bug#52206: Bug: rm -rf /*/*

[Bob Proulx]

Paul Eggert wrote:
> Robert Swinford wrote:
> > This seems like a bug: 
> > https://twitter.com/nixcraft/status/1465599844299411458
> 
> I don't see a coreutils bug there: rm operated as specified.

Agreed.  It's not an rm bug.  It's definitely unfortunate.  It's
similarly unfortunate to riding a bicycle into a lake.  But it isn't a
defect in the bicycle that it could not prevent someone from riding it
into a lake.

> > Interestingly, however, rm -rf // only does the following:
> 
> Yes, that's a special feature of GNU rm.

And apparently Bryan Cantrill reports that Solaris has the same
feature as GNU rm does for "rm -rf /" protection.

> > I believe illumos has already solved this problem in a POSIX compliant 
> > fashion
> 
> Not sure what you're talking about here. Could you be specific? Don't have
> time to watch videos.

I watched the cited video.  It features an interview with Bryan
Cantrill who very dynamically and entertainingly tells a story about a
scripted "rm -rf $1/$2" without checking if $1 and $2 were set or
unset resulting in "rm -rf /" being run by accident.  And therefore he
reports that Solaris implemented the prevention of running "rm -rf /".
This is said at time 1:27:00 in the video.  Which I note is the same
protection as GNU rm does.  So there isn't anything for GNU rm to
implement in order to match Solaris as it appears to be the same by
this report.

However $var1/$var2 expanding to / when those variables are not set is
a different case than /*/* expansion which has no variables and is
simply an error of usage.

> > BTW, zsh globbing doesn’t exhibit this behavior!  It seems it is only a 
> > problem in bash.
> 
> In that case, the bug (whatever it is) wouldn't be a coreutils bug.

I don't understand the comment that zsh doesn't expand the glob /*/*
and I tried it and verified that it does indeed expand that glob
sequence.

Bob