Re: [PATCH v1 2/4] pid: add pidctl()

[Daniel Colascione]

On Tue, Mar 26, 2019 at 9:46 AM Christian Brauner <christian@xxxxxxxxxx> wrote:
>
> On Tue, Mar 26, 2019 at 09:42:59AM -0700, Andy Lutomirski wrote:
> > On Tue, Mar 26, 2019 at 9:34 AM Christian Brauner <christian@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Mar 26, 2019 at 05:31:42PM +0100, Christian Brauner wrote:
> > > > On Tue, Mar 26, 2019 at 05:23:37PM +0100, Christian Brauner wrote:
> > > > > On Tue, Mar 26, 2019 at 09:17:07AM -0700, Daniel Colascione wrote:
> > > > > > Thanks for the patch.
> > > > > >
> > > > > > On Tue, Mar 26, 2019 at 8:55 AM Christian Brauner <christian@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > The pidctl() syscalls builds on, extends, and improves translate_pid() [4].
> > > > > > > I quote Konstantins original patchset first that has already been acked and
> > > > > > > picked up by Eric before and whose functionality is preserved in this
> > > > > > > syscall:
> > > > > >
> > > > > > We still haven't had a much-needed conversation about splitting this
> > > > > > system call into smaller logical operations. It's important that we
> > > > > > address this point before this patch is merged and becomes permanent
> > > > > > kernel ABI.
> > > > >
> > > > > I don't particularly mind splitting this into an additional syscall like
> > > > > e.g.  pidfd_open() but then we have - and yes, I know you'll say
> > > > > syscalls are cheap - translate_pid(), and pidfd_open(). What I like
> > > > > about this rn is that it connects both apis in a single syscall
> > > > > and allows pidfd retrieval across pid namespaces. So I guess we'll see
> > > > > what other people think.
> > > >
> > > > There's something to be said for
> > > >
> > > > pidfd_open(pid_t pid, int pidfd, unsigned int flags);
> > > >
> > > > /* get pidfd */
> > > > int pidfd = pidfd_open(1234, -1, 0);
> > > >
> > > > /* convert to procfd */
> > > > int procfd = pidfd_open(-1, 4, 0);
> > > >
> > > > /* convert to pidfd */
> > > > int pidfd = pidfd_open(4, -1, 0);
> > >
> > > probably rather:
> > >
> > > int pidfd = pidfd_open(-1, 4, PIDFD_TO_PROCFD);
> >
> > Do you mean:
> >
> > int procrootfd = open("/proc", O_DIRECTORY | O_RDONLY);
> > int procfd = pidfd_open(procrootfd, pidfd, PIDFD_TO_PROCFD);
> >
> > or do you have some other solution in mind to avoid the security problem?
>
> Yes, we need the proc root obviously. I just jotted this down.
>
> We probably would need where one of the fds can refer to the proc root.
>
> pidfd_open(pid_t, int fd, int fd, 0)

Indeed. This is precisely the pidfd-procfd translation API I proposed
in the last paragraph of [1].

[1] https://lore.kernel.org/lkml/CAKOZuetCFgu0B53+mGmQ3+539MPT_tiu-PACx2ATvihHrrmUKg@xxxxxxxxxxxxxx/