path: root/client/wallet.pod

=head1 NAME

wallet - Client for retrieving secure data from a central server

=for stopwords
-hv srvtab arg keytabs metadata keytab ACL PTS kinit klist remctl PKINIT
acl timestamp autocreate backend-specific setacl enctypes enctype ktadd
KDC appdefaults remctld Allbery nul uuencode getacl backend

=head1 SYNOPSIS

B<wallet> [B<-hv>] [B<-c> I<command>] [B<-f> I<file>]
    [B<-k> I<principal>] [B<-p> I<port>] [S<B<-s> I<server>>]
    [B<-S> I<srvtab>] [B<-u> I<principal>] I<command> [I<arg> ...]

=head1 DESCRIPTION

B<wallet> is a client for the wallet system, which stores or creates
secure information such as Kerberos keytabs, associates them with ACLs and
other metadata, and allows clients to view and download them.  This client
provides the user interface to the wallet system for both users and wallet
administrators.

The B<wallet> command-line client takes a command and optional arguments
on the command line, authenticates to the wallet server using Kerberos,
and sends that command and arguments to server.  It then reads the results
and displays them to the user or stores them in a file.  The client itself
does not know which commands are valid and which aren't; apart from some
special handling of particular commands, it sends all commands to the
server to respond to appropriately.  This allows additional commands to be
added to the wallet system without changing all of the clients.

The primary commands of the wallet system are C<get>, which retrieves some
secure data from the wallet, C<store>, which stores some secure data in
the wallet, and C<show>, which stores the metadata about an object stored
in the wallet.  Each object in the wallet has a type, which determines
what data the object represents and may determine special handling when
downloading or storing that object, and a name.  For example, a wallet
object for the C<host/example.com> Kerberos keytab would have a type of
C<keytab> and a name of C<host/example.com>.  The meaning of the name is
specific to each type of object.

Most other wallet commands besides those three are only available to
wallet administrators.  The exception is attribute commands; see
L<ATTRIBUTES>.  The other commands allow setting ownership and ACLs on
objects, creating and destroying objects, creating and destroying ACLs,
and adding and removing entries from ACLs.  An ACL consists of one or more
entries, each of which is a scheme and an identifier.  A scheme specifies
a way of checking whether a user is authorized.  An identifier is some
data specific to the scheme that specifies which users are authorized.
For example, for the C<krb5> scheme, the identifier is a principal name
and only that principal is authorized by that ACL entry.

To run the wallet command-line client, you must already have a Kerberos
ticket.  You can obtain a Kerberos ticket with B<kinit> and see your
current Kerberos tickets with B<klist>.  The wallet client uses the remctl
protocol to talk to the wallet server.

=head1 OPTIONS

=over 4

=item B<-c> I<command>

The command prefix (remctl type) to use.  Normally this is an internal
implementation detail and the default (C<wallet>) should be fine.  It may
sometimes be useful to use a different prefix for testing a different
version of the wallet code on the server.  This option can also be set in
F<krb5.conf>; see L<CONFIGURATION> below.

=item B<-f> I<file>

This flag is only used in combination with the C<get> and C<store>
commands.  For C<get>, rather than sending the secure data to standard
output (the default), the secure data will be stored in I<file>.  For
C<store>, the data to be stored will be read from I<file>.

With C<get>, if the object being retrieved is not a keytab object, any
current file named I<output> is renamed to F<I<outout>.bak> before the new
file is created.  F<I<outout>.new> is used as a temporary file and any
existing file with that name will be deleted.

If the object being retrieved is a keytab object and the file I<output>
already exists, the downloaded keys will be added to the existing keytab
file I<output>.  Old keys are not removed; you may wish to run C<kadmin
ktremove> or an equivalent later to clean up old keys.  F<I<output>.new>
is still used as a temporary file and any existing file with that name
will be deleted.

C<store> does not yet support nul bytes in I<file> (or in any other way of
specifying the data to be stored).  To store binary files in the wallet,
you will need to encode them with uuencode, base64, or some similar scheme
and then decode them after retrieval.

=item B<-k> I<principal>

The service principal of the wallet server.  The default is to use the
C<host> principal for the wallet server.  The principal chosen must match
one of the keys in the keytab used by B<remctld> on the wallet server.
This option can also be set in F<krb5.conf>; see L<CONFIGURATION> below.

=item B<-h>

Display a brief summary of options and exit.  All other valid options and
commands are ignored.

=item B<-p> I<port>

The port to connect to on the wallet server.  The default is the default
remctl port.  This option can also be set in F<krb5.conf>; see
L<CONFIGURATION> below.

=item B<-S> I<srvtab>

This flag is only used in combination with the C<get> command on a
C<keytab> object, and must be used in conjunction with the B<-f> flag.
After the keytab is saved to the file specified by B<-f>, the DES key for
that principal will be extracted and written as a Kerberos v4 srvtab to
the file I<srvtab>.  Any existing contents of I<srvtab> will be
destroyed.

The Kerberos v4 principal name will be generated from the Kerberos v5
principal name using the krb5_524_conv_principal() function of the
Kerberos libraries.  See its documentation for more information, but
briefly (and in the absence of special configuration), the Kerberos v4
principal name will be the same as the Kerberos v5 principal name except
that the components are separated by C<.> instead of C</>; the second
component is truncated after the first C<.> if the first component is one
of the recognized host-based principals (generally C<host>, C<imap>,
C<pop>, or C<smtp>); and the first component is C<rcmd> if the Kerberos v5
principal component is C<host>.  The principal name must not contain more
than two components.

=item B<-s> I<server>

The wallet server to connect to.  The default may be set when compiling
the wallet client.  If it isn't, either B<-s> must be given or the server
must be set in F<krb5.conf>.  See L<CONFIGURATION> below.

=item B<-u> I<principal>

Rather than using the user's existing ticket cache for authentication,
authenticate as I<principal> first and use those credentials for
authentication to the wallet server.  B<wallet> will prompt for the
password for I<principal>.  Non-password authentication methods such as
PKINIT aren't supported; to use those, run B<kinit> first and use an
existing ticket cache.

=item B<-v>

Display the version of the B<wallet> client and exit.  All other valid
options and commands are ignored.

=back

=head1 COMMANDS

As mentioned above, most commands are only available to wallet
administrators.  The exceptions are C<get>, C<store>, C<show>, C<destroy>,
C<flag clear>, C<flag set>, C<getattr>, C<setattr>, and C<history>.  All
of those commands have their own ACLs except C<getattr> and C<history>,
which use the C<show> ACL, and C<setattr>, which uses the C<store> ACL.
If the appropriate ACL is set, it alone is checked to see if the user has
access.  Otherwise, C<get>, C<store>, C<show>, C<getattr>, C<setattr>, and
C<history> access is permitted if the user is authorized by the owner ACL
of the object.

Administrators can run any command on any object or ACL except for C<get>
and C<store>.  For C<get> and C<show>, they must still be authorized by
either the appropriate specific ACL or the owner ACL.

If the locked flag is set on an object, no commands can be run on that
object that change data except the C<flags> commands, nor can the C<get>
command be used on that object.  C<show>, C<history>, C<getacl>,
C<getattr>, and C<owner> or C<expires> without an argument can still be
used on that object.

For more information on attributes, see L<ATTRIBUTES>.

=over 4

=item acl add <id> <scheme> <identifier>

Adds an entry with <scheme> and <identifier> to the ACL <id>.  <id> may be
either the name of an ACL or its numeric identifier.

=item acl create <name>

Create a new, empty ACL with name <name>.  When setting an ACL on an
object with a set of entries that don't match an existing ACL, first
create a new ACL with C<acl create>, add the appropriate entries to it
with C<acl add>, and then set the ACL on an object with the C<owner> or
C<setacl> commands.

=item acl destroy <id>

Destroy the ACL <id>.  This ACL must no longer be referenced by any object
or the ACL destruction will fail.  The special ACL named C<ADMIN> cannot
be destroyed.

=item acl history <id>

Display the history of the ACL <id>.  Each change to the ACL (not
including changes to the name of the ACL) will be represented by two
lines.  The first line will have a timestamp of the change followed by a
description of the change, and the second line will give the user who made
the change and the host from which the change was made.

=item acl remove <id> <scheme> <identifier>

Remove the entry with <scheme> and <identifier> from the ACL <id>.  <id>
may be either the name of an ACL or its numeric identifier.  The last
entry in the special ACL C<ADMIN> cannot be removed to protect against
accidental lockout, but administrators can remove themselves from the
C<ADMIN> ACL and can leave only a non-functioning entry on the ACL.  Use
caution when removing entries from the C<ADMIN> ACL.

=item acl show <id>

Display the name, numeric ID, and entries of the ACL <id>.

=item autocreate <type> <name>

Create a new object of type <type> with name <name>.  The user must be
listed in the default ACL for an object with that type and name, and the
object will be created with that default ACL set as the object owner.

Normally, there's no need to run this command directly.  It's
automatically run when trying to get or store an object that doesn't
already exist.

=item check <type> <name>

Check whether an object of type <type> and name <name> already exists.  If
it does, prints C<yes>; if not, prints C<no>.

=item create <type> <name>

Create a new object of type <type> with name <name>.  With some backends,
this will trigger creation of an entry in an external system as well.
The new object will have no ACLs and no owner set, so usually the
administrator will want to then set an owner with C<owner> so that the
object will be usable.

=item destroy <type> <name>

Destroy the object identified by <type> and <name>.  With some backends,
this will trigger destruction of an object in an external system as well.

=item expires <type> <name> [<expires>]

If <expires> is not given, displays the current expiration of the object
identified by <type> and <name>, or C<No expiration set> if none is set.
The expiration will be displayed in seconds since epoch.

If <date> is given, sets the expiration on the object identified by <type>
and <name> to <date> and (if given) <time>.  <date> must be in the format
C<YYYY-MM-DD> and <time> in the format C<HH:MM:SS>.  If <date> is the empty
string, clears the expiration of the object.

Currently, the expiration of an object is not used.

=item flag clear <type> <name> <flag>

Clears the flag <flag> on the object identified by <type> and <name>.

=item flag set <type> <name> <flag>

Sets the flag <flag> on the object identified by <type> and <name>.
Recognized flags are C<locked>, which prevents all further actions on that
object until the flag is cleared, and C<unchanging>, which tells the object
backend to not generate new data on get but instead return the same data as
previously returned.  The C<unchanging> flag is not meaningful for objects
that do not generate new data on the fly.

=item get <type> <name>

Prints to standard output the data associated with the object identified
by <type> and <name>, or stores it in a file if the B<-f> option was
given.  This may trigger generation of new data and invalidate old data
for that object depending on the object type.

If an object with type <type> and name <name> does not already exist when
this command is issued (as checked with the check interface), B<wallet>
will attempt to automatically create it (using autocreate).

=item getacl <type> <name> <acl>

Prints the ACL <acl>, which must be one of C<get>, C<store>, C<show>,
C<destroy>, or C<flags>, for the object identified by <type> and <name>.
Prints C<No ACL set> if that ACL isn't set on that object.  Remember that
if the C<get>, C<store>, or C<show> ACLs aren't set, authorization falls
back to checking the owner ACL.  See the C<owner> command for displaying
or setting it.

=item getattr <type> <name> <attr>

Prints the object attribute <attr> for the object identified by <type> and
<name>.  Attributes are used to store backend-specific information for a
particular object type, and <attr> must be an attribute type known to the
underlying object implementation.  The attribute values, if any, are
printed one per line.  If the attribute is not set on this object, nothing
is printed.

=item history <type> <name>

Displays the history for the object identified by <type> and <name>.
This human-readable output will have two lines for each action that
changes the object, plus for any get action.  The first line has the
timestamp of the action and the action, and the second line gives the user
who performed the action and the host from which they performed it.

=item owner <type> <name> [<owner>]

If <owner> is not given, displays the current owner ACL of the object
identified by <type> and <name>, or C<No owner set> if none is set.  The
result will be the name of an ACL.

If <owner> is given, sets the owner of the object identified by <type> and
<name> to <owner>.  If <owner> is the empty string, clears the owner of
the object.

=item setacl <type> <name> <acl> <id>

Sets the ACL <acl>, which must be one of C<get>, C<store>, C<show>,
C<destroy>, or C<flags>, to <id> on the object identified by <type> and
<name>.  If <id> is the empty string, clears that ACL on the object.

=item setattr <type> <name> <attr> <value> [<value> ...]

Sets the object attribute <attr> for the object identified by <type> and
<name>.  Attributes are used to store backend-specific information for a
particular object type, and <attr> must be an attribute type known to the
underlying object implementation.  To clear the attribute for this object,
pass in a <value> of the empty string (C<''>).

=item show <type> <name>

Displays the current object metadata for the object identified by <type>
and <name>.  This human-readable output will show the object type and
name, the owner, any specific ACLs set on the object, the expiration if
any, and the user, remote host, and time when the object was created, last
stored, and last downloaded.

=item store <type> <name> [<data>]

Stores <data> for the object identified by <type> and <name> for later
retrieval with C<get>.  Not all object types support this.  If <data> is
not specified on the command line, it will be read from the file specified
with B<-f> (if given) or from standard input.

Currently, the stored data must not contain nul characters and may
therefore not be binary data.  Its length is also limited by the maximum
command line length of the operating system of the wallet server.  These
restrictions will be lifted in the future.

If an object with type <type> and name <name> does not already exist when
this command is issued (as checked with the check interface), B<wallet>
will attempt to automatically create it (using autocreate).

=back

=head1 ATTRIBUTES

Object attributes store additional properties and configuration
information for objects stored in the wallet.  They are displayed as part
of the object data with C<show>, retrieved with C<getattr>, and set with
C<setattr>.

=head2 Keytab Attributes

Keytab objects support the following attributes:

=over 4

=item enctypes

Restricts the generated keytab to a specific set of encryption types.  The
values of this attribute must be enctype strings recognized by Kerberos
(strings like C<aes256-cts-hmac-sha1-96> or C<des-cbc-crc>).  Note that
the salt should not be included; since the salt is irrelevant for keytab
keys, it will always be set to C<normal> by the wallet.

If this attribute is set, the specified enctype list will be passed to ktadd
when get() is called for that keytab.  If it is not set, the default set in
the KDC will be used.

This attribute is ignored if the C<unchanging> flag is set on a keytab.
Keytabs retrieved with C<unchanging> set will contain all keys present in
the KDC for that Kerberos principal and therefore may contain different
enctypes than those requested by this attribute.

=back

=head1 CONFIGURATION

B<wallet> can optionally be configured in the system F<krb5.conf>.  It
will read the default F<krb5.conf> file for the Kerberos libraries with
which it was compiled.  To set an option, put the option in the
[appdefaults] section.  B<wallet> will look for options either at the top
level of the [appdefaults] section or in a subsection named C<wallet>.
For example, the following fragment of a F<krb5.conf> file would set the
default port to 4373 and the default server to C<wallet.example.org>.

    [appdefaults]
        wallet_port = 4373
        wallet = {
            wallet_server = wallet.example.org
        }

The supported options are:

=over 4

=item wallet_principal

The service principal of the wallet server.  The default is to use the
C<host> principal for the wallet server.  The principal chosen must match
one of the keys in the keytab used by B<remctld> on the wallet server.
The B<-k> command-line option overrides this setting.

=item wallet_port

The port to connect to on the wallet server.  The default is the default
remctl port.  The B<-p> command-line option overrides this setting.

=item wallet_server

The wallet server to connect to.  The B<-s> command-line option overrides
this setting.  The default may be set when compiling the wallet client.
If it isn't, either B<-s> must be given or this parameter must be present
in in F<krb5.conf>.

=item wallet_type

The command prefix (remctl type) to use.  Normally this is an internal
implementation detail and the default (C<wallet>) should be fine.  It may
sometimes be useful to use a different prefix for testing a different
version of the wallet code on the server.  The B<-c> command-line option
overrides this setting.

=back

=head1 SEE ALSO

kadmin(8), kinit(1), krb5.conf(5), remctl(1), remctld(8)

This program is part of the wallet system.  The current version is available
from L<http://www.eyrie.org/~eagle/software/wallet/>.

B<wallet> uses the remctl protocol.  For more information about remctl,
see L<http://www.eyrie.org/~eagle/software/remctl/>.

=head1 AUTHOR

Russ Allbery <rra@stanford.edu>

=cut