Patent Application
20120284795
The present invention relates to intrusion prevention and detection systems. In particular, the invention is directed to updating host-intrusion-protection configurations according to varying host state and changing intrusion patterns while minimizing processing effort.
An intrusion detection system (IDS) is a system devised to detect many types of malicious network traffic that can't be detected by a conventional firewall. This includes network attacks against vulnerable services, data driven attacks on applications, unauthorized logins, access to sensitive files, viruses, Trojan horses, and worms.
An IDS comprises sensors which generate security events and a central Engine that records events logged by the sensors in a database and uses a system of rules to generate alerts from security events received. In a host-based system, the sensor usually consists of a software agent, which monitors activity of the host on which it is installed and identifies intrusions.
An intrusion prevention system (IPS) responds to a suspicious activity by resetting a connection or by reprogramming a firewall to block network traffic from the suspected malicious source either automatically or in response to a command of an operator. This is traditionally achieved by examining network communications and identifying patterns (known as signatures) of common computer attacks using a process known as “deep packet inspection” (DPI). DPI is a process of packet filtering where the data content of a packet, not just the packet's header, is examined to determine if the packet meets predefined criteria. An examined packet may be modified, discarded, redirected, or marked. A DPI device may also identify data flows.
A host intrusion prevention system (HIPS) prevents malicious behavior from occurring on a host (server or desktop computer). Unlike Network intrusion prevention system (NIPS), HIPS uses software-based filters, and the filters are deployed on the host itself, closest to the applications and data to be protected. Each filter, when added to the HIPS, incrementally adds to the load of the system as a whole, resulting in slowing down the operation of the system.
Accordingly, there is a need in the industry for developing a method for optimally provisioning filters to each host supported by HIPS and determining whether a filter can be removed from a host security configuration. There is also a need for minimizing the resulting processing effort while ensuring appropriate filter provisioning.
It is an object of the invention to provide a method for dynamically providing protection-software to a host in a host-intrusion-prevention system (HIPS).
It is another object of the invention to provide a method for dynamically removing redundant protection-software components from the HIPS.
It is a further object of the invention to provide a method of minimizing processing effort in the HIPS while ensuring appropriate provisioning of protection software.
The main components of host-intrusion-protection system according to the present invention include a central server maintaining protection software, a plurality of local servers, each functioning as a deep security manager (DSM) receiving protection-software from the central server and supporting a respective plurality of hosts, and software agents (hereinafter referenced as agents for brevity) residing in each host.
The protection software comprises filters and detection rules (hereinafter referenced as “rules” for brevity) for matching each filter to a host based on known intrusion patterns, host operational configuration, and host state. The rules rely on “detection expressions” (hereinafter referenced as expressions) which contain scripts that can process information gathered from a host. Based on host operational configuration and events, filters that become redundant are dynamically removed from a host's security configuration to improve performance of the HIPS engine.
A recommendation engine associated with each DSM uses a set of detection rules and associated detection expressions to determine host configuration elements which match specific host attributes. The host-configuration elements include intrusion-prevention filters. The DSM sends queries to hosts to obtain the information relevant to each filter. The DSM also maintains a profile of each subtending host. An agent provided at each host processes received queries and sends relevant information to a respective recommendation engine.
The recommendation engine processes query responses related to host configuration and events, and may modify a host's security configuration accordingly. The recommendation engine may recommend modifying the host's security configuration by adding new filters to a host and/or removal of current filters from the host.
The host-intrusion-protection system according to the present invention is characterized by: using specific and reusable scripts for a centralized scripting engine, the scripts having the capability of retrieving configuration and event information from any host in the system; deriving a set of host queries dynamically from the scripts; dynamically determining required filters for each host; and determining if a filter may be removed from the host's security configuration.
Thus, by using a dynamic, deterministic, and intelligent mechanism for protection-software provisioning and determining when a filter may be removed from a host, the performance of the host is improved.
In accordance with an aspect, the present invention provides an intrusion-protection system for protecting a plurality of hosts. The system comprises: a plurality of agents each associated with one of the hosts; a plurality of local servers each local server communicatively coupled to each host in a respective subset of the plurality of hosts; and a central server maintaining a software library comprising a plurality of encoded filters and a plurality of encoded rules. The central server shares the software library with the each local server. Each local server communicates with an agent, from among the plurality of agents, associated with each host to acquire metadata of each host and prescribes a subset of filters for each host according to the metadata. The metadata comprises a plurality of data elements characterizing configuration of each host and processes executed within each host.
The software library further comprises a set of rules and a set of queries, where each query corresponds to a specific data element from among the plurality of data elements. A rule may require host metadata, in which case execution of the rule comprises processing at least one query. A data element acquired from processing a query from among the at least one query determines either a requirement for a subsequent data element to execute a specific rule or acquisition of all data elements needed to execute the specific rule and determine a corresponding action.
Execution of a rule in the plurality of rules determines at least one of: selection of a filter from among the plurality of filters for installation in an examined host from among the plurality of hosts; selection of a filter already installed in the examined host for removal from the examined host; and ascertaining that filters already installed in the examined host provide adequate intrusion protection to the examined host.
Preferably, each local server maintains a database storing, for each host and for every query processed, a record of every query, a record of a response acquired from each host, an indication of a last execution time, and an indication of a recommended succeeding execution time. Preferably, each local server maintains historical data related to changes in queries responses.
In accordance with another aspect, the present invention provides a method of intrusion prevention implemented at a server supporting a plurality of hosts. The method comprises steps of: maintaining a set of filters, each filter being a set of instructions; prompting a target host, from among the plurality of hosts, to provide metadata characterizing the target host, where the metadata includes a plurality of data elements; receiving the metadata from the target host; determining a subset of rules, from among a predefined set of rules according to the metadata; executing the subset of rules to determine applicable filters from among the set filters; and transmitting the applicable filters to the target host.
The method comprises further steps of: repeating the prompting step; receiving updated metadata from the target host; and comparing new metadata to previous metadata. Metadata discrepancy determined from the step of comparing invokes a further step of executing a new set of rules to either ascertain adequacy of already installed filters in the target host or recommend changes. The changes may include installing new filters in the target host and/or removal of already installed filters from the target host.
With the metadata comprising a plurality of data elements, the method further comprises maintaining for the target host: a record of a time of receiving each data element of the metadata; content of each data element; identifiers of the subset of rules; and identifiers of the applicable filters.
In a preferred embodiment, the method comprises steps of: associating each rule in the predefined set of rules with a respective data element from among the plurality of data elements; selecting the respective data element as a current data element; and executing each rule using the current data element to determine either a result of executing each rule or designating a subsequent data element as a current data element and continuing the step of executing the each rule. The result of executing the each rule may comprise either retaining a current security configuration of the target host or updating the current security configuration of the target host. Updating the security configuration may comprise steps of installing at least one new filter in the target host or removing at least one already installed filter from the target host.
The method further comprises steps of receiving new rules from a central server and placing the new rule in a list of obligatory rules that are executed for a host regardless of the state of the host.
In accordance with a further aspect, the present invention provides an intrusion-protection method for protecting a plurality of hosts. The method comprises steps of: identification of intrusion patterns; devising a set of data filters, each data filter corresponding to at least one of the intrusion patterns; formulating a set of descriptors for characterizing the plurality of hosts; determining a set of rules, each rule associated with a respective data filter in the set of filters and with a subset of descriptors from among the set of descriptors; executing, for a selected host, a selected rule from among the set of rules using content of a respective subset of descriptors; and ascertaining relevance of a specific data filter associated with the selected rule to the selected host according to a result of executing the selected rule.
The method comprises a further step of either installing the specific data filter or removing the specific data filter from the selected host. The method preferably associates each rule with a respective subset of hosts based on static host characteristics.
In accordance with a further aspect, the present invention provides an intrusion-prevention server comprising: an interface communicatively coupled to a plurality of hosts; a plurality of data filters, each data filter corresponding to at least one of intrusion patterns from among a set of intrusion patterns; a plurality of encoded descriptors for characterizing the plurality of hosts; a plurality of encoded rules each rule associated with a subset of descriptors from among the plurality of encoded descriptors; and a recommendation engine for applying a subset of the encoded rules to a selected host from among the plurality of hosts to recommend a subset of the data filters for inclusion in the selected host.
The intrusion-prevention server further comprises an interface communicatively coupled to a central server for receiving the data filters, the encoded descriptors, and the encoded rules. The intrusion-prevention server further comprises a database maintaining a profile of each host in the plurality of hosts.
In accordance with a further aspect, the present invention provides an intrusion-prevention server supporting a plurality of hosts Hk, 0≦k<v, where v is an integer greater than 0. The server comprises: a first data store holding a set of μ encoded filters {Fm, 0≦m<μ}; a second data store holding a global set D of Q descriptors {dj, 0≦j<Q}; an interface with the plurality of hosts for acquiring metadata Ck for characterizing a selected host Hk, the metadata Ck comprising data elements having a one-to-one correspondence to descriptors in a domain of descriptors of the global set D; and an engine for executing a set of rules {Γm(Ck), 0≦m<μ} to determine a binary indicator Φm,k where a value of Φm,k equal to 1 assigns filter Fm to host Hk and a value of Φm,k equal to 0 excludes filter Fm from host Hk. If Φm,k=1 and filter Fm is already installed in host Hk, then nothing need be done. If Φm,k=1 and filter Fm is not installed in host Hk, then the server sends filter Fm to host Hk. If Φm,k=0 and filter Fm is already installed in host Hk, then filter Fm is removed from host Hk. If Φm,k=0 and filter Fm is not installed in host Hk, then nothing need be done.
A third data store holds chronological metadata of the selected host Hk. A program store holds instructions for determining the domain of descriptors using the chronological metadata.
The engine selects a subset of descriptors from within the domain of descriptors according to a current state of the selected host and determines the binary indicator based solely on the subset of descriptors. Each rule Γm(Ck) may be associated with a tree of descriptors, in which case the subset of descriptors are descriptors along a tree path between a root descriptor and a leaf descriptor of the tree of descriptors.
In accordance with a further aspect, the present invention provides a method, implemented at a server supporting a plurality of hosts, for providing intrusion-protection software to the hosts. The method comprises steps of: devising a superset of rules for selectively assigning intrusion-protection software to the hosts; formulating a superset of descriptors for characterizing the hosts; selecting a target host from among the plurality of hosts; acquiring from the target host, a first set of data elements defining a first rule path for a first rule; executing the first rule according to the first set of data elements; acquiring from the target host, a second set of data elements defining a part of a second rule path for a second rule where the second rule path intersects the first rule path in at least one common descriptor; and executing the second rule according to the second set of data elements and a subset of the first set of data elements corresponding to the at least one common descriptor.
Rather than using the entire superset of rules for each host, a host-specific set of rules determined from among the superset of rules and comprising rules applicable to each host individually may be defined to reduce the processing effort. Accordingly, the first rule and the second rule belong to a set of rules specific to the target host.
Data elements acquired from the target host for an individual rule are shared among all rules applicable to the target host. Thus, the total acquired data elements correspond to a union of descriptors defining the rule paths of the individual rules.
A domain of descriptors is defined for each rule. A domain comprises descriptors from the superset of descriptors relevant to each rule. Thus, the first rule path belongs to a domain of the first rule and the second rule path belongs to a domain of the second rule. The domain of descriptors of each rule may be arranged in a tree structure having a root descriptor, inner descriptors, and leaf descriptors. Accordingly, the method comprises steps of: setting a current descriptor to equal the root descriptor; starting a rule path of each rule for the target host from the root descriptor; and repetitively determining a subsequent descriptor along the rule path according to a data element of the target host corresponding to the current descriptor until a leaf descriptor is encountered. When the subsequent descriptor is an inner descriptor, the subsequent descriptor becomes the current descriptor. A rule uses a rule path from the root descriptor to the leaf descriptor to determine if a data filter corresponding to the rule is needed for the target host.
In accordance with another aspect, the present invention provides a method for providing protection software to a plurality of hosts. The method comprises steps of: determining a current time indicator; sending, from a server, a set of queries to a target host from among the plurality of hosts to acquire current characterizing data elements from the target host; and comparing the current characterizing data elements with prior characterizing data elements of the target host. A current protection-software configuration is determined for the target host if at least one current characterizing data element differs from a corresponding prior characterizing data element.
If the current protection-software configuration, just determined, differs from a prior protection-software configuration, a host-reconfiguration time indicator is set to equal the current time indicator, and the current protection-software configuration is transmitted to the target host. The current characterizing data elements are retained for subsequent use as prior characterizing data elements, and the current protection-software configuration is retained for subsequent use as prior protection-software configuration.
Each host is associated with a host type, from among a predefined set of host types, and a set of descriptors, from among a predefined superset of descriptors, is associated with each host type. Thus, the set of queries sent to the target host is associated with a specific set of descriptors associated with a type of the target host.
A superset of rules, where each rule corresponds to an element of the protection-software, is predefined. To determine the current protection-software configuration, selected rules from among the superset of rules are executed according to the current characterizing data elements.
The queries in the set of queries may be assigned different activation intervals.
Accordingly, the method comprises further steps of: associating an activation interval with each query in the set of queries; identifying particular queries in the set of queries where, for each of the particular queries, a sum of a respective activation interval and a prior time indicator associated with the target host exceeds the current time indicator; removing the particular queries from the set of queries; and retaining the current time indicator for subsequent use as a prior time indicator associated with the target host.
Host-specific monitoring periods may be dynamically updated to reflect changing host operational conditions. Accordingly, the method comprises further steps of: determining a current host-reconfiguration period as the current time indicator minus a prior host-reconfiguration time indicator; determining a current monitoring period for the target host as an arithmetic mean of the current host-reconfiguration period and a prior monitoring period; and retaining the monitoring period for subsequent use as a prior monitoring period.
In accordance with a further aspect, the present invention provides a recommendation engine associated with a server for providing intrusion protection to a plurality of hosts communicatively connected to the server. The recommendation engine comprises: means for associating a monitoring period with each host from among the plurality of hosts; intrusion-protection software for determining a current host-protection configuration for each host; means for installing the current host-protection configuration in the target host upon determining discrepancy between the current host-protection configuration and a prior host-protection configuration of each host; means for recording successive host-reconfiguration periods where a host reconfiguration period is a difference between successive instants of time at each of which a current host-protection configuration differs from a prior host-protection configuration; means for updating the monitoring period according to a value of at least one of the successive host-reconfiguration periods for each host; and a scheduler for activating the intrusion-protection software according to the monitoring period.
The intrusion-protection software comprises: a set of queries for acquiring data elements from each host; and a set of rules which execute expressions for determining the current host-protection configuration based on the data elements.
In accordance with an embodiment, the means for updating the monitoring period updates the monitoring period associated with each host as an arithmetic mean value of a current host-reconfiguration period in the successive host-reconfiguration periods and a preceding monitoring period of each host.
In accordance with another embodiment, the means for updating the monitoring period updates the monitoring period associated with each host as an arithmetic mean value of a predetermined number of host-reconfiguration periods in the successive host-reconfiguration periods. The monitoring period associated with each host may also be updated to equal the arithmetic mean value minus a standard deviation of the predetermined number of host-reconfiguration periods subject to a condition that the monitoring period exceeds a predefine lower bound.
In accordance with a further aspect, the present invention provides a method of regulating communications between a server and a plurality of hosts. The server provides intrusion-protection software to the plurality of hosts. The method comprises:
associating a monitoring time τ* with a target host from among the plurality of hosts; and executing a process at the server for determining a current host-protection configuration for the target host. Where the current host-protection configuration differs from a prior host-protection configuration, the method comprises: installing the current host-protection configuration in the target host; recording a reconfiguration time as a current time indicator; determining a current reconfiguration period t as a difference between a current reconfiguration-time indicator and a prior reconfiguration-time indicator; updating the monitoring period τ* as τ*←(τ*+τ)/2 and; scheduling a subsequent execution of the process according to the monitoring period.
The monitoring period τ* is subject to a constraint τ*≦τb*, where τb* is an upper bound specific to the target host.
The process for determining the current host-protection configuration comprises a step of executing a set of rules to determine the current host-protection configuration, where executing each rule comprises steps of: selecting a set of queries from a superset of queries according to a current state of the target host; sending the set of queries to the target host; and receiving from the target host a data element in response to each query in the set of queries.
The step of selecting the set of queries comprises for each query, starting with a root query, determining a subsequent query according to a data element received from the target host in response to each query, where a null subsequent query completes formation of the set of queries.
Each host in the plurality of hosts is selected as the target host at least once during a cyclic global monitoring period.
In accordance with a further aspect, the present invention provides a method of regulating communications between a server and a plurality of hosts. The server provides intrusion-protection software to the plurality of hosts. The method comprises: initializing to zero each of a first sum Σ1, a second sum Σ2, entry m of a vector Vm, 0≦m<κ, and entry m a vector Wm, 0≦m<κ, where κ>1 is a predefined parameter; initializing a cyclic event counter j to −1; and executing a process for determining a current host-protection configuration for the target host.
Where the current host-protection configuration differs from a prior host-protection configuration, the method comprises: installing the current host-protection configuration in the target host; recording an event time as a current time indicator; for j≧0, determining a current reconfiguration period τ as a difference between a current reconfiguration-time indicator and a prior reconfiguration-time indicator; performing the operations j←(j+1)modulo
In accordance with an embodiment, updating the monitoring period comprises further steps of: determining a mean reconfiguration period <τ> as Σ1/κ; determining a reconfiguration-period standard deviation as σ=(Σ2/κ−<τ>2)1/2; and setting the monitoring period as τ*=<τ>−α×σ, α being a predetermined design parameter. The monitoring period is constrained to remain between a lower bound τa* and an upper bound τb*, so that τa*≦τ*≦τb*. The lower bound and the upper bound are specific to the target host.
The method comprises a further step of determining a global monitoring period during which each host in the plurality of hosts is selected at least once as the target host.
Embodiments of the present invention will be further described with reference to the accompanying exemplary drawings, in which:
The terms used in the specification are listed below.
Host: A computing device, such as a personal computer or a server, connecting to a communications network is conventionally referenced as a “host”.
Agent: An agent comprises selected software modules, including data filters (also called “filters”) and rules, installed in a host for monitoring host activities and identifying intrusions. An agent also comprises firewall modules and deep-packet-inspection modules.
Intrusion-detection system (IDS): An IDS is a system that detects suspicious activity in a computer.
Intrusion-prevention system (IPS): An IPS is a system that responds to suspicious activity in a computer.
Host intrusion-prevention system (HIPS): A host intrusion prevention system (HIPS) protects hosts, such as server or personal computers from malicious intrusions.
Local server: In the context of the present invention, a local server is a computer configured to provide intrusion-protection software to a respective set of hosts.
Deep packet inspection (DPI): DPI is a process of packet filtering where the data content of a packet, not just the packet's header, is examined to determine if the packet contains undesirable data.
Deep-Security Manager (DSM): A local server providing deep-packet inspection software to subtending hosts is referenced as a deep-security manager.
Central server: A central server in the security system of the present invention stores and updates protection software and distributes the protection software to a plurality of local servers, functioning as deep-security managers (DSMs), which—in turn—selectively disseminate host-specific protection software.
Host operational configuration: The term refers to configuration of software used for normal operation of a host. This includes an operating system as well as application programs.
Host security configuration: The term refers to configuration of protection software provided to a host by a local server functioning as a security manager.
Host descriptor: A host descriptor is an encoded descriptor defined in a DSM to characterize one aspect of a host. For brevity, a host descriptor is hereinafter referenced as a descriptor. A host may be characterized by numerous descriptors. A host descriptor is a generic designation (such as ‘IP address’) applicable to any host from among a plurality of hosts.
Data element: A data element is the content of a descriptor in a specific host. For example, the data element “127.000.000.001” may be associated with the descriptor “IP address” of a first host and the data element “127.000.000.063” may be associated with the descriptor “IP address” of a second host. A data element may provide static information, quasi-static information or dynamic information. Static, or quasi-static, information may comprise the host's processor type, IP-address, storage capacity, etc. Dynamic information may comprise current application software, processes being run, errors logged, etc. Changes in the configuration of a host and/or the host's processing activities are reflected in the data elements communicated by the host. The terms “acquiring a descriptor” and “acquiring a data element corresponding to a descriptor” may be used synonymously.
Host descriptors are defined at a central server and distributed to local servers (DSMs). Data elements are acquired from each host through an agent residing in the host.
Metadata: The term “metadata” is generally used in the art to refer to concise data that characterize the content of monitored data in a data-transport medium or data-storage medium. In the context of a computer-communication system, the monitored data may be the content of computer files. Metadata about an individual data block might typically include the name of a field containing the data block and its length. Metadata about a computer file might typically include the name of the file, the type of file, and the name of the data administrator. In the context of the present application, metadata related to a specific host may comprise data elements characterizing the host.
Rule: A rule (also called a detection rule) is an encoded set of functions each function associated with a descriptor to determine a course of action for a target host, such as specifying a filter or a number of filters. In general, a rule may apply to each host in a plurality of hosts. However, based on static descriptors of a host, a DSM may determine a priori that a given rule is not applicable to the host. A function of a specific rule applied to a data element of a specific host results in either a pointer to another function within the set of functions defining the rule or in determining completion of the execution of the specific rule for the specific host.
Domain of a rule: The domain of a rule is a set of all descriptors that may be needed by the rule. The domain of a rule is conveniently represented as a tree with a starting descriptor as a root of the tree.
Rule path: A sequence of descriptors within a domain of a rule starting with a root descriptor and ending with a leaf in the tree representation of the rule is a rule path. A rule path always starts at the root descriptor but may end in any of the leaves depending on the time-varying data elements of a host.
Host-specific rule: A rule may be tailored to a specific host or a specific host type by eliminating descriptors within the domain of the rule that are known a priori to be inapplicable to the specific host or specific host type.
Query: Descriptors are a property of a DSM while data elements are a property of a host. A data element associated with a descriptor directs a rule path to a subsequent descriptor. A DSM acquires a data element corresponding to a descriptor by sending a query to an agent residing at the host. A root query corresponds to a root descriptor.
Intersecting rules: Two rules are said to be intersecting if there is at least one descriptor which is common in the domains of the two rules.
Protection software: In the context of the present invention, the term protection software refers to data filters for handling different patterns of intrusion, and rules for specifying relevant filters to a host based on metadata acquired from the host.
Expression: A rule executes functional expressions (also called “detection expressions” or “expressions”) for data-elements' acquisition and processing. An expression may be invoked by more than one rule.
Host state: The term refers to current operational-software installed in a host as well as current processes executed by the host, host registry in a window operating system, open listen-ports“, and files in a normal filing system.
Host monitoring: Host monitoring is a process of acquiring a current state of a host for determining an appropriate security configuration of the host.
Host reconfiguration time indicator: A host-reconfiguration time indicator is an indicator of an instant of time at which a security configuration of a host changes.
Host reconfiguration period: A host reconfiguration period is a time interval between successive changes of a security configuration of a host.
Host monitoring period: A host monitoring period is a recommended waiting time between successive executions of a monitoring process for a specific host.
Global monitoring period: A global monitoring period is a period of time, specific to a DSM, during which each host supported by the DSM is monitored at least once. The host monitoring period may vary significantly from one host to another and a host may be monitored several times during a global monitoring period of the DSM to which the host belongs.
An intrusion-protection system 100 for protecting a plurality of computer devices 160 is illustrated in
The intrusion-protection system 100 comprises a central server 120 which maintains encoded intrusion-protection filters and detection rules devised for host-specific provisioning of intrusion-protection filters. The detection rules use detection expressions and each rule bases its filter selection on a set of data elements characterizing each host. The central server 120 securely communicates with a plurality of local servers 140 through links 122 to a private or public network 125. Each local server 140 supports a number of hosts 160. In the context of an Intrusion-Prevention System, the local server 140 may perform a function of a Deep Security Manager (DSM).
Intrusion information acquired by some intrusion-information means 170 is analyzed to identify intrusion patterns. Protection software 190 is devised, based on acquired intrusion patterns, for installation in the hosts. The protection software preferably comprises filters individually matching intrusion patterns. Given the variety of host hardware and software configurations, a large number of filters may be needed. The filters are maintained in the central server 120 to be distributed to individual local servers 140. A local server 140 may support a large number of hosts 160 in an enterprise and the host configurations may vary from one enterprise to another. A local server 140 may, therefore, need only a subset of the centrally maintained filters at the central server 120. The hosts 160 supported by a local server (DSM) 140 may have widely varying protection requirements and, therefore, each host may need only be provided with selected filters from among the filters supported by the local server.
The central server 120 may distribute protection software to individual local servers sequentially, in parallel, or in a combined sequential-parallel order. The central server 120 may comprise multiple servers arranged in a cooperative server farm. For the purpose of the present invention, the internal structure of central server 120 is irrelevant.
A local server 140 may evaluate the security-software configurations for subtending hosts 160 in sequential sessions and the tasks of rule processing are scheduled to maximize throughput. A local server 140 may have multiple processors.
The local server of interest functions as a deep-security manager (DSM). However, the methods of the present invention are also applicable to software distribution systems other than security-oriented systems.
A detection rule 740 uses expressions to process metadata acquired from a specific host. The rule executed by a specific DSM and applied to a specific host determines whether to install a new filter, keep an existing filter, or remove an installed filter.
A rule comprises:
(1) definition of host-characterizing information such as host name, host type, some additional text, etc.;
(2) a top level expression to call, and corresponding parameters to pass to the expression;
(3) an expected outcome of application of the rule; and
(4) an action, such as a host-configuration-element assignment where an intrusion-prevention filter is assigned to a host.
The rule executes the top-level expression and subsequent expressions, and if the execution result is in agreement with the expected outcome, the DSM performs the specified action.
As illustrated in
The division of the superset D (reference 820) of descriptors into host-specific descriptor sets 822, with further division of each host-specific set into rule domains, as illustrated in
The function Π(Kj, Sj) is evaluated either according to an encoded algorithm or by examining a respective lookup table (not illustrated).
As illustrated in
In operations DSM 140 supporting a plurality of ν>0 hosts 160(0) to 160(ν-1), denoted Hk, 0≦k<ν, selects a host 160(k) from among the ν hosts according to a predetermined criterion or a schedule. In accordance with an embodiment of the present invention, the DSM 140 may send the entire superset D of descriptors to the selected host 160(k) and the host 160(k) responds by sending metadata Ck(D) which comprise a data element 760 corresponding to each descriptor dj0≦j<Q in the superset D. A “recommendation engine” within the DSM 140, to be described with reference to
The DSM 140 executes the rules one at a time and the DSM sends the domain of descriptors associated with a rule being processed to the selected host 160.
At a given host state, only a subset of descriptors from the rule domain may be applicable. Therefore, in a preferred embodiment, the DSM sends a first query indicating only a first descriptor (root descriptor) to the selected host 160 and determines a second descriptor according to a first data element 760 the DSM receives in response to the first query. The DSM sends a second query indicating the second descriptor to the selected host and receives a second data element 760. If the second data element completes the rule's requirement, another rule, if any, may be executed.
Otherwise, the DSM determines a third descriptor, according to the second data element and sends a third query indicating the third descriptor to the selected host which responds with a third data element 760, and so on. The rule execution is complete when the DSM determines that no further data elements are required. With the rule encoded in a tree structure, execution of the rule terminates when the DSM receives a data element corresponding to a leaf descriptor. The process is repeated for each other host 160.
The function of a DSM 140 is illustrated in steps 1742 to 1750 in
Each host 160 is provided with a security configuration comprising a selected subset of the filters 620 determined according to the host's characteristics and observed events. In step 1744, the local server 140 selects a rule from the set of rules 740 acquired in step 1742. In step 1746, a set of descriptors (and corresponding set of queries) needed to execute the selected rule 740 are identified. Each host data element 760, corresponding to a descriptor 750 of a selected host, is acquired by the DSM 140 by sending a respective query to the selected host 160 as indicated by communication 1747. The selected host 160 processes the query (step 1762) and sends the required data element to the DSM 140 as indicated by communication 1763 to be processed in step 1748.
A rule 740 may be encoded as a rule tree as described with reference to
The profile of a host includes descriptors, and corresponding queries, that are relevant to the host, as well as chronological data of query responses (metadata) acquired from the host and decisions made by the recommendation engine 1860 based on the query responses.
The DSM 140 sends a set of queries over a link 1828 to a specific host 160. The specific host responds with metadata comprising specific data elements 760 that characterize the host's type and its present software content. The DSM 140 receives host responses through incoming link 1824 and activates the recommendation engine 1860 to determine whether or not to perform an action.
The engine 1860 may maintain software instructions for classifying hosts according to predefined host types. The engine 1860 analyzes the hosts' profiles and associates each host with a host type. The engine may associate a subset of descriptors, herein referenced as host-specific descriptors, with each host-type. The engine may also have software instructions for associating a subset of descriptors with each rule, thus creating rule-specific subsets of descriptors, to reduce the processing effort at both the DSM 140 and the hosts 160 supported by the DSM. The engine may further have software instructions for determining an intersection of each host-specific subset of descriptors with each rule-specific subset of descriptors to determine a rule domain that is specific to a host 160.
The DSM 140 may also comprise a scheduler, which may be embedded in the engine 1860, for determining a time table for examining each host 160 supported by the DSM 140.
The recommendation engine 1860 may be activated to handle a host 160 in one of two ways: prompting by a security administrator; or scheduling for automatic runs every predetermined period of time.
The recommendation engine 1860 comprises a set of instructions. The instructions implement detection rules 740. Each detection rule requires execution of certain expressions. If the result of executing a rule meets a corresponding expected outcome, a corresponding action is performed.
The detection rules are basically treated as data to be analyzed by the recommendation engine and perform functions specified within the rules. Detection expressions are also treated as data. The detection rules and detection expressions are passed from the central server 120 to the DSMs 140 as data.
In the set of detection expressions, any expression can be a top-level expression. If the underlying language is java, the return values are java types. They can be simple/primitive types, or complex java objects. For example, four “Expected Results”: {null, NOT null, true, false} may be expected. The expected results may comprise other types such as fixed strings, integers, real numbers, or complex values. The corresponding logic can be contained within the expressions. When the recommendation engine runs, it evaluates a rule, determines and executes the expression, and produces a result. If the result is exactly equal to a predefined outcome (such as “EXPECTEDRESULT”) then a corresponding action is performed.
Prior to assigning filters to hosts, the recommendation engine performs two functions. The first is to gather an optimal set of detection rules. The second is to determine host metadata queries. The two functions may be based on specific monitoring parameters. Using monitoring parameters, the recommendation engine can determine a reduced set of rules to process and queries to send.
The monitoring parameters are the complete set of queries ever asked of a given agent. The first time the recommendation engine operates on a host, the engine gathers and stores all the queries executed on the agent residing in the host. The queries are processed during each run for a given host. If all the results are the same as those obtained in a previous run, and if none of the detection rules has changed, then the rules need not be executed. Hence, by monitoring the queries and the results, a considerable amount of processing may be avoided.
Consider multiple rules, having intersecting descriptors, to be applied to an examined host. A process of acquiring a descriptor from a host is a process of acquiring from the host a data element corresponding to the descriptor. A first set of descriptors relevant to a first rule is acquired from the examined host and used to execute the first rule. A second set of descriptors relevant to a second rule is found to intersect the first set of descriptors in at least one common descriptor. A subset of the second set of descriptors which excludes any descriptors common with the first set of descriptors is acquired from the examined host. Executing the second rule may still be based on the entire second set of descriptors, but the process of acquiring common descriptors from the host is performed only once. Likewise, a third set of descriptors relevant to a third rule may have descriptors common with the first set of descriptors and the second set of descriptors and data elements, of the examined host, corresponding to such common descriptors are readily available. Generally, any set of descriptors relevant to a rule, other than the first rule, may have descriptors common with descriptors relevant to preceding rules and data elements, of the examined host, corresponding to such common descriptors would be readily available.
A rule is executed for a target host when (1) the rule itself is modified in the central server 120 and/or (2) when the state of the host changes, i.e., when current metadata relevant to the rule differs for a last-acquired metadata of the host.
A rule that is modified at the central server 120 may be executed “immediately” for each of the hosts supported by a DSM. Alternatively, the rule may be executed for a given host when the host is polled according to a schedule or any other criterion.
The procedure of
In step 2720, the DSM 140 identifies rules that have been added or modified since the immediately preceding examination. The identified rules are added to rule-set-1. In step 2722, queries applicable to the host under consideration are identified and placed in a set termed “query-set”. In step 2724, rules associated with the queries in the query-set are identified and placed in rule-set-2. In step 2730, the query-set and rule-set-1 are examined to determine if both are empty. If the two sets are empty, the procedure is directed to step 2732 for action. Otherwise, the queries in query set are processed in step 2740. Each rule in rule-set-2 corresponding to a query of changing response is transferred from rule-set-2 to rule-set-1. The result of each query and an indicator of the query's response are recorded at the DSM 140 (step 2744). In step 2750, each rule in rule-set-1 is executed to determine whether the rule's processing is complete, i.e., whether a leaf in the rules domain has been reached. When a rule is processed to completion and a recommended action, if any, is taken or recorded for subsequent implementation, the rule is deleted from rule-set-1. If a rule's processing is not yet complete, steps 2730, 2740, 2742, 2744, and 2750 are repeated. When step 2730 indicates that rule-set-1 is empty, i.e., all the relevant rules to the host 160 under consideration have been executed, step 2732 is executed to perform the recommendations produced by executing the rules. The recommendations could be leaving the security configuration of a target host 160 unchanged or modifying the security configuration by adding new filters to the host and/or removing already installed filters from the host.
The DSM stores a profile and corresponding metadata for each of its subtending hosts. Each host is monitored every “monitoring period” to determine a current host-protection configuration. A default monitoring period may be assigned to a host 160 and updated as the state of the host changes.
The recommendation engine 1860 associated with a DSM 140 (
The engine may determine an updated monitoring period associated with a host as an arithmetic mean value of a current host-reconfiguration period in the successive host-reconfiguration periods and a preceding monitoring period of said each host.
Alternatively, the engine may determine the monitoring period as an arithmetic mean value or a geometric mean value of a predetermined number of host-reconfiguration periods in the successive host-reconfiguration periods. The monitoring period determined as an arithmetic mean value may be further reduced by a standard deviation, or a standard deviation multiplied by a predefined value, of said predetermined number of host-reconfiguration periods subject to a condition that the monitoring period exceeds a predetermined lower bound.
A process for determining a current host-protection configuration for the monitored host comprises steps 3130, 3132, and 3134. In step 3130, the DSM 140 sends queries to the monitored host 160 to acquire respective data elements. Each query corresponds to a descriptor. In step 3132, the DSM receives data elements from the monitored host 160 and in step 3134 the DSM applies a set of rules applicable to the monitored host to determine a current host-protection configuration. The steps 3130, 3132, and 3134 may be interleaved.
As described earlier, a DSM stores a profile for each of its subtending hosts 160. In step 3136, the DSM compares the current host-protection configuration derived in step 3134 with a previous host-protection configuration. If the current and previous host-protection configurations are equivalent, the monitoring period is unchanged and the DSM schedules a subsequent monitoring session of the same host based on the same monitoring period. If step 3136 determines that the current and previous host-protection configurations are different, the DSM starts a process of reconfiguring the monitored host and records a reconfiguration instant of time T (step 3138), which is a current time indicator. For the first reconfiguration event with respect to the monitored host, as determined in step 3139, the DSM does not have data to determine a host reconfiguration period. Thus, step 3146 is executed to increase the event counter j to zero and record the value of T as a prior reconfiguration instant T−. The monitoring period is still unchanged and is used in step 3148 to schedule a subsequent monitoring session for the same host. In the subsequent session, steps 3130, 3132, and 3134 are executed and when step 3136 determines a subsequent reconfiguration event, the DSM starts a reconfiguration process of the monitored host and records a current reconfiguration instant T. A current reconfiguration period τ is determined in step 3140 as a difference between T and T−. The values of Σ1, Σ2, entry Vj, and Wj are updated in step 3142 as:
Σ1←Σ1+(τ−Vj); Σ2←Σ2+(τ2−Wj); Vj←τ; and Wj_77 τ2.
It is noted that in actual implementation, τ2 is computed once for use in updating Wj and Σ2.
With continuous operation of the intrusion-protection system, the number of reconfiguration events may grow. However, reconfiguration periods are retained for only κ most-recent reconfigurations. The event counter j is a circular counter increased by one, modulo κ, as indicated in step 3146, with each new reconfiguration event. A number of κ successive reconfiguration periods τ may be stored in circular indices of vector Vj, 0≦j<κ, and similarly, successive values of τ2 are stored in circular indices of vector Wj, 0≦j<κ. Initially, in a transient period before accumulating κ reconfiguration periods, a smaller number λ≦κ of reconfigurations periods may be used for updating the host monitoring period. In step 3144, the host monitoring period τ* may be updated to equal the mean value <τ>=Σ1/λ, or the mean value <τ> minus a multiplier α of the standard deviation σ of the reconfiguration periods; preferably, 0<α<2. The standard deviation σ is determined in a conventional manner as σ=(Σ2/λ−<τ>)1/2.
The host monitoring period τ* is constrained to be at least equal to lower bound τa* and to not exceed an upper bound τb*, i.e., τa*≦τ*≦τb*. The lower bound and the upper bound may be host specific.
The hosts 160 may have significantly different host monitoring periods. A DSM 140 schedules the monitoring processes of its subtending hosts 160 so that each host is monitored at least once during a predefined global monitoring period.
A DSM 140 stores a profile for each of its subtending hosts 160. In Step 3236, the DSM compares the current host-protection configuration derived in step 3234 with a previous host-protection configuration. If the current and previous host-protection configurations are equivalent, the monitoring period is unchanged and the DSM schedules a subsequent monitoring session of the same host based on the same monitoring period (step 3244). If step 3236 determines that the current and previous host-protection configurations are different, the DSM starts a process of reconfiguring the monitored host and records a reconfiguration instant of time T (step 3238), which is a current time indicator. In step 3240, the DSM determines a current reconfiguration period τ as a difference (T−T−), and the value T− of a previous-reconfiguration instant is updated to equal T. In step 3242, a new value of the host monitoring period is determined as τ*←(τ*+τ)/2. In step 3244, the DSM 140 schedules a subsequent monitoring session for the same target host according to the new value of τ*.
It is noted that a time indicator is likely to be cyclic, for example expressed as a number between 0 and (232-1), and the cyclic nature should be taken into account in computing a reconfiguration period τ by determining the difference (T−T−) as [T−T−]={T−T−}modulo y, Y being the time-indicator period; for example Y=232. For example, successive reconfiguration instants of 4,294,967,290 and 0,000,000,020 time units yield a reconfiguration period of (0,000,000,020−4,294,967,290+232)=26 time units.
In a transient period after a first reconfiguration event, step 3242 may determine the current monitoring period τ* to equal the predefined nominal value of τ*, as illustrated in
When the recommendation engine is run to examine a specific host 160, it first gathers a subset of rules that should be run. Once the subset of rules that should be run is established based on data and timestamp requirements, the engine processes a first set of queries against the agent residing in a host. A first set of queries is derived based on previous runs of the subset of rules and their monitor period with respect to the specific host. Query processing is minimized by processing a query only once regardless of the number of rules associated with the query.
Once the first set of queries has been run, the expressions associated with the set of rules are then evaluated. Once all expressions have been processed, the engine decides what to do next:
The rule execution ends when none of the rules produces additional new queries to execute. Once the execution is complete, the rule's actions are performed. The following outlines steps of an exemplary detection rule run instance against a given host.
The detection rules 740 and expressions 1620 are defined at the central server 120. They contain script and directives that are processed by the recommendation engine 1860. In order for the recommendation engine to remove unnecessary Filters, it processes all detection rules including those that are associated with Filters that are already applied to an Agents security configuration.
A first pass assesses the included detection expressions complexity and determines a first set of agent queries to execute. The recommendation engine then communicates with the agent and executes the queries. The query results are processed, and another run through the detection rules is performed if necessary. This process continues until the recommendation engine determines that it can gather no more useful information from the Agent in order to process the detection rules.
The expressions have access to any data or software tool in the DSM. The functions of the expressions include:
The recommendation engine 1860 checks the results of executing the detection rules, using the gathered information from the queries. A detection rule may relate to filter removal and the DSM may either display a recommendation that a specific filter be removed or automatically remove the Filter from the Agent's security configuration.
In the former case, the DSM administrator manually removes the filter, and instructs the DSM to apply a security configuration update. In the latter case, the DSM immediately applies a new security configuration to the agent of a host 160 under consideration with the specific filter excluded.
The steps leading to filter removal from a host 160 are illustrated by means of the simple example below.
Another example of filter removal is given below with a specific rule already available on the DSM. The filter prevents buffer overflow from occurring in a Hypertext Preprocessor (PHP) module designed to plug-in to IIS.
Thus, by using a combination of statically provided logic, host configuration and events, and derived host queries, a method for dynamically assigning filters to a host's security configuration is provided.
Although specific embodiments of the invention have been described in detail, it should be understood that the described embodiments are intended to be illustrative and not restrictive. Various changes and modifications of the embodiments shown in the drawings and described in the specification may be made within the scope of the following claims without departing from the scope of the invention in its broader aspect.
The present application is a Continuation of U.S. patent application Ser. No. 13/166,991 to Durie, filed on Jun. 23, 2011, which is itself a Continuation of U.S. patent application Ser. No. 11/875,500 to Durie et al., filed on Oct. 19, 2007, entitled “System for Regulating Host Security Configuration”, entire contents of which are incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13166991 | Jun 2011 | US |
| Child | 13545932 | US | |
| Parent | 11875500 | Oct 2007 | US |
| Child | 13166991 | US |
