CVE-2024-49756

AshPostgres is the PostgreSQL data layer for Ash Framework. Starting in version 2.0.0 and prior to version 2.4.10, in certain very specific situations, it was possible for the policies of an update action to be skipped. This occurred only on "empty" update actions (no changing fields), and would allow their hooks (side effects) to be performed when they should not have been. Note that this does not allow reading new data that the user should not have had access to, only triggering a side effect a user should not have been able to trigger.

To be vulnerable, an affected user must have an update action that is on a resource with no attributes containing an "update default" (updated_at timestamp, for example); can be performed atomically; does not have require_atomic? false; has at least one authorizer (typically Ash.Policy.Authorizer); and has at least one change (on the resource's changes block or in the action itself). This is where the side-effects would be performed when they should not have been.

This problem has been patched in 2.4.10 of ash_postgres. Several workarounds are available. Potentially affected users may determine that none of their actions are vulnerable using a script the maintainers provide in the GitHub Security Advisory, add require_atomic? false to any potentially affected update action, replace any usage of Ash.update with Ash.bulk_update for an affected action, and/or add an update timestamp to their action.

AshPostgres is the PostgreSQL data layer for Ash Framework. Starting in version 2.0.0 and prior to version 2.4.10, in certain very specific situations, it was possible for the policies of an update action to be skipped. This occurred only on "empty" update actions (no changing fields), and would allow their hooks (side effects) to be performed when they should not have been. Note that this does not allow reading new data that the user should not have had access to, only triggering a side effect a user should not have been able to trigger.

To be vulnerable, an affected user must have an update action that is on a resource with no attributes containing an "update default" (updated_at timestamp, for example); can be performed atomically; does not have require_atomic? false; has at least one authorizer (typically Ash.Policy.Authorizer); and has at least one change (on the resource's changes block or in the action itself). This is where the side-effects would be performed when they should not have been.

This problem has been patched in 2.4.10 of ash_postgres. Several workarounds are available. Potentially affected users may determine that none of their actions are vulnerable using a script the maintainers provide in the GitHub Security Advisory, add require_atomic? false to any potentially affected update action, replace any usage of Ash.update with Ash.bulk_update for an affected action, and/or add an update timestamp to their action.

Impact

What kind of vulnerability is it? Who is impacted?

In certain very specific situations, it was possible for the policies of an update action to be skipped. This occurred only on "empty" update actions (no changing fields), and would allow their hooks (side effects) to be performed when they should not have been. Note that this does not allow reading new data that the user should not have had access to, only triggering a side effect a user should not have been able to trigger.

You must have an update action that:

• Is on a resource with no attributes containing an "update default" (updated_at timestamp, for example)
• can be performed atomically.
• Does not have require_atomic? false
• Has at least one authorizer (typically Ash.Policy.Authorizer)
• Has at least one change (on the resource's changes block or in the action itself) This is where the side-effects would be performed when they should not have been.

• Is there ever a place where you call this action manually, using Ash.update. Note that AshGraphql and AshJsonApi action calls are not affected as they use Ash.bulk_update.
• If so, is there ever a case where you call the action with zero inputs, and have it produce zero changing fields.
• If so, could it then produce a side effect. This means you'd have an after_action hook that calls some other resource.
• If so, does that side effect bypass another resource's policies, i.e using authorize?: false, or not providing the same actor.

Everything above the line can be checked with the provided script. Everything below it, must be checked manually. The script for checking this is available in the "Might I be affected" section.

The script can have false positives, but will not have any false negatives. So if you run the script, and it says "No potential vulnerabilities found", then all you need to do is update ash_postgres.

Patches

This problem has been patched in 2.4.10 of ash_postgres.

Workarounds

You could:

1. Determine that none of your actions are vulnerable using the script.
2. Add require_atomic? false to any potentially affected update action
3. Replace any usage of Ash.update with Ash.bulk_update for an affected action
4. add an update timestamp to your action.

Might I be affected

This gist provides a script you can run to detect if you are potentially vulnerable.

https://gist.github.com/zachdaniel/e49166b765978c48dfaf998d06df436e

References

Original Report/discovery: https://elixirforum.com/t/empty-update-action-with-policies/66954 Fix commit: https://github.com/ash-project/ash_postgres/commit/1228fcd851f29a68609e236f7d6a2622a4b5c4ba

AshPostgres is the PostgreSQL data layer for Ash Framework. Starting in version 2.0.0 and prior to version 2.4.10, in certain very specific situations, it was possible for the policies of an update action to be skipped. This occurred only on "empty" update actions (no changing fields), and would allow their hooks (side effects) to be performed when they should not have been. Note that this does not allow reading new data that the user should not have had access to, only triggering a side effect a user should not have been able to trigger.

To be vulnerable, an affected user must have an update action that is on a resource with no attributes containing an "update default" (updated_at timestamp, for example); can be performed atomically; does not have require_atomic? false; has at least one authorizer (typically Ash.Policy.Authorizer); and has at least one change (on the resource's changes block or in the action itself). This is where the side-effects would be performed when they should not have been.

This problem has been patched in 2.4.10 of ash_postgres. Several workarounds are available. Potentially affected users may determine that none of their actions are vulnerable using a script the maintainers provide in the GitHub Security Advisory, add require_atomic? false to any potentially affected update action, replace any usage of Ash.update with Ash.bulk_update for an affected action, and/or add an update timestamp to their action.

AshPostgres is the PostgreSQL data layer for Ash Framework. Starting in version 2.0.0 and prior to version 2.4.10, in certain very specific situations, it was possible for the policies of an update action to be skipped. This occurred only on "empty" update actions (no changing fields), and would allow their hooks (side effects) to be performed when they should not have been. Note that this does not allow reading new data that the user should not have had access to, only triggering a side effect a user should not have been able to trigger.

To be vulnerable, an affected user must have an update action that is on a resource with no attributes containing an "update default" (updated_at timestamp, for example); can be performed atomically; does not have require_atomic? false; has at least one authorizer (typically Ash.Policy.Authorizer); and has at least one change (on the resource's changes block or in the action itself). This is where the side-effects would be performed when they should not have been.

This problem has been patched in 2.4.10 of ash_postgres. Several workarounds are available. Potentially affected users may determine that none of their actions are vulnerable using a script the maintainers provide in the GitHub Security Advisory, add require_atomic? false to any potentially affected update action, replace any usage of Ash.update with Ash.bulk_update for an affected action, and/or add an update timestamp to their action.

Impact

What kind of vulnerability is it? Who is impacted?

In certain very specific situations, it was possible for the policies of an update action to be skipped. This occurred only on "empty" update actions (no changing fields), and would allow their hooks (side effects) to be performed when they should not have been. Note that this does not allow reading new data that the user should not have had access to, only triggering a side effect a user should not have been able to trigger.

You must have an update action that:

• Is on a resource with no attributes containing an "update default" (updated_at timestamp, for example)
• can be performed atomically.
• Does not have require_atomic? false
• Has at least one authorizer (typically Ash.Policy.Authorizer)
• Has at least one change (on the resource's changes block or in the action itself) This is where the side-effects would be performed when they should not have been.

• Is there ever a place where you call this action manually, using Ash.update. Note that AshGraphql and AshJsonApi action calls are not affected as they use Ash.bulk_update.
• If so, is there ever a case where you call the action with zero inputs, and have it produce zero changing fields.
• If so, could it then produce a side effect. This means you'd have an after_action hook that calls some other resource.
• If so, does that side effect bypass another resource's policies, i.e using authorize?: false, or not providing the same actor.

Everything above the line can be checked with the provided script. Everything below it, must be checked manually. The script for checking this is available in the "Might I be affected" section.

The script can have false positives, but will not have any false negatives. So if you run the script, and it says "No potential vulnerabilities found", then all you need to do is update ash_postgres.

Patches

This problem has been patched in 2.4.10 of ash_postgres.

Workarounds

You could:

1. Determine that none of your actions are vulnerable using the script.
2. Add require_atomic? false to any potentially affected update action
3. Replace any usage of Ash.update with Ash.bulk_update for an affected action
4. add an update timestamp to your action.

Might I be affected

This gist provides a script you can run to detect if you are potentially vulnerable.

https://gist.github.com/zachdaniel/e49166b765978c48dfaf998d06df436e

References

Original Report/discovery: https://elixirforum.com/t/empty-update-action-with-policies/66954 Fix commit: https://github.com/ash-project/ash_postgres/commit/1228fcd851f29a68609e236f7d6a2622a4b5c4ba