maandag 21 juli 2008


Explain newrole versus sudo to: transition user domain.

in f10 selinux-policy-targeted newrole is no longer encouraged to be used for domain transitions. in policy mls it is still used (but mls is not common)

sudo allows root privileges without actually knowing the root password.

su with newrole still requires a user to enter a root password to gain root privileges.

not needing a root password anymore is a big advantage!


from the wiki:
SELinux and virtualization (relabeling images if images are not in /etc/xen/).

this is not a domain specific issue. customized types. the solution should be man virt_selinux or man xen_selinux like man httpd_selinux. however there is no man xen_selinux yet etc.
this may no longer be an issue in fedora 10 , virt is working on a solution to be implemented in f10


make sure to use proper selinux terminology consequent in an effort to keep it simple.

subject object, security context, domain types, file types, port types, security level, categories, domain transition, executable file types, domain, application domain, user domain, init daemon, classes, attributes, scontext, tcontext, access vector cache, type enforcement, multi level security, multi category security.


from the wiki:

• Do mount points need to be mnt_t?

boolean mount_any_file. this is also kind of a customized type issue. use semanage boolean -l to list all booleans and their explanation. for example:

/usr/sbin/semanage boolean -l | grep mount

sh-3.2# /usr/sbin/semanage boolean -l | grep mount
allow_mount_anyfile -> off Allow the mount command to mount any directory or file.
xguest_mount_media -> off Allow xguest users to mount removable media

consideration: you should consider to not mention get,setsebool. semanage also provides this functionality. keep it simple


from the wiki:
mislabeled files, relabeled but still problems, touch /.autorelabel (Dans journal).
aplain how touch /.autorelabel && reboot relates to fixfiles relabel


different policies:
first there was strict and optional mls (for dod). strict was in fedora2. it was introduced too soon. users were too restricted. targeted was introduced to avoid restriction on users.unconfined domain is a domain exempted fro m most selinux policy. unconfined is a property of targeted policy. targeted policy only targets a select group. strict targets everything on a system. targeted plus multi category security (mcs) which is a (poor) implementation of confidentiality(its discretionary users can chcat aslong as they are member of the cat. mcs in policy mls is part of security level. in policy mls, mcs is is strict plus BLP, plus MCS, plus MLS. strict no longer maintained, merged with targeted. (strict plus unconfined domain) if you remove the unconfined domain then you use what use to be strict. at the moment there are two selinux policy models maintained: policy targeted and policy-mls. mls aims to enforces confidentiality BLP.

SELinux policy and dependencies.

A policy module has 3 files. Here is the explaination of the 3 files.

mydomain.te (.te) (type enforcement file) it has PRIVATE policy for the "mydomain" policy module.

mydomain.if (.if) (interface file) it has PUBLIC policy for the "mydomain" policy module.

mydomain.fc (.fc) (file context file) it has file contexts for the "mydomain" policy module.

The type enforcement file.

This file has private policy. Policy that is, in my example, related to "mydomain"

for example, you might find a rule like this in the mydomain.te file:

This policy was provided by apache.if to "mydomain". You can look it up in the apache.if file. It is really a template or interface with rules for how to read apaches user content. We are using (instantiating) that interface that apache policy module provides in it's apache.if file, in our mydomain.te file.

Let us refer to interfaces and templates as blocks of public policy. Public policy blocks should be prefixed by the policy module name of the domain that facilitates it in it's .if (interface file).

for example, just by looking at the following interface call in mydomain.te i know: 1. which module provided the interface 2. where to roughly find it. 3. where to find what policy te interface provides. 4. which domain instantiates the block of public policy:


1. provided by the alsa policy module.
2. can be found in alsa.if
3. Summary: Read alsa writable config files

allow $1 alsa_etc_rw_t:dir list_dir_perms;

4. this policy is instantiated by mydomain_t domain.

So you can easily from looking at a .te file know the modules dependencies by parsing each called interface prefix. as each called interface is prefixed by the domain that made it available in its interface file.

important note regarding public policy.

creating a quick policy module package(.pp) can be very handy for implementing quick policy. but it is also limited.

to compile policy one need selinux-devel. it has development files for each module that is used by the compiler to see if the policy that we want to compile is valid.

when you compile and install a seperate policy package with semodule -i mydomain.pp for example. there will not be a devel package installed.

interfaces files are therefore rendered useless for seperate policy module packages. for the reason that other modules will not be able too instantiate any public policy for that module.

the reason is that when you try to compile your module that has a call to a public policy block of a module that was installed with semodule, the compiler will nnot find that interface/ template in its devel files because non were installed!

This is important to know!
do you want to develop and implement much policy, then do not use policy module packages with semodule but instead integrate your module into the selinux-policy source provided upstream, rebuild it and reinstall it.

by rebuilding selinux-policy, a new selinux-policy-devel package is created. this selinux-policy-devel package DOES include the public policy for the domain that you integrated and thus is usable as opposed to using a .pp with semodule.

Basic access control models ( DAC , MAC ) ( not so basic MDAC )

explain discretionary
explain the dac model attributes: user group permission bits
explain why dac acl is not sufficient. example privilege escalation
explain the mac model attributes: security context
explain mandatory
explain that MAC is ACL layer on top of the DAC ACL layer
explain Type enforcement
explain Role Based AC
explain Multi Level Security
Explain Multi Category/Compartment Security

compare a selinux system to a submarine with compartments. if one compartment has a leak, the water will be contained to that compartment and will not be able to spread ( escalate) . submarine will not sink

Security context / SELinux attributes

explain the security context tuple and how to read it (explain the fields)
explain user ( which SELinux user (group) created the object? )
explain type is the attribute for type enforcement (TE)
explain role is the attribute for role enforcement (RBAC)
explain security level is the attribute for security level enforcement (MLS)
explain categories/compartments is the attribute for security level enforcement or category/compartment enforcement (MLS or MCS)

Subjects and objects ( processes and "files" )

explain that everything in a system is a object
explain that even subjects in a system are represented as objects in proc mountpoint
explain subjects and objects
explain subjects are processes (ps auxZ)
explain objects are "files" (ls -alZ)
- file objects ( files , lnk files, dirs, fifo files, sock files etc)
- port objects
- interface objects
- node objects
- objects available by other programs ACE access control extension: XACE, sepostgesql, SEDBUS, mscd, etc.
- explain object is a class defined in kernel :process :file :tcp_socket
example of a class: process. example of a class: file
explain domain type is the attribute of a process ( user_t is (user) domain type/attribute of "user"
explain object type is the attribute of a object or "file". do not mistake files with file objects/file types. a "file" is any object
explain that a object type can never be a scontext ( source context ) in a avc denail
explain that processes (subjects) generally operate on files (objects)
explain that processes (subjects) also operate on other processes (subjects) example: process ( sigchld ) if a user processes spawns a program process.
explain that "files" ( objects ) do not operate. they get operated on by subjects ( processes )
explain permissions that define how to operate on subjects and objects ( classes ) are defined in the kernel and are attributes of classes
explain classes and their attributes are static defined in kernel:
- example of a file object class and its attributes:
+ file read
+ dir write
+ lnk_file getattr
- example of a subject class and its attributes:
+ process sigchld
- example of a object available by other programs ACL
+ dbus send_msg
explain that although classes and their attributes are defined in the kernel, that one can assign "types" to subjects and objects, and that one can define policy for these types can interact using the object classes and their attributes supplied by the kernel.


scontext/domain type/subject | tcontext/file type/object | "object" class | "object" permissions/attributes
user_t | user_home_t | dir | getattr
httpd_t | httpd_sys_content_ra_t | file | read
user_t | mozilla_t | process | sigchld
user_t | self | process | transition
mozilla_t | httpd_port_t | tcp_socket | connect
unconfined_t | cupsd_t | dbus | send_msg

How to find out if selinux is supported /enabled:
enabled?: getenforce /selinux/config sestatus

explain selinux framework and selinux policy. explain the selinux framework is responsible for enforcing policy.
explain the access vector cache.
perruse selinux packages ( rpm -ql ) and discuss important locations : /etc/selinux , /selinux

How to disable SELinux: i refer to dwalsh blog. some highlights selinux=0 , enforcing=0, setenforce 0, system-config-selinux, semanage

system-config-selinux is a GUI for semanage. semanage is THE central managing point for SELinux administration:
label file objects ( semanage fcontect -a)
label port objects ( semanage port -a) etc
explain each optipn of semanage and system-config-selinux: label interfaces, set booleans, add , modify, delete selinux user (groups) and SELinux logins.
explain translation ( requires mcstransd )
explain what mcstransd does
explain what restorecond does
explain auditd connection to selinux ( explain ausearch /auctl )

show some pratical examples for managing users. add a unconfined user , add a confined user , add a staff users, assign mcs categories to user (ranges)
create custom selinux user groups
create custom selinux logins

explain booleans
explain customizable types
mention manual pages for targeted daemons.

explain audit2allow
explain audit2why
explain sesearch and how you can use this to make decisions
explain semodule, sestatus , restorecon , semanage, setenforce , getenforce
explain limitations of chcon
explain advantage of chcon
explain chcat

explain selinux-policy-devel ( /usr/share/selinux/devel/Makefile )
show example how to make a custom policy module
explain the limitations of a policy module package
explain the advantages of a policy module package

explain role base access control and derrived types.

explain star and selinux tar support (exmaples)

important: Possible problems caused from running in permissive mode, such as having permissions to mislabel files.
important: Copying Vs moving files.

explain avc denials field by field.
explain advantage and limitation of sealert/setroublehoot and how this relates to audit.

explain file_t, unlabeled_t
explain initrc_t
explain unconfined_t
explain sepolgen and gui

explain why /tmp will not be relabled:

read selinux by example book

explain the MLS vs TARGETED
explain mcs role in targetted versus mcs role in mls

Subjects and object.
a system is really just files. we call files objects. there are different kind of files. There are devices, ports, interfaces, regular files etc. There are also file that can be executed and that spawn a process. A process is also represented as file in the proc file system. processes are called subjects. There are different kind of processes. like there are different kind of files. for example a user process is spawned from a TTY or PTS. these are also represented on a file system as a file. a user program process is spawned from a application executable file and a init daemon is started from a init script which is also represented as a file on the files system. so you see everything just really is a file on your system and processes spawned of these files. files are objects, process are subjects.

SElinux lets us label every file and thus everything in a system. SElinux provides classes and permissions. a class for a subject is process and a permission could be signal, to signal the process. a class for a object could be file or dir or tcp_socket etc each type of file has a class and each class has a set of permission that one can use when enforcing flexible selinux.

so theres everything(files), files are either files or processes.(objects or subjects) then there are different kind of objects (file objects , port objects), and there are also different kind of subjects.(user domains, application domains etc).

these different files have classes defined in the kernel and each class has its own set of permissions. one can assign types to each file and specify how each of these types may interact with the other types.

a real user logins into the system by executing a terminal. we labeled the terminal file with a type: tty_terminal_exec_type. this is a executable file. it spawns a user process. one must specify in which domain the user process should run. lets call the user process, or as we also call process: domains, user_t.

first we should assign the type to the object. we can use chcon or semanage to assign types to file objects and interface objects and port objects. then we should define the rules so that when the user runs the tty terminal file it transition to a user domain.

this is fiction:

file_type tty_terminal_exec_type; # declare a type for your object
userdomain_transition(user_type,tty_terminal_exec_type) # this macro would suggest that when the object with type tty_terminal_type gets executed, the user process (subject) should transtion to the user_type domain.

now we assigned a user domain type (user_type) to a user process. In SElinux world all access between the types is denied by default. so unless there is policy is specified this user domain will not be able to access any other object other then the tty object it executed to get here. lets give this user domain type access to a file type. file type is a name for a set of different kind of files including a general file and a directory file.

lets give directory ~/test a file type of test_dir_type and a file ~/test/file.txt a file type of test_file_type. we use chcon or semanage to assign the file type to the dir and file. and specify how our user domain type may interact with these file types

the class for our user domain type is process. this is static in the kernel. the classes for a file type are dir for the directory file and file for the regular file. also static in the kernel. permissions specific to the file type class are for examples getattr search read write etc. lets write policy

(me executing the tty_terminal with executable file type tty_terminal_exec_t)

allow me to get attribute of class dir with file type test_dir_t
allow user_type test_dir_type:dir getattr;

allow me to manage class dir with file type test_dir_t
allow user_type test_dir_type:dir manage_dir_perms;

the manage_dir_perms is a macro that represents all permissions needed to manage a dir class.

allow me to read class file with file type test_file_type
allow user_type test_file_type:file read;

now we can manage the dir object with file type test_dir_t and class dir and we (subject with a user domain type) can read the file object with file_type test_file_t and class file

can you imagine the flexibility? can you imagine the ammount of rules? how to keep this managable? answer: macros.


explain permissive mode and how that relates to selinux=0. permissive logs would be denials, selinux=0 or the option to disable it in /etc/selinux/config will fully disable selinux.
concentrate on semanage. this is the most important tool for users. if a users knows all options in semanage and how/when to apply them then the user knows alot about selinux. maybe treat semanage as an entrypoint.
if a user knows semanage then the user will have no trouble with system_config_selinux. but focus on semanage and not system config selinux (but ofcourse do mention it)

Geen opmerkingen: