Outlined in this document are the practices and policies that npm applies to help ensure that we release stable/secure software, and react appropriately to security threats when they arise.
We review all security reports within one business day. Note that the npm staff is generally offline for most US holidays, but please do not delay your report! Our off-hours support staff can fix many issues, and will alert our security point of contact if needed.
npm's CTO Laurie Voss is the current point of contact for all security related issues.
Any emails sent to firstname.lastname@example.org will be escalated to the security point of contact, who will delegate incident response activities as appropriate.
All new technical hires are introduced to our security policy as part of the onboarding process.
We learn about critical software updates and security threats from a variety of sources:
Along with keeping an eye out for critical security updates, automatic security updates are enabled on all of our production servers allowing patches to be applied immediately without human intervention.
When a security threat is identified, we have the following process in place:
security-all, which is used to prioritize and coordinate responses to security threats.
P0: Drop everything and fix!
P1: High severity, schedule work within 7 days.
P2: Medium severity, schedule work within 30 days.
P3: Low severity, fix within 180 days.
Along with reacting to security notifications as they happen, we proactively pen-test and audit software.
We perform regular penetration testing and code audits with the security firm Lift Security.
While working on features at npm, all engineers coordinate security audits with the Security Point of Contact.
Documents from this process are available, and can be provided to customers when requested.
The cloud hosting platforms that we use provide options for automated vulnerability scanning.
We should opt for alternative authentication methods when possible:
SSH keys should be rolled out selectively, providing developers access to only the severs that they require access to.
In the next section of the document, we discuss the design methodologies that we use to build stable and secure software.
Logs are important for both debugging applications and detecting security breaches in our software -- ask CJ for a speech about logging.
All applications should contain logging for
operation, and a
unique request identifier.
We use common-log-string internally to standardize this:
At least 90 days of logs should be kept for each service. On high traffic hosts this may require backing-up logs in cloud storage on a regular basis.
On the servers that we manage for other companies, we should audit logs on a regular basis.
TODO: We plan to build automated anomaly detection systems in place for our logs see internal issue #381.
Logs should not contain any sensitive user information, e.g., passwords.
The module hide-secrets is used to help with this.
Micro-services should only have access to databases and files that they need access to.
With our docker-based infrastructure (npm Enterprise) this is achieved by having containers only mount folders on the root host that they require access to.
In our production environment, this is achieved by partitioning services across multiple hosts.
Security-groups, or Zones in the case of SoftLayer, are used to limit the network connectivity between hosts.
When deploying a service, ask: "what other services does this actually need to connect to?"
Any sensitive user information should be encrypted at rest. Using encrypted EBS drives, or an equivalent, is a great way to achieve this.
Communication between services on the same host can be performed via HTTP.
All inter-service communication between two hosts is performed using TLS.
npm has a well-defined, security-focused, development process:
No code goes into production unless it is reviewed by at least one other developer.
The onus is on the reviewer to ask hard questions: "what are the ramifications of opening up port-X?", "why is this connection being made over HTTP instead of HTTPS?"
We love testing at npm:
The design process, and management techniques vary from team to team at npm. Across the board, however, we strive to have continuous deployments. Releasing many small features as they become production ready.
Security is taken into account during all phases of the software development life-cycle: unit tests think about potential threats; when testing on staging, we attempt to test potential exploits, etc.
On our managed Ubuntu hosts, we run the ClamAV AntiVirus software.
The infected server should be retired, and a new server should be provisioned from scratch.
This is a living document and may be updated from time to time. Please refer to the git history for this document to view the changes.
This document may be reused under a Creative Commons Attribution-ShareAlike License.