Towards Pervasive Cryptographic Access Control Models
Mikko Kiviharju
Finnish Defence Forces Technical Research Centre, Lakiala, Finland
Keywords: Cryptographic Access Control, CAC, Permissions, Access Control, CBIS.
Abstract: Access control lies at the heart of any technical information security and information assurance system.
Access control is traditionally enforced by reference monitors, which are assumed to be able to reliably
monitor and mediate all traffic from users to objects. An alternative view to enforcement is cryptography,
referred to as cryptographic access control (CAC). CAC has gained popularity with the emergence of
distributed computing, especially cloud computing and “everything as a service”. CAC is not a formal
model, but an enforcement paradigm. In this paper we propose an extension to the current CAC framework
and discuss the limits, where it is in general feasible to extend CAC as a paradigm over reference monitors.
1 INTRODUCTION
Cryptographic access control (CAC) paradigm refers
to replacing reference monitors (RM) by
cryptography in enforcing access control. A
reference monitor is a common name for an active
process checking all access requests aimed at
protected resources. First explicit examples of the
CAC-paradigm date back three decades, and some
significant advances have been made technologically
since then, notably the introduction of attribute-
based cryptography (Bethencourt, 2007); (Khader,
2007). However, the profound difference between a
reference monitor as an active component and
cryptography as a passive technique suggests rather
far-reaching conclusions.
CAC has gained popularity due to the rise of
distributed computing systems, described by the lack
of ability to place reference monitors close to the
data (or at all reachable from the data). Examples
include digital rights management (DRM), cloud
computing and high-assurance environments:
In a typical cloud computing environment, the
data owner has to place her trust concerning the
security of her information on the cloud service
provider, whose systems are usually not possible to
be audited clearly. However, distributed information
can be nearly impossible to erase, as is the case for
example with different personal information in the
internet.
Distributors of copyrighted material cannot
control the access for their content with trusted
reference monitors, once the content is purchased, as
the processing environment is owned by the user of
the material.
Military environments with classified
information have high assurance needs, yet high-
assurance reference monitors are expensive, rigid,
highly application-specific and prone to become
bottlenecks.
Current CAC-schemes, on the other hand, are
lacking mainly in two senses:
The “depth” of cryptographic enforcement – data
might be cryptographically protected, but first-layer
of key management still requires reference monitors;
Access rights formalization, resulting in human
responsibility in defining, designing and
implementing more complex access right types;
In this paper we propose an extended framework for
CAC aimed at the possibilities for extending the
paradigm and lessening the dependence on reference
monitors. The framework is based on the
decomposition of access rights into different levels
of metadata and basic types of access rights.
This paper is organized as follows: chapter two
reviews some of the related work in the area; chapter
three motivates our setting and explains the scenario;
in the fourth chapter we present our model and
finally chapter five concludes the paper.
239
Kiviharju M..
Towards Pervasive Cryptographic Access Control Models.
DOI: 10.5220/0004042502390244
In Proceedings of the International Conference on Security and Cryptography (SECRYPT-2012), pages 239-244
ISBN: 978-989-8565-24-2
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
2 RELATED WORK
The basic idea of CAC is simple – almost every
piece of information is encrypted and signed. The
simplicity of the idea is why models, schemes,
implementations and even standards abound. The
previous work on CAC can be roughly divided in
two: symmetric and asymmetric key methods.
Asymmetric key methods, such as XML encryption,
do not generally formalize their access control
models.
Symmetric-key CAC-schemes stem from the
world of DRM and broadcast encryption, distributed
file servers, outsourced data and cloud computing.
Symmetric-key CAC-schemes use formalization to
show the optimality of their metrics (Naor, 2001);
(De Capitani di Vimercati, 2007) or that the
described method actually does implement the
desired policy (De Capitani di Vimercati, 2007).
Formalizations are not on the access control concept
level, the way MAC and RBAC have been
formalized (Bell, 1973); (Ferraiolo, 1992).
A major advance within the CAC paradigm in
asymmetric-key methods was achieved with the
introduction of attribute-based cryptography.
Attribute-based cryptogrpahy can enforce read-
rights very flexibly via ABE and (in principle) write-
rights through ABGS.
ABE encodes subject properties, called attributes
(such as rank, clearance, etc.) to a set of keys, and
the object access control list is used as a logical
formula to encrypt the actual plaintext.
The CBIS-concept (Content-based Information
Security) experimented by US DoD between 2000
and 2005 as an Advanced Concept and Technology
Demonstrator (McGovern, 2001); (Savoie, 2004)
and researched later e.g. in (Kiviharju, 2010) was
based on similar threat model we use here.
In conjunction with CBIS it was presented the
idea of using metadata in helping to extend the CAC
paradigm to additional security services, but the idea
was not elaborated further, nor applied to specific
types of rights.
The recent interest in FHE - (fully)
homomorphic encryption (Gentry, 2009) for cloud
computing solves many confidentiality problems for
performing computations remotely without the data
storage being able to decipher the meaning of the
computations. FHE, however, does not cover a wide
area in the access control domain: as a program
using FHE is able to run normally without
decrypting any of its input or output, it would seem
that the execute-permission is covered by FHE as
well. However, this is not the case: even with
sufficiently efficient FHE the encryption scheme
does not stop anyone from actually running the code,
or make the execution flow itself encrypted.
3 THE CASE FOR EXTENDED
CAC MODELS
Our main setting considers a cloud-based storage
service with typical dynamic subjects and
collaboratively modifiable objects. The main focus
here is on the properties of the cloud: it is assumed
to be reliable in the sense of availability (as
documents are distributed, backed up, shared and
synchronized, it becomes less and less likely to
actually lose information accidentally or by
malevolent interaction).
Specifically, we lay out the following
assumptions:
A server (in the cloud) acts as a storage or an
execution platform only, focusing on availability and
speed. It may have the capability to remove (all its
copies of the) files on an authorized request. The
server does not have the capability to perform key-
management or cryptographic duties related to the
stored content.
A storage-server is almost always assumed to be
able to provide at least one “clean” copy of the
requested data, although it may not have the ability
to identify the correct instance.
There are no unpassable reference monitors
“close” to the data. For authorized users, some RM-
functionality is expected, but it is also possible to
read and write (including deletion) the data by
bypassing these RMs.
We propose to extend the current CAC-paradigm to
cover the full spectrum of access control rights by:
Including most conventional reference-monitor-
enforced permissions.
Extending the cryptographic enforcement further
from the content / data, in terms of metadata.
If more types of access rights than just read on data
and selected metadata are to be protected, there is
need to encode the whole access control matrix
somehow to be enforced cryptographically. This is,
however, more complex than just encrypting and
signing the whole access control matrix as metadata
to the protected data. On the other hand, metadata
encryption itself seems a viable option.
One of the main promises of some of the
cryptographically enforced access control
implementations seems to be that since data is
encrypted, there is no need for reference monitors,
SECRYPT2012-InternationalConferenceonSecurityandCryptography
240
as there is no way to read the contents without the
cryptographic keys. This is, however, an
oversimplification of the whole access control
concept:
Types of access rights form a hugely varied
category beyond data confidentiality;
Cryptographic keys need to be guarded as well as
data;
Access control entails many dynamic supporting
functions (bookkeeping, adding and deleting users,
objects and rights) that need to be accomplished
somehow in addition to the actual access control,
requiring access to different levels of metadata.
However old-fashioned or unscalable reference
monitors may seem to be, they can still emulate the
intuitive functions of access control easily, for
example enforcing access right revocation: they
simply close and further disallow the connection
between subject and object – which is possible since
by definition every access should pass through the
RM.
In the traditional view of information security,
cryptography is able to answer properly to
confidentiality and integrity only. When this is
mirrored to access control:
Disable read = encrypt (“make text
incomprehensible”, not “disable viewing”)
Disable write = sign (“make it possible to detect
modification”, not “disable bit-flips, deletions and
insertions”)
Cryptographic techniques themselves are not able to
enforce much anything, since they are passive. This
passiveness leads to a shift in responsibility from
data or service provider to its consumer. The
responsibility shift means, for example, that
outsourced data server may not be responsible for
the data integrity; instead the data consumer should
check the validity of the data if she so chooses.
4 THE EXTENDED CAC MODEL
4.1 Dimensions in Rights and Data
Our main contribution in this paper is an extended
view on cryptographic access control, based on data-
metadata hierarchy and decomposition of RM-based
types of access rights into combinations of read
and write, applied to different types of data
enforceable with cryptographic methods.
The access control matrix (ACM) uses a flat
model to lay out objects, and does not fix any types
of access rights. In ACM, subjects and object
metadata can all be equally objects along with more
traditional files and directories.
However, in real systems, types of access rights
or permissions tend to be more varied and complex.
Permissions are specified on top of each other, e.g.
who is allowed to change or audit permissions.
Permissions and object form data-metadata
hierarchies, which can be used to our advantage in
reducing permission types to those usable in CAC.
Our model performs permission type
decomposition according to three parameters, shown
as axes: r/w, metalevel and data-type axis. The r/w-
axis shows, which CAC-enforceable type
(encryption for read or signing for write) should
be used; data-type specifies on which data the
permission applies to; and metalevel-axis shows,
whether the type of the data type is data, metadata
on data or even metadata of metadata.
The data-type axis has several types, but the
CAC model only cares about whether the protected
data is content (“payload”) or access-control related.
Other types are optional, and shown only to show
the connection to table 1. The dimensions are shown
in Figure 1.
The benefits of the decomposition are
straightforward: read- and write-permissions can
readily be enforced cryptographically, provided that
there is something to encrypt or sign. Dividing the
targets into metadata-levels conceptually places
more abstract functions into the data-plane and
enables their representation with known methods,
such as structured data (e.g. XML-documents). This
in turn enables different classes of data
administrators to perform their duties independent of
their rights to the payload content.
The underlying idea in using hierarchy is to
represent most of the access rights as existing
(structured) data, with each type of metadata placed
parallel to the actual content node. Access control
metadata would typically include encrypted
symmetric keys (along with their metadata) and
signatures.
Each conventional access right is assumed to be
able to be represented by a “small” number of points
in the decomposition space. Enumerating and
canonizing the rights this way avoids the translation
issues between permission types expressed in natural
language between different systems, and clearly
states, what is expected of the cryptographic scheme
proposed to protect that particular permission type.
As an example, consider the own(Take
Ownership)-permission from Bell-LaPadula
(Windows): it can be seen as both read- and
TowardsPervasiveCryptographicAccessControlModels
241
write-
control,
lies on
p
ermissi
defined
b
Figure 1
:
It s
h
merely
s
into a s
e
read/wri
t
cry
p
tog
r
server
p
p
ointles
s
We
p
ermissi
and
w
r
i
general
machine
comple
m
informa
t
We
efficient
monitor
s
RMs in
s
4.2
The mo
d
access r
i
goal is t
w
1) Sho
w
reduced
rights in
2) Sho
w
access r
i
sufficie
n
with a r
e
N
ot all
their r/
w
we dee
m
permission
p
and as it is a
d
the metadat
a
on would be
b
y access co
n
:
Decompositio
n
h
ould be not
e
s
how that a
p
e
t of reads a
n
t
e-operation
r
aphy can hel
p
p
rocess int
e
s
).
do not e
x
on can be re
d
ite-rights.
models for c
, use only
m
ented by
t
ion (i.e. meta
d
also concur
and / or pra
c
s
, even consi
d
s
uitable locat
i
M
odelin
g
A
del needs to
i
ghts can be e
n
w
ofol
d
:
w
ing that a
to a combi
n
general;
w
ing that en
fo
i
ghts type at
t
n
t to realize t
h
e
ference moni
t
permission t
y
w
-decomposit
i
m
ed someho
w
p
ertaining to
a
d
ditional info
r
a
level. Th
u
set of two
p
n
trol and meta
n
model for ty
p
e
d that it is
p
ermission ty
p
n
d writes:
i
is not
p
only in pri
n
e
rnal structu
r
x
pect that
a
d
uced to a s
e
However,
w
omputing, s
u
r
ead- and
w
decisional
l
d
ata).
that some
c
tical to enfo
r
d
ering that e
s
i
ons may be e
x
A
ccess Ri
gh
show that
d
n
forced crypt
o
type of acc
e
n
ation of re
a
fo
rcing each
d
t
he data cons
u
h
e same kin
d
t
or.
y
pes are pra
c
i
on. We selec
t
w
general an
d
a
datatype a
c
r
mation on da
t
u
s that parti
c
p
oints on the
data.
p
es of access ri
g
not sufficie
n
p
e can be br
o
i
f the target o
f
persistent
d
n
ciple (to en
c
r
es is prob
a
ny concie
v
e
t of mere
re
w
e note that
u
ch as the T
u
w
rite operat
i
l
ogic and
s
rights are
m
r
ce with refer
e
s
tablishing tr
u
x
pensive.
ts
d
ifferent type
o
graphically.
T
e
ss right ca
n
ad
- and
w
r
i
d
ecompositio
n
u
mer’s doma
i
d
of protectio
n
c
tical to turn
t
ed a set of r
i
d
relevant to
c
cess
t
a, it
c
ular
line
g
hts.
n
t to
o
ken
f
the
d
ata,
c
rypt
b
ably
v
able
e
a
d
-
the
u
ring
i
ons,
s
tate
m
ore
e
nce
u
sted
e
s of
This
n
be
i
te-
n
of
i
n is
n as
into
i
ghts
o
the
co
n
the
ge
n
fro
m
Wi
n
p
er
m
mo
d
res
p
Ta
b
cry
p
T
y
lo
au
d
T
A
t
/w-
t
sui
t
gro
La
P
Se
r
the
ho
s
dat
a
ess
e
an
o
n
sumer’s co
mp
model clos
e
n
eral types,
w
m
MS SQL
S
n
dows XP
/
m
issions (M
i
d
el (Bell, 19
7
p
ectively and
b
le 1: Types of
p
tographically
e
y
peofaccess
right
create
delete
o
g actions
u
dit logs
d
elegate
execute
search
revoke,
grant,
control
own
T
raverse
Folder
Read
t
tributes
The permissi
o
t
ypes for sele
t
able target ar
e
In table 1,
d
u
ped accord
i
P
adula, Win
d
v
er 2000.
We excluded
table due to
t
s
t domain (
s
a
base statem
e
e
ntially dupl
i
o
ther type or
p
It should be
p
uting envir
o
e
r to real s
y
w
e included
t
S
erver 2000
d
/
Windows
crosoft, 200
7
7
3). This tot
a
including the
a
ccess rights a
n
e
nforcable type
Source r/w
gen. w
gen. w
gen. r
w
gen. r
gen. w
gen. r
w
gen. r
r
BLP w
BLP r
w
Win7 r
Win7 r
o
ns and their
c
ted permissi
o
e
shown in ta
b
d
ifferent type
s
ng to their
s
d
ows XP /
7
most of the i
n
t
heir trivialit
y
s
trongly at
t
e
nt permissi
o
i
cate the dec
o
p
ermission.
noted that ea
o
nment. In or
y
stems, in a
d
t
ypes of acc
e
d
atabase ma
n
7 - called
7
) - and Bel
l
a
ls 11+11+10
general type
s
n
d their decom
p
e
s and targets.
thetarget
o
datastorage
datastorage
systema
meta
d
systeml
o
systeml
o
accesscon
t
metadata(wi
basedrest
r
executable
c
AN
D
programexe
c
&metada
meaningf
u
indexingme
t
data(inam
wa
y
accesscon
t
meta
d
dataandal
metadat
a
dataandal
meta
d
datastorage
filemet
a
decompositi
o
on types join
b
le 1.
s of access
r
source: gene
r
7
and Micro
s
n
vestigated ri
g
y
(e.g. read,
t
he storage,
o
ns), or bec
a
omposition t
y
a
ch permissio
n
d
er to tie
d
dition to
e
ss rights
n
agement,
“special”
-laPadula
+
4 types,
s
.
p
osition to
fther/w
metadata
metadata
ctions
d
ata
o
gdata
o
gdata
t
roltable
t
hsubject
r
ictions)
c
odedata
D
c
utiondata
ta(ina
u
lway)
t
adataOR
e
aningful
y
)
t
roltable
d
ata
lrelated
a
AND
lrelated
d
ata
metadata
a
data
o
n into r-
ed with a
r
ights are
r
al, Bell-
s
oft SQL
g
hts from
write),
such as
a
use they
y
pe from
n
is itself
SECRYPT2012-InternationalConferenceonSecurityandCryptography
242
metadata of the target it addresses. Thus
cryptographic protection of a permission is always
one ladder higher on the metadata axis than its
target.
Delegate (grant in BLP) is a permission on
a permission, or a meta-permission. The
delegate-right here means that a subject has a
permission to grant selected rights further, subject to
a set of additional restrictions.
The enforcement of the search-right depends
on the actual implementation of the search: if an
indexing structure is built during the search, that
structure can be encrypted; if there is no indexing
structure, one may use searchable encrypted
databases, using searchable encryption techniques or
homomorphic encryption.
The BLP-model addresses permissions that
affect the access control matrix itself. We call these
rights access control metadata, also referred to as
security-related metadata (Kiviharju, 2010).
Permissions such as grant and revoke address
the questions of who is allowed to change the access
control enforcement function, i.e. the security
administrator role. This implies that access control
metadata needs to be addressed differently than
other metadata. For metadata other than access
control, there is no need to elevate the metadata-
level for more than one, and CAC adds just another
layer on top of traditional structured documents’
protection.
4.3 Responsibility Shift
The shifting of responsibility becomes essential in
those types of access rights, for which the misuse
can directly result in breach of a security attribute
outside the data consumer’s domain (such as service
disruptions).
We assumed that the data storage facility is able
to provide at least one instance of data “clean” that
has otherwise been somehow deleted or modified via
unauthorized channels. This implies that some rights
can be left for the responsibility of the data storage
in the cloud:
delete and update (in such a way that it
erases original data altogether)
extensive append and create (to create a
DoS attack)
control and revoke
By our assumptions, consumer processes data in a
platform distinct from the storage server. However,
some rights affect metadata, which is needed by the
server in order to perform the storage function.
Rights that may affect operation inside the
storage domain include the permissions listed above
and most notably execute. Execution refers to
data that is interpreted as program code, and run on
some platform. In order to be able to execute the
code, the subject has to have the general right to
access the code altogether (read rights) and then
have an access to the execution system on a specific
processor (write rights). Read can be
accomplished by encrypting the executable file, but
write rights to the processor are more complex to
accomplish cryptographically. It is not meaningful to
sign the code, since the platform owner is not
necessarily responsible for verifying signatures.
Alternative approach could be e.g. instruction set
encryption per platform, in which case the
executable would be twice encrypted: first the
instruction set permutation and then block
encryption. The instruction set encryption’s key
management should be the responsibility of the
execution platform owner
4.4 Restrictions on Permissions
Reference monitor-based access control systems are
able to restrict access based on different criteria,
such as time, history (based on subject’s previous
actions), location and role. The basis of the
restriction is often trusted, as the reference monitor
(assumed to be non-corruptible) can also be assumed
to be able to synchronize to a reliable time source
reasonably accurately and possess a complete set of
information concerning the exercise of the rights (as
all permission usage should travel via the RM) for
auditing purposes.
In CAC, as the encryptor/verifier generally has
little or no control over the decryptor’s/signer’s
environment, the existence and use of permission
restrictions based on environmental factors cannot
be trusted.
In CAC-schemes the restrictions are almost
always based on key-management: the subject keys
are assumed to be available only if the intended
restriction does not apply. However, the consumer is
only bound by the possession of these attributes
(keys), not by the particular circumstances related to
the use of the attributes.
In a more controlled environment it could be
possible to enforce some portions of the computing
platform to have higher-than-usual assurance in their
integrity, by using TPM or other tamper-proof
hardware. In these cases ABE could be used to place
environment-based restrictions by dividing the
attribute-keys between the user and the platform:
TowardsPervasiveCryptographicAccessControlModels
243
Attribute-keys pertaining to the user are given to
the user (as files or tokens);
Attribute-keys pertaining to the environment are
embedded into the tamperproof hardware modules;
The encryption / signing is performed with a
combination of these two types of keys’ public parts.
Some restrictions are needed to express certain
rights at all, for example Traverse Folder
requires that read rights are checked also for other
objects in the folder hierarchy above the target.
5 CONCLUSIONS AND FUTURE
WORK
In this paper we have discussed the current status of
cryptographic access control paradigm from the
point of view of access control models, and
suggested extensions to the current CAC-paradigm.
The proposal concerned extending the
cryptographic enforcement to multiple permission
types other than just just read- and write. We
accomplished this to a practical set of rights by
decomposing conventional permissions to a set of
read- and write-operations on certain data, and
showing that for most of them it is feasible and
practical to enforce them cryptographically as well.
There still remain many open questions and
problems, such as efficient revocation. However,
less elusive problems, such as cryptographical role
binding are natural cryptographic efforts to tackle;
from the modeling point of view it would be
instructive to create actual XML-schemas or formal
cryptographic RBAC-models.
REFERENCES
Bell, D., LaPadula, L., 1973. Secure Computer Systems:
Mathematical Foundations. MITRE Technical Report
2547, vol. 1. MITRE.
Bethencourt, J., Sahai, A., Waters, B., 2007. Ciphertext-
Policy Attribute-Based Encryption. In IEEE
Symposium on Security and Privacy 2007, IEEE
Computer Society. pp. 321 – 334.
De Capitani di Vimercati, S., Foresti, S., Jajodia, S.,
Paraboschi, S., Samarati, P., 2007. Over-encryption:
Management of Access Control Evolution on
Outsourced Data. In Very Large Databases (VLDB)
2007 – Conference Proceedings, ACM. pp. 123-134.
Ferraiolo, D., Kuhn, D., 1992. Role Based Access Control.
In 15
th
National Computer Security Conference –
Conference Proceedings. pp. 554-563.
Gentry, C., 2009. A Fully Homomorphic Encryption
Scheme. PhD Dissertation in Stanford University.
[Online], Available http://crypto.stanford.edu/craig/
[11.3.2012].
Khader, D., 2007. Attribute Based Group Signature
Scheme. [Online], Available: http://eprint.iacr.
org/2007/159 [8.3.2012].
Kiviharju, M., 2010. Content-Based Information Security
(CBIS): Definitions, Requirements and Cryptographic
Architecture, FDF Technical Research Centre.
McGovern, S., 2001. Information Security Requirements
for a Coalition Wide Area Network Thesis in Naval
Postgraduate School, Monterey, California,
NPS/CISR.
Microsoft, 2007. How to set, view, change, or remove
special permissions for files and folders in Windows
XP. [Online], Available: http://support.microsoft.com/
kb/308419 [11.3.2012].
Naor, D., Naor, M., Lotspiech, J., 2001. Revocation and
Tracing Schemes for Stateless Receivers. In CRYPTO
2001 – Conference Proceedings, Springer-Verlag. pp.
41-62.
Savoie, J., 2004. A Strong three-factor authentication
device: TrustedDAVE and the new Generic Content-
Based Information Security (CBIS) architecture,
Technical Memorandum TM 2004-198, DRDC
Ottawa.
SECRYPT2012-InternationalConferenceonSecurityandCryptography
244