On Tue, Jul 21, 2009 at 07:19:44PM -0700, U.V. Ravindra wrote:

> The daemon can be run by hand, but it would be pointless.
> Running faked is only useful if there's a fakeroot at the
> other end telling it what to fake.

Let's not put Private executables in /usr/bin.

> >>Imported Interfaces Classification  Comment
> >>=================== ==============  ===========================
> >>SUNWlibms           Committed       Math&  Microtasking Libraries (Usr)
> >
> >What math functions does this use?
> 
> None, really.

So why have you declared these libraries as imported interfaces?

> >>Project Private
> >>===============
> >>/usr/lib/libfakeroot.so.0   Library
> >>/usr/lib/libfakeroot.so             Symbolic link
> >>/usr/lib/64/libfakeroot.so.0        library (64bit)
> >>/usr/lib/64/libfakeroot.so  Symbolic link (64bit)
> >
> >If these are project private, why have compilation symlinks?
> 
> fakeroot ld_preloads libfakeroot.so to achieve its desired end.
> Presumably this is for future "expansion"; should there be another
> version of the library in a future release, say, libfakeroot.so.1,
> fakeroot can continue to ld_preload libfakeroot.so as it does now.

That only makes sense if there's more than one thing on the system that
would use the library.  Which there can't be, because you're marking the
libraries Project Private.  If only fakeroot will ever use this library,
then it can make all the changes it likes in it, regardless of
compatibility, since the command and the library ship as a unit.  But if it
decided to move to .so.1 for whatever reason, then the command would just
change to reflect that.

So there's little point in doing versioning.  You should probably deliver
the .so.0 versions, and remove the .so symlinks, though I'll bet it's
easier to do it the other way around (i.e., you won't have to patch the
script).

> >Has any thought been given to having fake elevated privileges, or fake
> >role assumption?
> 
> Apologies: I am a bit unsure as to the meaning of the question.

Right now, it simulates the raising of all privileges.  I'm asking about
whether it can simulate the raising of some subset of all privileges --
say, file_dac_write, but not sys_mount.  That could help you debug a
privilege-aware program.  Similarly for assuming a role.

I'm not suggesting you should add this functionality if it's not currently
there, but as part of the whole "own it like a puppy" mantra, if these are
desirable things, it would be worth exploring the implementation with the
upstream folks.

Danek

Reply via email to