Re: [PATCH] clone: fix CLONE_PIDFD support

From: Christian Brauner <christian@brauner.io>
To: "Dmitry V. Levin" <ldv@altlinux.org>
Cc: Anatoly Pugachev <matorola@gmail.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] clone: fix CLONE_PIDFD support
Date: Sun, 14 Jul 2019 16:23:05 +0200	[thread overview]
Message-ID: <20190714142304.3uihy4vepmxgdqha@brauner.io> (raw)
In-Reply-To: <20190714141007.GA9131@altlinux.org>

On Sun, Jul 14, 2019 at 05:10:08PM +0300, Dmitry V. Levin wrote:
> On Sun, Jul 14, 2019 at 02:17:25PM +0200, Christian Brauner wrote:
> > On Sun, Jul 14, 2019 at 03:02:06PM +0300, Dmitry V. Levin wrote:
> > > The introduction of clone3 syscall accidentally broke CLONE_PIDFD
> > > support in traditional clone syscall on compat x86 and those
> > > architectures that use do_fork to implement clone syscall.
> > > 
> > > This bug was found by strace test suite.
> > > 
> > > Link: https://strace.io/logs/strace/2019-07-12
> > > Fixes: 7f192e3cd316 ("fork: add clone3")
> > > Bisected-and-tested-by: Anatoly Pugachev <matorola@gmail.com>
> > > Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
> > 
> > Good catch! Thank you Dmitry.
> > 
> > One change request below.
> > 
> > > ---
> > >  arch/x86/ia32/sys_ia32.c | 1 +
> > >  kernel/fork.c            | 1 +
> > >  2 files changed, 2 insertions(+)
> > > 
> > > diff --git a/arch/x86/ia32/sys_ia32.c b/arch/x86/ia32/sys_ia32.c
> > > index 64a6c952091e..98754baf411a 100644
> > > --- a/arch/x86/ia32/sys_ia32.c
> > > +++ b/arch/x86/ia32/sys_ia32.c
> > > @@ -239,6 +239,7 @@ COMPAT_SYSCALL_DEFINE5(x86_clone, unsigned long, clone_flags,
> > >  {
> > >  	struct kernel_clone_args args = {
> > >  		.flags		= (clone_flags & ~CSIGNAL),
> > > +		.pidfd		= parent_tidptr,
> > >  		.child_tid	= child_tidptr,
> > >  		.parent_tid	= parent_tidptr,
> > >  		.exit_signal	= (clone_flags & CSIGNAL),
> > > diff --git a/kernel/fork.c b/kernel/fork.c
> > > index 8f3e2d97d771..2c3cbad807b6 100644
> > > --- a/kernel/fork.c
> > > +++ b/kernel/fork.c
> > > @@ -2417,6 +2417,7 @@ long do_fork(unsigned long clone_flags,
> > >  {
> > >  	struct kernel_clone_args args = {
> > >  		.flags		= (clone_flags & ~CSIGNAL),
> > > +		.pidfd		= parent_tidptr,
> > >  		.child_tid	= child_tidptr,
> > >  		.parent_tid	= parent_tidptr,
> > >  		.exit_signal	= (clone_flags & CSIGNAL),
> > > -- 
> > 
> > Both of these legacy clone helpers need to make CLONE_PIDFD and
> > CLONE_PARENT_SETTID incompatible, i.e. could you please add a helper to
> > kernel/fork.c:
> > 
> > bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
> > {
> > 	/* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
> > 	if ((kargs->flags & CLONE_PIDFD) && (kargs->flags & CLONE_PARENT_SETTID))
> > 		return false;
> > }
> > 
> > and export it and use it in ia32 too?
> 
> copy_process already performs the check, isn't this enough?

No it doesn't anymore. clone3() allows CLONE_PIDFD + CLONE_PARENT_SETTID
since struct clone_args has a dedicated pidfd return argument.

> Also, the check in sys_clone looks redundant and I was going to suggest
> its removal.

See above.