ArgumentOutOfRangeException in NtProcessInfoHelper.GetProcessInfos when retrieving Process.SessionId

marius-bughiu commented 5 days ago

Description

Hi team 👋 , I'm running into the following ArgumentOutOfRangeException on all of the machines of a certain customer, when making a simple call to retrieve the current process' SessionId. For context, we're talking about a .NET 8.0.0 (sdk: 8.0.100) WPF application.

System.ArgumentOutOfRangeException: Specified argument was out of the range of valid values.
   at System.Diagnostics.NtProcessInfoHelper.GetProcessInfos(ReadOnlySpan`1 data, Nullable`1 processIdFilter, String processNameFilter)
   at System.Diagnostics.NtProcessInfoHelper.GetProcessInfos(Nullable`1 processIdFilter, String processNameFilter)
   at System.Diagnostics.ProcessManager.GetProcessInfo(Int32 processId, String machineName)
   at System.Diagnostics.Process.EnsureState(State state)
   at System.Diagnostics.Process.get_SessionId()

The exception occurs when trying to Slice the ReadOnlySpan.

0:000> !clrstack2
DbgId ThreadId Apartment Kind CLR           GC Mode    GC Suspending?
    0     3c74 STA            v8.0.23.53103 Preemptive no

SP               IP               Function                                                                                                                                                                                                            Source
000000d307f79038 0000000000000000 HelperMethodFrame                                                                                                                                                                                                   
000000d307f79130 00007ffab8f6bd36 System.ThrowHelper.ThrowArgumentOutOfRangeException()                                                                                                                                                               /_/src/libraries/System.Private.CoreLib/src/System/ThrowHelper.cs @ 82
000000d307f79170 00007ffb17098329 System.ReadOnlySpan`1[[System.Byte, System.Private.CoreLib]].Slice(Int32)                                                                                                                                           /_/src/libraries/System.Private.CoreLib/src/System/ReadOnlySpan.cs @ 351
000000d307f791a0 00007ffb1b1b13ff System.Diagnostics.NtProcessInfoHelper.GetProcessInfos(System.ReadOnlySpan`1<Byte>, System.Nullable`1<Int32>, System.String)                                                                                        /_/src/libraries/System.Diagnostics.Process/src/System/Diagnostics/ProcessManager.Win32.cs @ 334
000000d307f792e0 00007ffb1b1b129a System.Diagnostics.NtProcessInfoHelper.GetProcessInfos(System.Nullable`1<Int32>, System.String)                                                                                                                     /_/src/libraries/System.Diagnostics.Process/src/System/Diagnostics/ProcessManager.Win32.cs @ 307

This is the code triggering the exception:

using var process = Process.GetCurrentProcess();
CurrentProcessId = process.Id;
CurrentSessionId = process.SessionId;

We cannot debug on the customer's machine, but we were able to capture a memory dump of the process when it crashed. Based on that dump, I've managed to extract the byte[] returned by NtQuerySystemInformation and what happens is:

the buffer size gets created for 881112
by adding NextEntryOffset to processInformationOffset we reach an offset of 881296 at some point, which is outside the buffer
Splice throws an ArgumentOutOfRangeException

I've turned the dotnet code upside down and on all sides, and I cannot figure out a code path on which this could possibly happen. Interop.NtDll.NtQuerySystemInformation looks to provide a snapshot in time of the processes, so there should be no overrun due to newly created processes. The only other option is that Interop.NtDll.NtQuerySystemInformation returns invalid data.

_Some other things we've tried:

reproducing the issue in blank console application - does not reproduce
temporarily disabling security software on the machine - no effect_

Thoughts on why this might happen / how we could troubleshoot this further? Any help would be greatly appreciated, thank you!

Reproduction Steps

using var process = Process.GetCurrentProcess();
CurrentProcessId = process.Id;
CurrentSessionId = process.SessionId;

Expected behavior

No exception is thrown and the correct SessionId is returned.

Actual behavior

ArgumentOutOfRangeException is thrown when retrieving the SessionId.

Regression?

On the same machines, a .NET 6 version of the same application works without any issues. Though this might no be very relevant, since we're not able to reproduce this with a .NET 8 console application either.

Known Workarounds

No response

Configuration

WPF application, using .NET 8.0.0, built using SDK: 8.0.100
Windows x64

Other information

No response

dotnet-policy-service[bot] commented 5 days ago

Tagging subscribers to this area: @dotnet/area-system-diagnostics-process See info in area-owners.md if you want to be subscribed.

jkotas commented 4 days ago

Thoughts on why this might happen / how we could troubleshoot this further?

Since this happens on all of the machines of a certain customer, I would double check for a buggy security monitoring software:

Check that there are no .dlls related to any security software loaded in the process. There may be more than one security monitoring software installed on the machine. Also, disabling it may not actually disable it completely.
Check the disassembly of NtQuerySystemInformation to verify that it was not modified by the security software.

The security monitoring software often intercept Windows OS APIs and make the APIs behave incorrectly. (For example, hang tracked by https://github.com/dotnet/runtime/issues/108144 is a different issue caused by a buggy security software.)

dotnet / runtime