Method and a system for secure execution of workflow tasks in a distributed workflow management system within a decentralized network system

Abstract

There are provided a method, a system and an initiator server for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern in a distributed workflow management system within a decentralized network system with a plurality of servers (b0, b1, . . . , bn) including at least an initiator server and at least a number of groups of servers of the plurality of servers. Each group satisfies a policy (poli) of a vertex, and thus, knows a corresponding policy key pair including a policy private key (SKpoli) and a policy public key (PKpoli), respectively. Each vertex denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key and a vertex public key. The vertex private keys and the policy public keys are jointly encrypted within a first onion structure, the first onion structure being built up of a number of onion layers representing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex thus revealing the corresponding vertex private key.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application claims the benefit of European Patent Application Serial No. 07 290 413.9, filed on Apr. 4, 2007, which is incorporated by reference in its entirety.

TECHNICAL FIELD

This description refers to a distributed workflow management system within a decentralized network system, particularly to a secure execution of workflow tasks of a workflow in a distributed workflow management system within a decentralized network system.

DESCRIPTION OF THE RELATED ART

In computer science workflow management systems are used to distribute the execution of workflow tasks to a number of servers within a network system. The infrastructure either can be a centralized workflow management system comprising a trusted centralized point of coordination or a more flexible decentralized workflow management system. Existing distributed workflow management systems eliminate on the one hand the need for a centralized coordinator that can be a performance bottleneck in some business scenarios. This flexibility introduced by decentralized workflow management systems on the other hand raises new security requirements like integrity of workflow execution in order to assure the compliance of the overall sequence of operations with a pre-defined workflow execution plan. Moreover, as opposed to usual centralized workflow management systems, the distributed execution of workflows can not rely on a trusted centralized coordination mechanism to manage the most basic execution primitives such as message routing between servers. Yet, existing decentralized workflow management systems appear to be limited when it comes to integrating security mechanisms that meet these specific requirements in addition to the ones identified in the centralized setting. Even though some recent research efforts in the field of distributed workflow security have indeed been focusing on issues related to the management of rights in server assignment or detecting conflicts of interests, basic security issues related to the security of the overall workflow execution such as integrity and evidence of execution have not yet been addressed.

Security of cross-organizational workflows in both centralized and decentralized settings has been an active research field over the passed years mainly focusing on access control, and separation of duty in conflict of interests issues. However, in the decentralized setting issues related to the integrity of workflow execution and workflow instance forging have been left aside.

SUMMARY

Therefore, it is an object to provide new security mechanisms supporting a secure execution of workflows in a decentralized setting. The proposed mechanisms assure the integrity of the distributed execution of workflows and prevent servers from being involved in a workflow instance forged by a malicious peer. Therefore, it is capitalized in the following on onion encryption techniques and security policy models in the design of these mechanisms.

In one aspect, there is provided a method for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern in a distributed workflow management system with a plurality of servers, each server satisfying a policy of a vertex of the workflow, and thus, knowing a corresponding policy private key and a corresponding policy public key, respectively, the method comprising, starting at an i'th server which is determined at runtime in accord with the execution pattern to perform the i'th vertex of the workflow, the policy of which the i'th server satisfies, at least the following operations:

• • receiving a first onion structure being built up of a number of onion layers representing the execution pattern with an outermost layer including an i'th vertex private key assigned to the i'th vertex and encrypted with the i'th policy public key,
  • receiving the workflow originating from an initiator server and including workflow data encoded by a corresponding i'th vertex public key,
  • decoding the i'th vertex private key by using the i'th policy private key, and
  • decoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,
  • executing the i'th vertex and processing the decoded workflow data accordingly, and
  • updating the workflow according to the executed i'th vertex and the processed workflow data.

In another aspect, a method for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern in a distributed workflow management system within a decentralized network system is provided. The network system comprises a plurality of servers including at least an initiator server and at least a number of groups of servers of the plurality of servers, wherein each group satisfies a policy of a vertex, and thus, knows a corresponding policy key pair including a policy private key and a policy public key, respectively. Further, each vertex denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key and a vertex public key. The vertex private keys and the policy public keys are jointly encrypted within a first onion structure, the first onion structure being built up of a number of onion layers reproducing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex thus revealing the corresponding vertex private key. The proposed method comprises, starting at an i'th server as the one server of one of the groups of servers which is determined at runtime in accord with the execution pattern to perform one of the vertices of the workflow, the policy of which the i'th server satisfies, called herein the i'th vertex, at least the following operations:

• • receiving the first onion structure with an outermost layer including the i'th vertex private key and encrypted with the i'th policy public key,
  • receiving the workflow originating from the initiator server and including workflow data encoded by the i'th vertex public key,
  • decoding the i'th vertex private key by using the known i'th policy private key, and
  • decoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,
  • executing the i'th vertex and processing the decoded workflow data accordingly, and
  • updating the workflow according to the executed i'th vertex and the processed workflow data.

It is possible that the operation of updating the workflow comprises encoding at least the processed workflow data with the (i+1)'th vertex public key corresponding to the succession of vertices and adding those workflow data to the workflow, and that the method further comprises the following operation:

• • sending the updated workflow and the partly decoded first onion structure further to at least one (i+1)'th server as one server of the group of servers satisfying the policy of an (i+1)'th vertex as the next vertex in the succession of vertices.

Thereby, the i'th server can choose the (i+1)'th server at runtime to execute the (i+1)'th vertex as the vertex following the i'th vertex in the succession of vertices.

It is supposable that the processed workflow data are included in a data block consisting of two fields, the processed workflow data that have last been modified during execution of the i'th vertex and a signature, and wherein the data block is associated with a set of signatures that is computed by the i'th server assigned to the i'th vertex at runtime.

The data block and the set of signatures can be determined according to the following formulas:

B _k ^a=(d_k,sign_a(d_k))

sign_a(d_k)={h₁(d_k)}_SKa

wherein d_k describes the actual data, h₁ a first hash function, and SK_a the private vertex key of vertex a.

Generally, it is possible that the first onion structure is determined by the initiator server.

According to a further aspect, along with the execution of the workflow, a second onion structure is built at each execution step with vertex private keys based on the workflow execution pattern, the second onion structure being initialized by the initiator server, so that the i'th server receives the second onion structure with (i−1) layers and encrypts its upper layer with the i'th vertex private key to extend the second onion structure up to i layers which is sent to the (i+1)'th server upon completion of the i'th vertex.

Afterwards, the i'th server sends a workflow message to the (i+1)'th server upon execution of the i'th vertex and updating of the workflow, the workflow message including the updated workflow, the first onion structure and the extended second onion structure so that the (i+1)'th server can first retrieve the (i+1)'th vertex private key from the first onion structure, can verify compliance of the workflow execution with the workflow pattern using the extended second onion structure and finally can execute the (i+1)'th vertex and process the workflow data accordingly.

It is possible that the second onion structure is initiated by the initiator server who computes the first layer as the most inner layer by determining a hash value of a workflow policy using the first hash function and encoding the hash value by the policy private key of the first vertex according to the succession of vertices.

The workflow policy can be defined by the initiator server.

It can be provided that each vertex key pair assigned to exactly one vertex is defined for a single instance of the workflow so that it cannot be reused during any other workflow instance.

The first onion structure can adopt and represent a wide variety of different execution patterns, also called workflow patterns.

In one possible aspect, the first onion structure O_d represents a sequence workflow pattern which can be described by the following formula:

$O_d:\{\begin{array}{c}{l}_1=\left\{{\mathrm{SK}}_n\right\}\\ l_i=\left\{{\left\{l_{i-1}\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_{n-i+2}}},{\mathrm{SK}}_{n-i+1}\right\}\mathrm{for}i\in \left[2,n\right]\\ l_{n+1}=\left\{{\left\{l_n\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_1}}\right\}\end{array}$

wherein l_j describes the j'th layer of the onion structure O_d, SK_j describes the j'th vertex private key, and PK_polj describes the j'th policy public key, wherein jε[1,n].

Alternatively, it is possible that the first onion structure O_d represents a AND-SPLIT workflow pattern which can be described by the following formula:

$O_{d1}=\left\{{\mathrm{SK}}_1,O_{d2},O_{d3}\dots ,O_{\mathrm{dn}}\right\}$ $O_{\mathrm{di}}=\left\{{\left\{{\mathrm{SK}}_i\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_i}}\right\}\mathrm{for}i\in \left[2,n-1\right]$

wherein O_dj describes an onion as part of the onion structure O_d, SK_j describes the j'th vertex private key, and PK_polj describes the j'th policy public key, wherein jε[1,n].

Furthermore, it is possible that the first onion structure O_d represents a AND-JOIN workflow pattern which can be described by the following formula:

$O_{d_1}=\left\{{\left\{\lambda ,{\mathrm{SK}}_{n_1}\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_n}}\right\}$ $O_{d_i}=\left\{{\left\{{\mathrm{SK}}_{n_i}\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_n}}\right\}\mathrm{for}i\in \left[2n-1\right]$

wherein O_dj describes an onion as part of the onion structure O_d, PK_polj describes the j'th policy public key, the vertex private key SK_n is divided into n−1 parts and defined by SK_n=SK_n1⊕SK_n2⊕ . . . ⊕SK_nn−1, describes the onion structure, wherein jε[1,n−1].

The first onion structure O_d can also represent a OR-SPLIT workflow pattern which can be described by the following formula;

$O_{d_1}=\left\{{\mathrm{SK}}_1,O_{d_2},O_{d_3},\dots ,O_{d_n}\right\}$ $O_{d_i}=\left\{{\left\{{\mathrm{SK}}_i\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_i}}\right\}\mathrm{for}i\in \left[2,n-1\right]$

wherein O_dj describes an onion as part of the onion structure O_d, PK_polj describes the j'th policy public key, and SK_j describes the j'th vertex private key, wherein jε[1,n−1].

It is also possible that the first onion structure O_d represents a OR-JOIN workflow pattern which can be described by the following formula:

$O_d:\{\begin{array}{c}{l}_1=\left\{\lambda ,{\mathrm{SK}}_n\right\}\\ l_2=\left\{{\left\{l_1\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_n}}\right\}\end{array}$

wherein PK_poln describes the n'th policy public key, and SK_n describes the n'th vertex private key, λ describes the onion structure, and l₁ and l₂ layers of the onion structure O_d.

The first onion structure can also represent a combination of any of the above mentioned workflow patterns. It is possible that the first onion structure enabling the vertex private keys distribution during the execution of the workflow can represent a workflow pattern consisting of a plurality of different branches. Starting from the above mentioned basic workflow patterns the procedure towards building a first onion structure corresponding to a more complicated workflow pattern is rather straightforward and will be explained in more detail below.

According to a further aspect, the vertex key pairs are defined as follows:

iε[1,n]

$\{\begin{array}{c}{\mathrm{PK}}_i=h_1\left(W_{\mathrm{iid}}\oplus S_W\oplus v_i\right)\\ {\mathrm{SK}}_i=s\times h_2\left({\mathrm{PK}}_i\right)\end{array}$

wherein sε for a prime q, s is called a master key held by the initiator server, h₂ is a second hash function, W_iid is a string, called workflow instance identifier, S_W denotes a workflow specification, v_i is the i'th vertex, and the vertex key pairs (PK_i,SK_i) are calculated upon creation of the workflow pattern.In another aspect, there is provided a system configured to be used for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern within a decentralized network system, the system including at least an initiator server and a plurality of servers, each server satisfying a policy of a vertex of the workflow, and thus, knowing a corresponding policy key pair including a policy private key and a policy public key, respectively, wherein each vertex is assigned a vertex key pair including a vertex private key and a vertex public key, wherein an i'th server as at least one server which is to be determined at runtime of the workflow in accord with the execution pattern to perform one of the vertices of the workflow, the policy of which the i'th server satisfies, called herein the i'th vertex, is configured to perform at least the following operations:

• • receiving the first onion structure being built up of a number of onion layers representing the execution pattern with a most upper layer including the i'th vertex private key and being encrypted with the i'th policy public key,
  • receiving the workflow originating from the initiator server and including workflow data encoded by the i'th vertex public key,
  • decoding the i'th vertex private key by using the known i'th policy private key, and
  • decoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,
  • executing the i'th vertex and processing the decoded workflow data accordingly, and
  • updating the workflow according to the executed i'th vertex and the processed workflow data.

In still another aspect, a system is provided which is usable for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern within a decentralized network system with a plurality of servers. The system includes at least an initiator server and at least a number of groups of servers of the plurality of servers, each group satisfying a policy of a vertex, and thus, knowing a corresponding policy key pair including a policy private key and a policy public key, respectively. Each vertex denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key and a vertex public key. The vertex private keys and the policy public keys are jointly encrypted within a first onion structure, the first onion structure being built up of a number of onion layers representing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex thus revealing the corresponding vertex private key. Thereby, an i'th server as at least one server of one of the groups of servers which is to be determined at runtime of the workflow in accord with the execution pattern to perform one of the vertices of the workflow, the policy of which the i'th server satisfies, called herein the i'th vertex, is configured to perform at least the following operations:

• • receiving the first onion structure with an outermost upper layer including the i'th vertex private key and being encrypted with the i'th policy public key,
  • receiving the workflow originating from the initiator server and including workflow data encoded by the i'th vertex public key,
  • decoding the i'th vertex private key by using the known i'th policy private key, and
  • decoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,
  • executing the i'th vertex and processing the decoded workflow data accordingly, and
  • updating the workflow according to the executed i'th vertex and the processed workflow data.

It is possible that the operation of updating the workflow comprises encoding at least the processed workflow data with the (i+1)'th vertex public key corresponding to the succession of vertices and adding those workflow data to the workflow, and that the i'th server is configured to further execute at least the following operation:

• • sending the updated workflow and the partly decoded first onion structure further to at least one (i+1)'th server as one server of the group of servers satisfying the policy of an (i+1)'th vertex as the next vertex in the succession of vertices.

It can be arranged that the i'th server is configured to choose the (i+1)'th server at runtime to execute the (i+1)'th vertex as the vertex following the i'th vertex in the succession of vertices.

As already mentioned with respect to the proposed method, it is also possible with respect to the system that the processed workflow data are included in a data block consisting of two fields, the processed workflow data that have last been modified during execution of the i'th vertex and a signature, and wherein the data block is associated with a set of signatures that is to be computed by the i'th server assigned to the i'th vertex at runtime.

The data block and the set of signatures can be determined according to the following formulas:

B _k ^a=(d_k,sign_a(d_k))

sign_a(d_k)={h₁(d_k)}_SKa

wherein d_k describes the actual data, h₁ a first hash function, and SK_a the vertex private key of vertex a.

In one possible aspect of the system, the first onion structure is determined by the initiator server.

Furthermore, it is possible that, along with the execution of the workflow, a second onion structure is to be built at each execution step with vertex private keys based on the workflow execution pattern, the second onion structure being initialized by the initiator server, so that the i'th server can receive the second onion structure with (i−1) layers and can encrypt its upper layer with the i'th vertex private key to extend the second onion structure up to i layers which is to be sent to the (i+1)'th server upon completion of the i'th vertex.

It can be provided that the i'th server can send a workflow message to the (i+1)'th server upon execution of the i'th vertex and updating of the workflow, the workflow message including the updated workflow, the first onion structure and the extended second onion structure so that the (i+1)'th server can first retrieve the (i+1)'th vertex private key from the first onion structure, can verify compliance of the workflow execution with the workflow pattern using the extended second onion structure and finally can execute the (i+1)'th vertex and process the workflow data accordingly.

Further it can be established that the second onion structure is to be initiated by the initiator server who can compute the first layer as the most inner layer by determining a hash value of a workflow policy using the first hash function and encoding the hash value by the policy private key of the first vertex according to the succession of vertices.

Thereby, it is possible that the workflow policy is to be defined by the initiator server.

It can also be determined that each vertex key pair assigned to exactly one vertex is to be defined for a single instance of the workflow so that it cannot be reused during any other workflow instance.

As already mentioned above, the first onion structure can represent a wide variety of different workflow patterns. There are some basic workflow patterns, such as a sequence workflow pattern, an AND-SPLIT workflow pattern, an AND-JOIN workflow pattern, an OR-SPLIT workflow pattern and an OR-JOIN workflow pattern, which can also be combined to form much more complex workflow patterns.

According to another aspect, an initiator server for initiating a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern within a decentralized network system with a plurality of servers is provided. The system includes at least a number of groups of servers of the plurality of servers, each group satisfying a policy of a vertex, and thus, knowing a corresponding policy key pair including a policy private key and a policy public key, respectively, wherein each vertex denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key and a vertex public key. The proposed initiator server is configured to encrypt the vertex private keys and the policy public keys jointly within a first onion structure, wherein the first onion structure is built up of a number of onion layers representing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex thus revealing the corresponding vertex private key. The initiator server is further configured to determine at least one server of one of the groups of servers at runtime of the workflow in accord with the execution pattern to perform one of the vertices of the workflow.

The initiator server can be one server of at least one of the groups of servers, each group satisfying a policy of a vertex.

The initiator server can further be configured to initialize, along with the execution of the workflow, a second onion structure which is to be extend at each execution step with vertex private keys based on the workflow execution pattern, so that the i'th server can receive the second onion structure with (i−1) layers and can encrypt its upper layer with the i'th vertex private key to extend the second onion structure up to i layers which is to be sent to the (i+1)'th server upon completion of the i'th vertex.

It is also possible that the initiator server can initialize the second onion structure by computing the first layer as the most inner layer by determining a hash value of a workflow policy using the first hash function and encoding the hash value by the policy private key of the first vertex according to the succession of vertices.

Furthermore, it is possible that the initiator server is configured to define the workflow policy.

In another aspect, a computer program product is provided with a computer-readable medium and a computer program stored on the computer-readable medium with a program code which is suitable for carrying out a method for securely executing workflow tasks in a distributed workflow management system within a decentralized network system as described before when the computer program is run on a computer.

Another implementation provides a computer program with a program code which is suitable for carrying out a method for securely executing workflow tasks in a distributed workflow management system within a decentralized network system as described before when the computer program is run on a computer.

A computer-readable medium with a computer program stored thereon is also provided, the computer program comprising a program code which is suitable for carrying out a method for securely executing workflow tasks in a distributed workflow management system within a decentralized network system as described before when the computer program is run on a computer.

Further features and embodiments will become apparent from the description and the accompanying drawings.

For the purpose of clarity, the present discussion refers to an abstract example of a method, a system and an initiator server, respectively. However, implementations of the method, the system and the initiator server may operate with a wide variety of types of network systems including networks and communication systems dramatically different from the specific example as illustrated in the following drawings.

It should be understood that while details of one or more implementations are described in terms of a specific system, further implementations may have applications in a variety of communication systems, such as advanced cable-television systems, advanced telephone-networks or any other communication systems that would benefit from the system or the method. It is intended that the system as used in this specification and claims is realizable within any communication system unless the context requires otherwise.

Implementations are schematically illustrated in the drawings by way of an example embodiment and explained in detail with reference to the drawings. It is understood that the description is in no way limiting and is merely an illustration of various implementations.

It will be understood that the features mentioned above and those described hereinafter can be used not only in the combination specified but also in other combinations or on their own, without departing from the scope of the present disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

In the drawings,

FIG. 1 is a block diagram showing an example for a workflow, comprising a sequence-, an AND-Split- and an AND-JOIN-workflow pattern,

FIG. 2 is showing the principle of a first onion structure, namely a key distribution scheme O_d and a second onion structure, namely an execution proof mechanism O_p,

FIG. 3 is a block diagram of the accessibility to workflow data, stored within a specific data block,

FIG. 4 is a block diagram of a sequence workflow pattern,

FIG. 5 is a block diagram of an AND-SPLIT workflow pattern,

FIG. 6 is a block diagram of an AND-JOIN workflow pattern,

FIG. 7 is a block diagram of a workflow massage structure,

FIG. 8 shows a sequence of operations for a workflow initiation, and

FIG. 9 shows a sequence of operations for a workflow message processing.

DETAILED DESCRIPTION

A workflow management system was designed in F. Montagut and R. Molva. Enabling pervasive execution of workflows. In Proceedings of the i^st IEEE International Conference on Collaborative Computing: Networking, Applications and Worksharing, CollaborateCom, 2005. and is used to support the following description. This model introduces a workflow management system supporting the execution of business processes in environments without infrastructure. This workflow management system features a distributed architecture characterized by two objectives:

1. Fully decentralized architecture: the workflow management task is carried out by a set of devices in order to cope with the lack of dedicated infrastructure.2. Dynamic assignment of servers to workflow tasks: the actors can be discovered at runtime.

Having designed an abstract representation of the workflow in form of a workflow pattern whereby servers, also designated as business partners, are not yet assigned to workflow tasks, an initiator server launches the execution and executes a first set of workflow tasks. Then the initiator server searches for a server able to perform a next set of workflow tasks. Once the discovery phase is complete, a workflow message including all workflow data is sent by the workflow initiator server, also called initiator server, to the newly discovered server and the workflow execution further proceeds with the execution of the next set of workflow tasks and a new discovery procedure. The execution pattern is stateless so that upon completion of required workflow tasks each server sends all workflow data to the next server involved and thus does not have to remain online till the end of the workflow execution. In this decentralized setting, the data transmitted amongst servers include all workflow data. W denotes in the following the abstract representation of a distributed workflow defined by W={(v_i)_iε[1,n],δ} where v_i denotes a vertex which is a set of workflow tasks that are performed by a server b_i assigned to v_i and δ is the set of execution dependencies between those vertices.

${\left(M_{i->j_p}\right)}_{p\in \left[1,z_j\right]}$

is noted as a set of workflow messages issued by b_i to the z_i servers assigned to the vertices

${\left(v_{j_p}\right)}_{p\in \left[1,z_i\right]}$

executed right after the completion of v_i. An instance of W wherein servers have been assigned to vertices is denoted W_b={W_iid,(b_i)_iε[1,n]}, where W_iid is a string called workflow instance identifier.

This model is depicted in FIG. 1. The present description only mentions a subset of execution dependencies or workflow patterns, namely sequence workflow patterns, AND-SPLIT workflow patterns, AND-JOIN workflow patterns, OR-SPLIT workflow patterns and OR-JOIN workflow patterns. A variety of other workflow patterns or combinations thereof are also possible.

FIG. 1 shows a possible workflow execution pattern. There are seven vertices v₁ to v₇ to be executed according to the given execution pattern. Each vertex v_i is assigned to a server b_i which has to execute the vertex v_i when appealed for in accord with the execution pattern. Upon execution of vertex v₁, servers b₂ and b₄ assigned to vertices v₂ and v₄, respectively, are contacted concurrently by b₁ assigned to v₁ as indicated by the bifurcation “AND-Split”. Server b₂ contacts server b₃ assigned to vertex v₃ upon execution of vertex v₂. Server b₄ executes vertex v₄ and contacts then server b₅ assigned to vertex v₅. Vertex v₆ is executed by server b₆ if server b₆ receives a message form server b₃ upon execution of v₃ and a further message from server b₅ upon execution of v₅ as indicated by block “AND-Join”. The workflow is terminated by server b₇ assigned to vertex v₇ which is contacted by server b₆ upon execution of vertex v₆.

Such a distributed execution of workflows raises security constraints due to the lack of dedicated infrastructure assuring the management and control of the workflow execution. As a result, basic security features such as compliance of the workflow execution with the predefined plan are no longer assured so far.

Three main categories of security requirements for a distributed workflow system are: authorization, proof of execution and workflow data protection.

Concerning authorization, the main security requirement for a workflow management system is to ensure that only authorized servers are assigned to workflow tasks throughout a workflow instance. In the decentralized setting, the assignment of workflow tasks is managed by servers themselves relying on a service discovery mechanism in the case of runtime assignment. In this case, the server assignment procedure enforces a matchmaking procedure whereby the security credentials for the servers are matched against security requirements for workflow tasks.

Concerning execution proofs, a decentralized workflow management system does not offer any guarantee regarding the compliance of actual execution of workflow tasks with the predefined execution plan. Without any trusted coordinator to refer to, a server b_i assigned to a vertex v_i needs to be able to verify that the vertices scheduled to be executed beforehand were actually executed according to the workflow pattern. This is a crucial requirement to prevent any malicious peer from forging a workflow instance.

A need for workflow data protection occurs particularly in the case of decentralized workflow execution, wherein the set of workflow data denoted D=(d_k)_kε[1,j] is transferred from one server to another. This raises major requirements for workflow data security in terms of integrity, confidentiality and access control as follows:

1. Data confidentiality: for each vertex v_i, the server b_i assigned to v_i should only be authorized to read a subset D_i^r of D. In the example of FIG. 1 server b₂ assigned to v₂ and server b₅ assigned to v₅ are only authorized to read d₁, respectively, as indicated by the fact that d₁εD₂^r; and d₁εD₅^r, respectively.2. Data integrity: for each vertex v_i, the server b_i assigned to v_i should only be authorized to modify a subset D_i^w of D. In the example of FIG. 1 server b₁ assigned to v₁ is authorized to modify d₁ since d₁εD₁^w. The same applies for server b₃ assigned to v₃ and server b₆ assigned to v₆.3. Access control: the subsets D_i^r and D_i^w associated with each vertex v_i should be determined based on the security policy of the workflow.

Therefore, according to one proposed approach two types of key pairs are introduced. Each vertex v_i is first associated with a policy pol_i defining a set of credentials a candidate server needs to satisfy in order to be assigned to v_i. The policy pol_i is mapped to a key pair (PK_poli,SK_poli), where SK_poli is the policy private key and PK_poli the policy public key. Thus satisfying the policy pol_i is equivalent to knowing the policy private key SK_poli. The policy private key SK_poli can be distributed by a single key distribution server based on the compliance of servers with policy pol_i or by means of a more sophisticated cryptographic scheme such as group key distribution. Second, vertex key pairs (PK_i,SK_i)_iε[1,n] are introduced to protect the access to workflow data. A key distribution scheme is suggested wherein a server b_i whose identity is a priori unknown retrieves the vertex private key SK_i upon his assignment to the vertex v_i. Onion encryption techniques with policy public keys PK_poli are used to distribute vertex private keys SK_i.

Furthermore, execution proofs have to be issued along with the workflow execution in order to ensure the compliance of the execution with the pre-defined plan. To that effect, onion encryption techniques are introduced in order to built an onion structure with vertex private keys to assure the integrity of the workflow execution. The suggested key distribution scheme (O_d) and the execution proof mechanism (O_p) are depicted in FIG. 2 and specified in more detail later on. In the following denotes a message space, C a ciphertext and K a key space. An encryption of a message mε with a key KεK is noted {m}_i and h₁, h₂ denote a first and a second one-way hash function.

FIG. 2 clearly shows the principle according to which a first onion structure O_d and a second onion structure O_p are transported, updated and processed between servers b_i, each server b_i being assigned to a respective vertex v_i which he has to execute according to a given execution pattern. The two onion structures are generally included in the messages which are transported between the servers. In FIG. 2 the two onion structures are illustrated due to clarity separately from one another. In the case shown in FIG. 2 the messages are transported sequentially between servers b_i−1, b_i and b_i+1. Server b_i−1 which is assigned to vertex v_i−1 receives the message including the first onion structure O_d. The server b_i−1 peels off the first onion structure O_d by using the policy public key PK_poli−1. By decrypting the outermost layer of the first onion structure O_d server b_i−1 can retrieve the vertex private key SK_i−1. By means of the vertex private key SK_i−1 server b_i−1 can get access to workflow data which he received from server b_i−2 upon execution of vertex v_i−2 right before vertex v_i−1. Along with the workflow execution the second onion structure O_p is built at each execution step with the appropriate vertex private key in order to allow servers to verify the integrity of the workflow execution. The second onion structure O_p is initialized by the initiator server who computes the most inner layer. Thereby, a public parameter is computed by the workflow initiator server and that is available to the servers being involved in the execution of the workflow. The second onion structure O_p is initialized this way so that it cannot be replayed as it is defined for a specific instance of a workflow specification. At step i−1 of the workflow execution server b_i−1 receives O_pi−2 and encrypts its upper layer with the vertex private key SK_i−1 to build an extended second onion structure O_pi−1 which he sends further to server b_i upon completion of vertex v_i−1 as indicated in the upper part of FIG. 2. The next server b_i assigned to vertex v_i receives now the first onion structure O_d partly peeled off up to the onion layer which is encoded by the policy public key PK_poli. Server b_i assigned to v_i knows the policy private key SK_poli since server b_i satisfies the policy pol_i of the vertex v_i. Therefore, server b_i can retrieve from the first onion structure O_d the vertex private key SK_i by using the policy private key SK_poli. As already described with respect to the previous server b_i−1 server b_i can use the vertex private key SK_i to decode the workflow data he has to process during execution of vertex v_i and also to encrypt the upper layer of the second onion structure O_p with SK_i to build an updated second onion structure O_p, namely by building a next layer O_pi so that he can send the extended second onion structure O_p to the next server b_i+1 together with the updated workflow data and the first onion structure O_d partly peeled off up to the next inner layer which is encoded by the policy public key PK_poli+1. Server b_i+1 can now retrieve from the first onion structure O_d the vertex private key SK_i+1 by using the policy private key SK_poli+1. The vertex private key SK_i+1 can again be used to decode the encrypted workflow data and also to update the second onion structure O_p by building the next onion layer O_pi+1 as indicated in the upper part of FIG. 2.

Concerning the data protection, the role of a server b_i assigned to a vertex v_i consists in processing the workflow data that are granted read-only and read-write access during the execution of vertex v_i. In FIG. 3 a specific structure is depicted called data block to protect workflow data accordingly. Each data block consists of two fields: the actual data d_k and a signature sign_a(d_k)={h₁(d_k)}_SKa. The data block is denoted as B_k^a=(d_k,sign_a(d_k)) including a data segment d_k that has last been modified during the execution of vertex v_a. The data block B_k^a is also associated with a set of signatures denoted H_k^a that is computed by b_a assigned to v_a.

$H_k^a=\left\{{\left\{h_1\left({\left\{B_k^a\right\}}_{{\mathrm{PK}}_i}\right)\right\}}_{{\mathrm{SK}}_a}|l\in R_k^a\right\}$

where R_k^a denotes the set defined by R_k^a={lε[1,n]|(d_kεD_l^r−D_l^w) and (v_l is executed after v_a) and (v_l is executed before the first vertex v_pa following v_a such that d_kεD_pa^w and that is located on the same branch of the workflow as v_a and v_l)}. For instance, consider the example of FIG. 1 whereby d₁ belongs to D₁^w, D₂^r, D₃^w, D₅^r and D₆^w, R₁¹={2, 5}.

When a server b_i receives the data block B_k^a encrypted with PK_i (i.e. he is granted read access on d_k), he decrypts the structure using SK_i in order to get access to d_k and sign_a(d_k). b_i is then able to verify the integrity of d_k using PK_a i.e. that d_k was last modified after the execution of vertex v_a. Further, if b_i is granted write access on d_k, he can update the value of d_k and compute sign_i(d_k) yielding a new data block B_kⁱ and a new set H_kⁱ. If on the contrary server b_i receives data block B_k^a encrypted with PK_m, whereby in this case vertex v_m is executed after v_i, b_i can verify the integrity of {B_k^a}_PKm by matching h_i({B_k^a}_PKm) against the value contained in H_k^a. The integrity and confidentiality of data access thus relies on the fact that the vertex private key SK_i is made available to server b_i only prior to the execution of vertex v_i.

The objective of the vertex private key distribution mechanism is to ensure that only a server b_i assigned to v_i at runtime and whose identity is a priory unknown can access the vertex private key SK_i. The basic idea behind this mechanism is to map the workflow structure in terms of execution patterns with an onion structure O_d so that at each step of the workflow execution a layer of O_d is peeled off using SK_poli and SK_i is revealed.

When X is a set, an onion O is a multi-layered structure composed of a set of n subsets of X (l_k)_kε[1,n] such that ∀kε[1,n]l_k+1⊂l_k. The elements of (l_k)_kε[1,n] are called layers of O, in particular, l₁ and l_n are the lowest and upper layers of O, respectively. l_p(O) is the layer p of an onion O.

When A=(α_k)_kε[1,j] and B=(b_k)_kε[1,l] are two onion structures, A is said to be wrapped by B, denoted A⊂B when ∃kε[1,l] such that a_j⊂b_k. The vertex private keys are distributed to servers with respect to various workflow patterns including SEQUENCE, AND-SPLIT, AND-JOIN, OR-SPLIT and OR-JOIN being defined as follows.

Using the sequence workflow pattern the vertex private keys are sequentially distributed to servers. In this case, an onion structure assuring the distribution of vertex private keys is sequentially peeled off by servers. Considering a sequence of n vertices (v_i)_iε[1,n], b₁ assigned to v_l initiates the workflow execution with the onion structure O_d defined as follows.

$O_d\text{:}\{\begin{array}{c}{l}_1=\left\{{\mathrm{SK}}_n\right\}\\ l_i=\left\{{\left\{l_{i-1}\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_{n-i+2}}},{\mathrm{SK}}_{n-i+1}\right\}\mathrm{for}i\in \left[2,n\right]\\ l_{n+1}=\left\{{\left\{l_n\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_1}}\right\}\end{array}$

The workflow execution further proceeds as depicted in FIG. 4. For iε[2,n−1] the server b_i assigned to the vertex v_i receives

${\left\{l_{n-i+1}\left(O_d\right)\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_l}},$

peels one layer off by decrypting it using SK_poli, reads l_n−i+1(O_d) to retrieve SK_i and sends {l_n−i(O_d)}PK_poli+1 to b_i+1.

In the case of an AND-SPLIT workflow pattern, the servers (b_i)_iε[2,n] assigned to the vertices (v_i)_iε[2,n] are contacted concurrently by b₁ assigned to the vertex v₁. In this case, n−1 vertex private keys should be delivered to (b_i)_iε[2,n] and the upper layer of the onion O_d1 available to b₁ therefore wraps SK₁ and n−1 onions (O_di)_iε[2,n] to be sent to (b_i)_iε[2,n] as depicted in FIG. 5.

$O_{d}=\left\{{\mathrm{SK}}_1,O_{d2},O_{d3}\dots ,O_{\mathrm{dn}}\right\}$ $O_{\mathrm{di}}=\left\{{\left\{{\mathrm{SK}}_i\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_i}}\right\}\mathrm{for}i\in \left[2,n-1\right]$

In the case of a single workflow initiator, an AND-JOIN workflow pattern is preceded in the workflow by an AND-SPLIT workflow pattern. In this case, the vertex v_n is executed by the server b_n if and only if the latter receives n−1 messages as depicted in FIG. 6. The vertex private key SK_n is thus divided into n−1 parts and defined by SK_n=SK_n1⊕SK_n2⊕ . . . ⊕SK_nn−1. The onion O_di sent by b_i thus includes SK_ni. Besides, in order to avoid redundancy, the onion structure δ associated with the sequel of the workflow execution right after v_n is only included in one of the onions received by b_n. Each (b_i)_iε[1,n−1] therefore sends O_di to b_n where

$O_{d1}=\left\{{\left\{\lambda ,{\mathrm{SK}}_{n_1}\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_n}}\right\}$ $O_{\mathrm{di}}=\left\{{\left\{{\mathrm{SK}}_{n_j}\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_n}}\right\}\mathrm{for}i\in \left[2,n-1\right]$

The OR-SPLIT workflow pattern is an exclusive choice, v₁ sends one message to an appropriate participant.

$O_{d1}=\left\{{\mathrm{SK}}_1,O_{d2},O_{d3},\dots ,O_{\mathrm{dn}}\right\}$ $O_{\mathrm{di}}=\left\{{\left\{{\mathrm{SK}}_i\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_i}}\right\}\mathrm{for}i\in \left[2,n-1\right]$

O_d1 is available to the server assigned to v₁. This is the same structure as the AND-SPLIT workflow pattern, yet the latter only sends the appropriate O_di to v_i depending on the result of the OR-SPLIT condition.

In an OR-JOIN workflow pattern, there is a single workflow initiator server. Therefore, the OR-JOIN is preceded in the workflow by an OR-SPLIT workflow pattern. The server assigned to v_n receives in any cases a single message thus a single vertex private key is required that is sent by one of the (b_i)_[1,n−1] depending on the choice made of the previous OR-SPLIT in the workflow. b_n thus receives in any cases:

$O_d\text{:}\{\begin{array}{c}{l}_1=\left\{\lambda ,{\mathrm{SK}}_n\right\}\\ l_2=\left\{{\left\{l_1\right\}}_{{\mathrm{PK}}_{{\mathrm{pol}}_n}}\right\}\end{array}$

where δ is an onion structure associated with the sequel of the workflow execution right after v_n.

The key distribution is processed by building an onion structure corresponding to the workflow execution pattern. This is rather straightforward and better sketched throughout an example according to the workflow depicted in FIG. 1. The onion O_d enabling the vertex private keys distribution during the execution of the workflow is defined as follows:

The onions associated with the two branches forming the AND-SPLIT workflow pattern are wrapped by the layer corresponding to v₁. Only the first AND-SPLIT branch includes the sequel of the workflow after v₆.

For creation of an execution proof along the workflow execution, an onion structure O_pi is built at each execution step i with vertex private keys in order to allow servers to verify the integrity of the workflow execution. The onion structure O_p is initialized by the server b₁ assigned to v₁ who computes

$O_{p1}={\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{Sk}}_{{\mathrm{pol}}_1}}$

where P_W is called workflow policy and is defined as follows. A workflow specification S_W denotes the set S_W={W,(J_i^r,J_i^w,pol_i)_iε[1,n],h₁} where J_i^r={kε[1,j]|d_kεD_i^r} and J_i^w={kε[1,j]|d_kεD_i^w} (J_i^r and J_i^w basically specify for each vertex the set of data that are granted read-only and read-write access, respectively). S_W is defined at workflow design phase.

The workflow policy P_W denotes the set P_W=S_W∪{W_iid,h₂}∪{PK_i|iε[1,n]}, wherein P_W is a public parameter computed by the workflow initiator server b₁ and that is available to the servers being involved in the execution of the workflow W.

The onion structure O_p is initialized this way so that it cannot be replayed as it is defined for a specific instance of a workflow specification. At the step i of the workflow execution, b_i receives O_pi−1 and encrypts its upper layer with SK_i to build an onion O_pi which he sends to b_i+1 upon completion of v_i. Considering a set (v_i)_[1,n] of vertices executed in sequence the following definitions are resulting:

$O_{p_1}:l_1=\left\{{\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}\right\}$ $O_{p_2}:l_1=\left\{{\left\{l_1\left(O_{p_1}\right)\right\}}_{{\mathrm{SK}}_2}\right\}$ $O_{p_i}:l_1=\left\{{\left\{l_1\left(O_{p_{i-1}}\right)\right\}}_{{\mathrm{SK}}_i}\right\}\mathrm{for}i\in \left[3,n\right]$

The building process of O_pi is based on workflow execution patterns yet since it is built at runtime contrary to the onion structure O_d, this is straightforward. First, there is no specific rule for OR-SPLIT and OR-JOIN workflow patterns. Second, when encountering an AND-SPLIT workflow pattern, the same structure O_pi is concurrently sent while in case of an AND-JOIN workflow pattern, the n−1 onions received by a server b_n are wrapped by a single structure:

O_pn:l₁={{O_p1, O_p2, . . . , O_pn−1}_SKn}.

In order to verify that the workflow execution is compliant with the pre-defined plan when he starts the execution of the vertex v_i the server b_i assigned to v_i just peels off the layers of O_pi−1 using the vertex public keys of the vertices previously executed based on S_W. Doing so he retrieves the value

${\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}$

that should be equal to the one he can compute given P_W, if the workflow execution has been so far executed according to the plan.

Considering the example depicted in FIG. 1, at the end of the workflow execution the onion O_p is defined as follows.

$O_p=\hspace{1em}\left\{{\left\{\underset{\underset{\mathrm{First}\mathrm{AND}-\mathrm{SPLIT}\mathrm{branch}}{}}{\{{\left\{{\left\{{\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}\right\}}_{{\mathrm{SK}}_2}\right\}}_{{\mathrm{SK}}_3}},\underset{\underset{\mathrm{Second}\mathrm{AND}-\mathrm{SPLIT}\mathrm{branch}}{}}{{\hspace{1em}{\left\{{\left\{{\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}\right\}}_{{\mathrm{SK}}_4}\right\}}_{{\mathrm{SK}}_5}\}}_{{\mathrm{SK}}_6}}\right\}}_{{\mathrm{SK}}_7}\right\}{\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}\hspace{1em}$

is sent by b₁ assigned to v₁ to both b₂ and b₄ assigned to v₂ and v₄ respectively. The onion structure associated with the two branches forming the AND-SPLIT workflow pattern thus includes

${\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}$

twice.

The vertex key pairs have to be generated and defined for a single instance of a workflow specification in order to avoid replay attacks. To that effect, it might be proposed to capitalize on ID-based encryption techniques, in the specification of the set (PK_i,SK_i)_iε[1,n]. For all iε[1,n] (PK_i,SK_i) is defined by:

$\{\begin{array}{c}{\mathrm{PK}}_i=h_1\left(W_{\mathrm{iid}}\oplus S_W{\oplus }_{v_i}\right)\\ {\mathrm{SK}}_i=s\times h_2\left({\mathrm{PK}}_i\right)\end{array}\hspace{1em}$

where sε for a prime q. s is called master key and is held by the private vertex keys generator which is in this case the workflow initiator server. This vertex key pair specification has a double advantage. First vertex key pairs cannot be reused during any other workflow instance and second vertex public keys can be directly retrieved from W and W_iid when verifying the integrity of workflow data or peeling off the second onion structure O_p.

In order to support a coherent execution of the mechanisms presented so far, workflow messages exchanged between servers consist of the set of information that is depicted in FIG. 7. Workflow data (d_k)_kε[1,j] are all transported between servers and satisfy the data block specification. A single message may include several copies of the same data block structure that are encrypted with different vertex public keys based on the execution plan. This can be the case with AND-SPLIT workflow patterns. Besides, workflow data can be stored in two different ways depending on the requirements for the execution of the workflow. Either the iterations of data resulting from each modification in workflow messages is kept till the end of the execution or the data content is simply replaced upon completion of a vertex. The bandwidth requirements are higher in the first case since the size of messages increases as the workflow execution proceeds further.

P_W is required to retrieve vertex and policy public keys and specifies the workflow execution plan. The two onion structures, namely the first onion structure O_d and the second onion structure O_p, are also included in the message.

Upon receipt of the message depicted in FIG. 7 a server b₁ assigned to v₁ retrieves first the vertex private key from the first onion structure O_d. He then checks that P_W is genuine i.e. that it was initialized by the initiator server of the workflow assigned to v₁. He is later on able to verify the compliance of the workflow execution with the plan using the second onion structure O_p and finally he can process workflow data.

The mechanisms presented so far can be combined to support the secure execution of the workflow in the decentralized setting.

Integrating security mechanisms to enforce the security requirements of the decentralized workflow execution requires a process strongly coupled with both workflow design and runtime specifications. At the workflow design phase, the workflow specification S_W is defined in order to specify for each vertex the sets of data that are accessible in read and write access and the credentials required by potential servers to be assigned to workflow vertices. At workflow initiation phase, the workflow policy P_W is specified and the first onion structure O_d is built. The workflow initiator server builds then the first set of workflow messages to be sent to the next servers involved. This message generation process consists of the initialization of the data blocks and that of the second onion structure O_p.

At runtime, a server b_i chosen to execute a vertex v_i receives a set of workflow messages. Those messages are proceeded to retrieve SK_i from the first onion structure O_d and to access workflow data. Once the vertex execution is complete b_i builds a set of workflow messages to be dispatched to the next servers involved in the execution. In this message building process, the data and the second onion structure O_p are updated.

The set of functional operations composing the workflow initiation and runtime specifications is precisely specified later. In the following N_kⁱ denotes a set defined by N_kⁱ={lε[1,n]|d_kεD_l^r and v_l is executed right after v_i}. considering the example of FIG. 1: d₁ is accessed during the execution of the vertices v₁, v₂ and v₅ thus N_l¹={2, 5}.

FIG. 8 shows a sequence of operations for a workflow initiation. The workflow is initiated by initiator server b₁ assigned to vertex v₁ who issues a first set of workflow messages

${\left(M_{1->j_p}\right)}_{p\in \left[1,z_1\right]}.$

The workflow initiation consists of the following operations:1. Workflow policy specification: generate (PK_i,SK_i)_iε[1,n]2. Initialization of the first onion structure O_d 3. Data block initialization: compute ∀kε[1,j]sign₁(d_k)4. Data block encryption: ∀kε[1,j] determine N_k¹ and compute ∀kε[1,j], ∀l εN_k¹{B_k¹}PK₁ 5. Data block hash sets: ∀kε[1,j] determine R_k¹ and compute ∀kε[1,j], ∀lεR_k¹h₁({B_k¹}_PK1)}_SK1 6. Initialization of the second onion structure O_p: compute O_p1 7. Message generation based on W and (N_k¹)_kε[1,j].

The workflow messages are generated with respect to the specification defined in FIG. 7 and sent to the next servers involved. This includes the initialization of the second onion structure O_p and that of data blocks which are encrypted with appropriate vertex public keys.

FIG. 9 shows a sequence of operations for a workflow message processing. A server b_i being assigned to a vertex v_i proceeds as follows upon receipt of the set of workflow messages (M_jp→i)_pε[1,ki] sent by k_i servers assigned to the vertices (v_jp)_pε[1,ki] executed right before v_i.

1. Retrieve SK_i from the first onion structure O_d 2. Data block decryption with SK_i based on J_i^r 3. Execution proofs verification: peel off the second onion structure O_p 4. Data integrity check based on W and P_W 5. Vertex execution6. Data block update: compute ∀kεJ_i^wsign_i(d_k) and update d_k content7. Data block encryption: ∀kεJ_i^r determine N_kⁱ and compute ∀kεJ_i^r, ∀lεN_kⁱ{B_kⁱ}PK₁ 8. Data block hash sets: ∀kεJ_i^w determine R_kⁱ and compute ∀kεJ_i^w, ∀lεR_kⁱh₁({B_kⁱ}PK₁)}_SKi 9. Second onion structure O_p update: compute O_pi 10. Message generation based on W and (N_kⁱ)_kε[1,j]

After having retrieved SK_i from the first onion structure O_d, server b_i verifies the integrity of workflow data and that the execution of the workflow up to his workflow is consistent with the second onion structure O_p. Workflow data are then processed during the execution of v_i and data blocks are updated and encrypted upon completion. Finally, server b_i computes O_pi and issues the set of workflow messages (M_i→j)_jε[1,zi] to the intended servers in accord with the execution pattern of the workflow.

For the security of the workflow execution, there are several alternatives with respect to the management of the policy key pair (PK_poli,SK_poli), including single key distribution based on the policy compliance, group key management or policy-based cryptography. Amongst those alternatives, only the policy based cryptography scenario as part of the security evaluation of the object is discussed herein. In the following two assumptions are made:

1. IND-PB-CCA: the policy-based encryption scheme used in the specification of (PK_poli,SK_poli)_[1,n] is semantically secure against a chosen ciphertext attack for a policy-based encryption and the associated policy-based signature scheme achieves signature unforgeability.2. IND-CCA: the public key encryption scheme used in the specification of (PK_i,SK_i)_[1,n] is semantically secure against a chosen ciphertext attack the associated signature scheme achieves signature unforgeability.

The integrity of the distributed workflow execution is ensured. This basically means that workflow data are accessed and modified by authorized servers based on the pre-defined plan specified by means of the sets J_i^r and J_i^w.

This property is ensured by the first onion structure O_d which assures the vertex key distribution used in the access to workflow data based on the workflow execution plan. Assuming that a workflow initiator server builds O_d based on the methodology specified before and under IND-PB-CCA, thus it is not feasible for an adversary A to extract the vertex private key SK_i from O_d if A does not satisfy the set of policies (pol_ik)_kε[1,l] associated with the set of vertices (v_ik)_kε[1,l] executed prior to v_i in workflow W. This is true as the structure of O_d is mapped to workflow W.

Upon receipt of a workflow message, a server is sure that the workflow has been properly executed so far provided that he trusts the servers satisfying the policy pol_i.

This means that an adversary that does not verify a policy that is trusted by some servers cannot forge a workflow instance, i.e. that he cannot produce a workflow message faking a valid workflow instance. This property is enforced by the second onion structure O_p. Assuming that a workflow initiator server builds the second onion structure O_p based on the methodology specified above and under IND-PB-CCA, the second onion structure O_p unforgeable. To assure the unforgeability property, it has to be verified that:

1. A genuine second onion structure O_p built during a previous instance of a workflow cannot be replayed.2. A second onion structure O_p cannot he built by an adversary that is not trusted by servers.

The first property is enforced by the fact that a second onion structure O_p properly built by trustworthy peers is bound to a specific workflow policy P_W and thus cannot be reused during an attempt to execute a malicious workflow instance. The second property is straightforward under IND-PB-CCA as the policy-based signature scheme achieves signature unforgeability. Thus, an adversary cannot produce a valid onion

$O_{p_1}:l_1=\left\{{\left\{h_1\left(P_W\right)\right\}}_{{\mathrm{SK}}_{{\mathrm{pol}}_1}}\right\}.$

Assuming servers involved in a workflow instance do not share vertex private keys they retrieve from the first onion structure O_d, the approach achieves the following data integrity properties:

1. Data truncation and insertion resilience: any server can detect the deletion or the insertion of a piece of data in a workflow message.2. Data content integrity: any server can detect the integrity violation of a data block content in a workflow message.

The first property is ensured as the set of workflow data blocks that should be present in a workflow message is specified in P_W, the workflow message formatting has thus to be compliant with the workflow specification. The second property is assured by the fact that an adversary cannot modify a given data block without providing a valid signature on this data block. This property relies on the unforgeability of the signature scheme used in the data block and hash set specifications.

These three security properties enable a coherent and secure execution of distributed workflows. Yet the approach can still be optimized to avoid the replication workflow messages. A server may indeed send the same workflow message several times to different servers satisfying the same security policy resulting in concurrent executions of the given workflow instance. An approach based on a stateful service discovery mechanism can be envisioned to cope with this problem.

In prior technical work that describes preventing information leakage within workflows that execute among competing organizations, and/or that describe a chinese wall security model for decentralized workflow systems, mechanisms are proposed for the management of conflicts of interest during the distributed execution of workflows. These pieces of work specify solutions in the design of access control policies to prevent servers from accessing data that are not part of their classes of interest. These approaches do not address the issue policy enforcement with respect to integrity of execution in fully decentralized workflow management systems. Nonetheless, the access control policy models described in this work can be used to augment the above approach especially in the specification of the sets J_i^r and J_i^w at workflow design time.

Onion encryption techniques have been introduced in a variety of technical publications. In contrast to this the above described approach maps onion structures with workflow execution patterns in order to built proofs of execution and enforce access control on workflow data. As a result, more complex business scenarios are supported by the present approach than usual onion routing solutions. Furthermore, combined with policy encryption techniques, the present approach provides a secure runtime environment for the execution of fully decentralized workflows supporting runtime assignment of servers, a feature which had not been tackled so far.

Finally the present approach is suitable for any business scenarios in which business roles can be mapped to security policies that can be associated with key pairs. It can thus be easily integrated into existing security policy models such as chinese wall security model.

In the above, mechanisms were presented towards meeting the security requirements raised by the execution of workflows in a decentralized setting. The present approach, capitalizing on onion encryption techniques and security policy models, protects the access to workflow data with respect to a pre-defined workflow execution pattern/plan and provides proofs of execution to servers. Those mechanisms can easily be integrated into the runtime specification of decentralized workflow management systems and are further suitable for fully decentralized workflow supporting the runtime assignment of servers to workflow tasks. These mechanisms will foster the development of dynamic business applications whereby workflow actors do not need to rely on a dedicated infrastructure to provide their resources as one of the major flaws slowing down this trend was the lack of security.

Claims

1. A method for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern in a distributed workflow management system within a decentralized network system with a plurality of servers (b0, b1, . . . , bn) including at least an initiator server and at least a number of groups of servers of the plurality of servers, each group satisfying a policy (poli) of a vertex (vi), and thus, knowing a corresponding policy key pair including a policy private key (SKpoli) and a policy public key (PKpoli), respectively, wherein each vertex (vi) denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key (SKi) and a vertex public key (PKi), and wherein the vertex private keys and the policy public keys are jointly encrypted within a first onion structure (Od), the first onion structure (Od) being built up of a number of onion layers representing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex (vi) thus revealing the corresponding vertex private key (SKi), the method comprising, starting at an i'th server as the one server of one of the groups of servers which is determined at runtime in accord with the execution pattern to perform one of the vertices of the workflow, the policy of which the i'th server satisfies, called herein the i'th vertex, at least the following operations: receiving the first onion structure (Od) with an outermost layer including the i'th vertex private key (SKi) and encrypted with the i'th policy public key (PKpoli),receiving the workflow originating from the initiator server and including workflow data encoded by the i'th vertex public key (PKi),decoding the i'th vertex private key (SKi) by using the known i'th policy private key (PKpoli), anddecoding the workflow data encoded by the i'th vertex public key (PKi) from the received workflow by using the i'th vertex private key (SKi),executing the i'th vertex (vi) and processing the decoded workflow data accordingly, andupdating the workflow according to the executed i'th vertex (vi) and the processed workflow data.

2. The method according to claim 1, wherein the operation of updating the workflow comprises encoding at least the processed workflow data with the (i+1)'th vertex public key (PKi+1) corresponding to the succession of vertices and adding those workflow data to the workflow, and the method further comprising the following operation: sending the updated workflow and the partly decoded first onion structure (Od) further to at least one (i+1)'th server as one server of the group of servers satisfying the policy (poli+1) of an (i+1)'th vertex (vi+1) as the next vertex in the succession of vertices.

3. The method according to claim 2, wherein the i'th server (bi) chooses the (i+1)'th server (bi+1) at runtime to execute the (i+1)'th vertex (vi+1) as the vertex following the i'th vertex (vi) in the succession of vertices.

4. The method according to claim 2, wherein the processed workflow data are included in a data block consisting of two fields, the processed workflow data that have last been modified during execution of the i'th vertex (vi) and a signature, and wherein the data block is associated with a set of signatures that is computed by the i'th server (bi) assigned to the i'th vertex (vi) at runtime.

5. The method according to claim 4, wherein the data block and the set of signatures can be determined according to the following formulas: Bka=(dk,signa(dk))signa(dk)={h1(dk)}SKa

6. The method according to claim 1, wherein the first onion structure (Od) is determined by the initiator server.

7. The method according to claim 1, wherein along with the execution of the workflow a second onion structure (Op) is built at each execution step with vertex private keys based on the workflow execution pattern, the second onion structure (Op) being initialized by the initiator server, so that the i'th server (bi) receives the second onion structure (Op) with (i−1) layers, and encrypts its upper layer with the i'th vertex private key (SKi) to extend the second onion structure (Op) up to i layers which is sent to at least one (i+1)'th server as one server of the group of servers satisfying the policy (poli+1) of an (i+1)'th vertex (vi+1) as the next vertex in the succession of vertices upon completion of the i'th vertex.

8. The method according to claim 7, wherein the i'th server (bi) sends a workflow message to the (i+1)'th server (bi+1) upon execution of the i'th vertex (vi) and updating of the workflow, the workflow message including the updated workflow, the partly decoded first onion structure (Od) and the extended second onion structure (Op) so that the (i+1)'th server can first retrieve the (i+1)'th vertex private key (SKi+1) from the first onion structure (Od), can verify compliance of the workflow execution with the workflow pattern using the extended second onion structure (Op) and finally can execute the (i+1)'th vertex (vi+1) and process the workflow data accordingly.

9. The method according to claim 7, wherein the second onion structure (Op) is initiated by the initiator server which computes the first layer as the most inner layer by determining a hash value of a workflow policy using the first hash function and encoding the hash value by the policy private key of the first vertex according to the succession of vertices.

10. The method according to claim 9, wherein the workflow policy is defined by the initiator server.

11. The method according to claim 1, wherein each vertex key pair (SKi, PKi) assigned to exactly one vertex (vi) is defined for a single instance of the workflow so that it cannot be reused during any other workflow instance.

12. The method according to claim 1, wherein the first onion structure (Od) represents a sequence workflow pattern which can be described by the following formula:

13. The method according to claim 1, wherein the first onion structure (Od) represents a AND-SPLIT workflow pattern which can be described by the following formula:

14. The method according to claim 1, wherein the first onion structure (Od) represents a AND-JOIN workflow pattern which can be described by the following formula:

15. The method according to claim 1, wherein the first onion structure (Od) represents a OR-SPLIT workflow pattern which can be described by the following formula:

16. The method according to claim 1, wherein the first onion structure (Od) represents a OR-JOIN workflow pattern which can be described by the following formula:

17. The method according to claim 1, wherein the vertex key pairs are defined as follows: iε[1,n], n being an integer value

18. A system configured to be used for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern within a decentralized network system with a plurality of servers (b0, b1, . . . , bn), the system including at least an initiator server and at least a number of groups of servers of the plurality of servers, each group satisfying a policy (poli) of a vertex, and thus, knowing a corresponding policy key pair including a policy private key (SKpoli) and a policy public key (PKpoli), respectively, wherein each vertex (vi) denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key (SKi) and a vertex public key (PKi), and wherein the vertex private keys and the policy public keys are jointly encrypted within a first onion structure (Od), the first onion structure (Od) being built up of a number of onion layers representing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex (vi) thus revealing the corresponding vertex private key (SKi), wherein an i'th server as at least one server of one of the groups of servers which is to be determined at runtime of the workflow in accord with the execution pattern to perform one of the vertices of the workflow, the policy of which the i'th server satisfies, called herein the i'th vertex, is configured to perform at least the following operations: receiving the first onion structure (Od) with an most upper layer including the i'th vertex private key (SKi) and being encrypted with the i'th policy public key (PKpoli),receiving the workflow originating from the initiator server and including workflow data encoded by the i'th vertex public key (PKi),decoding the i'th vertex private key (SKi) by using the known i'th policy private key, anddecoding the workflow data encoded by the i'th vertex public key (PKi) from the received workflow by using the i'th vertex private key (SKi),executing the i'th vertex (vi) and processing the decoded workflow data accordingly, andupdating the workflow according to the executed i'th vertex (vi) and the processed workflow data.

19. An initiator server for initiating a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern within a decentralized network system with a plurality of servers (b0, b1, . . . , bn) including at least a number of groups of servers of the plurality of servers, each group satisfying a policy (poli) of a vertex, and thus, knowing a corresponding policy key pair including a policy private key (SKpoli) and a policy public key (PKpoli), respectively, wherein each vertex (vi) denotes a set of workflow tasks to be executed in accord with the execution pattern and is assigned a vertex key pair including a vertex private key (SKi) and a vertex public key (PKi), and wherein the initiator server is configured to encrypt the vertex private keys and the policy public keys jointly within a first onion structure (Od), the first onion structure (Od) being built up of a number of onion layers representing the execution pattern which defines a succession of vertices such that each layer is decryptable by using the policy private key of exactly one vertex (vi) thus revealing the corresponding vertex private key (SKi), wherein the initiator server is further configured to determine at least one server of one of the groups of servers at runtime of the workflow in accord with the execution pattern to perform one of the vertices of the workflow.

20. The initiator server according to claim 19, which is one server of at least one of the groups of servers, each group satisfying a policy (poli) of a vertex.

21. The initiator server according to claim 19, which is configured to initialize, along with the execution of the workflow, a second onion structure (Op) which is to be extend at each execution step with vertex private keys based on the workflow execution pattern, so that an i'th server determined at runtime to perform the i'th vertex can receive the second onion structure (Op) with (i−1) layers and can encrypt its upper layer with the i'th vertex private key (SKi) to extend the second onion structure (Op) up to i layers which is to be sent to at least one (i+1)'th server as one server of the group of servers satisfying the policy (poli+1) of an (i+1)'th vertex (vi+1) as the next vertex in the succession of vertices upon completion of the i'th vertex.

22. The initiator server according to claim 21, wherein the initiator server can initialize the second onion structure (Op) by computing the first layer as the most inner layer by determining a hash value of a workflow policy using the first hash function and encoding the hash value by the policy private key of the first vertex according to the succession of vertices.

23. The initiator server according to claim 22, wherein the initiator server is configured to define the workflow policy.

24. A method for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern in a distributed workflow management system with a plurality of servers, each server satisfying a policy of a vertex of the workflow, and thus, knowing a corresponding policy private key and a corresponding policy public key, respectively, the method comprising, starting at an i'th server which is determined at runtime in accord with the execution pattern to perform the i'th vertex of the workflow, the policy of which the i'th server satisfies, at least the following operations: receiving a first onion structure being built up of a number of onion layers representing the execution pattern with an outermost layer including an i'th vertex private key assigned to the i'th vertex and encrypted with the i'th policy public key,receiving the workflow originating from an initiator server and including workflow data encoded by a corresponding i'th vertex public key,decoding the i'th vertex private key by using the i'th policy private key, anddecoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,executing the i'th vertex and processing the decoded workflow data accordingly, andupdating the workflow according to the executed i'th vertex and the processed workflow data.

25. A computer program product with a computer-readable medium and a computer program stored on the computer-readable medium with a program code which is suitable for carrying out a method for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern in a distributed workflow management system with a plurality of servers, each server satisfying a policy of a vertex of the workflow, and thus, knowing a corresponding policy private key and a corresponding policy public key, respectively, the method when the computer program is run on a computer comprising, starting at an i'th server which is determined at runtime in accord with the execution pattern to perform the i'th vertex of the workflow, the policy of which the i'th server satisfies, at least the following operations: receiving a first onion structure being built up of a number of onion layers representing the execution pattern with an outermost layer including an i'th vertex private key assigned to the i'th vertex and encrypted with the i'th policy public key,receiving the workflow originating from an initiator server and including workflow data encoded by a corresponding i'th vertex public key,decoding the i'th vertex private key by using the i'th policy private key, anddecoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,executing the i'th vertex and processing the decoded workflow data accordingly, andupdating the workflow according to the executed i'th vertex and the processed workflow data.

26. A system configured to be used for a secure execution of workflow tasks of a workflow to be executed according to a given execution pattern within a decentralized network system, the system including at least an initiator server and a plurality of servers, each server satisfying a policy of a vertex of the workflow, and thus, knowing a corresponding policy key pair including a policy private key and a policy public key, respectively, wherein each vertex is assigned a vertex key pair including a vertex private key and a vertex public key, wherein an i'th server as at least one server which is to be determined at runtime of the workflow in accord with the execution pattern to perform one of the vertices of the workflow, the policy of which the i'th server satisfies, called herein the i'th vertex, is configured to perform at least the following operations: receiving the first onion structure being built up of a number of onion layers representing the execution pattern with a most upper layer including the i'th vertex private key and being encrypted with the i'th policy public key,receiving the workflow originating from the initiator server and including workflow data encoded by the i'th vertex public key,decoding the i'th vertex private key by using the known i'th policy private key, anddecoding the workflow data encoded by the i'th vertex public key from the received workflow by using the i'th vertex private key,executing the i'th vertex and processing the decoded workflow data accordingly, andupdating the workflow according to the executed i'th vertex and the processed workflow data.