Re: 2.6.22-rc6(mm1) Unable to handle kernel NULL pointer dereference - git-bisect result

[Linus Torvalds <torvalds@linux-foundation.org>]

On Sun, 8 Jul 2007, Ulrich Drepper wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Linus Torvalds wrote:
> > notify_change() does *not* do permission checks for 
> > ATTR_CTIME/MTIME/ATIME.
> 
> Then I don't understand
> 
>         /* Check for setting the inode time. */
>         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
>                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
>                         goto error;
>         }
> 
> in inode_change_ok.  This seems to me exactly like the check needed.

No it's not. It really allows only the _owner_ to change MTIME/ATIME.

And the thing is, other people _can_ change MTIME/ATIME, but only to the 
current time!

This is why ATTR_MTIME_SET ("set it to a _specific_ time") is so different 
from ATTR_MTIME ("set it to the _current_ time").

The former is for times() with times != NULL, and is strictly limited to 
only the owner (or root).

The latter is a "anybody who can change the inode can implicitly also 
change the inode timestamps, but only to the _current_ time".

In other words, the ATTR_A/MTIME_SET thing allows people to _lie_ about 
the time, and only the owner is allowed to do that. Which is why they are 
very different.

		Linus