Re: [PATCH v3] capabilities.7, prctl.2: Document ambient capabilities

On Fri, Dec 4, 2015 at 7:08 AM, Michael Kerrisk (man-pages)
[off-list ref] wrote:

Hi Andy,

I have applied your patch (below). Thanks for writing it.
But I have a question or two and a request.

===

In the capabilities(7) page tehre is the longstanding text:

       An  application  can use the following call to lock itself, and
       all of its descendants, into an environment where the only  way
       of  gaining capabilities is by executing a program with associ‐
       ated file capabilities:

           prctl(PR_SET_SECUREBITS,
                   SECBIT_KEEP_CAPS_LOCKED |
                   SECBIT_NO_SETUID_FIXUP |
                   SECBIT_NO_SETUID_FIXUP_LOCKED |
                   SECBIT_NOROOT |
                   SECBIT_NOROOT_LOCKED);

As far as I can estimate, no changes are needed here to include
SECBIT_NO_CAP_AMBIENT_RAISE and SECBIT_NO_CAP_AMBIENT_RAISE_LOCKED
in the above prctl() call, but could you confirm please?

Correct.  I'll probably write up a patch to suggest that doing this is
a poor idea on a conventional distro, though, and I'll explain why.  I
suppose than deleting this would be an option, too.

===

In the message for kernel commit
58319057b7847667f0c9585b9de0e8932b0fdb08
you included this text:

[[
Because capability inheritance is so
broken, setting KEEPCAPS, using setresuid to switch to nonroot uids, and
then calling execve effectively drops capabilities.  Therefore,
setresuid from root to nonroot conditionally clears pA unless
SECBIT_NO_SETUID_FIXUP is set.  Processes that don't like this can
re-add bits to pA afterwards.
]]

I'm struggling to understand the significance of this text,
especially as your man-pages patch makes no mention of this point.

The thing is, that text ("Therefore...") implies that there's
something special going on beyond the rules already documented
elsewhere. I mean, according to the rules aly documented elsewhere
in the page:

Whoops, I forgot to add that to the manpage.

(1) If a process with UIDs of 0 sets all its UIDs
nonzero, then, the permitted and effective sets are cleared
(that's the classical behavior), and because the permitted
set is cleared, then so is the ambient set.

(2) And if we set SECBIT_NO_SETUID_FIXUP then
a UID 0 ==> nonzero transition doesn't clear permitted and
effective sets, and then of course the ambient set is not
cleared.

So, what additional point were you meaning to convey in
the commit message? (Maybe it was just cruft in the commit
message, but if not, can you explain precisely the arguments
for setresuid() that are supposed to generate the special
behavior described by the above text.)

It's case (1b), which is like (1) but with KEEPCAPS set.  The
permitted set doesn't get cleared, but the ambient set is still
cleared.

I'll write a manpage patch.

--Andy

===

I did quite a bit of tweaking of the text that you added for
the capabilities page. Could you please check the following
to make sure I added no errors:

       Ambient (since Linux 4.3):
              This  is a set of capabilities that are preserved across
              an execve(2) of a program that is not  privileged.   The
              ambient capability set obeys the invariant that no capa‐
              bility can ever be ambient if it is not  both  permitted
              and inheritable.

              The  ambient  capability  set  can  be directly modified
              using prctl(2).  Ambient capabilities are  automatically
              lowered  if  either  of  the  corresponding permitted or
              inheritable capabilities is lowered.

              Executing a program that changes UID or GID due  to  the
              set-user-ID  or set-group-ID bits or executing a program
              that has any file capabilities set will clear the  ambi‐
              ent  set.  Ambient capabilities are added to the permit‐
              ted set and assigned to the effective set when execve(2)
              is called.

Looks good (other than the preexisting gotcha above).

--Andy