| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20200325112652.sx66bhad7cqdsatm@wittgenstein>
Date: Wed, 25 Mar 2020 12:26:52 +0100
From: Christian Brauner <christian.brauner@...ntu.com>
To: Adrian Reber <areber@...hat.com>
Cc: Andrei Vagin <avagin@...il.com>, Arnd Bergmann <arnd@...db.de>,
Eric Biederman <ebiederm@...ssion.com>,
Pavel Emelyanov <ovzxemul@...il.com>,
Oleg Nesterov <oleg@...hat.com>,
Dmitry Safonov <0x7f454c46@...il.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Mike Rapoport <rppt@...ux.ibm.com>,
Radostin Stoyanov <rstoyanov1@...il.com>,
Michael Kerrisk <mtk.manpages@...il.com>,
Cyrill Gorcunov <gorcunov@...nvz.org>,
Thomas Gleixner <tglx@...utronix.de>,
Aleksa Sarai <cyphar@...har.com>,
Linux API <linux-api@...r.kernel.org>
Subject: Re: clone3: allow creation of time namespace with offset
On Wed, Mar 25, 2020 at 08:58:36AM +0100, Adrian Reber wrote:
> On Tue, Mar 24, 2020 at 06:56:49PM +0100, Christian Brauner wrote:
> > On Tue, Mar 24, 2020 at 05:25:46PM +0100, Adrian Reber wrote:
> > > On Tue, Mar 24, 2020 at 05:09:45PM +0100, Christian Brauner wrote:
> > > > On Fri, Mar 20, 2020 at 11:33:55AM -0700, Andrei Vagin wrote:
> > > > > On Thu, Mar 19, 2020 at 11:29:55AM +0100, Christian Brauner wrote:
> > > > > > On Thu, Mar 19, 2020 at 09:16:43AM +0100, Arnd Bergmann wrote:
> > > > > > > On Thu, Mar 19, 2020 at 9:11 AM Adrian Reber <areber@...hat.com> wrote:
> > > > > > >
> > > > > > > > With Arnd's idea of only using nanoseconds, timens_offset would then
> > > > > > > > contain something like this:
> > > > > > > >
> > > > > > > > struct timens_offset {
> > > > > > > > __aligned_s64 monotonic_offset_ns;
> > > > > > > > __aligned_s64 boottime_offset_ns;
> > > > > > > > };
> > > > > > > >
> > > > > > > > I kind of prefer adding boottime and monotonic directly to struct clone_args
> > > > > > > >
> > > > > > > > __aligned_u64 tls;
> > > > > > > > __aligned_u64 set_tid;
> > > > > > > > __aligned_u64 set_tid_size;
> > > > > > > > + __aligned_s64 monotonic_offset_ns;
> > > > > > > > + __aligned_s64 boottime_offset_ns;
> > > > > > > > };
> > > > > > >
> > > > > > > I would also prefer the second approach using two 64-bit integers
> > > > > > > instead of a pointer, as it keeps the interface simpler to implement
> > > > > > > and simpler to interpret by other tools.
> > > > > >
> > > > > > Why I don't like has two reasons. There's the scenario where we have
> > > > > > added new extensions after the new boottime member and then we introduce
> > > > > > another offset. Then you'd be looking at:
> > > > > >
> > > > > > __aligned_u64 tls;
> > > > > > __aligned_u64 set_tid;
> > > > > > __aligned_u64 set_tid_size;
> > > > > > + __aligned_s64 monotonic_offset_ns;
> > > > > > + __aligned_s64 boottime_offset_ns;
> > > > > > __aligned_s64 something_1
> > > > > > __aligned_s64 anything_2
> > > > > > + __aligned_s64 sometime_offset_ns
> > > > > >
> > > > > > which bothers me just by looking at it. That's in addition to adding two
> > > > > > new members to the struct when most people will never set CLONE_NEWTIME.
> > > > > > We'll also likely have more features in the future that will want to
> > > > > > pass down more info than we want to directly expose in struct
> > > > > > clone_args, e.g. for a long time I have been thinking about adding a
> > > > > > struct for CLONE_NEWUSER that allows you to specify the id mappings you
> > > > > > want the new user namespace to get. We surely don't want to force all
> > > > > > new info into the uppermost struct. So I'm not convinced we should here.
> > > > >
> > > > > I think here we can start thinking about a netlink-like interface.
> > > >
> > > > I think netlink is just not a great model for an API and I would not
> > > > want us to go down that route.
> > > >
> > > > I kept thinking about this for a bit and I think that we will end up
> > > > growing more namespace-related functionality. So one thing that came to
> > > > my mind is the following layout:
> > > >
> > > > struct {
> > > > struct {
> > > > __s64 monotonic;
> > > > __s64 boot;
> > > > } time;
> > > > } namespaces;
> > > >
> > > > struct _clone_args {
> > > > __aligned_u64 flags;
> > > > __aligned_u64 pidfd;
> > > > __aligned_u64 child_tid;
> > > > __aligned_u64 parent_tid;
> > > > __aligned_u64 exit_signal;
> > > > __aligned_u64 stack;
> > > > __aligned_u64 stack_size;
> > > > __aligned_u64 tls;
> > > > __aligned_u64 set_tid;
> > > > __aligned_u64 set_tid_size;
> > > > __aligned_u64 namespaces;
> > > > __aligned_u64 namespaces_size;
> > > > };
> > > >
> > > > Then when we end up adding id mapping support for CLONE_NEWUSER we can
> > > > extend this with:
> > > >
> > > > struct {
> > > > struct {
> > > > __aligned_u64 monotonic;
> > > > __aligned_u64 boot;
> >
> > s/__aligned_u64/__s64/g
> >
> > Sorry, leftover from my first draft.
> >
> > > > } time;
> > > >
> > > > struct {
> > > > /* id mapping members */
> > > > } user;
> > > > } namespaces;
> > > >
> > > > Thoughts? Other ideas?
> > >
> > > Works for me.
> > >
> > > If we add the user namespace id mappings and then at some point a third
> > > element for the time namespace appears it would also start to be mixed.
> > > Just as you mentioned that a few mails ago.
> >
> > I think you misunderstand me or I'm misunderstanding you. That new time
> > namespace member would go into struct time {} so
> >
> > struct {
> > struct {
> > __s64 monotonic;
> > __s64 boot;
> > __s64 someothertime;
> > } time;
> >
> > struct {
> > /* id mapping members */
> > } user;
> > } namespaces;
>
> My question was about how does the kernel know how 'struct namespaces'
> is structured. How can an older kernel (which only is aware of two
> clocks) deal with a, like in this example, third clock. Will the size
> '__aligned_u64 namespaces_size' be used for versioning?
Yes, that would be the idea.
I don't want to give the impression that I think this is the best
solution. It's one solution that I think is feasible. But if we have
something better that is also future proof I'm happy to hear ideas.
Christian
Powered by blists - more mailing lists