Re: [rdiff-backup-users] [PATCH] Backing up Windows ACLs

[Andrew Ferguson]

On Jun 22, 2008, at 1:25 AM, Josh Nisly wrote:

> Attached is a first-draft patch to add support for native Windows  
> permissions, including owner and ACLs.
>
> First, let me say that Fred Gansevles is responsible for the most  
> interesting part of this - the code to load and save permissions  
> information to files. Thanks, Fred!
>
> There are a couple of places where I've "cheated" with initializing  
> ACL loading to True instead of determining whether ACLs are  
> supported, but I wanted to get the patch out so that other people  
> could take a look at it and comment on it. At this point, it should  
> work for both local<->local and local<->remote, as long as the  
> remote end has these changes.
>
> Probably the most controversial change is make Windows ACLs  
> completely separate in the code from POSIX ACLs, including saving  
> them to a separate file, etc. The reason that I did this is because  
> they are conceptually and practically two completely different  
> things. If you back up from Windows and restore to linux, you don't  
> want to try to apply the Windows ACLs to the linux filesystem, and  
> vice-versa. I see them as similar to OS X's carbon files, that is,  
> they are a os-specific file information.

Indeed, that makes perfect sense. No controversy here. :-)

An exciting thing about this approach is that we might be able to re- 
use some of the work you and Fred have done here for Mac OS X ACLs.  
The Mac ACLs are sort of a hybrid of Windows and POSIX ACLs --- you  
use the POSIX-defined API to access ACL entries which look identical  
to Windows and (at first study) look quite incompatible with the Linux  
ones.

> The one thing I don't like about this patch is that it is not  
> backwards compatible with existing rdiff-backup servers. This is  
> because an RPath object now contains a win_acls.ACL object, which  
> causes the remote end to throw when unpickling the object. I don't  
> know of an easy way around this. One way would be to serialize the  
> object to string, but I'm not sure I like that.

You have to serialize it to string to write it to the repository file  
anyway, no? I'm just guessing here -- I haven't read the patch yet.


looking forward to reading it and many thanks to both of you,
Andrew