Privacy hardening, remote-peer IP masking and review fixes

- Add privacy projection (core/privacy.py), private report storage, bounded
  process capture; omit identities, free text and unapproved paths
- Mask remote peer IPs: only loopback/host/wildcard stay numeric, others are
  reduced to remote_scope private/public (schema + scoring updated)
- Map ss TCP states to canonical names; ASCII-only numeric fields; fix
  "Microsoft Visual C++" product matching (Python and PowerShell)
- Merge 7fed3b9: -Diagnostic switch, serialization exit code 3, PS7 module
  path, UI error detail; keep hardened task/certificate collectors
- Add audit docs, privacy docs and regression tests

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-04 18:50:24 -08:00
parent 7fed3b9ab9
commit 610a3bb447
51 changed files with 2719 additions and 604 deletions

View File

@@ -1,29 +1,38 @@
# Local Windows Server inventory
# Local Windows Server collection
Targets: Windows Server 2016/2019/2022/2025, Windows PowerShell 5.1, 64-bit execution. **Real Windows host validation is still pending in this Linux development workspace.** No remoting or credentials are used.
Read-only collection commands for supported Server release builds 2016/2019/2022/2025.
The Python workflow launches fixed sections using system Windows PowerShell 5.1,
NoProfile, fixed argument lists, bounded memory output and time/cancellation limits.
No remoting/configuration, execution-policy bypass, elevation or credential storage.
Actual Windows runtime/least-privilege acceptance is still pending.
`Get-ServerInventory.ps1` is the fixed JSON driver. With no arguments it emits a full v1 inventory envelope. `-Section system|roles|services|network|tasks|software|storage|shares|certificates` emits an internal per-section JSON result. The Python collector invokes these independently to enforce time/output limits while preserving other sections after failures.
The bundled `samsara/assets/privacy_policy.json` and `privacy_catalogue.json` are
required. Protect the deployment against unauthorized changes. Name/path projection
occurs before PowerShell JSON output and again at Python UI/report boundaries.
| Module | Local read-only source |
| --- | --- |
| Get-ServerSystem.ps1 | CIM OS/product type/build, boot, domain membership, adapter configuration |
| Get-ServerRoles.ps1 | Installed Get-WindowsFeature entries |
| Get-ServerServices.ps1 | Win32_Service metadata and service identities; no PathName/arguments |
| Get-ServerNetwork.ps1 | Get-NetTCPConnection and Get-Process names/IDs |
| Get-ServerTasks.ps1 | Get-ScheduledTask visible definitions, accounts, safe executable basenames, selected triggers |
| Get-ServerSoftware.ps1 | HKLM uninstall registrations using explicit Registry32/Registry64 read-only handles |
| Get-ServerStorage.ps1 | Get-Volume, Get-SmbShare and aggregate Get-SmbOpenFile share-root correlation; volumes/shares share the existing storage inventory category |
| Get-ServerCertificates.ps1 | Public metadata in accessible LocalMachine certificate stores |
| Inventory.Common.ps1 | Section timing, safe errors and permission limitations |
| Module | Default queries | Privacy/coverage limits |
| --- | --- | --- |
| System | Local OS/build/boot, domain-joined boolean, adapter IPs | Fixed OS labels, opaque identities; no domain string/adapter description requested |
| Roles | Get-WindowsFeature installed identifiers | Canonical/opaque names; no active-use assertion |
| Services | Win32_Service Name, State, StartMode, ProcessId | No StartName, DisplayName, PathName; names canonical/opaque |
| Network | Get-Process names/PIDs, Get-NetTCPConnection | No command lines; names canonical/opaque; historical direction unverified |
| Tasks | MSFT_ScheduledTask TaskName, TaskPath, State only | Opaque identities/category; no Actions, Principals, Triggers, Settings, XML or arguments requested; partial |
| Software | Read-only HKLM uninstall registry views | DoNotExpandEnvironmentNames; canonical/opaque display name, numeric version; no installer query |
| Storage | Fixed Win32_LogicalDisk size/free/filesystem/drive metadata | No labels, mount-only volumes or child paths; partial |
| Shares | Get-SmbShare configured roots | Only exact reviewed local roots retained; no Get-SmbOpenFile or session enumeration; partial |
| Certificates | No provider call | not_collected pending a separately approved minimal design |
No Win32_Product, installer operations, private-key export, task execution, service changes, registry writes or remoting setup. No execution-policy bypass; use approved script signing/distribution if enterprise policy blocks scripts. Non-elevated sensitive checks are conservatively partial even if their visible query succeeds. Elevation is never attempted automatically. Domain Administrator privileges are not required; assess with the least-privileged authorized account and treat omitted evidence explicitly.
Provider-returned strings may enter process memory before projection; explicit
properties do not prove that native provider internals process no other metadata.
The application cannot promise PHI-free memory or zero incidental OS writes.
Permission failures have categorical errors, timestamps and limitations; no raw
exception messages are returned. Missing evidence must never imply inactivity.
The Windows Server gate uses CIM ProductType 2/3 plus the release build map; Windows 10/11 are not Server targets even where their kernel builds overlap. The gate is enforced by Python and by full-envelope PowerShell use. `-Section system` can return failed/partial identity details for diagnostics, but Python will not accept an unverified identity as an assessment.
Prefer the Python CLI and its configured private local report folder. Direct driver
output is not supervised by Python's bounds or secure writer: do not use arbitrary
shell redirects on production hosts. See [privacy setup and native validation](../docs/PRIVACY_HARDENING.md).
Direct full-script use has no Python timeout/cancellation supervisor; use the Python CLI or GUI for bounded execution. Both paths honor local script execution policy. Every script emits only selected public/administrative metadata, never raw CIM objects, exception text, task XML, task arguments or service command lines.
See README.md for exact GUI, headless and real-host validation instructions. `tests/powershell/Test-Collectors.ps1` parses scripts and mocks their data sources; this is synthetic testing, not real host validation.
Implementation references: [Get-WindowsFeature](https://learn.microsoft.com/en-us/powershell/module/servermanager/get-windowsfeature), [Win32_Service](https://learn.microsoft.com/en-us/windows/win32/cimwin32prov/win32-service), [Get-NetTCPConnection](https://learn.microsoft.com/en-us/powershell/module/nettcpip/get-nettcpconnection), [Get-ScheduledTask](https://learn.microsoft.com/en-us/powershell/module/scheduledtasks/get-scheduledtask), [registry software inventory](https://learn.microsoft.com/en-us/powershell/scripting/samples/working-with-software-installations), [certificate provider](https://learn.microsoft.com/en-us/powershell/module/microsoft.powershell.security/about/about_certificate_provider).
Scoring v2 excludes default administrative shares, but keeps them in inventory. Business-share client confirmation requires share-root-correlated open-file counts, not just TCP 445 activity. The read-only activity query exports no file names, session users or client names; same-root administrative aliases remain ambiguous. See docs/SCORING.md. This new query still requires native Windows validation.
`tests/powershell/Test-Collectors.ps1` parses ASTs and invokes synthetic mocked
providers only. It makes forbidden certificate, open-file, full-task and label
enumeration fail. Run on an authorized lab host with its existing script policy;
no real collector runs in that test. A passing mock test is not real-host acceptance.