On 05/05/2017 22:28, Eric W. Biederman wrote:
> Al Viro <viro@ZenIV.linux.org.uk> writes:
> 
>> On Thu, May 04, 2017 at 08:46:49PM -0700, Linus Torvalds wrote:
>>> On Thu, May 4, 2017 at 7:47 PM, Jann Horn <jannh@google.com> wrote:
>>>>
>>>> Thread 1 starts an AT_BENEATH path walk using an O_PATH fd
>>>> pointing to /srv/www/example.org/foo; the path given to the syscall is
>>>> "bar/../../../../etc/passwd". The path walk enters the "bar" directory.
>>>> Thread 2 moves /srv/www/example.org/foo/bar to
>>>> /srv/www/example.org/bar.
>>>> Thread 1 processes the rest of the path ("../../../../etc/passwd"), never
>>>> hitting /srv/www/example.org/foo in the process.
>>>>
>>>> I'm not really familiar with the VFS internals, but from a coarse look
>>>> at the patch, it seems like it wouldn't block this?
>>>
>>> I think you're right.
>>>
>>> I guess it would be safe for the RCU case due to the sequence number
>>> check, but not the non-RCU case.
>>
>> 	Yes and no...  FWIW, to exclude that it would suffice to have
>> mount --rbind /src/www/example.org/foo /srv/www/example.org/foo done first.
>> Then this kind of race will end up with -ENOENT due to path_connected()
>> logics in follow_dotdot_rcu()/follow_dotdot().  I'm not sure about the
>> intended applications, though - is that thing supposed to be used along with
>> some horror like seccomp, or...?
> 
> As I recall the general idea is that if you have an application like a
> tftp server or a web server that gets a path from a possibly dubious
> source.  Instead of implementing an error prone validation logic in
> userspace you can use AT_BENEATH and be certain the path resolution
> stays in bounds.
> 
> As you can do stronger things as root this seems mostly targeted at
> non-root applications.
> 
> I seem to recall part of the idea was to sometimes pair this to seccomp
> to be certain your application can't escape a sandbox.  That plays to
> seccomp limitations that it can inspect flags as they reside in
> registers but seccomp can't follow pointers.

Here is the code and tests from David Drysdale:
https://github.com/google/capsicum-linux/commits/openat-v2
...and the latest patch: https://lkml.org/lkml/2015/3/9/407
The O_BENEATH flag have also been discussed for FreeBSD to support Capsicum.

> 
> Which all suggests that we would want something similar to is_subdir
> when AT_BENEATH is specified that we check every time we follow ..
> that would verify that on the same filesystem we stay below and
> that we also stay on a mount that is below.  mount --move has
> all of the same challenges for enforcing you stay within bounds
> as rename does.

FYI, I'm working on a new LSM [1] to work around the limitations of
seccomp-bpf, especially the pointer checks. The idea is to enable some
filtering as seccomp-bpf can do but instead of checking at the syscall
level, Landlock take advantage of LSM hooks. I had a first PoC of an
eBPF function and map type to check if a file was beneath another [2]. I
plan to create a new one that record a "snapshot" of the current mount
tree into an eBPF map to be able to check if a file is beneath or a
parent of another one.

[1] https://lkml.kernel.org/r/20170328234650.19695-1-mic@digikod.net
[2] https://lkml.kernel.org/r/20161026065654.19166-9-mic@digikod.net