client-auth.sgml

<!-- $Header: /cvsroot/pgsql/doc/src/sgml/client-auth.sgml,v 1.3 2000/07/15 21:35:47 petere Exp $ -->

<chapter id="client-authentication">
 <title>Client Authentication</title>

 <para>
  User names from the operating system and from a
  <productname>Postgres</productname> database installation are
  logically separate. When a client application connects, it specifies
  which database user name it wants to connect as, similar to how one
  logs into a Unix computer. Within the SQL environment the active
  database user name determines various access privileges to database
  objects -- see <xref linkend="user-manag"> for more information
  about that. It is therefore obviously essential to restrict what
  database user name a given client can connect as.
 </para>

 <para>
  <firstterm>Authentication</firstterm> is the process by which the
  database server establishes the identity of the client, and by
  extension determines whether the client application (or the user
  which runs the client application) is permitted to connect with the
  user name that was requested.
 </para>

 <para>
  <productname>Postgres</productname> offers client authentication by
  (client) host and by database, with a number of different
  authentication methods available.
 </para>

 <sect1 id="pg-hba.conf">
  <title>The <filename>pg_hba.conf</filename> file</title>

  <para>
   Client authentication is controlled by the file
   <filename>pg_hba.conf</filename> in the data directory, e.g.,
   <filename>/usr/local/pgsql/data/pg_hba.conf</filename>. (HBA =
   host-based authentication) A default file is installed when the
   data area is initialized by <application>initdb</application>.
  </para>

  <para>
   The general format of the <filename>pg_hba.conf</filename> file is
   of a set of records, one per line. Blank lines and lines beginning
   with a hash character (<quote>#</quote>) are ignored. A record is
   made up of a number of fields which are separated by spaces and/or
   tabs.
  </para>

  <para>
   A record may have one of the two formats
   <synopsis>
local <replaceable>database</replaceable> <replaceable>authentication-method</replaceable> [ <replaceable>authentication-option</replaceable> ]
host <replaceable>database</replaceable> <replaceable>IP-address</replaceable> <replaceable>IP-mask</replaceable> <replaceable>authentication-method</replaceable> [ <replaceable>authentication-option</replaceable> ]
    </synopsis>
   The meaning of the fields is as follows:

   <variablelist>
    <varlistentry>
     <term><literal>local</literal></term>
     <listitem>
      <para>
       This record pertains to connection attempts over Unix domain
       sockets.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term><literal>host</literal></term>
     <listitem>
      <para>
       This record pertains to connection attempts over TCP/IP
       networks. Note that TCP/IP connections are completely disabled
       unless the server is started with the <option>-i</option> or
       the equivalent configuration parameter is set.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term><replaceable>database</replaceable></term>
     <listitem>
      <para>
       Specifies the database that this record applies to. The value
       <literal>all</literal> specifies that it applies to all
       databases.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term><replaceable>IP address</replaceable></term>
     <term><replaceable>IP mask</replaceable></term>
     <listitem>
      <para>
       These two fields control to which hosts a
       <literal>host</literal> record applies, based on their IP
       address. (Of course IP addresses can be spoofed but this
       consideration is beyond the scope of
       <productname>Postgres</productname>.) The precise logic is that
       <blockquote>
        <informalfigure>
         <programlisting>(<replaceable>actual-IP-address</replaceable> xor <replaceable>IP-address-field</replaceable>) and <replaceable>IP-mask-field</replaceable></programlisting>
        </informalfigure>
       </blockquote>
       must be zero for the record to match.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term><replaceable>authentication method</replaceable></term>
     <listitem>
      <para>
       Specifies the method a user must use to authenticate themselves
       when connecting to that database.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term><replaceable>authentication option</replaceable></term>
     <listitem>
      <para>
       This field is interpreted differently depending on the
       authentication method.
      </para>
     </listitem>
    </varlistentry>
   </variablelist>

   The first record that matches a connection attempt is used. Note
   that there is no <quote>fall-through</quote> or
   <quote>backup</quote>, that is, if one record is chosen and the
   authentication fails, the following records are not considered. If
   no record matches, the access will be denied.
  </para>

  <para>
   The <filename>pg_hba.conf</filename> file is re-read before each
   connection attempt. It is therefore easily possible to modify
   access permissions while the server is running.
  </para>

  <para>
   An example of a <filename>pg_hba.conf</filename> file is shown in
   <xref linkend="example-pg-hba.conf">. See below for details on the
   different authentication methods.

   <example id="example-pg-hba.conf">
    <title>An example <filename>pg_hba.conf</filename> file</title>
<programlisting>
# Trust any connection via Unix domain sockets.
local	trust
# Trust any connection via TCP/IP from this machine.
host	all	127.0.0.1	255.255.255.255		trust
# We don't like this machine.
host	all	192.168.0.10	255.255.255.0		reject
# This machine can't encrypt so we ask for passwords in clear.
host	all	192.168.0.3	255.255.255.0		password
# The rest of this group of machines should provide encrypted passwords.
host	all	192.168.0.0	255.255.255.0		crypt
# Authenticate these networks using ident
host    all     192.168.1.0     255.255.255.0           ident usermap
host    all     192.168.2.0     255.255.255.0           ident othermap
</programlisting>
   </example>
  </para>
 </sect1>

 <sect1 id="auth-methods">
  <title>Authentication methods</title>
  <para>
   The following authentication methods are supported. They are
   descibed in detail below.

   <variablelist>
    <varlistentry>
     <term>trust</term>
     <listitem>
      <para>
       The connection is allowed unconditionally. This method allows
       any user that has login access to the client host to connect as
       any user whatsoever. Use with care.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term>reject</term>
     <listitem>
      <para>
       The connection is rejected unconditionally. This is mostly
       useful to <quote>filter out</quote> certain hosts from a group.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term>password</term>
     <listitem>
      <para>
       The client is required to supply a password with the connection
       attempt which is required to match the password that was set up
       for the user.
      </para>
      <para>
       An optional file name may be specified after the
       <literal>password</literal> keyword. This file is expected to
       contain a list of users that this record pertains to, and
       optionally alternative passwords.
      </para>
      <para>
       The password is sent over the wire in clear text. For better
       protection, use the <literal>crypt</literal> method.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term>crypt</term>
     <listitem>
      <para>
       Like the <literal>password</literal> method, but the password
       is sent over the wire encrypted using a simple
       challenge-response protocol. This is still not
       cryptographically secure but it protects against incidental
       wire-sniffing. The name of a file may follow the
       <literal>crypt</literal> keyword that contains a list of users
       that this record pertains to.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term>krb4</term>
     <listitem>
      <para>
       Kerberos V4 is used to authenticate the user. This is only
       available for TCP/IP connections.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term>krb5</term>
     <listitem>
      <para>
       Kerberos V5 is used to authenticate the user. This is only
       available for TCP/IP connections.
      </para>
     </listitem>
    </varlistentry>

    <varlistentry>
     <term>ident</term>
     <listitem>
      <para>
       The ident server on the client host is asked for the identity
       of the connecting user. <productname>Postgres</productname>
       then verifies whether the so identified operating system user
       is allowed to connect as the database user that is requested.
       The <replaceable>authentication option</replaceable> following
       the <literal>ident</> keyword specifies the name of an
       <firstterm>ident map</firstterm> that specifies which operating
       system users equate with which database users. See below for
       details.
      </para>
     </listitem>
    </varlistentry>
   </variablelist>
  </para>

  <sect2>
   <title>Password authentication</title>
   <para>
    <productname>Postgres</> database passwords are separate from any
    operating system user passwords. Ordinarily, the password for each
    database user is stored in the pg_shadow system catalog table.
    Passwords can be managed with the query language commands
    <command>CREATE USER</command> and <command>ALTER USER</command>,
    e.g., <userinput>CREATE USER foo WITH PASSWORD
    'secret';</userinput>. By default, that is, if no password has
    explicitly been set up, the stored password is <quote>NULL</quote>
    and password authentication will always fail for that user.
   </para>

   <para>
    To restrict the set of users that are allowed to connect to
    certain databases, list the set of users in a separate file (one
    user name per line) in the same directory that
    <filename>pg_hba.conf</> is in, and mention the (base) name of the
    file after the <literal>password</> or <literal>crypt</> keyword,
    respectively, in <filename>pg_hba.conf</>. If you do not use this
    feature, then any user that is known to the database system can
    connect to any database (as long as he passes password
    authentication, of course).
   </para>

   <para>
    These files can also be used a apply a different set of passwords
    to a particular database or set thereof. In that case, the files
    have a format similar to the standard Unix password file
    <filename>/etc/passwd</filename>, that is,
<synopsis>
<replaceable>username</replaceable>:<replaceable>password</replaceable>
</synopsis>
    Any extra colon separated fields following the password are
    ignored. The password is expected to be encrypted using the
    system's <function>crypt()</function> function. The utility
    program <application>pg_passwd</application> that is installed
    with <productname>Postgres</productname> can be used to manage
    these password files.
   </para>

   <para>
    Lines with and without passwords can be mixed in secondary
    password files. Lines without password indicate use of the main
    password in <literal>pg_shadow</> that is managed by
    <command>CREATE USER</> and <command>ALTER USER</>. Lines with
    passwords will cause that password to be used. A password entry of
    <quote>+</quote> also means using the pg_shadow password.
   </para>

   <para>
    Alternative passwords cannot be used when using the
    <literal>crypt</> method. The file will still be evaluated as
    usual but the password field will simply be ignored and the
    <literal>pg_shadow</> password will be used.
   </para>

   <para>
    Note that using alternative passwords like this means that one can
    no longer use <command>ALTER USER</command> to change one's
    password. It will still appear to work but the password one is
    actually changing is not the password that the system will end up
    using.
   </para>

  </sect2>

  <sect2>
   <title>Kerberos authentication</title>

   <para>
    <productname>Kerberos</productname> is an industry-standard secure
    authentication system suitable for distributed computing over a
    public network. A description of the
    <productname>Kerberos</productname> system is far beyond the scope
    of this document; in all generality it can be quite complex (yet
    powerful). The <ulink
    url="http://www.nrl.navy.mil/CCS/people/kenh/kerberos-faq.html">Kerberos
    <acronym>FAQ</></ulink> or <ulink
    url="ftp://athena-dist.mit.edu">MIT Project Athena</ulink> can be
    a good starting point for exploration. Several sources for
    <productname>Kerberos</> distributions exist.
   </para>

   <para>
    In order to use <productname>Kerberos</>, support for it must be
    enable at build time. Both Kerberos 4 and 5 are supported
    (<literal>./configure --with-krb4</> or <literal>./configure
    --with-krb5</> respectively).
   </para>

   <para>
    <productname>Postgres</> should operate like a normal Kerberos
    service. The name of the service principal is normally
    <literal>postgres</literal>, unless it was changed during the
    build. Make sure that your server keytab file is readable (and
    preferrably only readable) by the Postgres server account (see
    <xref linkend="postgres-user">). The location of the keytab file
    is specified at build time; by default it is
    <filename>/etc/srvtab</filename> in Kerberos 4 and
    <filename>FILE:/usr/local/pgsql/etc/krb5.keytab</filename> in
    Kerberos 5.
   </para>

   <para>
    To generate the keytab file, use for example (with version 5)
<screen>
kadmin% <userinput>ank -randkey postgres/server.my.domain.org</>
kadmin% <userinput>ktadd -k krb5.keytab postgres/server.my.domain.org</>
</screen>
    Read the <productname>Kerberos</> documentation for defails.
   </para>

   <para>
    In the <productname>Kerberos</> 5 hooks, the following assumptions
    are made about user and service naming:

    <itemizedlist>
     <listitem>
      <para>
       User principal names (anames) are assumed to contain the actual
       Unix/<productname>Postgres</> user name in the first component.
      </para>
     </listitem>
     <listitem>
      <para>
       The <productname>Postgres</> service is assumed to be have two
       components, the service name and a hostname, canonicalized as
       in Version 4 (i.e., with all domain suffixes removed).
      </para>
     </listitem>
    </itemizedlist>

    <informaltable>
     <tgroup cols="2">
      <thead>
       <row>
        <entry>Parameter</>
	<entry>Example</>
       </row>
      </thead>
      <tbody>
       <row>
	<entry>user</>
	<entry>frew@S2K.ORG</>
       </row>
       <row>
	<entry>user</>
	<entry>aoki/HOST=miyu.S2K.Berkeley.EDU@S2K.ORG</>
       </row>	   
       <row>
	<entry>host</>
	<entry>postgres_dbms/ucbvax@S2K.ORG</>
       </row>
      </tbody>
     </tgroup>
    </informaltable>
   </para>

   <para>
    If you use mod_auth_krb and mod_perl on your Apache web server,
    you can use AuthType KerberosV5SaveCredentials with a mod_perl
    script. This gives secure database access over the web, no extra
    passwords required.
   </para>

  </sect2>

  <sect2>
   <title>Ident-based authentication</title>

   <para>
    The <quote>Identification Protocol</quote> is described in
    <citetitle>RFC 1413</citetitle>. Virtually every Unix-like
    operating systems ships with an ident server that listens on TCP
    port 113 by default. The basic functionality of an ident server
    is to answer questions like <quote>What user initiated the
    connection that goes out of your port <replaceable>X</replaceable>
    and connects to my port <replaceable>Y</replaceable>?</quote>.
    Since <productname>Postgres</> knows both <replaceable>X</> and
    <replaceable>Y</> when a physical connection is established, it
    can interrogate the ident server on the host of the connecting
    client and could theoretically determine the operating system user
    for any given connection this way.
   </para>

   <para>
    The drawback of this procedure is that it depends on the integrity
    of the client: if the client machine is untrusted or compromised
    an attacker could run just about any program on port 113 and
    return any user name he chooses. This authentication method is
    therefore only appropriate for closed networks where each client
    machine is under tight control and where the database and system
    administrators operate in close contact. Heed the warning:
    <blockquote>
     <attribution>RFC 1413</attribution>
     <para>
      The Identification Protocol is not intended as an authorization
      or access control protocol.
     </para>
    </blockquote>
   </para>

   <para>
    When using ident-based authentication, after having determined the
    operating system user that initiated the connection,
    <productname>Postgres</productname> determines as what database
    system user he may connect. This is controlled by the ident map
    argument that follows the <literal>ident</> keyword in the
    <filename>pg_hba.conf</filename> file. The simplest ident map is
    <literal>sameuser</literal>, which allows any operating system
    user to connect as the database user of the same name (if the
    latter exists). Other maps must be created manually.
   </para>

   <para>
    Ident maps are held in the file <filename>pg_ident.conf</filename>
    in the data directory, which contains lines of the general form:
<synopsis>
<replaceable>map-name</> <replaceable>ident-username</> <replaceable>database-username</>
</synopsis>
    Comments and whitespace are handled in the usual way.
    The <replaceable>map-name</> is an arbitrary name that will be
    used to refer to this mapping in <filename>pg_hba.conf</filename>.
    The other two fields specify which operating system user is
    allowed to connect as which database user. The same
    <replaceable>map-name</> can be used repeatedly to specify more
    user-mappings. There is also no restriction regarding how many
    database users a given operating system may correspond to and vice
    versa.
   </para>

   <para>
    A <filename>pg_ident.conf</filename> file that could be used in
    conjunction with the <filename>pg_hba.conf</> file in <xref
    linkend="example-pg-hba.conf"> is shown in <xref
    linkend="example-pg-ident.conf">. In that example setup, anyone
    logged in to a machine on the 192.168.1 network that does not have
    the a user name joe, robert, or ann would not be granted access.
    Unix user robert would only be allowed access when he tries to
    connect as <quote>bob</quote>, not as <quote>robert</quote> or
    anyone else. <quote>ann</quote> and <quote>joe</quote> would only
    be allowed to connect <quote>as themselves</quote>. On the
    192.168.2 network, however, a user <quote>ann</quote> would not be
    allowed to connect at all, only the user <quote>bob</> can connect
    as <quote>bob</> and some user <quote>karl</> can connect as
    <quote>joe</> as well.
   </para>

   <example id="example-pg-ident.conf">
    <title>An example <filename>pg_ident.conf</> file</title>
<programlisting>
usermap         joe     joe
# bob has username robert on these machines
usermap         robert  bob
usermap         ann     ann

othermap        joe     joe
othermap        bob     bob
othermap        karl    joe
</programlisting>
   </example>
  </sect2>
 </sect1>

  <sect1 id="client-authentication-problems">
   <title>Authentication problems</title>

   <para>
    Genuine authentication failures and related problems generally
    manifest themselves through error messages like the following.
   </para>

   <para>
<ProgramListing>
No pg_hba.conf entry for host 123.123.123.123, user joeblow, database testdb
</ProgramListing>
    This is what you are most likely to get if you succeed in
    contacting the server, but it doesn't want to talk to you. As the
    message suggests, the server refused the connection request
    because it found no authorizing entry in its <filename>pg_hba.conf</filename>
    configuration file.
   </para>

   <para>
<ProgramListing>
Password authentication failed for user 'joeblow'
</ProgramListing>
    Messages like this indicate that you contacted the server, and
    it's willing to talk to you, but not until you pass the
    authorization method specified in the
    <filename>pg_hba.conf</filename> file. Check the password you're
    providing, or check your Kerberos or IDENT software if the
    complaint mentions one of those authentication types.
   </para>

   <para>
<ProgramListing>
FATAL 1:  SetUserId: user 'joeblow' is not in 'pg_shadow'
</ProgramListing>
    This is the fancy way of saying that the user doesn't exist at all.
   </para>

   <para>
<ProgramListing>
FATAL 1:  Database testdb does not exist in pg_database
</ProgramListing>
    The database you're trying to connect to doesn't exist. Note that
    if you don't specify a database name, it defaults to the database
    user name, which may or may not be the right thing.
   </para>
  </sect1>

 </chapter>