[Tyrie Vella (tyrielv)]

Problem and Context

VFS for Git requires the PrjFlt ProjFS minifilter driver on every volume. The driver's load behavior is governed by its service Start type. A security-hardening baseline, an anti-virus/EDR product, Group Policy, or manual troubleshooting can set the service to Disabled. When that happens, VFS for Git cannot start the driver.

Today ProjFSFilter.TryStartService calls ServiceController.Start() on the disabled service. The call fails with ERROR_SERVICE_DISABLED (1058), which is caught by a generic Win32Exception handler. EnableAndAttachProjFSHandler.TryEnablePrjFlt then reports the dead-end message "Failed to start prjflt service" — with no indication that the driver is disabled and no hint at the fix. No other code path detects or reports this state. (The existing AutoLogger Start handling is an unrelated ETW logging key, not the driver service.)

This change detects the disabled state and reports it with an actionable message. It does not re-enable the driver: a disabled driver is often disabled deliberately by an administrator or a security tool, so silently re-enabling it would change the machine's security posture without consent.

Changes

• ProjFSFilter.TryStartService now checks ServiceController.StartType. When the service is Disabled, it returns a specific error that names the driver and gives the exact elevated command to re-enable it (sc.exe config prjflt start= auto). It also maps ERROR_SERVICE_DISABLED from a failed Start() as a fallback for the disable-after-check race.
• TryStartService now reports its failure through an out string error parameter instead of returning only a bool, so the caller can surface the specific message.
• EnableAndAttachProjFSHandler.TryEnablePrjFlt propagates that message instead of the generic "Failed to start prjflt service", and records the start-failure reason in the TryEnablePrjFlt_Summary health telemetry so a disabled driver is detectable in aggregate.
• Added a pure internal GetStartServiceFailureError(int nativeErrorCode) seam and unit tests in ProjFSFilterTests covering the disabled error code and other start-failure codes.

[Nathan Baird (ShiningMassXAcc)]

Review pass over this PR. 5 comment(s).

[Nathan Baird (ShiningMassXAcc)]

Re-review of the amended commit. All five of my earlier threads are genuinely addressed — thank you. Two new findings below; the first one I think inverts the PR's premise, so flagging it before this lands.

CI is green across every leg.

[Nathan Baird (ShiningMassXAcc)]

The current head is unchanged since the prior review. Requesting changes for the existing High finding: wrapped native service-start failures still enter the InvalidOperationException handler and are reported as "service was not found," bypassing the intended Win32 error mapping. Please address the existing thread and cover the wrapped-exception shape before merge:
#2122 (comment)

The existing Medium request for stable telemetry reason codes also remains:
#2122 (comment)

These are references to the original findings, not new duplicate inline threads.

[Keith Klein (KeithIsSleeping)]

[KeithBot]: 🟡 Medium · Design — Installer drops this message: exit code 4 still shows a dead-end fatal error

Location: /GVFS/GVFS.Installers/Setup.iss:563

After install or upgrade, Setup.iss runs gvfs service --mount-all; with PrjFlt disabled it exits 4 and prints this new remedy. The exit-4 branch raises a fixed 'Fatal: Could not configure and start Windows Projected File System.' and never shows the captured MountOutput, so installs and reinstalls on such machines still end in the dead-end error.

Evidence:
gvfs service --mount-all (built at c248802), stdout redirected like ExecWithResult, against a stub service pipe returning the handler's disabled-driver error:
exit code: 4
stdout: Failed to enable PrjFlt: The PrjFlt driver service is disabled and cannot be started. [...] sc.exe config prjflt start= auto
Setup.iss:563-565 then runs RaiseException('Fatal: Could not configure and start Windows Projected File System.'); MountOutput is only shown by the non-4 branch (567-570).

Fix: Include the captured output in the FilterError branch, e.g. RaiseException('Fatal: Could not configure and start Windows Projected File System.' + #13#10 + MountOutput);. Optionally also Log(MountOutput) so silent installs keep the remedy in the setup log.

_{🤖 KeithBot · 1/3 models · verified locally}

[Keith Klein (KeithIsSleeping)]

[KeithBot]: ⚠️ 1 issue · iteration 1 (c248802)

1. 🟡 Medium Installer drops this message: exit code 4 still shows a dead-end fatal error

Validated: ✅ Build GVFS.UnitTests at source + base · ✅ ProjFSFilterTests 17/17 at source · ⚠️ Full unit suite 984/996 at source · ⚠️ Mutation of fix parts vs new tests · ✅ Real-SCM harness, read-only · ⚠️ Mocked-SCM error matrix 25/25 · ✅ gvfs service --mount-all via stub pipe · ✅ NativeAOT keeps enum names · ⏭️ Installer (Inno Setup) exit-4 path · ⏭️ Functional tests (ServiceTests)

_{🤖 KeithBot · claude-opus-5.5, gpt-6.1-sol, gemini-4-argon · reply on a thread to dispute}