ITS#6052

[openldap] / doc / guide / admin / access-control.sdf

# $OpenLDAP$
# Copyright 1999-2009 The OpenLDAP Foundation, All Rights Reserved.
# COPYING RESTRICTIONS APPLY, see COPYRIGHT.

H1: Access Control

H2: Introduction

As the directory gets populated with more and more data of varying sensitivity, 
controlling the kinds of access granted to the directory becomes more and more
critical. For instance, the directory may contain data of a confidential nature 
that you may need to protect by contract or by law. Or, if using the directory 
to control access to other services, inappropriate access to the directory may 
create avenues of attack to your sites security that result in devastating 
damage to your assets.

Access to your directory can be configured via two methods, the first using
{{SECT:The slapd Configuration File}} and the second using the {{slapd-config}}(5) 
format ({{SECT:Configuring slapd}}).

The default access control policy is allow read by all clients. Regardless of 
what access control policy is defined, the {{rootdn}} is always allowed full 
rights (i.e. auth, search, compare, read and write) on everything and anything.

As a consequence, it's useless (and results in a performance penalty) to explicitly 
list the {{rootdn}} among the {{<by>}} clauses.

The following sections will describe Access Control Lists in more details and 
follow with some examples and recommendations. 

H2: Access Control via Static Configuration

Access to entries and attributes is controlled by the
access configuration file directive. The general form of an
access line is:

>    <access directive> ::= access to <what>
>        [by <who> [<access>] [<control>] ]+
>    <what> ::= * |
>        [dn[.<basic-style>]=<regex> | dn.<scope-style>=<DN>]
>        [filter=<ldapfilter>] [attrs=<attrlist>]
>    <basic-style> ::= regex | exact
>    <scope-style> ::= base | one | subtree | children
>    <attrlist> ::= <attr> [val[.<basic-style>]=<regex>] | <attr> , <attrlist>
>    <attr> ::= <attrname> | entry | children
>    <who> ::= * | [anonymous | users | self
>            | dn[.<basic-style>]=<regex> | dn.<scope-style>=<DN>] 
>        [dnattr=<attrname>]
>        [group[/<objectclass>[/<attrname>][.<basic-style>]]=<regex>]
>        [peername[.<basic-style>]=<regex>]
>        [sockname[.<basic-style>]=<regex>]
>        [domain[.<basic-style>]=<regex>]
>        [sockurl[.<basic-style>]=<regex>]
>        [set=<setspec>]
>        [aci=<attrname>]
>    <access> ::= [self]{<level>|<priv>}
>    <level> ::= none | disclose | auth | compare | search | read | write | manage
>    <priv> ::= {=|+|-}{m|w|r|s|c|x|d|0}+
>    <control> ::= [stop | continue | break]

where the <what> part selects the entries and/or attributes to which
the access applies, the {{EX:<who>}} part specifies which entities
are granted access, and the {{EX:<access>}} part specifies the
access granted. Multiple {{EX:<who> <access> <control>}} triplets
are supported, allowing many entities to be granted different access
to the same set of entries and attributes. Not all of these access
control options are described here; for more details see the
{{slapd.access}}(5) man page.


H3: What to control access to

The <what> part of an access specification determines the entries
and attributes to which the access control applies.  Entries are
commonly selected in two ways: by DN and by filter.  The following
qualifiers select entries by DN:

>    to *
>    to dn[.<basic-style>]=<regex>
>    to dn.<scope-style>=<DN>

The first form is used to select all entries.  The second form may
be used to select entries by matching a regular expression against
the target entry's {{normalized DN}}.   (The second form is not
discussed further in this document.)  The third form is used to
select entries which are within the requested scope of DN.  The
<DN> is a string representation of the Distinguished Name, as
described in {{REF:RFC4514}}.

The scope can be either {{EX:base}}, {{EX:one}}, {{EX:subtree}},
or {{EX:children}}.  Where {{EX:base}} matches only the entry with
provided DN, {{EX:one}} matches the entries whose parent is the
provided DN, {{EX:subtree}} matches all entries in the subtree whose
root is the provided DN, and {{EX:children}} matches all entries
under the DN (but not the entry named by the DN).

For example, if the directory contained entries named:

>    0: o=suffix
>    1: cn=Manager,o=suffix
>    2: ou=people,o=suffix
>    3: uid=kdz,ou=people,o=suffix
>    4: cn=addresses,uid=kdz,ou=people,o=suffix
>    5: uid=hyc,ou=people,o=suffix

\Then:
. {{EX:dn.base="ou=people,o=suffix"}} match 2;
. {{EX:dn.one="ou=people,o=suffix"}} match 3, and 5;
. {{EX:dn.subtree="ou=people,o=suffix"}} match 2, 3, 4, and 5; and
. {{EX:dn.children="ou=people,o=suffix"}} match 3, 4, and 5.


Entries may also be selected using a filter:

>    to filter=<ldap filter>

where <ldap filter> is a string representation of an LDAP
search filter, as described in {{REF:RFC4515}}.  For example:

>    to filter=(objectClass=person)

Note that entries may be selected by both DN and filter by
including both qualifiers in the <what> clause.

>    to dn.one="ou=people,o=suffix" filter=(objectClass=person)

Attributes within an entry are selected by including a comma-separated
list of attribute names in the <what> selector:

>    attrs=<attribute list>

A specific value of an attribute is selected by using a single
attribute name and also using a value selector:

>    attrs=<attribute> val[.<style>]=<regex>

There are two special {{pseudo}} attributes {{EX:entry}} and
{{EX:children}}.  To read (and hence return) a target entry, the
subject must have {{EX:read}} access to the target's {{entry}}
attribute.  To perform a search, the subject must have
{{EX:search}} access to the search base's {{entry}} attribute.
To add or delete an entry, the subject must have
{{EX:write}} access to the entry's {{EX:entry}} attribute AND must
have {{EX:write}} access to the entry's parent's {{EX:children}}
attribute.  To rename an entry, the subject must have {{EX:write}}
access to entry's {{EX:entry}} attribute AND have {{EX:write}}
access to both the old parent's and new parent's {{EX:children}}
attributes.  The complete examples at the end of this section should
help clear things up.

Lastly, there is a special entry selector {{EX:"*"}} that is used to
select any entry.  It is used when no other {{EX:<what>}}
selector has been provided.  It's equivalent to "{{EX:dn=.*}}"


H3: Who to grant access to

The <who> part identifies the entity or entities being granted
access. Note that access is granted to "entities" not "entries."
The following table summarizes entity specifiers:

!block table; align=Center; coltags="EX,N"; \
    title="Table 6.3: Access Entity Specifiers"
Specifier|Entities
*|All, including anonymous and authenticated users
anonymous|Anonymous (non-authenticated) users
users|Authenticated users
self|User associated with target entry
dn[.<basic-style>]=<regex>|Users matching a regular expression
dn.<scope-style>=<DN>|Users within scope of a DN
!endblock

The DN specifier behaves much like <what> clause DN specifiers.

Other control factors are also supported.  For example, a {{EX:<who>}}
can be restricted by an entry listed in a DN-valued attribute in
the entry to which the access applies:

>    dnattr=<dn-valued attribute name>

The dnattr specification is used to give access to an entry
whose DN is listed in an attribute of the entry (e.g., give
access to a group entry to whoever is listed as the owner of
the group entry).

Some factors may not be appropriate in all environments (or any).
For example, the domain factor relies on IP to domain name lookups.
As these can easily be spoofed, the domain factor should be avoided.


H3: The access to grant

The kind of <access> granted can be one of the following:

!block table; colaligns="LRL"; coltags="EX,EX,N"; align=Center; \
    title="Table 6.4: Access Levels"
Level        Privileges    Description
none        =0             no access
disclose    =d             needed for information disclosure on error
auth        =dx            needed to authenticate (bind)
compare     =cdx           needed to compare
search      =scdx          needed to apply search filters
read        =rscdx         needed to read search results
write       =wrscdx        needed to modify/rename
manage      =mwrscdx       needed to manage
!endblock

Each level implies all lower levels of access. So, for example,
granting someone {{EX:write}} access to an entry also grants them
{{EX:read}}, {{EX:search}}, {{EX:compare}}, {{EX:auth}} and
{{EX:disclose}} access.  However, one may use the privileges specifier
to grant specific permissions.


H3: Access Control Evaluation

When evaluating whether some requester should be given access to
an entry and/or attribute, slapd compares the entry and/or attribute
to the {{EX:<what>}} selectors given in the configuration file.
For each entry, access controls provided in the database which holds
the entry (or the global access directives if not held in any database) apply
first, followed by the global access directives. However, when dealing with 
an access list, because the global access list is effectively appended 
to each per-database list, if the resulting list is non-empty then the 
access list will end with an implicit {{EX:access to * by * none}} directive. 
If there are no access directives applicable to a backend, then a default 
read is used.

Within this
priority, access directives are examined in the order in which they
appear in the config file.  Slapd stops with the first {{EX:<what>}}
selector that matches the entry and/or attribute. The corresponding
access directive is the one slapd will use to evaluate access.

Next, slapd compares the entity requesting access to the {{EX:<who>}}
selectors within the access directive selected above in the order
in which they appear. It stops with the first {{EX:<who>}} selector
that matches the requester. This determines the access the entity
requesting access has to the entry and/or attribute.

Finally, slapd compares the access granted in the selected
{{EX:<access>}} clause to the access requested by the client. If
it allows greater or equal access, access is granted. Otherwise,
access is denied.

The order of evaluation of access directives makes their placement
in the configuration file important. If one access directive is
more specific than another in terms of the entries it selects, it
should appear first in the config file. Similarly, if one {{EX:<who>}}
selector is more specific than another it should come first in the
access directive. The access control examples given below should
help make this clear.



H3: Access Control Examples

The access control facility described above is quite powerful.  This
section shows some examples of its use for descriptive purposes.

A simple example:

>    access to * by * read

This access directive grants read access to everyone.

>    access to *
>        by self write
>        by anonymous auth
>        by * read

This directive allows the user to modify their entry, allows anonymous
to authentication against these entries, and allows all others to
read these entries.  Note that only the first {{EX:by <who>}} clause
which matches applies.  Hence, the anonymous users are granted
{{EX:auth}}, not {{EX:read}}.  The last clause could just as well
have been "{{EX:by users read}}".

It is often desirable to restrict operations based upon the level
of protection in place.  The following shows how security strength
factors (SSF) can be used.

>    access to *
>        by ssf=128 self write
>        by ssf=64 anonymous auth
>        by ssf=64 users read

This directive allows users to modify their own entries if security
protections have of strength 128 or better have been established,
allows authentication access to anonymous users, and read access
when 64 or better security protections have been established.  If
client has not establish sufficient security protections, the
implicit {{EX:by * none}} clause would be applied.

The following example shows the use of a style specifiers to select
the entries by DN in two access directives where ordering is
significant.

>    access to dn.children="dc=example,dc=com"
>         by * search
>    access to dn.children="dc=com"
>         by * read

Read access is granted to entries under the {{EX:dc=com}} subtree,
except for those entries under the {{EX:dc=example,dc=com}} subtree,
to which search access is granted.  No access is granted to
{{EX:dc=com}} as neither access directive matches this DN.  If the
order of these access directives was reversed, the trailing directive
would never be reached, since all entries under {{EX:dc=example,dc=com}}
are also under {{EX:dc=com}} entries.

Also note that if no {{EX:access to}} directive matches or no {{EX:by
<who>}} clause, {{B:access is denied}}.  That is, every {{EX:access
to}} directive ends with an implicit {{EX:by * none}} clause. When dealing
with an access list, because the global access list is effectively appended 
to each per-database list, if the resulting list is non-empty then the access 
list will end with an implicit {{EX:access to * by * none}} directive. If
there are no access directives applicable to a backend, then a default read is
used.

The next example again shows the importance of ordering, both of
the access directives and the {{EX:by <who>}} clauses.  It also
shows the use of an attribute selector to grant access to a specific
attribute and various {{EX:<who>}} selectors.

>    access to dn.subtree="dc=example,dc=com" attrs=homePhone
>        by self write
>        by dn.children="dc=example,dc=com" search
>        by peername.regex=IP:10\..+ read
>    access to dn.subtree="dc=example,dc=com"
>        by self write
>        by dn.children="dc=example,dc=com" search
>        by anonymous auth

This example applies to entries in the "{{EX:dc=example,dc=com}}"
subtree. To all attributes except {{EX:homePhone}}, an entry can
write to itself, entries under {{EX:example.com}} entries can search
by them, anybody else has no access (implicit {{EX:by * none}})
excepting for authentication/authorization (which is always done
anonymously).  The {{EX:homePhone}} attribute is writable by the
entry, searchable by entries under {{EX:example.com}}, readable by
clients connecting from network 10, and otherwise not readable
(implicit {{EX:by * none}}).  All other access is denied by the
implicit {{EX:access to * by * none}}.

Sometimes it is useful to permit a particular DN to add or
remove itself from an attribute. For example, if you would like to
create a group and allow people to add and remove only
their own DN from the member attribute, you could accomplish
it with an access directive like this:

>    access to attrs=member,entry
>         by dnattr=member selfwrite

The dnattr {{EX:<who>}} selector says that the access applies to
entries listed in the {{EX:member}} attribute. The {{EX:selfwrite}} access
selector says that such members can only add or delete their
own DN from the attribute, not other values. The addition of
the entry attribute is required because access to the entry is
required to access any of the entry's attributes.

!if 0
For more details on how to use the {{EX:access}} directive,
consult the {{Advanced Access Control}} chapter.
!endif


H3: Configuration File Example

The following is an example configuration file, interspersed
with explanatory text. It defines two databases to handle
different parts of the {{TERM:X.500}} tree; both are {{TERM:BDB}}
database instances. The line numbers shown are provided for
reference only and are not included in the actual file. First, the
global configuration section:

E:  1.    # example config file - global configuration section
E:  2.    include /usr/local/etc/schema/core.schema
E:  3.    referral ldap://root.openldap.org
E:  4.    access to * by * read
 
Line 1 is a comment. Line 2 includes another config file
which contains {{core}} schema definitions.
The {{EX:referral}} directive on line 3
means that queries not local to one of the databases defined
below will be referred to the LDAP server running on the
standard port (389) at the host {{EX:root.openldap.org}}.

Line 4 is a global access control.  It applies to all
entries (after any applicable database-specific access
controls).

The next section of the configuration file defines a BDB
backend that will handle queries for things in the
"dc=example,dc=com" portion of the tree. The
database is to be replicated to two slave slapds, one on
truelies, the other on judgmentday. Indices are to be
maintained for several attributes, and the {{EX:userPassword}}
attribute is to be protected from unauthorized access.

E:  5.    # BDB definition for the example.com
E:  6.    database bdb
E:  7.    suffix "dc=example,dc=com"
E:  8.    directory /usr/local/var/openldap-data
E:  9.    rootdn "cn=Manager,dc=example,dc=com"
E: 10.    rootpw secret
E: 11.    # indexed attribute definitions
E: 12.    index uid pres,eq
E: 13.    index cn,sn,uid pres,eq,approx,sub
E: 14.    index objectClass eq
E: 15.    # database access control definitions
E: 16.    access to attrs=userPassword
E: 17.        by self write
E: 18.        by anonymous auth
E: 19.        by dn.base="cn=Admin,dc=example,dc=com" write
E: 20.        by * none
E: 21.    access to *
E: 22.        by self write
E: 23.        by dn.base="cn=Admin,dc=example,dc=com" write
E: 24.        by * read

Line 5 is a comment. The start of the database definition is marked
by the database keyword on line 6. Line 7 specifies the DN suffix
for queries to pass to this database. Line 8 specifies the directory
in which the database files will live.

Lines 9 and 10 identify the database {{super-user}} entry and associated
password. This entry is not subject to access control or size or
time limit restrictions.

Lines 12 through 14 indicate the indices to maintain for various
attributes.

Lines 16 through 24 specify access control for entries in this
database. For all applicable entries, the {{EX:userPassword}} attribute is writable
by the entry itself and by the "admin" entry.  It may be used for
authentication/authorization purposes, but is otherwise not readable.
All other attributes are writable by the entry and the "admin"
entry, but may be read by all users (authenticated or not).

The next section of the example configuration file defines another
BDB database. This one handles queries involving the
{{EX:dc=example,dc=net}} subtree but is managed by the same entity
as the first database.  Note that without line 39, the read access
would be allowed due to the global access rule at line 4.

E: 33.    # BDB definition for example.net
E: 34.    database bdb
E: 35.    suffix "dc=example,dc=net"
E: 36.    directory /usr/local/var/openldap-data-net
E: 37.    rootdn "cn=Manager,dc=example,dc=com"
E: 38.    index objectClass eq
E: 39.    access to * by users read

H2: Access Control via Dynamic Configuration

Access to slapd entries and attributes is controlled by the
olcAccess attribute, whose values are a sequence of access directives.
The general form of the olcAccess configuration is:

>    olcAccess: <access directive>
>    <access directive> ::= to <what>
>        [by <who> [<access>] [<control>] ]+
>    <what> ::= * |
>        [dn[.<basic-style>]=<regex> | dn.<scope-style>=<DN>]
>        [filter=<ldapfilter>] [attrs=<attrlist>]
>    <basic-style> ::= regex | exact
>    <scope-style> ::= base | one | subtree | children
>    <attrlist> ::= <attr> [val[.<basic-style>]=<regex>] | <attr> , <attrlist>
>    <attr> ::= <attrname> | entry | children
>    <who> ::= * | [anonymous | users | self
>            | dn[.<basic-style>]=<regex> | dn.<scope-style>=<DN>] 
>        [dnattr=<attrname>]
>        [group[/<objectclass>[/<attrname>][.<basic-style>]]=<regex>]
>        [peername[.<basic-style>]=<regex>]
>        [sockname[.<basic-style>]=<regex>]
>        [domain[.<basic-style>]=<regex>]
>        [sockurl[.<basic-style>]=<regex>]
>        [set=<setspec>]
>        [aci=<attrname>]
>    <access> ::= [self]{<level>|<priv>}
>    <level> ::= none | disclose | auth | compare | search | read | write | manage
>    <priv> ::= {=|+|-}{m|w|r|s|c|x|d|0}+
>    <control> ::= [stop | continue | break]

where the <what> part selects the entries and/or attributes to which
the access applies, the {{EX:<who>}} part specifies which entities
are granted access, and the {{EX:<access>}} part specifies the
access granted. Multiple {{EX:<who> <access> <control>}} triplets
are supported, allowing many entities to be granted different access
to the same set of entries and attributes. Not all of these access
control options are described here; for more details see the
{{slapd.access}}(5) man page.


H3: What to control access to

The <what> part of an access specification determines the entries
and attributes to which the access control applies.  Entries are
commonly selected in two ways: by DN and by filter.  The following
qualifiers select entries by DN:

>    to *
>    to dn[.<basic-style>]=<regex>
>    to dn.<scope-style>=<DN>

The first form is used to select all entries.  The second form may
be used to select entries by matching a regular expression against
the target entry's {{normalized DN}}.   (The second form is not
discussed further in this document.)  The third form is used to
select entries which are within the requested scope of DN.  The
<DN> is a string representation of the Distinguished Name, as
described in {{REF:RFC4514}}.

The scope can be either {{EX:base}}, {{EX:one}}, {{EX:subtree}},
or {{EX:children}}.  Where {{EX:base}} matches only the entry with
provided DN, {{EX:one}} matches the entries whose parent is the
provided DN, {{EX:subtree}} matches all entries in the subtree whose
root is the provided DN, and {{EX:children}} matches all entries
under the DN (but not the entry named by the DN).

For example, if the directory contained entries named:

>    0: o=suffix
>    1: cn=Manager,o=suffix
>    2: ou=people,o=suffix
>    3: uid=kdz,ou=people,o=suffix
>    4: cn=addresses,uid=kdz,ou=people,o=suffix
>    5: uid=hyc,ou=people,o=suffix

\Then:
. {{EX:dn.base="ou=people,o=suffix"}} match 2;
. {{EX:dn.one="ou=people,o=suffix"}} match 3, and 5;
. {{EX:dn.subtree="ou=people,o=suffix"}} match 2, 3, 4, and 5; and
. {{EX:dn.children="ou=people,o=suffix"}} match 3, 4, and 5.


Entries may also be selected using a filter:

>    to filter=<ldap filter>

where <ldap filter> is a string representation of an LDAP
search filter, as described in {{REF:RFC4515}}.  For example:

>    to filter=(objectClass=person)

Note that entries may be selected by both DN and filter by
including both qualifiers in the <what> clause.

>    to dn.one="ou=people,o=suffix" filter=(objectClass=person)

Attributes within an entry are selected by including a comma-separated
list of attribute names in the <what> selector:

>    attrs=<attribute list>

A specific value of an attribute is selected by using a single
attribute name and also using a value selector:

>    attrs=<attribute> val[.<style>]=<regex>

There are two special {{pseudo}} attributes {{EX:entry}} and
{{EX:children}}.  To read (and hence return) a target entry, the
subject must have {{EX:read}} access to the target's {{entry}}
attribute.  To perform a search, the subject must have
{{EX:search}} access to the search base's {{entry}} attribute.
To add or delete an entry, the subject must have
{{EX:write}} access to the entry's {{EX:entry}} attribute AND must
have {{EX:write}} access to the entry's parent's {{EX:children}}
attribute.  To rename an entry, the subject must have {{EX:write}}
access to entry's {{EX:entry}} attribute AND have {{EX:write}}
access to both the old parent's and new parent's {{EX:children}}
attributes.  The complete examples at the end of this section should
help clear things up.

Lastly, there is a special entry selector {{EX:"*"}} that is used to
select any entry.  It is used when no other {{EX:<what>}}
selector has been provided.  It's equivalent to "{{EX:dn=.*}}"


H3: Who to grant access to

The <who> part identifies the entity or entities being granted
access. Note that access is granted to "entities" not "entries."
The following table summarizes entity specifiers:

!block table; align=Center; coltags="EX,N"; \
    title="Table 5.3: Access Entity Specifiers"
Specifier|Entities
*|All, including anonymous and authenticated users
anonymous|Anonymous (non-authenticated) users
users|Authenticated users
self|User associated with target entry
dn[.<basic-style>]=<regex>|Users matching a regular expression
dn.<scope-style>=<DN>|Users within scope of a DN
!endblock

The DN specifier behaves much like <what> clause DN specifiers.

Other control factors are also supported.  For example, a {{EX:<who>}}
can be restricted by an entry listed in a DN-valued attribute in
the entry to which the access applies:

>    dnattr=<dn-valued attribute name>

The dnattr specification is used to give access to an entry
whose DN is listed in an attribute of the entry (e.g., give
access to a group entry to whoever is listed as the owner of
the group entry).

Some factors may not be appropriate in all environments (or any).
For example, the domain factor relies on IP to domain name lookups.
As these can easily be spoofed, the domain factor should be avoided.


H3: The access to grant

The kind of <access> granted can be one of the following:

!block table; colaligns="LRL"; coltags="EX,EX,N"; align=Center; \
    title="Table 5.4: Access Levels"
Level        Privileges    Description
none         =0            no access
disclose     =d            needed for information disclosure on error
auth         =dx           needed to authenticate (bind)
compare      =cdx          needed to compare
search       =scdx         needed to apply search filters
read         =rscdx        needed to read search results
write        =wrscdx       needed to modify/rename
manage       =mwrscdx      needed to manage
!endblock

Each level implies all lower levels of access. So, for example,
granting someone {{EX:write}} access to an entry also grants them
{{EX:read}}, {{EX:search}}, {{EX:compare}}, {{EX:auth}} and
{{EX:disclose}} access.  However, one may use the privileges specifier
to grant specific permissions.


H3: Access Control Evaluation

When evaluating whether some requester should be given access to
an entry and/or attribute, slapd compares the entry and/or attribute
to the {{EX:<what>}} selectors given in the configuration.  For
each entry, access controls provided in the database which holds
the entry (or the global access directives if not held in any database) apply
first, followed by the global access directives (which are held in
the {{EX:frontend}} database definition). However, when dealing with 
an access list, because the global access list is effectively appended 
to each per-database list, if the resulting list is non-empty then the 
access list will end with an implicit {{EX:access to * by * none}} directive. 
If there are no access directives applicable to a backend, then a default 
read is used.

Within this priority,
access directives are examined in the order in which they appear
in the configuration attribute.  Slapd stops with the first
{{EX:<what>}} selector that matches the entry and/or attribute. The
corresponding access directive is the one slapd will use to evaluate
access.

Next, slapd compares the entity requesting access to the {{EX:<who>}}
selectors within the access directive selected above in the order
in which they appear. It stops with the first {{EX:<who>}} selector
that matches the requester. This determines the access the entity
requesting access has to the entry and/or attribute.

Finally, slapd compares the access granted in the selected
{{EX:<access>}} clause to the access requested by the client. If
it allows greater or equal access, access is granted. Otherwise,
access is denied.

The order of evaluation of access directives makes their placement
in the configuration file important. If one access directive is
more specific than another in terms of the entries it selects, it
should appear first in the configuration. Similarly, if one {{EX:<who>}}
selector is more specific than another it should come first in the
access directive. The access control examples given below should
help make this clear.



H3: Access Control Examples

The access control facility described above is quite powerful.  This
section shows some examples of its use for descriptive purposes.

A simple example:

>    olcAccess: to * by * read

This access directive grants read access to everyone.

>    olcAccess: to *
>        by self write
>        by anonymous auth
>        by * read

This directive allows the user to modify their entry, allows anonymous
to authenticate against these entries, and allows all others to
read these entries.  Note that only the first {{EX:by <who>}} clause
which matches applies.  Hence, the anonymous users are granted
{{EX:auth}}, not {{EX:read}}.  The last clause could just as well
have been "{{EX:by users read}}".

It is often desirable to restrict operations based upon the level
of protection in place.  The following shows how security strength
factors (SSF) can be used.

>    olcAccess: to *
>        by ssf=128 self write
>        by ssf=64 anonymous auth
>        by ssf=64 users read

This directive allows users to modify their own entries if security
protections of strength 128 or better have been established,
allows authentication access to anonymous users, and read access
when strength 64 or better security protections have been established.  If
the client has not establish sufficient security protections, the
implicit {{EX:by * none}} clause would be applied.

The following example shows the use of style specifiers to select
the entries by DN in two access directives where ordering is
significant.

>    olcAccess: to dn.children="dc=example,dc=com"
>         by * search
>    olcAccess: to dn.children="dc=com"
>         by * read

Read access is granted to entries under the {{EX:dc=com}} subtree,
except for those entries under the {{EX:dc=example,dc=com}} subtree,
to which search access is granted.  No access is granted to
{{EX:dc=com}} as neither access directive matches this DN.  If the
order of these access directives was reversed, the trailing directive
would never be reached, since all entries under {{EX:dc=example,dc=com}}
are also under {{EX:dc=com}} entries.

Also note that if no {{EX:olcAccess: to}} directive matches or no {{EX:by
<who>}} clause, {{B:access is denied}}.  When dealing with an access list, 
because the global access list is effectively appended to each per-database 
list, if the resulting list is non-empty then the access list will end with 
an implicit {{EX:access to * by * none}} directive. If there are no access 
directives applicable to a backend, then a default read is used.

The next example again shows the importance of ordering, both of
the access directives and the {{EX:by <who>}} clauses.  It also
shows the use of an attribute selector to grant access to a specific
attribute and various {{EX:<who>}} selectors.

>    olcAccess: to dn.subtree="dc=example,dc=com" attrs=homePhone
>        by self write
>        by dn.children=dc=example,dc=com" search
>        by peername.regex=IP:10\..+ read
>    olcAccess: to dn.subtree="dc=example,dc=com"
>        by self write
>        by dn.children="dc=example,dc=com" search
>        by anonymous auth

This example applies to entries in the "{{EX:dc=example,dc=com}}"
subtree. To all attributes except {{EX:homePhone}}, an entry can
write to itself, entries under {{EX:example.com}} entries can search
by them, anybody else has no access (implicit {{EX:by * none}})
excepting for authentication/authorization (which is always done
anonymously).  The {{EX:homePhone}} attribute is writable by the
entry, searchable by entries under {{EX:example.com}}, readable by
clients connecting from network 10, and otherwise not readable
(implicit {{EX:by * none}}).  All other access is denied by the
implicit {{EX:access to * by * none}}.

Sometimes it is useful to permit a particular DN to add or
remove itself from an attribute. For example, if you would like to
create a group and allow people to add and remove only
their own DN from the member attribute, you could accomplish
it with an access directive like this:

>    olcAccess: to attrs=member,entry
>         by dnattr=member selfwrite

The dnattr {{EX:<who>}} selector says that the access applies to
entries listed in the {{EX:member}} attribute. The {{EX:selfwrite}} access
selector says that such members can only add or delete their
own DN from the attribute, not other values. The addition of
the entry attribute is required because access to the entry is
required to access any of the entry's attributes.



H3: Access Control Ordering

Since the ordering of {{EX:olcAccess}} directives is essential to their
proper evaluation, but LDAP attributes normally do not preserve the
ordering of their values, OpenLDAP uses a custom schema extension to
maintain a fixed ordering of these values. This ordering is maintained
by prepending a {{EX:"{X}"}} numeric index to each value, similarly to
the approach used for ordering the configuration entries. These index
tags are maintained automatically by slapd and do not need to be specified
when originally defining the values. For example, when you create the
settings

>    olcAccess: to attrs=member,entry
>         by dnattr=member selfwrite
>    olcAccess: to dn.children="dc=example,dc=com"
>         by * search
>    olcAccess: to dn.children="dc=com"
>         by * read

when you read them back using slapcat or ldapsearch they will contain

>    olcAccess: {0}to attrs=member,entry
>         by dnattr=member selfwrite
>    olcAccess: {1}to dn.children="dc=example,dc=com"
>         by * search
>    olcAccess: {2}to dn.children="dc=com"
>         by * read

The numeric index may be used to specify a particular value to change
when using ldapmodify to edit the access rules. This index can be used
instead of (or in addition to) the actual access value. Using this 
numeric index is very helpful when multiple access rules are being managed.

For example, if we needed to change the second rule above to grant
write access instead of search, we could try this LDIF:

>    changetype: modify
>    delete: olcAccess
>    olcAccess: to dn.children="dc=example,dc=com" by * search
>    -
>    add: olcAccess
>    olcAccess: to dn.children="dc=example,dc=com" by * write
>    -

But this example {{B:will not}} guarantee that the existing values remain in
their original order, so it will most likely yield a broken security
configuration. Instead, the numeric index should be used:

>    changetype: modify
>    delete: olcAccess
>    olcAccess: {1}
>    -
>    add: olcAccess
>    olcAccess: {1}to dn.children="dc=example,dc=com" by * write
>    -

This example deletes whatever rule is in value #1 of the {{EX:olcAccess}}
attribute (regardless of its value) and adds a new value that is
explicitly inserted as value #1. The result will be

>    olcAccess: {0}to attrs=member,entry
>         by dnattr=member selfwrite
>    olcAccess: {1}to dn.children="dc=example,dc=com"
>         by * write
>    olcAccess: {2}to dn.children="dc=com"
>         by * read

which is exactly what was intended.

!if 0
For more details on how to use the {{EX:access}} directive,
consult the {{Advanced Access Control}} chapter.
!endif


H3: Configuration Example

The following is an example configuration, interspersed
with explanatory text. It defines two databases to handle
different parts of the {{TERM:X.500}} tree; both are {{TERM:BDB}}
database instances. The line numbers shown are provided for
reference only and are not included in the actual file. First, the
global configuration section:

E:  1.    # example config file - global configuration entry
E:  2.    dn: cn=config
E:  3.    objectClass: olcGlobal
E:  4.    cn: config
E:  5.    olcReferral: ldap://root.openldap.org
E:  6.    

Line 1 is a comment. Lines 2-4 identify this as the global
configuration entry.
The {{EX:olcReferral:}} directive on line 5
means that queries not local to one of the databases defined
below will be referred to the LDAP server running on the
standard port (389) at the host {{EX:root.openldap.org}}.
Line 6 is a blank line, indicating the end of this entry.

E:  7.    # internal schema
E:  8.    dn: cn=schema,cn=config
E:  9.    objectClass: olcSchemaConfig
E: 10.    cn: schema
E: 11.    

Line 7 is a comment. Lines 8-10 identify this as the root of
the schema subtree. The actual schema definitions in this entry
are hardcoded into slapd so no additional attributes are specified here.
Line 11 is a blank line, indicating the end of this entry.

E: 12.    # include the core schema
E: 13.    include: file:///usr/local/etc/openldap/schema/core.ldif
E: 14.    

Line 12 is a comment. Line 13 is an LDIF include directive which
accesses the {{core}} schema definitions in LDIF format. Line 14
is a blank line.

Next comes the database definitions. The first database is the
special {{EX:frontend}} database whose settings are applied globally
to all the other databases.

E: 15.    # global database parameters
E: 16.    dn: olcDatabase=frontend,cn=config
E: 17.    objectClass: olcDatabaseConfig
E: 18.    olcDatabase: frontend
E: 19.    olcAccess: to * by * read
E: 20.    

Line 15 is a comment. Lines 16-18 identify this entry as the global
database entry. Line 19 is a global access control. It applies to all
entries (after any applicable database-specific access controls).

The next entry defines a BDB backend that will handle queries for things
in the "dc=example,dc=com" portion of the tree. Indices are to be maintained
for several attributes, and the {{EX:userPassword}} attribute is to be
protected from unauthorized access.

E: 21.    # BDB definition for example.com
E: 22.    dn: olcDatabase=bdb,cn=config
E: 23.    objectClass: olcDatabaseConfig
E: 24.    objectClass: olcBdbConfig
E: 25.    olcDatabase: bdb
E: 26.    olcSuffix: "dc=example,dc=com"
E: 27.    olcDbDirectory: /usr/local/var/openldap-data
E: 28.    olcRootDN: "cn=Manager,dc=example,dc=com"
E: 29.    olcRootPW: secret
E: 30.    olcDbIndex: uid pres,eq
E: 31.    olcDbIndex: cn,sn,uid pres,eq,approx,sub
E: 32.    olcDbIndex: objectClass eq
E: 33.    olcAccess: to attrs=userPassword
E: 34.      by self write
E: 35.      by anonymous auth
E: 36.      by dn.base="cn=Admin,dc=example,dc=com" write
E: 37.      by * none
E: 38.    olcAccess: to *
E: 39.      by self write
E: 40.      by dn.base="cn=Admin,dc=example,dc=com" write
E: 41.      by * read
E: 42.    

Line 21 is a comment. Lines 22-25 identify this entry as a BDB database
configuration entry.  Line 26 specifies the DN suffix
for queries to pass to this database. Line 27 specifies the directory
in which the database files will live.

Lines 28 and 29 identify the database {{super-user}} entry and associated
password. This entry is not subject to access control or size or
time limit restrictions.

Lines 30 through 32 indicate the indices to maintain for various
attributes.

Lines 33 through 41 specify access control for entries in this
database. For all applicable entries, the {{EX:userPassword}} attribute is writable
by the entry itself and by the "admin" entry.  It may be used for
authentication/authorization purposes, but is otherwise not readable.
All other attributes are writable by the entry and the "admin"
entry, but may be read by all users (authenticated or not).

Line 42 is a blank line, indicating the end of this entry.

The next section of the example configuration file defines another
BDB database. This one handles queries involving the
{{EX:dc=example,dc=net}} subtree but is managed by the same entity
as the first database.  Note that without line 52, the read access
would be allowed due to the global access rule at line 19.

E: 43.    # BDB definition for example.net
E: 44.    dn: olcDatabase=bdb,cn=config
E: 45.    objectClass: olcDatabaseConfig
E: 46.    objectClass: olcBdbConfig
E: 47.    olcDatabase: bdb
E: 48.    olcSuffix: "dc=example,dc=net"
E: 49.    olcDbDirectory: /usr/local/var/openldap-data-net
E: 50.    olcRootDN: "cn=Manager,dc=example,dc=com"
E: 51.    olcDbIndex: objectClass eq
E: 52.    olcAccess: to * by users read


H3: Converting from {{slapd.conf}}(5) to a {{B:cn=config}} directory format

Discuss slap* -f slapd.conf -F slapd.d/  (man slapd-config)


H2: Access Control Common Examples

H3: Basic ACLs

Generally one should start with some basic ACLs such as:

>    access to attr=userPassword
>        by self =xw
>        by anonymous auth
>        by * none
>
>
>      access to *
>        by self write
>        by users read
>        by * none

The first ACL allows users to update (but not read) their passwords, anonymous 
users to authenticate against this attribute, and (implicitly) denying all 
access to others.

The second ACL allows users full access to their entry, authenticated users read 
access to anything, and (implicitly) denying all access to others (in this case, 
anonymous users). 


H3: Matching Anonymous and Authenticated users

An anonymous user has a empty DN. While the {{dn.exact=""}} or {{dn.regex="^$"}}
 could be used, {{slapd}}(8)) offers an anonymous shorthand which should be 
used instead.

>    access to *
>      by anonymous none
>      by * read

denies all access to anonymous users while granting others read. 

Authenticated users have a subject DN. While {{dn.regex=".+"}} will match any 
authenticated user, OpenLDAP provides the users short hand which should be used 
instead.

>    access to *
>      by users read
>      by * none

This ACL grants read permissions to authenticated users while denying others 
(i.e.: anonymous users).


H3: Controlling rootdn access

You could specify the {{rootdn}} in {{slapd.conf}}(5) or {[slapd.d}} without 
specifying a {{rootpw}}. Then you have to add an actual directory entry with 
the same dn, e.g.:

>    dn: cn=Manager,o=MyOrganization
>    cn: Manager
>    sn: Manager
>    objectClass: person
>    objectClass: top
>    userPassword: {SSHA}someSSHAdata

Then binding as the {{rootdn}} will require a regular bind to that DN, which 
in turn requires auth access to that entry's DN and {{userPassword}}, and this 
can be restricted via ACLs. E.g.:

>    access to dn.base="cn=Manager,o=MyOrganization"
>      by peername.regex=127\.0\.0\.1 auth
>      by peername.regex=192\.168\.0\..* auth
>      by users none
>      by * none

The ACLs above will only allow binding using rootdn from localhost and 
192.168.0.0/24.


H3: Managing access with Groups

There are a few ways to do this. One approach is illustrated here. Consider the 
following DIT layout:

>    +-dc=example,dc=com
>    +---cn=administrators,dc=example,dc=com
>    +---cn=fred blogs,dc=example,dc=com 

and the following group object (in LDIF format):

>    dn: cn=administrators,dc=example,dc=com
>    cn: administrators of this region
>    objectclass: groupOfNames  (important for the group acl feature)
>    member: cn=fred blogs,dc=example,dc=com 
>    member: cn=somebody else,dc=example,dc=com

One can then grant access to the members of this this group by adding appropriate 
{{by group}} clause to an access directive in {{slapd.conf}}(5). For instance,

>    access to dn.children="dc=example,dc=com" 
>        by self write 
>        by group.exact="cn=Administrators,dc=example,dc=com" write  
>        by * auth

Like by {[dn}} clauses, one can also use {{expand}} to expand the group name 
based upon the regular expression matching of the target, that is, the to {{dn.regex}}). 
For instance,

>    access to dn.regex="(.+,)?ou=People,(dc=[^,]+,dc=[^,]+)$"
>             attrs=children,entry,uid
>        by group.expand="cn=Managers,$2" write
>        by users read
>        by * auth


The above illustration assumed that the group members are to be found in the 
{{member}} attribute type of the {{groupOfNames}} object class. If you need to 
use a different group object and/or a different attribute type then use the 
following {{slapd.conf}}(5) (abbreviated) syntax:

>    access to <what>
>            by group/<objectclass>/<attributename>=<DN> <access>

For example:

>    access to *
>      by group/organizationalRole/roleOccupant="cn=Administrator,dc=example,dc=com" write

In this case, we have an ObjectClass {{organizationalRole}} which contains the 
administrator DN's in the {{roleOccupant}} attribute. For instance:

>    dn: cn=Administrator,dc=example,dc=com
>    cn: Administrator
>    objectclass: organizationalRole
>    roleOccupant: cn=Jane Doe,dc=example,dc=com 

Note: the specified member attribute type MUST be of DN or {{NameAndOptionalUID}} syntax, 
and the specified object class SHOULD allow the attribute type.

Dynamic Groups are also supported in Access Control. Please see {{slapo-dynlist}}(5)
and the {{SECT:Dynamic Lists}} overlay section.


H3:  Granting access to a subset of attributes

You can grant access to a set of attributes by specifying a list of attribute names 
in the ACL {{to}} clause. To be useful, you also need to grant access to the 
{{entry}} itself. Also note how {{children}} controls the ability to add, delete, 
and rename entries.

>    # mail: self may write, authenticated users may read
>    access to attrs=mail
>      by self write
>      by users read
>      by * none
>    
>    # cn, sn: self my write, all may read
>    access to attrs=cn,sn
>      by self write
>      by * read
>    
>    # immediate children: only self can add/delete entries under this entry
>    access to attrs=children
>      by self write
>    
>    # entry itself: self may write, all may read
>    access to attrs=entry
>      by self write
>      by * read
>    
>    # other attributes: self may write, others have no access
>    access to *
>      by self write
>      by * none

ObjectClass names may also be specified in this list, which will affect 
all the attributes that are required and/or allowed by that {{objectClass}}. 
Actually, names in {{attrlist}} that are prefixed by {{@}} are directly treated 
as objectClass names. A name prefixed by {{!}} is also treated as an objectClass, 
but in this case the access rule affects the attributes that are not required 
nor allowed by that {{objectClass}}. 


H3: Allowing a user write to all entries below theirs

For a setup where a user can write to its own record and to all of its children:

>    access to dn.regex="(.+,)?(uid=[^,]+,o=Company)$"
>       by dn.exact,expand="$2" write
>       by anonymous auth

(Add more examples for above)


H3: Allowing entry creation

Let's say, you have it like this:

>        o=<basedn>
>            ou=domains
>                associatedDomain=<somedomain>
>                    ou=users
>                        uid=<someuserid>            
>                        uid=<someotheruserid>
>                    ou=addressbooks
>                        uid=<someuserid>
>                            cn=<someone>
>                            cn=<someoneelse>

and, for another domain <someotherdomain>:

>        o=<basedn>
>            ou=domains
>                associatedDomain=<someotherdomain>
>                    ou=users
>                        uid=<someuserid>            
>                        uid=<someotheruserid>
>                    ou=addressbooks
>                        uid=<someotheruserid>
>                            cn=<someone>
>                            cn=<someoneelse>

then, if you wanted user {{uid=<someuserid>}} to {{B:ONLY}} create an entry 
for its own thing, you could write an ACL like this:

>    # this rule lets users of "associatedDomain=<matcheddomain>"
>    # write under "ou=addressbook,associatedDomain=<matcheddomain>,ou=domains,o=<basedn>",
>    # i.e. a user can write ANY entry below its domain's address book;
>    # this permission is necessary, but not sufficient, the next 
>    # will restrict this permission further
>    
>    
>    access to dn.regex="^ou=addressbook,associatedDomain=([^,]+),ou=domains,o=<basedn>$" attrs=children
>            by dn.regex="^uid=([^,]+),ou=users,associatedDomain=$1,ou=domains,o=<basedn>$$" write
>            by * none
>    
>    
>    # Note that above the "by" clause needs a "regex" style to make sure
>    # it expands to a DN that starts with a "uid=<someuserid>" pattern
>    # while substituting the associatedDomain submatch from the "what" clause.
>    
>    
>    # This rule lets a user with "uid=<matcheduid>" of "<associatedDomain=matcheddomain>"
>    # write (i.e. add, modify, delete) the entry whose DN is exactly
>    # "uid=<matcheduid>,ou=addressbook,associatedDomain=<matcheddomain>,ou=domains,o=<basedn>"
>    # and ANY entry as subtree of it
>    
>    
>    access to dn.regex="^(.+,)?uid=([^,]+),ou=addressbook,associatedDomain=([^,]+),ou=domains,o=<basedn>$"
>            by dn.exact,expand="uid=$2,ou=users,associatedDomain=$3,ou=domains,o=<basedn>" write
>            by * none 
>    
>    
>    # Note that above the "by" clause uses the "exact" style with the "expand"
>    # modifier because now the whole pattern can be rebuilt by means of the
>    # submatches from the "what" clause, so a "regex" compilation and evaluation
>    # is no longer required.


H3: Tips for using regular expressions in Access Control 

Always use {{dn.regex=<pattern>}} when you intend to use regular expression 
matching. {{dn=<pattern>}} alone defaults to {{dn.exact<pattern>}}.

Use {{(.+)}} instead of {{(.*)}} when you want at least one char to be matched. 
{{(.*)}} matches the empty string as well.

Don't use regular expressions for matches that can be done otherwise in a safer 
and cheaper manner. Examples:

>    dn.regex=".*dc=example,dc=com"

is unsafe and expensive:

    * unsafe because any string containing {{dc=example,dc=com }}will match, 
not only those that end with the desired pattern; use {{.*dc=example,dc=com$}} instead.
    * unsafe also because it would allow any {{attributeType}} ending with {{dc}}
 as naming attribute for the first RDN in the string, e.g. a custom attributeType 
{{mydc}} would match as well. If you really need a regular expression that allows 
just {{dc=example,dc=com}} or any of its subtrees, use {{^(.+,)?dc=example,dc=com$}}, 
which means: anything to the left of dc=..., if any (the question mark after the 
pattern within brackets), must end with a comma;
    * expensive because if you don't need submatches, you could use scoping styles, e.g.

>    dn.subtree="dc=example,dc=com"

to include {{dc=example,dc=com}} in the matching patterns,

>    dn.children="dc=example,dc=com"

to exclude {{dc=example,dc=com}} from the matching patterns, or

>    dn.onelevel="dc=example,dc=com"

to allow exactly one sublevel matches only. 

Always use {{^}} and {{$}} in regexes, whenever appropriate, because 
{{ou=(.+),ou=(.+),ou=addressbooks,o=basedn}} will match 
{{something=bla,ou=xxx,ou=yyy,ou=addressbooks,o=basedn,ou=addressbooks,o=basedn,dc=some,dc=org}}

Always use {{([^,]+)}} to indicate exactly one RDN, because {{(.+)}} can 
include any number of RDNs; e.g. {{ou=(.+),dc=example,dc=com}} will match 
{{ou=My,o=Org,dc=example,dc=com}}, which might not be what you want.

Never add the rootdn to the by clauses. ACLs are not even processed for operations 
performed with rootdn identity (otherwise there would be no reason to define a 
rootdn at all).

Use shorthands. The user directive matches authenticated users and the anonymous
directive matches anonymous users.

Don't use the {{dn.regex}} form for <by> clauses if all you need is scoping 
and/or substring replacement; use scoping styles (e.g. {{exact}}, {{onelevel}}, 
{{children}} or {{subtree}}) and the style modifier expand to cause substring expansion.

For instance,

>    access to dn.regex=".+,dc=([^,]+),dc=([^,]+)$"
>      by dn.regex="^[^,],ou=Admin,dc=$1,dc=$2$$" write

although correct, can be safely and efficiently replaced by

>    access to dn.regex=".+,(dc=[^,]+,dc=[^,]+)$"
>      by dn.onelevel,expand="ou=Admin,$1" write

where the regex in the {{<what>}} clause is more compact, and the one in the {{<by>}} 
clause is replaced by a much more efficient scoping style of onelevel with substring expansion. 


H3: Granting and Denying access based on security strength factors (ssf)

You can restrict access based on the security strength factor (SSF)

>    access to dn="cn=example,cn=edu"
>          by * ssf=256 read

0 (zero) implies no protection,
1 implies integrity protection only,
56 DES or other weak ciphers,
112 triple DES and other strong ciphers,
128 RC4, Blowfish and other modern strong ciphers.

Other possibilities:

>    transport_ssf=<n>
>    tls_ssf=<n>
>    sasl_ssf=<n>

256 is recommended.

See {{slapd.conf}}(5) for information on {{ssf}}.


H3: When things aren't working as expected

Consider this example:

>    access to *
>      by anonymous auth
>    
>    access to *
>      by self write
>    
>    access to *
>      by users read 

You may think this will allow any user to login, to read everything and change 
his own data if he is logged in. But in this example only the login works and 
an ldapsearch returns no data. The Problem is that SLAPD goes through its access 
config line by line and stops as soon as it finds a match in the part of the 
access rule.(here: {{to *}})

To get what we wanted the file has to read:

>    access to *
>      by anonymous auth
>      by self write
>      by users read 

The general rule is: "special access rules first, generic access rules last"

See also {{slapd.access}}(8), loglevel 128 and {{slapacl}}(8) for debugging
information.


H2: Sets - Granting rights based on relationships

Sets are best illustrated via examples. The following sections will present 
a few set ACL examples in order to facilitate their understanding.

(Sets in Access Controls FAQ Entry: {{URL:http://www.openldap.org/faq/data/cache/1133.html}})

Note: Sets are considered experimental. 


H3: Groups of Groups

The OpenLDAP ACL for groups doesn't expand groups within groups, which are
groups that have another group as a member. For example:

> dn: cn=sudoadm,ou=group,dc=example,dc=com
> cn: sudoadm
> objectClass: groupOfNames
> member: uid=john,ou=people,dc=example,dc=com
> member: cn=accountadm,ou=group,dc=example,dc=com
>
> dn: cn=accountadm,ou=group,dc=example,dc=com
> cn: accountadm
> objectClass: groupOfNames
> member: uid=mary,ou=people,dc=example,dc=com

If we use standard group ACLs with the above entries and allow members of the
{{F:sudoadm}} group to write somewhere, {{F:mary}} won't be included:

> access to dn.subtree="ou=sudoers,dc=example,dc=com"
>         by group.exact="cn=sudoadm,ou=group,dc=example,dc=com" write
>         by * read

With sets we can make the ACL be recursive and consider group within groups. So
for each member that is a group, it is further expanded:

> access to dn.subtree="ou=sudoers,dc=example,dc=com"
>       by set="[cn=sudoadm,ou=group,dc=example,dc=com]/member* & user" write
>       by * read

This set ACL means: take the {{F:cn=sudoadm}} DN, check its {{F:member}}
attribute(s) (where the "{{F:*}}" means recursively) and intersect the result
with the authenticated user's DN. If the result is non-empty, the ACL is
considered a match and write access is granted.

The following drawing explains how this set is built:
!import "set-recursivegroup.png"; align="center"; title="Building a recursive group"
FT[align="Center"] Figure X.Y: Populating a recursive group set

First we get the {{F:uid=john}} DN. This entry doesn't have a {{F:member}}
attribute, so the expansion stops here.  Now we get to {{F:cn=accountadm}}.
This one does have a {{F:member}} attribute, which is {{F:uid=mary}}. The
{{F:uid=mary}} entry, however, doesn't have member, so we stop here again. The
end comparison is:

> {"uid=john,ou=people,dc=example,dc=com","uid=mary,ou=people,dc=example,dc=com"} & user

If the authenticated user's DN is any one of those two, write access is
granted. So this set will include {{F:mary}} in the {{F:sudoadm}} group and she
will be allowed the write access.

H3: Group ACLs without DN syntax

The traditional group ACLs, and even the previous example about recursive groups, require
that the members are specified as DNs instead of just usernames.

With sets, however, it's also possible to use simple names in group ACLs, as this example will
show.

Let's say we want to allow members of the {{F:sudoadm}} group to write to the
{{F:ou=suders}} branch of our tree. But our group definition now is using {{F:memberUid}} for
the group members:

> dn: cn=sudoadm,ou=group,dc=example,dc=com
> cn: sudoadm
> objectClass: posixGroup
> gidNumber: 1000
> memberUid: john

With this type of group, we can't use group ACLs. But with a set ACL we can
grant the desired access:

> access to dn.subtree="ou=sudoers,dc=example,dc=com"
>       by set="[cn=sudoadm,ou=group,dc=example,dc=com]/memberUid & user/uid" write
>       by * read

We use a simple intersection where we compare the {{F:uid}} attribute
of the connecting (and authenticated) user with the {{F:memberUid}} attributes
of the group. If they match, the intersection is non-empty and the ACL will
grant write access.

This drawing illustrates this set when the connecting user is authenticated as
{{F:uid=john,ou=people,dc=example,dc=com}}:
!import "set-memberUid.png"; align="center"; title="Sets with memberUid"
FT[align="Center"] Figure X.Y: Sets with {{F:memberUid}}

In this case, it's a match. If it were {{F:mary}} authenticating, however, she
would be denied write access to {{F:ou=sudoers}} because her {{F:uid}}
attribute is not listed in the group's {{F:memberUid}}.

H3: Following references

We will now show a quite powerful example of what can be done with sets. This
example tends to make OpenLDAP administrators smile after they have understood
it and its implications.

Let's start with an user entry:

> dn: uid=john,ou=people,dc=example,dc=com
> uid: john
> objectClass: inetOrgPerson
> givenName: John
> sn: Smith
> cn: john
> manager: uid=mary,ou=people,dc=example,dc=com

Writing an ACL to allow the manager to update some attributes is quite simple
using sets:

> access to dn.exact="uid=john,ou=people,dc=example,dc=com"
>    attrs=carLicense,homePhone,mobile,pager,telephoneNumber
>    by self write
>    by set="this/manager & user" write
>    by * read

In that set, {{F:this}} expands to the entry being accessed, so that
{{F:this/manager}} expands to {{F:uid=mary,ou=people,dc=example,dc=com}} when
john's entry is accessed.  If the manager herself is accessing John's entry,
the ACL will match and write access to those attributes will be granted.

So far, this same behavior can be obtained with the {{F:dnattr}} keyword. With
sets, however, we can further enhance this ACL. Let's say we want to allow the
secretary of the manager to also update these attributes. This is how we do it:

> access to dn.exact="uid=john,ou=people,dc=example,dc=com"
>    attrs=carLicense,homePhone,mobile,pager,telephoneNumber
>    by self write
>    by set="this/manager & user" write
>    by set="this/manager/secretary & user" write
>    by * read

Now we need a picture to help explain what is happening here (entries shortened
for clarity):

!import "set-following-references.png"; align="center"; title="Sets jumping through entries"
FT[align="Center"] Figure X.Y: Sets jumping through entries

In this example, Jane is the secretary of Mary, which is the manager of John.
This whole relationship is defined with the {{F:manager}} and {{F:secretary}}
attributes, which are both of the distinguishedName syntax (i.e., full DNs).
So, when the {{F:uid=john}} entry is being accessed, the
{{F:this/manager/secretary}} set becomes
{{F:{"uid=jane,ou=people,dc=example,dc=com"}}} (follow the references in the
picture):

> this = [uid=john,ou=people,dc=example,dc=com]
> this/manager = \
>   [uid=john,ou=people,dc=example,dc=com]/manager = uid=mary,ou=people,dc=example,dc=com
> this/manager/secretary = \
>   [uid=mary,ou=people,dc=example,dc=com]/secretary = uid=jane,ou=people,dc=example,dc=com

The end result is that when Jane accesses John's entry, she will be granted
write access to the specified attributes. Better yet, this will happen to any
entry she accesses which has Mary as the manager.

This is all cool and nice, but perhaps gives to much power to secretaries. Maybe we need to further
restrict it. For example, let's only allow executive secretaries to have this power:

> access to dn.exact="uid=john,ou=people,dc=example,dc=com"
>   attrs=carLicense,homePhone,mobile,pager,telephoneNumber
>   by self write
>   by set="this/manager & user" write
>   by set="this/manager/secretary & 
>           [cn=executive,ou=group,dc=example,dc=com]/member* & 
>           user" write
>   by * read

It's almost the same ACL as before, but we now also require that the connecting user be a member
of the (possibly nested) {{F:cn=executive}} group.