From: Rei Thiessen <rei.thiessen@gmail.com>
Date: Wed, 29 Jun 2011 18:48:40 GMT
Subject: Re: [PATCH] cygwin: set write permission before unlink
Message-ID: <BANLkTincfBvDBYYLGJe-m5hknvCVdmMUww@mail.gmail.com>
In-Reply-To: <09c0b1900a67bd1f701c0b23954a34ab.squirrel@mail.localhost.li>

Another point of concern is that the user might have specifically set
the read-only flag only certain files
to protect them from changes/deletion, but after the patch, git can
delete them with impunity.
But then, a file's permission isn't supposed to matter to unlink() anyways.
Interestingly, Cygwin's packaged unlink command-line utility will
delete read-only files,
so Cygwin's attempt to fake permissions through the read-only flag
when a filesystem is mounted with "noacl"
seems to be inconsistent.

I'll leave this issue up to Cygwin's package maintainer for git.

Regards,
Rei


2011/6/29 Christof Krüger <git@christof-krueger.de>:
>> +#undef unlink
>> +int cygwin_unlink(const char *pathname)
>> +{
>> +     /* "read-only" files can't be unlinked */
>> +     chmod(pathname, 0666);
>> +     return unlink(pathname);
>> +}
>
> I've no idea on how cygwin maps file permissions to the underlying
> filesystem, but the above raised my attention. Doesn't chmodding the file
> to 0666 open a small windows where "group" and "other" users have read
> access to the file? This might be unwanted by the user and could be
> exploited by some attacker listening for changes on that file.
> Or am I too paranoid?
>
> Regards,
>  Chris
>
>