Re: [PATCH v2] clear file privilege bits when mmap writing

[Jan Kara <jack@suse.cz>]

On Mon 07-12-15 16:40:14, Kees Cook wrote:
> On Mon, Dec 7, 2015 at 2:42 PM, Kees Cook <keescook@chromium.org> wrote:
> > On Thu, Dec 3, 2015 at 5:45 PM, yalin wang <yalin.wang2010@gmail.com> wrote:
> >>
> >>> On Dec 2, 2015, at 16:03, Kees Cook <keescook@chromium.org> wrote:
> >>>
> >>> Normally, when a user can modify a file that has setuid or setgid bits,
> >>> those bits are cleared when they are not the file owner or a member
> >>> of the group. This is enforced when using write and truncate but not
> >>> when writing to a shared mmap on the file. This could allow the file
> >>> writer to gain privileges by changing a binary without losing the
> >>> setuid/setgid/caps bits.
> >>>
> >>> Changing the bits requires holding inode->i_mutex, so it cannot be done
> >>> during the page fault (due to mmap_sem being held during the fault).
> >>> Instead, clear the bits if PROT_WRITE is being used at mmap time.
> >>>
> >>> Signed-off-by: Kees Cook <keescook@chromium.org>
> >>> Cc: stable@vger.kernel.org
> >>> —
> >>
> >> is this means mprotect() sys call also need add this check?
> >> mprotect() can change to PROT_WRITE, then it can write to a
> >> read only map again , also a secure hole here .
> >
> > Yes, good point. This needs to be added. I will send a new patch. Thanks!
> 
> This continues to look worse and worse.
> 
> So... to check this at mprotect time, I have to know it's MAP_SHARED,
> but that's in the vma_flags, which I can only see after holding
> mmap_sem.
> 
> The best I can think of now is to strip the bits at munmap time, since
> you can't execute an mmapped file until it closes.
> 
> Jan, thoughts on this?

Umm, so we actually refuse to execute a file while someone has it open for
writing (deny_write_access() in do_open_execat()). So dropping the suid /
sgid bits when closing file for writing could be plausible. Grabbing
i_mutex from __fput() context is safe (it gets called from task_work
context when returning to userspace).

That way we could actually remove the checks done for each write. To avoid
unexpected removal of suid/sgid bits when someone just opens & closes the
file, we could mark the file as needing suid/sgid treatment by a flag in
inode->i_flags when file gets written to or mmaped and then check for this
in __fput().

I've added Al Viro to CC just in case he is aware of some issues with
this...

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR