begin  quoting Gregory K. Ruiz-Ade as of Tue, May 13, 2008 at 11:11:38AM -0700:
> On May 12, 2008, at 5:51 PM, Andrew Lentvorski wrote:
> 
> >3) Root to install.  If software requires root for no good reason,  
> >it goes into the trash.
> 
> The only things that _need_ to be root to be installed tend to fall  
> into the following categories, in my experience:
> 
> ? The install process _really_ wants to chown/chgrp/chmod everything  
> it's installing to some value that may or may not be necessary.  For  
> some things, like network services which you are trying to keep  
> relatively self-contained for security reasons, can legitimately say  
> they require this, and generally only root can create a new "service"  
> user/group and chown the installed files.

This should be a warning, not an error worth aborting over. Tell me
"You should not do this:" and give me a list of steps. Or even a
list of commands to run when I get around to it.

> This seems marginally okay to me, and I get supremely annoyed when it  
> doesn't really need to do that.  Things I'll excuse for that behavior  
> are things like Postfix, which try very hard to keep to themselves  
> once installed and running.  In theory, you can install postfix as  
> your own user account if you really want to, and it should work just  
> fine (aside from being able to open ports 25, 465 and 587 without root  
> capability).  It's a grey area for me.

I believe this was the reason that OpenOffice used to use to justify
requiring administrator access.

> ? You're installing something that needs to start as root in order to  
> claim a low port, but then drops privileges to a non-root user.
> 
> In today's computing environment, This just seems outrageously stupid  
> to me.  It's not like being root is some magical, bureaucratically- 
> approved status.  Any idiot can have root on a machine thanks to  
> Linux, *BSD, Mac OS X, etc.  There's just nothing special about low  
> ports anymore, if you ask me, so it seems kind of silly to still  
> require that a process be owned by root in order to snag a listening  
> port < 1024.  This should be fixable with relative ease (Linux, *BSD),  
> but I've never looked into it.

I think Peter da Silva's correct:

http://groups.google.com/group/alt.folklore.computers/msg/aac01bd0f8e67f0e

EVERYTHING in the system should be controlled with the filesystem.

> ? You're installing something in a part of the filesystem hierarchy  
> where you, as a normal user, do not have sufficient write access to  
> copy the necessary files.

This is really only a problem when you can't specify WHERE to install
the program.  Hard-coded installation paths are just wrong.

> I'm not sure I see a way around this one, but if an installer assumes  
> that this is the case without testing first, I get really annoyed.   

Yes.

> Better to just fail with a "You can't write to /path/to/foo, so I  
> can't install BarMatic for you there" message, IMHO.

<voice character="Brain">Yes!</voice>

> ? You're installing third-party kernel extensions
> 
> This one really seems like the only one that should legitimately  
> demand to be root to do.  It's something that affects the core of the  
> system, with great potential for screwing up badly.

But there's no need for the *program* to check such things.

-- 
I need to play around with fakeroot.
Stewart Stremler


-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to