Wireshark-dev: Re: [Wireshark-dev] Npcap 0.01 call for test (2nd)
From: Graham Bloice <[email protected]>
Date: Thu, 23 Jul 2015 07:52:04 +0100

On 22 July 2015 at 22:13, Guy Harris <[email protected]> wrote:

On Jul 22, 2015, at 1:49 PM, Graham Bloice <[email protected]> wrote:

> On 22 July 2015 at 18:37, Guy Harris <[email protected]> wrote:
>> On Jul 22, 2015, at 9:37 AM, Graham Bloice <[email protected]> wrote:
>>> Most, if not all, will be running Wireshark unelevated, as this is a basic tenet of Wireshark use. There are millions of lines of code in Wireshark dissectors and they really shouldn't be given admin privs.
>> Does anybody know whether there exists, in Windows:
>>         1) an inter-process communications mechanism, either documented or reverse-engineered *and* likely to remain intact and usable from release to release and in future releases, over which a HANDLE can be passed;
> DuplicateHandle -  https://msdn.microsoft.com/en-us/library/windows/desktop/ms724251(v=vs.85).aspx

OK, so that's more than just UN*X dup()/dup2(), as it takes process handles and can affect another process's handles.

It says

        If the process that calls DuplicateHandle is not also the target process, the source process must use interprocess communication to pass the value of the duplicate handle to the target process.

which is the other part of this.

> A HANDLE to what though, the handle types that can be duplicated with that call are limited?
> If it's a socket HANDLE, then WSADuplicateSocket (https://msdn.microsoft.com/en-us/library/windows/desktop/ms741565(v=vs.85).aspx) is used.  This creates a structure that can be handed off to the target process by some IPC mechanism.

Nope, it's a handle to something opened with CreateFile(), although the path is a \\.\xxx symbolic link (in the Windows NT sense) to a device, the device in question being the one provided by the WinPcap driver.  See PacketOpenAdapterNPF() in packetNtx\Dll\Packet32.c in the WinPcap source.

DuplicateHandle() works for file handles.
> The IPC Mechanisms supported by Windows are listed here: https://msdn.microsoft.com/en-us/library/windows/desktop/aa365574(v=vs.85).aspx, pipes are commonly used.  I don't think there are issues with pipes between a non-elevated process and an elevated one, but I haven't personally tried that.

Sounds good (on UN*X, UNIX-domain sockets include functionality that's sort of the equivalent of DuplicateHandle() with the source process being the sending process and the target process being the receiving process).

>>         2) a mechanism by which a non-privileged process can request that a subprocess be run with elevated privileges - presumably requiring either user consent or something else to indicate trust - with such an IPC channel established between the non-privileged process and the privileged process?
> A way to elevate a subprocess is via a call to ShellExecuteEx() setting the lpVerb in the passed in SHELLEXECUTEINFO structure to "runas".  See http://blogs.msdn.com/b/vistacompatteam/archive/2006/09/25/771232.aspx.
> This will invoke UAC if enabled (a it should be).

Just out of curiosity:

        What happens if something you run from a command prompt in Windows invokes UAC - does it pop up a dialog in the GUI?

        If you were to ssh into a Windows box (using third-party ssh or Windows 10 ssh:


        ) are you running in a session with any access to the GUI and, if not, what happens with UAC?

I'll have to check what happens if I remote in using PowerShell.  Normally at work I do that with Domain Admin creds so I don't get UAC requests.  I suspect it won't work, and is the part of the reason for the PowerShell Invoke-Command (https://technet.microsoft.com/en-us/library/hh849719.aspx) that allows the user to supply credentials for the command(s) to be invoked.
>> UN*Xes that support libpcap generally have 1) in the form of UNIX-domain sockets (or, in newer versions of OS X, Mach messages, over which those newer versions of OS X support passing file descriptors), and probably have 2) in the form of, if nothing else, sudo or some GUI equivalent.
>> The idea here is to have libpcap - and WinPcap, if the answers to those questions are both "yes" - invoke a *small* helper process to do what work needs elevated privileges to open capture devices, turn on monitor mode, change channels, etc., so that programs using those libraries do not *themselves* require elevated privileges.
>> If the answer for the first question is "no", then do we have some way to run dumpcap with elevated privileges and have a pipe between it and Wireshark/TShark?
> That's what currently happens on Windows using a named pipe, without the elevation though.

That's what currently happens on all platforms, using anonymous pipes on UN*X (are you certain the pipes are named on Windows?  They're created with CreatePipe() - see the code in capchild/capture_sync.c).  On at least some UN*Xes, dumpcap's privileges are elevated, but not by virtue of a "run with elevated privileges" call; the executable image is marked as getting elevated privileges (set-UID root, set-GID to the appropriate group, or appropriate individual capabilities).

On Windows all pipes are named, even if the name is somewhat obscure.  From the CreatePipe function reference (https://msdn.microsoft.com/en-us/library/windows/desktop/aa365152(v=vs.85).aspx)

Anonymous pipes are implemented using a named pipe with a unique name. Therefore, you can often pass a handle to an anonymous pipe to a function that requires a handle to a named pipe.

I hadn't actually checked the code, only used ProcessExplorer to check for the pipe between Wireshark and dumpcap which shows the "named" pipe.

I'm trying to see whether I can, ultimately, get rid of the need to run dumpcap, as well as the need for as much code as there is in dumpcap ever running with elevated privileges.

It sounds as though the work done in NPCap (see the follow-up email from Yang),  is going down that path.

Graham Bloice