Windows absolute drive path detection incomplete

Hi,

I'd like to report a problem with path validation for Windows, and propose a fix.

Function `verify_path` first calls `has_dos_drive_prefix` in order to prevent absolute drive paths like "C:\xxx". The logic for that is implemented as follows:

#define has_dos_drive_prefix(path) \
	(isalpha(*(path)) && (path)[1] == ':' ? 2 : 0)

The problem is that Windows will still interpret a path like this as an absolute drive path even if the first character isn't alpha, and so it should still be rejected.

In fact, it's actually possible to create drives like that which bypass this filter on Windows (as Administrator), even if it's not an officially supported feature:

...

 Directory of 1:\

08/06/2019 06:19 PM <DIR> . 08/06/2019 06:19 PM <DIR> .. ...

With a drive like this present, a malicious server can write to wherever that drive points when you perform a git clone (to test this, just create a git repository containing :1/file):

Cloning into 'pathtest'...

...
 Directory of pathtest

08/08/2019  04:49 PM    <DIR>          .
08/08/2019  04:49 PM    <DIR>          ..
               0 File(s)              0 bytes

...
 Directory of C:\Users\x\Desktop

08/08/2019  04:49 PM                 9 file

If the drive doesn't exist, the clone will fail anyway:

The system cannot find the path specified.

Cloning into 'pathtest'... fatal: cannot create directory at '1:': No such file or directory warning: Clone succeeded, but checkout failed. You can inspect what was checked out with 'git status' and retry the checkout with 'git checkout -f HEAD'

So I'm proposing to remove the check for the drive letter being alpha in `has_dos_drive_prefix` macro:

#define has_dos_drive_prefix(path) \
	( (path)[1] == ':' ? 2 : 0)

As well as being a security patch for RCE under very unlikely circumstance, this patch is also a micro performance boost!

Thanks,

Christopher Ertl | MSRC Vulnerabilities & Mitigations | Microsoft Limited  Microsoft Limited (company number 01624297) is a company registered in England and Wales whose registered office is at Microsoft Campus, Thames Valley Park, Reading. RG6 1WG

On Thu, Aug 8, 2019 at 12:45 PM Christopher Ertl <Christopher.Ertl@microsoft.com> wrote:

Nit: This isn't safe and will access memory beyond end-of-string if
path is zero-length. Perhaps something like this would be better:

    #define has_dos_drive_prefix(path) \
        (*(path) && (path)[1] == ':' ? 2 : 0)

Thanks, Eric, you're absolutely right; good spot!

Christopher Ertl | MSRC Vulnerabilities & Mitigations | Microsoft Limited  Microsoft Limited (company number 01624297) is a company registered in England and Wales whose registered office is at Microsoft Campus, Thames Valley Park, Reading. RG6 1WG

-----Original Message-----
From: Eric Sunshine <sunshine@sunshineco.com> 
Sent: Thursday, August 8, 2019 5:51 PM
To: Christopher Ertl <Christopher.Ertl@microsoft.com>
Cc: git@vger.kernel.org
Subject: Re: Windows absolute drive path detection incomplete

On Thu, Aug 8, 2019 at 12:45 PM Christopher Ertl <Christopher.Ertl@microsoft.com> wrote:

Nit: This isn't safe and will access memory beyond end-of-string if path is zero-length. Perhaps something like this would be better:

    #define has_dos_drive_prefix(path) \
        (*(path) && (path)[1] == ':' ? 2 : 0)

On 2019-08-08 at 16:44:52, Christopher Ertl wrote:

Is the drive character required to be ASCII? If it can be non-ASCII Unicode, then this doesn't work, because the drive character will be encoded in UTF-8 and will be longer than one character.

An ABNF grammar of valid Windows path names would be helpful here.