Case study · Software supply chain

A query, not a hunt: building a live software inventory for an Indian stockbroker after Java4Shell

An Indian stockbroker ran its trading, identity and back-office systems on Java, some built in-house and some bought. When Java4Shell was disclosed, nobody could say quickly where every Java runtime was or which version it ran. The SBOMs produced for SEBI's cyber framework listed libraries, not runtimes. The goal was an inventory that answers that question in minutes, and a patch path that rebuilds rather than repairs.

Case studyCapital markets · IndiaSumanta Mukhopadhyay9 min read
#security#cloud#governance
CASE STUDY · CASE STUDY · SOFTWARE SUPPLY CHAIN19 Nov 2026 · 15:50 GST
At a glance
SectorCapital markets · India
My roleFractional CTO and principal engineer
EngagementIncident-driven assessment, architecture and delivery leadership
ScopeSBOM and runtime inventory, vendor software, container rebuilds, drift control, regulatory evidence
What changed
  • One inventory covering in-house, vendor and open-source software, including runtimes
  • Exposure questions answered by query instead of manual surveys
  • Patches shipped as rebuilt images, with drift from deployed state detected
  • Vendor products tracked with their own bundled components and fix dates
  • SBOM evidence for SEBI's framework generated from the same data
SBOM generation in CIRuntime discovery on hostsContainer image scanningInventory and graph storeVendor SBOM intakeKubernetes admission controlGolden base imagesVulnerability matching

Anonymised composite. Client names, identifying details and confidential figures are withheld; public facts are cited in the references.

The situation

A mid-sized Indian stockbroker ran most of its estate on Java: an in-house order management service, a customer portal, Keycloak for identity, Jenkins for builds, and several bought products for exchange connectivity, risk and back office, each shipping with its own Java. Some ran in Kubernetes, some on virtual machines, and a few on servers that only the vendor touched.

Under SEBI's Cybersecurity and Cyber Resilience Framework, the firm already produced software bills of materials for software supporting core operations, as regulated entities are expected to, covering in-house, third-party, open-source and SaaS software3. CERT-In's guidelines set out what those SBOMs should contain and have since been extended to other kinds of bill of materials4. On paper, the firm was well placed.

Why now. Java4Shell, a flaw in the Base64 decoder of OpenJDK's runtime, was disclosed in November 2026, with reports that it can enable unauthenticated remote code execution in applications such as Jira, Jenkins and Keycloak2. Updated Java builds arrived on 17 November, with other distributions following on their own schedules1. The first question from the board was simple: where are we exposed? The SBOMs could not answer it, because they listed the libraries each application declared, not the Java runtime underneath.

What had to be true

  • Runtimes in the inventory, not only application libraries: Java, Python, Node and the operating system packages beneath them.
  • Vendor software included, with the components it bundles and the vendor's fix status.
  • Answers by query: which systems run an affected version, who owns them, and are they internet-facing.
  • One patch path: rebuild and redeploy, so what runs matches what was scanned.
  • Regulatory evidence as a by-product, not a separate spreadsheet.

Architecture

Sources
CI buildsSBOM per artefactImage scansbase image and runtimeHost discoveryruntimes on VMsVendor SBOMsrequested at procurement
Inventory
Component graphartefact, image, host, ownerDeployment mapwhat runs whereExposure tagsinternet-facing, critical
Matching
Advisory feedsCVE and vendorVersion rulesper distributionExceptionsowner, expiry, mitigation
Remediation
Golden base imagesone place to patch JavaRebuild pipelineall dependantsAdmission controlblock old images, detect drift
Evidence
SBOM exportsfor auditsResponse timelineper advisoryException registerboard visible
SBOMs became one input to a live inventory, not the inventory itself.

The data flow, end to end

  1. BuildCI produces an SBOM and records the base image for every artefact
  2. DiscoverScans of images and hosts add runtimes the build never declared
  3. MapEach component is linked to where it runs, its owner and its exposure
  4. MatchA new advisory is matched by version and distribution in minutes
  5. RebuildGolden images are patched and every dependent image rebuilt and redeployed
  6. ProveThe inventory shows the old version gone, or an approved exception
Advisory in, query, rebuild, proof out: the same path for every runtime flaw.

The hard part: making the running estate match the inventory

An inventory is only as good as its link to what is actually running. Two things broke that link. First, Java arrived through many doors: base images from different teams, host packages, and vendor installers each bringing their own runtime. The fix was a small set of golden base images owned by the platform team, so patching Java meant changing one place and rebuilding everything that depended on it. Vendor products were tracked separately, with their bundled runtimes and the vendor's promised fix date recorded against them.

Second, people patched running systems by hand under pressure, which left containers that no longer matched their images. Kubernetes admission control can catch this: the Kubernetes blog describes a validating webhook that tags pods someone has run interactive commands in, and evicts them after a set interval so they are replaced from their declared state5. The team used the same idea, alongside a policy that only admits images built from current golden bases.

Sequence diagram: a developer runs kubectl exec, the API server calls a validating admission webhook that tags the pod as interacted, allows the request, and later evicts tagged pods after a set interval
Detecting drift: pods changed by hand are tagged and replaced from their declared state. Source: Kubernetes blog, Saifuding Diliyaer · CC BY 4.0

An SBOM tells you what you built. A runtime inventory tells you what you are running. For vulnerability response, only the second one matters.

Virtual machines and vendor-managed servers could not use admission control, so host discovery ran on a schedule and any runtime it found that was not in the inventory became a ticket with an owner. The approach builds on the legacy estate modernisation and vulnerability handling case studies, and on supply chain security patterns.

How it ran

Security and compliance
  1. Exposure triage for Java4Shell
  2. Exception register
  3. SBOM evidence for audits
Platform engineering
  1. Golden base images
  2. Rebuild pipeline and admission policy
  3. Host discovery
Vendor management
  1. Vendor runtime survey
  2. Fix dates in contracts
  3. SBOMs at procurement
Security asked the questions, platform made them answerable, vendor management closed the gaps outside the firm's control.
Phase 1Respond
  • Find internet-facing Java
  • Patch or isolate
  • Track vendor fixes
Phase 2Inventory
  • Image and host scans
  • Component graph
  • Owners and exposure
Phase 3Standardise
  • Golden base images
  • Rebuild pipeline
  • Drift detection
Phase 4Sustain
  • Advisory matching
  • Vendor SBOM intake
  • Regulatory evidence
Fix the incident first, then make the next one cheap.

Key decisions and trade-offs

DecisionAlternativeWhy
Scan images and hosts as well as buildsRely on build-time SBOMsRuntimes rarely appear in application manifests
A few golden base imagesEach team picks its own baseOne place to patch the runtime for everyone
Rebuild and redeployPatch running containersWhat runs stays identical to what was scanned
Vendor runtimes tracked with fix datesTreat vendor products as black boxesBundled Java is often the slowest part to fix
Time-limited exceptions with ownersOpen-ended risk acceptanceExceptions get reviewed instead of forgotten

What I'd do again, and differently

Again: treat the inventory as a product with an owner, and make regulatory SBOMs an export from it rather than a parallel effort. Differently: ask vendors for their bundled components and patch commitments at contract renewal, not during an incident. That was the slowest part of the response, and the only part the firm could not engineer its way out of. See vendor admin access as perimeter and the tech news for 19 November.

If you run a regulated business in India, my work with Indian companies explains how I approach it, or analyse your challenge and bring the result to a call.

References
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
    Using Admission Controllers to Detect Container Drift at RuntimeKubernetes blog (Saifuding Diliyaer) · 21 Dec 2021
FAQ

Common questions

Does SEBI require SBOMs?

Yes. SEBI's Cybersecurity and Cyber Resilience Framework expects regulated entities to maintain software bills of materials for software supporting core business operations, covering in-house, third-party, off-the-shelf, open-source and SaaS software.

Why did the existing SBOMs not answer the Java4Shell question?

They were generated from application build files, so they listed libraries the code declared. The Java runtime itself came from base images, host packages or vendor installers, and was not in them.

Why rebuild images instead of patching running containers?

A patched running container no longer matches its image, so the next restart brings the flaw back and nobody can prove what is running. Rebuilding the image keeps what is deployed, what is scanned and what is recorded the same thing.

Keep reading

Related

Open to advisory, fractional CTO & select builds

Let's talk about what you're building.

The fastest route is a free 30-minute call. Prefer to write? Send a message and I'll reply personally within two business days.

Start a conversation

How would you like to reach me?

Book a 30-min callFastest. Pick any open slot on my calendar.→ Write in GmailOpens a pre-filled draft in your browser.→ Use my email appOutlook, Apple Mail or your default client.→

I reply personally within two business days. Time zones from the Americas and Europe to the Middle East and Asia-Pacific.