| < wallet Setup | Russ Allbery > Software > wallet | wallet Objects and ACL Schemes > |
(Configuration handling for the wallet server)
use Wallet::Config;
my $driver = $Wallet::Config::DB_DRIVER;
my $info;
if (defined $Wallet::Config::DB_INFO) {
$info = $Wallet::Config::DB_INFO;
} else {
$info = "database=$Wallet::Config::DB_NAME";
$info .= ";host=$Wallet::Config::DB_HOST"
if $Wallet::Config::DB_HOST;
$info .= ";port=$Wallet::Config::DB_PORT"
if $Wallet::Config::DB_PORT;
}
my $dsn = "dbi:$driver:$info";
my $user = $Wallet::Config::DB_USER;
my $password = $Wallet::Config::DB_PASSWORD;
my $dbh = DBI->connect ($dsn, $user, $password);
Wallet::Config encapsulates all of the site-specific configuration for the wallet server. It is implemented as a Perl class that declares and sets the defaults for various configuration variables and then, if it exists, loads the file specified by the WALLET_CONFIG environment variable or /etc/wallet/wallet.conf if that environment variable isn't set. That file should contain any site-specific overrides to the defaults, and at least some parameters must be set.
This file must be valid Perl. To set a variable, use the syntax:
$VARIABLE = <value>;
where VARIABLE is the variable name (always in all-capital letters) and
<value> is the value. If setting a variable to a string and not a number,
you should normally enclose <value> in ''. For example, to set the
variable DB_DRIVER to MySQL, use:
$DB_DRIVER = 'MySQL';
Always remember the initial dollar sign ($) and ending semicolon
(;). Those familiar with Perl syntax can of course use the full range
of Perl expressions.
This configuration file should end with the line:
1;
This ensures that Perl doesn't think there is an error when loading the file.
Specifies the directory used to dump the database schema in formats for each possible database server. This also includes diffs between schema versions, for upgrades. The default value is /usr/local/share/wallet, which matches the default installation location.
Sets the Perl database driver to use for the wallet database. Common
values would be SQLite or MySQL. Less common values would be
Oracle, Sybase, or ODBC. The appropriate DBD::* Perl module for
the chosen driver must be installed and will be dynamically loaded by the
wallet. For more information, see DBI.
This variable must be set.
Sets the remaining contents for the DBI DSN (everything after the driver). Using this variable provides full control over the connect string passed to DBI. When using SQLite, set this variable to the path to the SQLite database. If this variable is set, DB_NAME, DB_HOST, and DB_PORT are ignored. For more information, see DBI and the documentation for the database driver you're using.
Either DB_INFO or DB_NAME must be set. If you don't need to pass any
additional information to DBI, set DB_INFO to the empty string ('').
If DB_INFO is not set, specifies the database name. The third part of the
DBI connect string will be set to database=DB_NAME, possibly with a
host and port appended if DB_HOST and DB_PORT are set. For more
information, see DBI and the documentation for the database driver
you're using.
Either DB_INFO or DB_NAME must be set.
If DB_INFO is not set, specifies the database host. ;host=DB_HOST will
be appended to the DBI connect string. For more information, see DBI
and the documentation for the database driver you're using.
If DB_PORT is not set, specifies the database port. ;port=DB_PORT will
be appended to the DBI connect string. If this variable is set, DB_HOST
should also be set. For more information, see DBI and the
documentation for the database driver you're using.
Specifies the user for database authentication. Some database backends, particularly SQLite, do not need this.
Specifies the password for database authentication. Some database backends, particularly SQLite, do not need this.
These configuration variables only need to be set if you intend to use the
duo object type (the Wallet::Object::Duo class).
If this configuration variable is set, its value should be an object that is call-compatible with LWP::UserAgent. This object will be used instead of LWP::UserAgent to make API calls to Duo. This is primarily useful for testing, allowing replacement of the user agent with a mock implementation so that a test can run without needing a Duo account.
The path to a file in JSON format that contains the key and hostname data
for the Duo Admin API integration used to manage integrations via wallet.
This file should be in the format expected by the key_file parameter
to the Net::Duo::Admin constructor. See Net::Duo::Admin for more
information.
DUO_KEY_FILE must be set to use Duo objects.
The type of integration to create. The default value is unix to create
UNIX integrations, since this was the first integration created and users
may rely on it to still be the default.
These configuration variables only need to be set if you intend to use the
file object type (the Wallet::Object::File class).
The directory into which to store file objects. File objects will be stored in subdirectories of this directory. See Wallet::Object::File for the full details of the naming scheme. This directory must be writable by the wallet server and the wallet server must be able to create subdirectories of it.
FILE_BUCKET must be set to use file objects.
The maximum size of data that can be stored in a file object in bytes. If this configuration variable is set, an attempt to store data larger than this limit will be rejected.
These configuration variables only need to be set if you intend to use the
password object type (the Wallet::Object::Password class). You will also
need to set the FILE_MAX_SIZE value from the file object configuration, as
that is inherited.
The directory into which to store password objects. Password objects will be stored in subdirectories of this directory. See Wallet::Object::Password for the full details of the naming scheme. This directory must be writable by the wallet server and the wallet server must be able to create subdirectories of it.
PWD_FILE_BUCKET must be set to use file objects.
The minimum length for any auto-generated password objects created when get is run before data is stored.
The maximum length for any auto-generated password objects created when get is run before data is stored.
These configuration variables only need to be set if you intend to use the
keytab object type (the Wallet::Object::Keytab class).
Specifies the keytab to use to authenticate to kadmind. The principal
whose key is stored in this keytab must have the ability to create,
modify, inspect, and delete any principals that should be managed by the
wallet. (In MIT Kerberos kadm5.acl parlance, this is admci
privileges.)
KEYTAB_FILE must be set to use keytab objects with any backend other than Active Directory.
These flags, if any, are passed to the addprinc command when creating a
new principal in the Kerberos KDC. To not pass any flags, set
KEYTAB_FLAGS to the empty string. The default value is -clearpolicy,
which clears any password strength policy from principals created by the
wallet. (Since the wallet randomizes the keys, password strength checking
is generally pointless and may interact poorly with the way addprinc
-randkey works when third-party add-ons for password strength checking
are used.)
This option is ignored when using Active Directory.
Specifies the host on which the kadmin or Active Directory service is running.
This setting overrides the admin_server setting in the [realms] section of
krb5.conf and any DNS SRV records and allows the wallet to run on a system
that doesn't have a Kerberos configuration for the wallet's realm.
The path to the kadmin command-line client. The default value is
kadmin, which will cause the wallet to search for kadmin on its
default PATH.
This option is ignored when using Active Directory.
The Kerberos KDC implementation type, chosen from AD, Heimdal, or MIT
(case-insensitive). KEYTAB_KRBTYPE must be set to use keytab objects.
The principal whose key is stored in KEYTAB_FILE. The wallet will authenticate as this principal to the kadmin service.
KEYTAB_PRINCIPAL must be set to use keytab objects unless Active Directory is the backend, at least until kadmin is smart enough to use the first principal found in the keytab it's using for authentication.
Specifies the realm in which to create Kerberos principals. The keytab object implementation can only work in a single realm for a given wallet installation and the keytab object names are stored without realm. KEYTAB_REALM is added when talking to the KDC via kadmin.
KEYTAB_REALM must be set to use keytab objects. ktadd doesn't always
default to the local realm and the Active Directory integration requires it.
A directory into which the wallet can write keytabs temporarily while
processing get commands from clients. The keytabs are written into
this directory with predictable names, so this should not be a system
temporary directory such as /tmp or /var/tmp. It's best to create a
directory solely for this purpose that's owned by the user the wallet
server will run as.
KEYTAB_TMP must be set to use keytab objects.
The following parameters are specific to generating keytabs from
Active Directory (KEYTAB_KRBTYPE is set to AD).
The base distinguished name of the Active Directory instance. This is used when Wallet uses LDAP directly to examine objects in Active Directory.
Specifies the ticket cache to use when manipulating Active Directory objects. The ticket cache must be for a principal able to bind to Active Directory and run msktutil.
The LDAP base DN for computer objects inside Active Directory. All
keytabs of the form host/<hostname> will be mapped to objects with a
samAccountName of the <hostname> portion under this DN.
AD_COMPUTER_RDN must be set if using Active Directory as the keytab backend.
If set to true, asks for some additional debugging information, such
as the msktutil command, to be logged to syslog. These debugging
messages will be logged to the local3 facility.
The path to the msktutil command-line client. The default value is
msktutil, which will cause the wallet to search for msktutil on its
default PATH.
The maximum length of a unique identifier, samAccountName, for Active
Directory keytab objects. If the identifier exceeds this length then it will
be truncated and an integer will be appended to the end of the identifier.
This parameter is here in hopes that at some point in the future Microsoft
will remove the limitation.
Used to limit the number of iterations used in attempting to find a unique account name for principals. Defaults to 999.
For service principals the AD_SERVICE_PREFIX will be combined with the principal identifier to form the account name, i.e. the CN, used to store the keytab entry in the Active Directory. Active Directory limits these CN's to a maximum of 20 characters. If the resulting CN is greater than 20 characters the CN will be truncated and an integer will be appended to it. The integer will be incremented until a unique CN is found.
The AD_SERVICE_PREFIX is generally useful only prevent name collisions when the service keytabs are store in branch of the DIT that also contains other similar objects.
The hostname of the Active Directory Domain Controller.
The LDAP base DN for user objects inside Active Directory. All keytabs of the
form service/<user> will be mapped to objects with a servicePrincipalName
matching the wallet object name under this DN.
AD_USER_RDN must be set if using Active Directory as the keytab backend.
Heimdal provides the choice, over the network protocol, of either downloading the existing keys for a principal or generating new random keys. Neither MIT Kerberos or Active Directory support retrieving an existing keytab; downloading a keytab over the kadmin protocol or using msktutil always rekeys the principal.
For MIT Kerberos, the keytab object backend therefore optionally supports
retrieving existing keys, and hence keytabs, for Kerberos principals by
contacting the KDC via remctl and talking to keytab-backend. This is
enabled by setting the unchanging flag on keytab objects. To configure
that support, set the following variables.
For Active Directory Kerberos, the keytab object backend supports storing the keytabs on the wallet server. This functionality is enabled by setting the configuration variable AD_KEYTAB_BUCKET. (This had not been implemented yet.)
This is not required for Heimdal; for Heimdal, setting the unchanging
flag is all that's needed.
Specifies the ticket cache to use when retrieving existing keytabs from
the KDC. This is only used to implement support for the unchanging
flag. The ticket cache must be for a principal with access to run
keytab retrieve via remctl on KEYTAB_REMCTL_HOST.
The host to which to connect with remctl to retrieve existing keytabs.
This is only used to implement support for the unchanging flag. This
host must provide the keytab retrieve command and KEYTAB_REMCTL_CACHE
must also be set to a ticket cache for a principal with access to run that
command.
The service principal to which to authenticate when retrieving existing
keytabs. This is only used to implement support for the unchanging
flag. If this variable is not set, the default is formed by prepending
host/ to KEYTAB_REMCTL_HOST. (Note that KEYTAB_REMCTL_HOST is not
lowercased first.)
The port on KEYTAB_REMCTL_HOST to which to connect with remctl to retrieve
existing keytabs. This is only used to implement support for the
unchanging flag. If this variable is not set, the default remctl port
will be used.
The path to store a copy of keytabs created. This is required for the support of unchanging keytabs with an Active Directory KDC. (This has not been implemented yet.)
These configuration variables only need to be set if you intend to use the
wakeyring object type (the Wallet::Object::WAKeyring class).
The directory into which to store WebAuth keyring objects. WebAuth keyring objects will be stored in subdirectories of this directory. See Wallet::Object::WAKeyring for the full details of the naming scheme. This directory must be writable by the wallet server and the wallet server must be able to create subdirectories of it.
WAKEYRING_BUCKET must be set to use WebAuth keyring objects.
The interval, in seconds, at which new keys are generated in a keyring. The object implementation will try to arrange for there to be keys added to the keyring separated by this interval.
It's useful to provide some interval to install the keyring everywhere that it's used before the key becomes inactive. Every keyring will therefore normally have at least three keys: one that's currently active, one that becomes valid in the future but less than WAKEYRING_REKEY_INTERVAL from now, and one that becomes valid between one and two of those intervals into the future. This means that one has twice this interval to distribute the keyring everywhere it is used.
Internally, this is implemented by adding a new key that becomes valid in twice this interval from the current time if the newest key becomes valid at or less than this interval in the future.
The default value is 60 * 60 * 24 (one day).
The interval, in seconds, from the key creation date after which keys are removed from the keyring. This is used to clean up old keys and finish key rotation. Keys won't be removed unless there are more than three keys in the keyring to try to keep a misconfiguration from removing all valid keys.
The default value is 60 * 60 * 24 * 90 (90 days).
This configuration variable is only needed if you intend to use the
external ACL type (the Wallet::ACL::External class). This ACL type
runs an external command to determine if access is granted.
Path to the command to run to determine whether access is granted. The first argument to the command will be the principal requesting access. The second and third arguments will be the type and name of the object that principal is requesting access to. The final argument will be the identifier of the ACL.
No other arguments are passed to the command, but the command will have access to all of the remctl environment variables seen by the wallet server (such as REMOTE_USER). For a full list of environment variables, see "ENVIRONMENT" in remctld(8).
The external command should exit with a non-zero status but no output to indicate a normal failure to satisfy the ACL. Any output will be treated as an error.
These configuration variables are only needed if you intend to use the
ldap-attr ACL type (the Wallet::ACL::LDAP::Attribute class). They
specify the LDAP server and additional connection and data model
information required for the wallet to check for the existence of
attributes.
The LDAP server name to use to verify LDAP ACLs. This variable must be set to use LDAP ACLs.
The base DN under which to search for the entry corresponding to a principal. Currently, the wallet always does a full subtree search under this base DN. This variable must be set to use LDAP ACLs.
The attribute used to find the entry corresponding to a principal. The
LDAP entry containing this attribute with a value equal to the principal
will be found and checked for the required attribute and value. If this
variable is not set, the default is krb5PrincipalName.
Specifies the Kerberos ticket cache to use when connecting to the LDAP server. GSS-API authentication is always used; there is currently no support for any other type of bind. The ticket cache must be for a principal with access to verify the values of attributes that will be used with this ACL type. This variable must be set to use LDAP ACLs.
Finally, depending on the structure of the LDAP directory being queried, there may not be any attribute in the directory whose value exactly matches the Kerberos principal. The attribute designated by LDAP_FILTER_ATTR may instead hold a transformation of the principal name (such as the principal with the local realm stripped off, or rewritten into an LDAP DN form). If this is the case, define a Perl function named ldap_map_principal. This function will be called whenever an LDAP attribute ACL is being verified. It will take one argument, the principal, and is expected to return the value to search for in the LDAP directory server.
For example, if the principal name without the local realm is stored in
the uid attribute in the directory, set LDAP_FILTER_ATTR to uid and
then define ldap_map_attribute as follows:
sub ldap_map_principal {
my ($principal) = @_;
$principal =~ s/\@EXAMPLE\.COM$//;
return $principal;
}
Note that this example only removes the local realm (here, EXAMPLE.COM). Any principal from some other realm will be left fully qualified, and then presumably will not be found in the directory.
These configuration variables are only needed if you intend to use the
netdb ACL type (the Wallet::ACL::NetDB class). They specify the remctl
connection information for retrieving user roles from NetDB and the local
realm to remove from principals (since NetDB normally expects unscoped
local usernames).
The wallet uses fully-qualified principal names (including the realm), but NetDB normally expects local usernames without the realm. If this variable is set, the given realm will be stripped from any principal names before passing them to NetDB. Principals in other realms will be passed to NetDB without modification.
Specifies the ticket cache to use when querying the NetDB remctl interface
for user roles. The ticket cache must be for a principal with access to
run netdb node-roles via remctl on KEYTAB_REMCTL_HOST. This variable
must be set to use NetDB ACLs.
The host to which to connect with remctl to query NetDB for user roles.
This host must provide the netdb node-roles command and
NETDB_REMCTL_CACHE must also be set to a ticket cache for a principal with
access to run that command. This variable must be set to use NetDB ACLs.
The service principal to which to authenticate when querying NetDB for
user roles. If this variable is not set, the default is formed by
prepending host/ to NETDB_REMCTL_HOST. (Note that NETDB_REMCTL_HOST is
not lowercased first.)
The port on NETDB_REMCTL_HOST to which to connect with remctl to query NetDB for user roles. If this variable is not set, the default remctl port will be used.
By default, only users in the ADMIN ACL can create new objects in the wallet. To allow other users to create new objects, define a Perl function named default_owner. This function will be called whenever a non-ADMIN user tries to create a new object and will be passed the type and name of the object. It should return undef if there is no default owner for that object. If there is, it should return a list containing the name to use for the ACL and then zero or more anonymous arrays of two elements each giving the type and identifier for each ACL entry.
For example, the following simple function says to use a default owner
named default with one entry of type krb5 and identifier
rra@example.com for the object with type keytab and name
host/example.com:
sub default_owner {
my ($type, $name) = @_;
if ($type eq 'keytab' and $name eq 'host/example.com') {
return ('default', [ 'krb5', 'rra@example.com' ]);
} else {
return;
}
}
Of course, normally this function is used for more complex mappings. Here is a more complete example. For objects of type keytab corresponding to various types of per-machine principals, return a default owner that sets as owner anyone with a NetDB role for that system and the system's host principal. This permits authorization management using NetDB while also allowing the system to bootstrap itself once the host principal has been downloaded and rekey itself using the old host principal.
sub default_owner {
my ($type, $name) = @_;
my %allowed = map { $_ => 1 }
qw(HTTP cifs host imap ldap nfs pop sieve smtp webauth);
my $realm = 'example.com';
return unless $type eq 'keytab';
return unless $name =~ m%/%;
my ($service, $instance) = split ('/', $name, 2);
return unless $allowed{$service};
my $acl_name = "host/$instance";
my @acl = ([ 'netdb', $instance ],
[ 'krb5', "host/$instance\@$realm" ]);
return ($acl_name, @acl);
}
The auto-created ACL used for the owner of the new object will, in the
above example, be named host/system where system is the
fully-qualified name of the system as derived from the keytab being
requested.
If the name of the ACL returned by the default_owner function matches an ACL that already exists in the wallet database, the existing ACL will be compared to the default ACL returned by the default_owner function. If the existing ACL has the same entries as the one returned by default_owner, creation continues if the user is authorized by that ACL. If they don't match, creation of the object is rejected, since the presence of an existing ACL may indicate that something different is being done with this object.
By default, wallet permits administrators to create objects of any name (unless the object backend rejects the name). However, naming standards for objects can be enforced, even for administrators, by defining a Perl function in the configuration file named verify_name. If such a function exists, it will be called for any object creation and will be passed the type of object, the object name, and the identity of the person doing the creation. If it returns undef or the empty string, object creation will be allowed. If it returns anything else, object creation is rejected and the return value is used as the error message.
This function is also called for naming audits done via Wallet::Report to find any existing objects that violate a (possibly updated) naming policy. In this case, the third argument (the identity of the person creating the object) will be undef. As a general rule, if the third argument is undef, the function should apply the most liberal accepted naming policy so that the audit returns only objects that violate all naming policies, but some sites may wish different results for their audit reports.
Please note that this return status is backwards from what one would normally expect. A false value is success; a true value is failure with an error message.
For example, the following verify_name function would ensure that any keytab objects for particular principals have fully-qualified hostnames:
sub verify_name {
my ($type, $name, $user) = @_;
my %host_based = map { $_ => 1 }
qw(HTTP cifs host imap ldap nfs pop sieve smtp webauth);
return unless $type eq 'keytab';
return unless $name =~ m%/%;
my ($service, $instance) = split ('/', $name, 2);
return unless $host_based{$service};
return "host name $instance must be fully qualified"
unless $instance =~ /\./;
return;
}
Objects that aren't of type keytab or which aren't for a host-based key
have no naming requirements enforced by this example.
The above demonstrates having a host-based naming convention, where we expect one part of an object name to be the name of the host that this object is for. The most obvious examples are those keytab objects above, where we want certain keytab names to be in the form of <service>/<hostname>. It's then also useful to provide a Perl function named is_for_host which then can be used to tell if a given object is a host-based keytab for a specific host. This function is then called by the objects_hostname in Wallet::Report to give a list of all host-based objects for a given hostname. It should return true if the given object is a host-based object for the hostname, otherwise false.
An example that matches the same policy as the last verify_name example would be:
sub is_for_host {
my ($type, $name, $hostname) = @_;
my %host_based = map { $_ => 1 }
qw(HTTP cifs host imap ldap nfs pop sieve smtp webauth);
return 0 unless $type eq 'keytab';
return 0 unless $name =~ m%/%;
my ($service, $instance) = split ('/', $name, 2);
return 0 unless $host_based{$service};
return 1 if $hostname eq $instance;
return 0;
}
Similar to object names, by default wallet permits administrators to create ACLs with any name. However, naming standards for ACLs can be enforced by defining a Perl function in the configuration file named verify_acl_name. If such a function exists, it will be called for any ACL creation or rename and will be passed given the new ACL name and the identity of the person doing the creation. If it returns undef or the empty string, object creation will be allowed. If it returns anything else, object creation is rejected and the return value is used as the error message.
This function is also called for naming audits done via Wallet::Report to find any existing objects that violate a (possibly updated) naming policy. In this case, the second argument (the identity of the person creating the ACL) will be undef. As a general rule, if the second argument is undef, the function should apply the most liberal accepted naming policy so that the audit returns only ACLs that violate all naming policies, but some sites may wish different results for their audit reports.
Please note that this return status is backwards from what one would normally expect. A false value is success; a true value is failure with an error message.
For example, the following verify_acl_name function would ensure that any
ACLs created contain a slash and the part before the slash be one of
host, group, user, or service.
sub verify_acl_name {
my ($name, $user) = @_;
return 'ACL names must contain a slash' unless $name =~ m,/,;
my ($first, $rest) = split ('/', $name, 2);
my %types = map { $_ => 1 } qw(host group user service);
unless ($types{$first}) {
return "unknown ACL type $first";
}
return;
}
Obvious improvements could be made, such as checking that the part after
the slash for a host/ ACL looked like a host name and the part after a
slash for a user/ ACL look like a user name.
If this environment variable is set, it is taken to be the path to the wallet configuration file to load instead of /etc/wallet/wallet.conf.
DBI(3), Wallet::Object::Keytab(3), Wallet::Server(3), wallet-backend(8)
This module is part of the wallet system. The current version is available from <https://www.eyrie.org/~eagle/software/wallet/>.
Russ Allbery <eagle@eyrie.org>
| < wallet Setup | Russ Allbery > Software > wallet | wallet Objects and ACL Schemes > |