Harbor opa agent #135
Harbor opa agent #135
Conversation
…ghlight the policy results persistence layers
…ed sequence diagrams for policy upload and evaluation
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
…ghlight the policy results persistence layers Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
…ed sequence diagrams for policy upload and evaluation Signed-off-by: Prahalad Deshpande <[email protected]>
Signed-off-by: Prahalad Deshpande <[email protected]>
…community into harbor_opa_agent
|
@xaleeks PTAL |
|
|
||
| ### Harbor Policy Agent User Workflow | ||
|
|
||
| Keeping adherance the multi-tenancy tenets of Harbor, OPA policies would be scoped at a per-user level and would be availble for use across **all** projects of which the user is a member. This level of scoping will satisfy the following scenarios: |
There was a problem hiding this comment.
scoped means users can author his own policies right? not enforced at the user level, that would be messy
There was a problem hiding this comment.
Correct. The term scoped has been used to refer to the fact that the user can author and upload/update his/her policies. It is not enforcement at the user level.
| ##Use Cases | ||
|
|
||
| ### Security Admin Persona | ||
| * What are the **Critical** vulnerabilities present in my **Harbor registry**? |
There was a problem hiding this comment.
this already exists but we need ability to display these across all projects that user is entitled do, instead of having to dig into each project to view
There was a problem hiding this comment.
Got it. Some of the use cases mentioned here can be satisfied with the current implementations in harbor too with enhancements to the user interface. I was enlisting the use cases from a completeness perspective with respect to policy integration.
There was a problem hiding this comment.
Disagree. IMO, OPA integration should focus on policy enforcement, not data querying or reorganization.
This can be covered by other tools/services.
There was a problem hiding this comment.
@steven-zou I agree with your point in principle. The above use case can be satisfied by re-organizing the data within the harbor database to confirm to the schema mentioned in the original document shared related to Cloud Native Common Vunerability Evaluation and Reporting Framework.I will create a separate proposal for that.
However, the very nature of OPA is that it allows users to write policies that need not be decision making based. I my understanding is correct, eventually Harbor would also function as a policy store for image compliance policies (@xaleeks please correct my understanding if wrong) and these policies need not be only those that return True or False or a single scalar value for decision making. Other tools and services can query the policy data store to also answer the type of questions in the use case above.
This type of integration would open up the possibility for the user to run custom analytics and queries against the Harbor OPA data store. @xaleeks please let us know what you think
There was a problem hiding this comment.
@steven-zou I agree with you that in in principle, the above use case can be achieved using data schema re-organization as was described in the original document shared related to Cloud-Native Common Vulnerability Specification.
That said, OPA allows users to write policies that return data other than scalars (true, false or a number). Since we would be storing this data within the Harbor OPA data store, other tools and services could query the Harbor OPA datastore to answer such queries and take further actions/decisions.
The use case was mentioned here with the intent of highlighting the possibility of user running ad-hoc queries and analytics against the data in the harbor OPA data store.
| * Which images are impacted by **CVE-12345** which has been flagged as business critical? | ||
| * Which Helm charts use images with **Critical** vulnerabilities? | ||
| * Where can I get access to a summary report on a regular cadence? | ||
| * What **Services** out in the field use an image containing **CVE-12345**? |
There was a problem hiding this comment.
I don't know if harbor itself can answer that question because we don't track things 'out in the field', this seems like a question for something like sonobuoy that can work in conjunction with Harbor
There was a problem hiding this comment.
Yes agreed. This is not the functionality of core Harbor. However, Harbor OPA integration would be storing the OPA policy evaluation results and hence could act as the data store that can be used by frameworks like Sonobuoy.
Similar to the pluggable scanner framework, Sonobuoy could query the Harbor OPA datastore with the name of the vulnerability and would be able to retrieve the results or image names that contain these vulnerabilities and then perform further action as appropriate
The report however for such a query would come from Sonobuoy.
There was a problem hiding this comment.
Seem concerns with @xaleeks.
I don't think Harbor has answers to those questions too.
There was a problem hiding this comment.
@steven-zou As mentioned in my response above - Harbor need not answer this as it's core functionality. However, the OPA data store in Harbor would be able to answer this query for external tools. @xaleeks please comment on what you think about this particular approach.
| * Where can I get access to a summary report on a regular cadence? | ||
| * What **Services** out in the field use an image containing **CVE-12345**? | ||
| * I have a set of enterprise wide "Acceptance for Use" critieria that must be satisified. How can I identify images that satisfy these criteria and those that do not? | ||
| * How can I share the best practice checks that I have designed with my enterprise partner organizations so all maintain the same level of compliance? |
There was a problem hiding this comment.
what is this referring to? replicating opa policies to another harbor instance?
There was a problem hiding this comment.
Yes. We can replicate the best practices and compliance checks across the board to maintain a uniform level of compliance across the board. It also allows for centralized policy authoring
| * What **Services** out in the field use an image containing **CVE-12345**? | ||
| * I have a set of enterprise wide "Acceptance for Use" critieria that must be satisified. How can I identify images that satisfy these criteria and those that do not? | ||
| * How can I share the best practice checks that I have designed with my enterprise partner organizations so all maintain the same level of compliance? | ||
| * I want to evaluate images against standard IT-GRC policies (PCI, HIPAA) and score images against these policies. |
There was a problem hiding this comment.
how would the scoring be accomplished?
There was a problem hiding this comment.
Using OPA policies. The user would write custom OPA policies using rego (a policy language used by OPA) where the scoring logic can implemented. The scores can then be retrieved from the Harbor OPA data store.
There was a problem hiding this comment.
It sounds statistic calculations. Need to depend on OPA?
There was a problem hiding this comment.
@steven-zou There is a need to depend upon OPA for this. While at the base level, the task involves querying data from the data store and aggregating then using some calculations; these calculations would need to be done in OPA on the fetched data if we want to come up with standard scoring policies that can be shared across the board.
The way I envision it working:
- Fetch data of compliance evaluation from scanners
- Evaluate OPA policies containing the statistical and mathematical calculations against this data
- Store the results of the evaluation in the Harbor OPA datastore.
The results in step 3 can then be queried by external tools and services to take further actions/decisions
|
|
||
| The below diagram depicts the user workflow to be followed for a image centric evaluation i.e. the image is a reference point against which policies are evaluated | ||
|
|
||
|  |
There was a problem hiding this comment.
got it, this is akin to the vulnerability scanning experience
There was a problem hiding this comment.
Yep, it is similar to the vulnerability scanning experience.
|
|
||
| #### Policy Evaluation Reporting Workflow | ||
|
|
||
|  |
There was a problem hiding this comment.
what does the report look like for a single policy, is this aggregated summary of all historical evaluations against this policy, across all projects for this particular user? what if I'm logged in as the security admin
There was a problem hiding this comment.
I have not yet come up with the UI representation of the report. But in terms of content it would be what you said. I was envisioning a simple tabular list enumerating the image, the corresponding project and a link to the latest evaluation report. Do we need historical results? I will update the content to be a more specific once we complete this initial discussion
There was a problem hiding this comment.
It seems you're using OPA policy to support some new functions.
Before checking this, I was thinking we can use OPA to do compliance checking in the middleware layer by the harbor system (not by users) and replace the current checking policies that are implemented with native code.
@xaleeks What do you think?
|  | ||
|
|
||
|
|
||
| The next sections detail the low level architecture and component view and interaction diagrams for the Harbor Policy Agent |
There was a problem hiding this comment.
please help review @steven-zou @reasonerjt
There was a problem hiding this comment.
Can we make the OPA itself a configurable service? That means I can leverage my existing OPA to do the work. The user just needs to specify an OPA endpoint in harbor side.
I don't think the policy agent needs to listen to the job service webhook, it can directly listen to the harbor webhook which includes the ScanComplete event.
The architecture is not too clear and may need to be refactored.
Actually, the most important thing is before reviewing the implementation design details we need to reach an agreement on the points in the previous sections, especially the personas/user stories.
I'm thinking whether we need to expand the OPA to be a policy engine as well as a metric engine or not. And whether we should provide entries for users to directly run policy evaluations and check results.
We can continue the offline discussions.
There was a problem hiding this comment.
@steven-zou Yes, the OPA instance being used could be something that already exists in the user environment and Harbor can be made to talk to that. In fact, all the PoCs that have been demonstrated until now, have been done using this assumption. I also think that this will result in a more cleaner integration; the OPA instance could be upgraded without impacting Harbor thereby allowing the user to use the latest and greatest features. Additionally, it also absolves Harbor from managing the configuration of OPA.
However, the results of the policy evaluation would be stored in Harbor using the schema proposed in the Common Vulnerability Evaluation and Reporting framework document. We can discuss this aspect offline to conclude on this aspect. Also, would discuss what parts of the architecture need to be refactored to make things more clear.
|
|
||
| The below sequence diagram depicts the interactions between the various components of the **Harbor Policy Agent** during the Policy evaluation process. It also depicts how Harbor interacts with the Policy Agent using the Policy HTTP Endpoint to trigger a policy evaluation. | ||
|
|
||
|  |
There was a problem hiding this comment.
where exactly does the agent reside?
There was a problem hiding this comment.
As mentioned that in the low level component diagram within the architecture, the Agent would be residing on a separate container image within the Harbor network.
|
|
||
| The **Harbor Policy Agent** provides policy evaluation and reporting capabilities within the Harbor ecosystem. A component view of the policy agent is shown below | ||
|
|
||
|  |
There was a problem hiding this comment.
please review this architecture @steven-zou @reasonerjt
There was a problem hiding this comment.
@xaleeks I have responded to your comments. Please take a look. I will also update the proposal with some initial details in the area related to reporting.
Awaiting review from @steven-zou and @reasonerjt
|
|
||
| ### Harbor Policy Agent User Workflow | ||
|
|
||
| Keeping adherance the multi-tenancy tenets of Harbor, OPA policies would be scoped at a per-user level and would be availble for use across **all** projects of which the user is a member. This level of scoping will satisfy the following scenarios: |
There was a problem hiding this comment.
Correct. The term scoped has been used to refer to the fact that the user can author and upload/update his/her policies. It is not enforcement at the user level.
| ##Use Cases | ||
|
|
||
| ### Security Admin Persona | ||
| * What are the **Critical** vulnerabilities present in my **Harbor registry**? |
There was a problem hiding this comment.
Got it. Some of the use cases mentioned here can be satisfied with the current implementations in harbor too with enhancements to the user interface. I was enlisting the use cases from a completeness perspective with respect to policy integration.
| * Which images are impacted by **CVE-12345** which has been flagged as business critical? | ||
| * Which Helm charts use images with **Critical** vulnerabilities? | ||
| * Where can I get access to a summary report on a regular cadence? | ||
| * What **Services** out in the field use an image containing **CVE-12345**? |
There was a problem hiding this comment.
Yes agreed. This is not the functionality of core Harbor. However, Harbor OPA integration would be storing the OPA policy evaluation results and hence could act as the data store that can be used by frameworks like Sonobuoy.
Similar to the pluggable scanner framework, Sonobuoy could query the Harbor OPA datastore with the name of the vulnerability and would be able to retrieve the results or image names that contain these vulnerabilities and then perform further action as appropriate
The report however for such a query would come from Sonobuoy.
| * Where can I get access to a summary report on a regular cadence? | ||
| * What **Services** out in the field use an image containing **CVE-12345**? | ||
| * I have a set of enterprise wide "Acceptance for Use" critieria that must be satisified. How can I identify images that satisfy these criteria and those that do not? | ||
| * How can I share the best practice checks that I have designed with my enterprise partner organizations so all maintain the same level of compliance? |
There was a problem hiding this comment.
Yes. We can replicate the best practices and compliance checks across the board to maintain a uniform level of compliance across the board. It also allows for centralized policy authoring
| * What **Services** out in the field use an image containing **CVE-12345**? | ||
| * I have a set of enterprise wide "Acceptance for Use" critieria that must be satisified. How can I identify images that satisfy these criteria and those that do not? | ||
| * How can I share the best practice checks that I have designed with my enterprise partner organizations so all maintain the same level of compliance? | ||
| * I want to evaluate images against standard IT-GRC policies (PCI, HIPAA) and score images against these policies. |
There was a problem hiding this comment.
Using OPA policies. The user would write custom OPA policies using rego (a policy language used by OPA) where the scoring logic can implemented. The scores can then be retrieved from the Harbor OPA data store.
|
|
||
| #### Policy Upload User Workflow | ||
|
|
||
|  |
|
|
||
| The below diagram depicts the user workflow to be followed for a policy centric evaluation i.e. the policy is a reference point against which images are evaluated | ||
|
|
||
|  |
There was a problem hiding this comment.
Yes, that is how it would be. Policies tab lists the policies and then the user can select the policies that need to be evaluated against images in the projects where they are an admin.
|
|
||
| The below diagram depicts the user workflow to be followed for a image centric evaluation i.e. the image is a reference point against which policies are evaluated | ||
|
|
||
|  |
There was a problem hiding this comment.
Yep, it is similar to the vulnerability scanning experience.
|
|
||
| #### Policy Evaluation Reporting Workflow | ||
|
|
||
|  |
There was a problem hiding this comment.
I have not yet come up with the UI representation of the report. But in terms of content it would be what you said. I was envisioning a simple tabular list enumerating the image, the corresponding project and a link to the latest evaluation report. Do we need historical results? I will update the content to be a more specific once we complete this initial discussion
|
|
||
| The below sequence diagram depicts the interactions between the various components of the **Harbor Policy Agent** during the Policy evaluation process. It also depicts how Harbor interacts with the Policy Agent using the Policy HTTP Endpoint to trigger a policy evaluation. | ||
|
|
||
|  |
There was a problem hiding this comment.
As mentioned that in the low level component diagram within the architecture, the Agent would be residing on a separate container image within the Harbor network.
proposals/new/opaintegration.md
Outdated
| ## Background | ||
| Harbor currently has support for scanning images for OS vulnerabilities using the [pluggable scanner framework](https://github.com/goharbor/pluggable-scanner-spec). Using the framework, end users can use OS vulnerability scanners of their choice to understand the OS vulnerabilities present within the system. However, reporting capabilities of Harbor with respect to the security and compliance posture of the images persisted within it is very minimal and is restricted to providing a summarized aggregate for the High, Medium and Low vulnerabilities within the images. There are some other crucial limitations such as: | ||
| * The scanning for vulnerabilities is scoped to images within a project of which the end user is a member | ||
| * The scanning is restricted to OS vulnerabilties. |
There was a problem hiding this comment.
This limitation is from scanner themselves, not related to any compliance checking. It seems not suitable to put it here.
There was a problem hiding this comment.
Got it. I have removed this line
proposals/new/opaintegration.md
Outdated
|
|
||
| The above limitations introduce shortcoming in addressing of critical security use cases of the enterprise security administrator who is concerned about the security and compliance posture of the applications being deployed and the container registry itself (including identification of projects containing the offending images) as opposed to individual project level granularities. | ||
| Additionally, from the dev-ops perspective; it is not possible to build a deployment criteria that prevents deploying a service based out of an image not satisfying the acceptance criteria or pre-empt the creation of deployment artifacts (e.g. Helm charts) when the deployments use images that do not satisfy the acceptance criteria | ||
| As can be seen from the use cases above; there is a requirement to persist the results of a scan or vulnerability evaluation in a format that supports ad-hoc querying as well as presentable within a report. |
There was a problem hiding this comment.
supports ad-hoc querying as well as presentable within a report should be covered by OPA? It seems a function of some metric service like Elastic search etc..
There was a problem hiding this comment.
This actually should be separate proposal. It is related to the Common Vulnerability Schema documentation that I had shared previously where-in the results of the compliance evaluation as well as policy evaluation would be stored in RDBMS or Elastic Search.. I will create another proposal for that. Removing this statement from the current proposal
| The above limitations introduce shortcoming in addressing of critical security use cases of the enterprise security administrator who is concerned about the security and compliance posture of the applications being deployed and the container registry itself (including identification of projects containing the offending images) as opposed to individual project level granularities. | ||
| Additionally, from the dev-ops perspective; it is not possible to build a deployment criteria that prevents deploying a service based out of an image not satisfying the acceptance criteria or pre-empt the creation of deployment artifacts (e.g. Helm charts) when the deployments use images that do not satisfy the acceptance criteria | ||
| As can be seen from the use cases above; there is a requirement to persist the results of a scan or vulnerability evaluation in a format that supports ad-hoc querying as well as presentable within a report. | ||
| Additionally, there also exists a critical requirement for the end user to be able to author complex policies that can evaluate the results of an image scan and produce an output that flags the image as matching or failing the acceptance criteria and also share the policies across departments to implement and enforce set of best practices uniformly. |
There was a problem hiding this comment.
I think this is the real requirement of this feature.
| ##Use Cases | ||
|
|
||
| ### Security Admin Persona | ||
| * What are the **Critical** vulnerabilities present in my **Harbor registry**? |
There was a problem hiding this comment.
Disagree. IMO, OPA integration should focus on policy enforcement, not data querying or reorganization.
This can be covered by other tools/services.
| * Which images are impacted by **CVE-12345** which has been flagged as business critical? | ||
| * Which Helm charts use images with **Critical** vulnerabilities? | ||
| * Where can I get access to a summary report on a regular cadence? | ||
| * What **Services** out in the field use an image containing **CVE-12345**? |
There was a problem hiding this comment.
Seem concerns with @xaleeks.
I don't think Harbor has answers to those questions too.
| Keeping adherance the multi-tenancy tenets of Harbor, OPA policies would be scoped at a per-user level and would be availble for use across **all** projects of which the user is a member. This level of scoping will satisfy the following scenarios: | ||
| * Users can author policies relevant to the projects that they manage or a member of and ensure that images within those projects can evaluated against the policies | ||
| * Helps easy migration to a cloud based deployment where-in each user would be a tenant and could own multiple projects against which the policies can be evaluated. | ||
| * To satisfy the use case of an uber registry wide or organization specific Security Admin persona, a new user account for the Security Admin can be created and all enterprise wide acceptance policies can be placed under the Security Admin account. Security Admins can then run evaluations against these policies either on demand or inline with a scan or using a periodinc schedule. |
There was a problem hiding this comment.
It's very different from the current harbor design. For applying to multiple projects, it's put at the system level. For a single project, it's put under the specified project. At present Harbor does not support managing resources at the same place for users with different permissions.
Actually, I like this management view. But it seems not doable based on current Harbor implementation.
proposals/new/opaintegration.md
Outdated
|
|
||
| #### Accessing Policy Functionality in Harbor | ||
|
|
||
| The below mockup shows the mechanism of accessing the Policy functionality within harbor |
There was a problem hiding this comment.
It's inconsistent with the workflow design shown above.
All the sections under Administration can be accessed only by the users with System Admin role.
|
|
||
| The below diagram depicts the user workflow to be followed for a policy centric evaluation i.e. the policy is a reference point against which images are evaluated | ||
|
|
||
|  |
There was a problem hiding this comment.
As rego policies are pure texts, maybe supporting online editor can also be an option.
There was a problem hiding this comment.
Sure. That would actually provide for an integrated experience :). But again it could be taken up in a later iteration IMHO, after implementing the core OPA integration pipeline.
|
|
||
| #### Policy Evaluation Reporting Workflow | ||
|
|
||
|  |
There was a problem hiding this comment.
It seems you're using OPA policy to support some new functions.
Before checking this, I was thinking we can use OPA to do compliance checking in the middleware layer by the harbor system (not by users) and replace the current checking policies that are implemented with native code.
@xaleeks What do you think?
|  | ||
|
|
||
|
|
||
| The next sections detail the low level architecture and component view and interaction diagrams for the Harbor Policy Agent |
There was a problem hiding this comment.
Can we make the OPA itself a configurable service? That means I can leverage my existing OPA to do the work. The user just needs to specify an OPA endpoint in harbor side.
I don't think the policy agent needs to listen to the job service webhook, it can directly listen to the harbor webhook which includes the ScanComplete event.
The architecture is not too clear and may need to be refactored.
Actually, the most important thing is before reviewing the implementation design details we need to reach an agreement on the points in the previous sections, especially the personas/user stories.
I'm thinking whether we need to expand the OPA to be a policy engine as well as a metric engine or not. And whether we should provide entries for users to directly run policy evaluations and check results.
We can continue the offline discussions.
| ##Use Cases | ||
|
|
||
| ### Security Admin Persona | ||
| * What are the **Critical** vulnerabilities present in my **Harbor registry**? |
There was a problem hiding this comment.
@steven-zou I agree with you that in in principle, the above use case can be achieved using data schema re-organization as was described in the original document shared related to Cloud-Native Common Vulnerability Specification.
That said, OPA allows users to write policies that return data other than scalars (true, false or a number). Since we would be storing this data within the Harbor OPA data store, other tools and services could query the Harbor OPA datastore to answer such queries and take further actions/decisions.
The use case was mentioned here with the intent of highlighting the possibility of user running ad-hoc queries and analytics against the data in the harbor OPA data store.
proposals/new/opaintegration.md
Outdated
| ## Background | ||
| Harbor currently has support for scanning images for OS vulnerabilities using the [pluggable scanner framework](https://github.com/goharbor/pluggable-scanner-spec). Using the framework, end users can use OS vulnerability scanners of their choice to understand the OS vulnerabilities present within the system. However, reporting capabilities of Harbor with respect to the security and compliance posture of the images persisted within it is very minimal and is restricted to providing a summarized aggregate for the High, Medium and Low vulnerabilities within the images. There are some other crucial limitations such as: | ||
| * The scanning for vulnerabilities is scoped to images within a project of which the end user is a member | ||
| * The scanning is restricted to OS vulnerabilties. |
There was a problem hiding this comment.
Got it. I have removed this line
proposals/new/opaintegration.md
Outdated
|
|
||
| The above limitations introduce shortcoming in addressing of critical security use cases of the enterprise security administrator who is concerned about the security and compliance posture of the applications being deployed and the container registry itself (including identification of projects containing the offending images) as opposed to individual project level granularities. | ||
| Additionally, from the dev-ops perspective; it is not possible to build a deployment criteria that prevents deploying a service based out of an image not satisfying the acceptance criteria or pre-empt the creation of deployment artifacts (e.g. Helm charts) when the deployments use images that do not satisfy the acceptance criteria | ||
| As can be seen from the use cases above; there is a requirement to persist the results of a scan or vulnerability evaluation in a format that supports ad-hoc querying as well as presentable within a report. |
There was a problem hiding this comment.
This actually should be separate proposal. It is related to the Common Vulnerability Schema documentation that I had shared previously where-in the results of the compliance evaluation as well as policy evaluation would be stored in RDBMS or Elastic Search.. I will create another proposal for that. Removing this statement from the current proposal
| * Which images are impacted by **CVE-12345** which has been flagged as business critical? | ||
| * Which Helm charts use images with **Critical** vulnerabilities? | ||
| * Where can I get access to a summary report on a regular cadence? | ||
| * What **Services** out in the field use an image containing **CVE-12345**? |
There was a problem hiding this comment.
@steven-zou As mentioned in my response above - Harbor need not answer this as it's core functionality. However, the OPA data store in Harbor would be able to answer this query for external tools. @xaleeks please comment on what you think about this particular approach.
| * What **Services** out in the field use an image containing **CVE-12345**? | ||
| * I have a set of enterprise wide "Acceptance for Use" critieria that must be satisified. How can I identify images that satisfy these criteria and those that do not? | ||
| * How can I share the best practice checks that I have designed with my enterprise partner organizations so all maintain the same level of compliance? | ||
| * I want to evaluate images against standard IT-GRC policies (PCI, HIPAA) and score images against these policies. |
There was a problem hiding this comment.
@steven-zou There is a need to depend upon OPA for this. While at the base level, the task involves querying data from the data store and aggregating then using some calculations; these calculations would need to be done in OPA on the fetched data if we want to come up with standard scoring policies that can be shared across the board.
The way I envision it working:
- Fetch data of compliance evaluation from scanners
- Evaluate OPA policies containing the statistical and mathematical calculations against this data
- Store the results of the evaluation in the Harbor OPA datastore.
The results in step 3 can then be queried by external tools and services to take further actions/decisions
| ### Dev Ops Persona | ||
|
|
||
| * Do not deploy a service S which uses images containing a **Critical** vulnerability **V** | ||
| * Fail the creation of Helm charts if they use an image with **Critical** vulnerabilities. |
There was a problem hiding this comment.
I was referring to the Helm Plug-ins (https://helm.sh/docs/topics/plugins/) which can be written in any language as per the documentation. Users have the ability to write their own plug-ins that will take actions based on the data queried from the Harbor OPA data store. Here is a workflow that can be achieved with Helm:
- Helm Plug-in queries Harbor OPA data store to check for any specific policy violations.
- If policy violations found do not pull image from Harbor.
As has been mentioned in the design below, a reporting API layer would expose the harbor OPA data store to be queried by the external world.
| * Pull images having that have a PCI compliance policy evaluation score greater than 8 out of 10 | ||
|
|
||
| ### Project Owner Persona | ||
| * Quarantine images whose PCI compliance policy evaluation score is less than 8 |
There was a problem hiding this comment.
@steven-zou by the term quarantine I meant isolating these images from other compliant images and making sure that images are never pulled or replicated. Such images would end up eating storage space and hence it would be also good if the Garbage Collector would simply delete them.
…ad policies. Modifying proposal to include the new image
There was a problem hiding this comment.
Further responses to review comments from @steven-zou. Will discuss these in an offline call.
| Keeping adherance the multi-tenancy tenets of Harbor, OPA policies would be scoped at a per-user level and would be availble for use across **all** projects of which the user is a member. This level of scoping will satisfy the following scenarios: | ||
| * Users can author policies relevant to the projects that they manage or a member of and ensure that images within those projects can evaluated against the policies | ||
| * Helps easy migration to a cloud based deployment where-in each user would be a tenant and could own multiple projects against which the policies can be evaluated. | ||
| * To satisfy the use case of an uber registry wide or organization specific Security Admin persona, a new user account for the Security Admin can be created and all enterprise wide acceptance policies can be placed under the Security Admin account. Security Admins can then run evaluations against these policies either on demand or inline with a scan or using a periodinc schedule. |
There was a problem hiding this comment.
@steven-zou The mechanism I am proposing here is the same as about is applied for CVE whitelists as @xaleeks has already mentioned. I also went through the documentation here https://goharbor.io/docs/1.10/administration/vulnerability-scanning/configure-system-whitelist/.
Basically, the intent is to have system-wide compliance policies that must be adhered to within the registry and per project based policies that can override or have additional policies.
The implementation of the security admin persona is not related to this proposal. It has been mentioned in this proposal for completeness. From what I understand after having read the documetation and @xaleeks comment, the new Security Admin role is not a requirement at the moment.
proposals/new/opaintegration.md
Outdated
|
|
||
| #### Accessing Policy Functionality in Harbor | ||
|
|
||
| The below mockup shows the mechanism of accessing the Policy functionality within harbor |
There was a problem hiding this comment.
@steven-zou - I placed it within that panel only for the purpose of illustration. Individual project owners who are not system admins can upload their authored policy bundle as mentioned in the new screenshot that I have provided.
|
|
||
| The below diagram depicts the user workflow to be followed for a policy centric evaluation i.e. the policy is a reference point against which images are evaluated | ||
|
|
||
|  |
There was a problem hiding this comment.
Sure. That would actually provide for an integrated experience :). But again it could be taken up in a later iteration IMHO, after implementing the core OPA integration pipeline.
|  | ||
|
|
||
|
|
||
| The next sections detail the low level architecture and component view and interaction diagrams for the Harbor Policy Agent |
There was a problem hiding this comment.
@steven-zou Yes, the OPA instance being used could be something that already exists in the user environment and Harbor can be made to talk to that. In fact, all the PoCs that have been demonstrated until now, have been done using this assumption. I also think that this will result in a more cleaner integration; the OPA instance could be upgraded without impacting Harbor thereby allowing the user to use the latest and greatest features. Additionally, it also absolves Harbor from managing the configuration of OPA.
However, the results of the policy evaluation would be stored in Harbor using the schema proposed in the Common Vulnerability Evaluation and Reporting framework document. We can discuss this aspect offline to conclude on this aspect. Also, would discuss what parts of the architecture need to be refactored to make things more clear.
This Pull Request submits the Harbor Policy Agent proposal for review and voting process