Re: [PATCH v8] fs: clear file privilege bits when mmap writing

[Konstantin Khlebnikov]

On Wed, Jan 13, 2016 at 7:09 PM, Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> On Wed, Jan 13, 2016 at 1:03 AM, Jan Kara <jack@xxxxxxx> wrote:
>> On Tue 12-01-16 11:09:04, Kees Cook 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 open time,
>>> or added at mprotect time.
>>>
>>> Since we can't do the check in the right place inside mmap (due to
>>> holding mmap_sem), we have to do it before holding mmap_sem, which
>>> means duplicating some checks, which have to be available to the non-MMU
>>> builds too.
>>>
>>> When walking VMAs during mprotect, we need to drop mmap_sem (while
>>> holding a file reference) and restart the walk after clearing privileges.
>>
>> ...
>>
>>> @@ -375,6 +376,7 @@ SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
>>>
>>>       vm_flags = calc_vm_prot_bits(prot);
>>>
>>> +restart:
>>>       down_write(&current->mm->mmap_sem);
>>>
>>>       vma = find_vma(current->mm, start);
>>> @@ -416,6 +418,28 @@ SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
>>>                       goto out;
>>>               }
>>>
>>> +             /*
>>> +              * If we're adding write permissions to a shared file,
>>> +              * we must clear privileges (like done at mmap time),
>>> +              * but we have to juggle the locks to avoid holding
>>> +              * mmap_sem while holding i_mutex.
>>> +              */
>>> +             if ((vma->vm_flags & VM_SHARED) && vma->vm_file &&
>>> +                 (newflags & VM_WRITE) && !(vma->vm_flags & VM_WRITE) &&
>>> +                 !IS_NOSEC(file_inode(vma->vm_file))) {
>>
>> This code assumes that IS_NOSEC gets set for inode once file_remove_privs()
>> is called. However that is not true for two reasons:
>>
>> 1) When you are root, SUID bit doesn't get cleared and thus you cannot set
>> IS_NOSEC.
>>
>> 2) Some filesystems do not have MS_NOSEC set and for those IS_NOSEC is
>> never true.
>>
>> So in these cases you'll loop forever.
>
> UUuugh.
>
>>
>> You can check SUID bits without i_mutex so that could be done without
>> dropping mmap_sem but you cannot easily call security_inode_need_killpriv()
>> without i_mutex as that checks extended attributes (IMA) and that needs
>> i_mutex to be held to avoid races with someone else changing the attributes
>> under you.
>
> Yeah, that's why I changed this from Konstantin's original suggestion.
>
>> Honestly, I don't see a way of implementing this in mprotect() which would
>> be reasonably elegant.
>
> Konstantin, any thoughts here?

Getxattr works fine without i_mutex: sys_getxattr/vfs_getxattr doesn't lock it.
If somebody changes xattrs under us we'll end up in race anyway.
But this still safe: setxattrs are sychronized.

>
> -Kees
>
> --
> Kees Cook
> Chrome OS & Brillo Security