyword ENCR or VERIFY can be used in the requirements field to require a level of encryption or a verified certificate. For example ENCR:168 requires a minimum of 168-bit encryption, although it does not require a verified certificate from the remote server. The VERIFY keyword does require a verified certificate and optionally can be combined with a numeric value to also require a level of encryption. For example, VERIFY:168 requires a verified certificate from the remote server and 168-bit encryption on the link. Recipe 8.7, Recipe 8.8, and Recipe 8.9 provide examples of the ENCR and VERIFY keywords. The sample TLS_Rcpt: entry just shown uses VERIFY:168 to require authentication and encryption.

The impact of the sample entry is seen when mail addressed to pat@wrotethebook.com is routed to the backup MX server mail.example.com:

# sendmail -Am -v -t
               To: pat@wrotethebook.com
               From: david@wrotethebook.net
               Subject: Test TLS_Rcpt:

Ctrl-D
pat@wrotethebook.com... Connecting to mail.example.com. via esmtp...
220 mail.example.com ESMTP Sendmail 8.12.9/8.12.9; Wed, 5 Feb 2003 15:33:02 -0500
>>> EHLO wrotethebook.net
250-mail.example.com Hello IDENT:24znrK/hAUFBK67n3St2d8DU/5bqb70s@chef [192.168.10.
8], pleased to meet you
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-EXPN
250-VERB
250-8BITMIME
250-SIZE
250-DSN
250-ETRN
250-AUTH GSSAPI DIGEST-MD5 CRAM-MD5
250-DELIVERBY
250 HELP
>>> MAIL From:<david@wrotethebook.net> SIZE=92 AUTH=david@wrotethebook.net
250 2.1.0 <david@wrotethebook.net>... Sender ok
pat@wrotethebook.com... Deferred
>>> RSET
250 2.0.0 Reset state
Closing connection to mail.example.com.
>>> QUIT
221 2.0.0 mail.example.com closing connection

The backup MX server mail.example.com does not offer the STARTTLS service in its response to the EHLO command. Therefore, the client does not issue a STARTTLS command. Instead, the client issues the MAIL From: command. The client then checks for a TLS_Rcpt: entry in the access database, finds it, and terminates the session with the message:

pat@wrotethebook.com ... Deferred

This message indicates that the mail has been queued and will be sent again during the next queue run, which means that sendmail is treating this delivery problem as a temporary error. By default, most TLS connection problems are treated as temporary failures. You can change by using the TLS_PERM_ERR define in the master configuration file, or you can override the default for a specific TLS_Rcpt: entry by adding PERM+ to the entry. For example, the following entry would make the failure shown above a permanent failure:

TLS_Rcpt:pat@wrotethebook.com     PERM+VERIFY:168

However, it is not a good idea to make this a permanent failure. Failures caused by TLS_Rcpt: entries should be temporary failures. The failure happened because the connection went to a backup MX server that doesn’t support STARTTLS. When the mail is resent during the queue run, it will probably go to the primary MX server, which does have STARTTLS, and therefore, the mail will be successfully delivered. Making the failure permanent undermines the error recovery potential inherent in the sendmail queue process. It is more likely that you would use TEMP+ to make the failure temporary if, for some reason, the default for TLS failures had been set to permanent using the TLS_PERM_ERR define. Recipe 8.7, Recipe 8.8, and Recipe 8.9 cover TLS_PERM_ERR, PERM+, and TEMP+.

The other access database entries covered in this chapter also use the ENCR:, VERIFY:, PERM+, and TEMP+ keywords. However, the TLS_Rcpt: record has more options than these other access entries. The requirements field of the TLS_Rcpt: entry can be followed by one or more optional values. These options are called suffixes because they are added to the TLS_Rcpt: record after the ENCR or VERIFY keyword. The first suffix is added after a single plus sign (+), and subsequent suffixes are added with two plus signs (++). The possible suffixes are:

Here is a sample TLS_Rcpt: entry with multiple suffixes:[4]

TLS_Rcpt:pat@wrotethebook.com VERIFY:168+CN++CI:/C=US/ST=Maryland/L=Gaithersburg/
O=WroteTheBook/CN=chef.wrotethebook.com/

This TLS_Rcpt: entry requires 168-bit encryption and a verified certificate from the remote host. The common name of the subject of the certificate must be the hostname of the server at the other end of the current connection, and the certificate must be signed by the private CA created on chef.wrotethebook.com in Recipe 8.1.

The various options available for the TLS_Rcpt: entry make it possible to handle most potential situations. Yet, it is also possible to create a custom ruleset to do your own TLS_Rcpt: processing using the Local_tls_rcpt ruleset hook.

See Also

Recipe 8.7, Recipe 8.8, and Recipe 8.9 show examples of related access database entries, and Recipe 8.5 and Recipe 8.6 describe other ways the access database is used with STARTTLS. The sendmail book covers ${cn_subject} in Section 21.9.24, ${server_name} in 21.9.84, ${cert_subject} in 21.9.15, ${cert_issuer} in 21.9.13, and TLS_Rcpt: in 10.10.8.3.

8.11. Refusing STARTTLS Service

See Also

Recipe 8.4 explains the STARTTLS defines used in this configuration. Recipe 8.5 to Recipe 8.10 show examples of using the access database with STARTTLS. The sendmail book covers ${server_addr} in Section 21.9.83, ${server_name} in 21.9.84, and Try_TLS: records in 10.10.8.4.

8.12. Selectively Advertising STARTTLS

Discussion

Use the Srv_Features: access database entry to control the extended SMTP features offered to connecting hosts. The return value of a Srv_Features: entry is one or more single character flags, some of which indicate whether an extended service should be advertised. (When more than one flag is used, the individual flags are separated by whitespace.) Lowercase flags enable SMTP extensions and uppercase flags disable extensions. The s flag selectively advertises STARTTLS, and the S flag selectively blocks advertising STARTTLS. These are the flags of particular interest for this recipe.[5]

Assume that you want to limit advertising STARTTLS to only those hosts in the wrotethebook.com domain. You could do that by adding the following two entries to the access database:

Srv_Features:wrotethebook.com             s
Srv_Features:                             S

In this case, STARTTLS is advertised to the hosts in the wrotethebook.com domain as directed by the s flag. The S flag in the second entry prevents STARTTLS from being advertised to any other host. A blank name field in a Srv_Features: entry indicates that the entry applies to every host for which there is not a more specific Srv_Features: entry. In effect, the second entry creates a default policy for advertising STARTTLS. This second entry is necessary because the normal default is for sendmail to advertise STARTTLS when STARTTLS is configured. In the absence of an applicable Srv_Features: entry, the default sendmail behavior applies.

Recipe 8.4 describes the basic STARTTLS configuration that needs to be done before this recipe is implemented. Recipe 7.9 and Recipe Recipe 8.13 provide other examples of using the Srv_Features: record. Recipe 8.5 through Recipe 8.13 show how the access database is used with STARTTLS. The sendmail book covers the Srv_Features: entry in Section 19.9.4.

8.13. Requesting Client Certificates

Discussion

Srv_Features: entries control the way in which a server interacts with its clients. See Recipe 8.12 and Recipe 7.9 for more general information about the Srv_Features: entry.

Use the v return value on a Srv_Features: entry to request the TLS client’s certificate. Use the V return value to specifically prevent sendmail from asking for the client’s certificate.

Adding the following entry to the access database requires a certificate from all hosts connecting to the local host from the dialin.wrotethebook.com domain:

Srv_Features:dialin.wrotethebook.com          v

By itself, the Srv_Features: entry does not do anything with the client certificate. It merely requests that the client present a certificate, which can then be used for special purposes such as authorizing relaying. Using the certificate for a special purpose requires additional configuration as shown in other recipes in this chapter.

Recipe 8.4 describes the STARTTLS defines used in this configuration. Recipe 7.9 provides other examples of using the Srv_Features: record. Recipe 8.5, Recipe 8.6, and Recipe 8.9 use client certificates to grant relaying privileges. The sendmail book covers the Srv_Features: entry in Section 19.9.4.



[1] For the SMTP connection, the TLS client is the system initiating the connection, and the TLS server is the system to which the connection is being made.

[2] The CA script is found in the misc subdirectory of the OpenSSL ssl directory. On our sample Red Hat Linux system, the full directory path is /usr/share/ssl/misc; on some other systems it is /usr/local/ssl/misc.

[3] Because of PIPELINING, which the EHLO response shows is offered on this system, some SMTP commands are sent together to save on TCP roundtrip times. For this reason, the response to the RCPT To: command follows the DATA command in this example.

[4] This is one long line with no line breaks.

[5] All of the flags are described in Table 7-2, which is shown in the Discussion section of Recipe 7.9.