Re: BUG_ON(nd->inode != parent->d_inode);

[ebiederm@xmission.com (Eric W. Biederman)]

Dave Jones <davej@redhat.com> writes:

> On Fri, Mar 08, 2013 at 09:56:31PM -0500, Dave Jones wrote:
>  > On Fri, Mar 08, 2013 at 09:26:23PM -0500, Dave Jones wrote:
>  >  > On Fri, Mar 08, 2013 at 06:08:52PM -0800, Linus Torvalds wrote:
>  >  >  > On Fri, Mar 8, 2013 at 6:03 PM, Dave Jones <davej@redhat.com> wrote:
>  >  >  > >
>  >  >  > > existing pathname + 'a' = fine.
>  >  >  > >
>  >  >  > > existing pathname + '/' + 'a' = boom.
>  >  >  > 
>  >  >  > Good.
>  >  >  > 
>  >  >  > > Looks like if I do this..
>  >  >  > >
>  >  >  > >        if (isdigit(newpath[len]) != 0) {
>  >  >  > >                 newpath[len] = '/';
>  >  >  > >                newpath[len+1] = 'A';
>  >  >  > >                newpath[len+2] = 0;
>  >  >  > >
>  >  >  > > no bug.
>  >  >  > 
>  >  >  > Well, but that will never trigger. newpath[len] will always be NUL, so
>  >  >  > you just disabled things entirely. Use "len-1".
>  >  >  > 
>  >  >  > So I don't think that did what you meant it to do.
>  >  > 
>  >  > Fixed that up, and even double checked my sanity by printing stuff out.
>  >  > 
>  >  > Confirmed that it's something that doesn't end in a number.
>  >  
>  > I've got a hunch that it's /proc/$$/ns/uts.
>  > 
>  > After 3-4 runs, that's the only common file in the last few that got mangled.
>  > 
>  > I'll do some more tests, but this might be the one.
>  
> confirmed.  A simple 
>
> mkdir /proc/self/ns/uts/A
>
> will trigger it.

This is a magic symlink similar to the other magic symlinks in proc so I
don't know if the test is wrong or my code is doing something to clever.

But I can reproduce it so I will look at it and see if I can make sense
of what is going on.

Eric