On Wed, May 16, 2018 at 9:09 PM Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> > In theory, we could instead allow people to set the values of
> > mm->{arg,env}_{start,end}, which is where /proc/<pid>/cmdline actually
> > finds the argument array.
> > HOWEVER, there are serious issues with that too, and right now we only
> > allow it for checkpointing and PR_SET_MM_MAP prctl, which does a lot
more
> > than that.
> Out of curiosity, what are the issues?
At least one issue is that we technically keep the argument space and the
environment space separate, but the *code* actually knows that they are
always adjacent.
So while we have what looks to be independent argument and environment
pointers, in actual fact we have arg_end == env_start, and we have some
code that "knows" that.
It would probably not be hard to fix up, but I worry about the cases I
don't even know about.
That said, getting it wrong is probably not the end of the world either, at
worst we'd show an odd command line through /proc. It's not like the kernel
really uses those fields for anything else.
So I guess it wouldn't be too nasty to just allow people to set the argv
start/len manually. We actually have the interfaces for it already, but
they are restricted to root-only (CAP_SYS_RESOURCE) and rather oddly laid
out. The existing interfaces are all built for checkpoint/restart, not for
generic use (see PR_SET_MM).
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
