[scponly] PATCH: support for multiple users with the same uid

[Steve Kehlet]

Hi Kaleb,

Sure thing, it's for a Linux-based appliance I'm creating.  There  
will be no shell access.  Users will scp/sftp stuff up into  
individual jailed areas, then a daemon process will read/write those  
files, possibly move them around (say from a user's "inbox" subfolder  
to "processed"), etc.  The jails keep users from mucking around with  
each other's stuff, and using the same uid for all the users and the  
daemon process saves a ton of headaches with permissions, umasks, etc.

Before this, I did take a long look at doing it "the right way", with  
separate uids, managing groups and appropriate group perms on all  
directories, then building a custom openssh with various available  
patches for chroot, restrictions on chmod so the users couldn't break  
any of the carefully set up permissions, running with a custom umask  
to enforce correct perms on new files/directories, adjusting the  
daemon's umask (possibly requiring modifying source code/startup  
files, developed by another team), etc, but after a while it seemed  
rather fragile and in fact pointless since there would never be shell  
access anyway.  Add that it's an appliance running outside of my  
control, and I wouldn't be able to log in and fix something if it  
broke.  Note: it has been deemed acceptable that the same (fairly big  
and expensive, java-based) daemon process handle all users' data.

That's the gist of it.  scp/sftp isn't the only upload mechanism, but  
it's a required one, and the only one with file browsing abilities,  
which I need jails for.

> so I'm curious if it's something I would consider a reasonable use  
> case.

I'd understand if you'd consider my case unique, and I can maintain  
my own patch no problem.  But I think my patch addresses a not-so- 
unreasonable, albeit uncommon, case where scponly can grant access  
outside a user's jail, in a way that only makes it safer.

Thanks!

Steve





On Jun 4, 2007, at 10:08 PM, Kaleb Pederson wrote:

> Steve,
>
> The patch itself seems fine, but I would like to know more about  
> why you need
> to have multiple users with the same uid?  Obviously, this prevents  
> the
> operating system from enforcing permissions based on uid, which  
> doesn't seem
> like a good idea to me -- so I'm curious if it's something I would  
> consider a
> reasonable use case.
>
> Thanks.
>
> --Kaleb
>
> On Monday 04 June 2007, Steve Kehlet wrote:
>> I acknowledge sharing uids in general is not considered a best
>> practice, but it's a nice solution for what I'm doing (I'll spare you
>> the details).  As far as I can tell, the above change seems safe,
>> provides arguably more correct/expected behavior, and may be useful
>> for others too.  But, I submit it here for review, so please post if
>> you see something I haven't or if you otherwise disagree :-).
>> Hopefully you'll like the patch and merge it in.  Many thanks, I
>> think scponly is a very handy package.