Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

[Miroslav Benes]

Hi,

> > Driver developers will simply have to open code these protections. In
> > light of what I see on LTP / fuzzing, I suspect the use case will grow
> > and we'll have to revisit this in the future. But for now, sure, we can
> > just open code the required protections everywhere to not crash on module
> > removal.
> 
> LTP and fuzzing too do not remove modules.  So I do not understand the
> root problem here, that's just something that does not happen on a real
> system.

If I am not mistaken, the issue that Luis tries to solve here was indeed 
found by running LTP.

> On Sat, Apr 03, 2021 at 08:13:23AM +0200, Greg KH wrote:
> > On Fri, Apr 02, 2021 at 06:30:16PM +0000, Luis Chamberlain wrote:
> > > On Fri, Apr 02, 2021 at 09:54:12AM +0200, Greg KH wrote:
> > > > No, please no.  Module removal is a "best effort",
> > > 
> > > Not for live patching. I am not sure if I am missing any other valid
> > > use case?
> > 
> > live patching removes modules?  We have so many code paths that are
> > "best effort" when it comes to module unloading, trying to resolve this
> > one is a valiant try, but not realistic.
> 
> Miroslav, your input / help here would be valuable. I did the
> generalization work because you said it would be worthy for you too...

Yes, we have the option to revert and remove the existing live patch from 
the system. I am not sure how (if) it is used in practice.

At least at SUSE we do not support the option. But we are only one of the 
many downstream users. So yes, there is the option.

Miroslav

Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

[Josh Poimboeuf]

On Tue, Apr 06, 2021 at 02:00:19PM +0200, Miroslav Benes wrote:
> Hi,
> 
> > > Driver developers will simply have to open code these protections. In
> > > light of what I see on LTP / fuzzing, I suspect the use case will grow
> > > and we'll have to revisit this in the future. But for now, sure, we can
> > > just open code the required protections everywhere to not crash on module
> > > removal.
> > 
> > LTP and fuzzing too do not remove modules.  So I do not understand the
> > root problem here, that's just something that does not happen on a real
> > system.
> 
> If I am not mistaken, the issue that Luis tries to solve here was indeed 
> found by running LTP.
> 
> > On Sat, Apr 03, 2021 at 08:13:23AM +0200, Greg KH wrote:
> > > On Fri, Apr 02, 2021 at 06:30:16PM +0000, Luis Chamberlain wrote:
> > > > On Fri, Apr 02, 2021 at 09:54:12AM +0200, Greg KH wrote:
> > > > > No, please no.  Module removal is a "best effort",
> > > > 
> > > > Not for live patching. I am not sure if I am missing any other valid
> > > > use case?
> > > 
> > > live patching removes modules?  We have so many code paths that are
> > > "best effort" when it comes to module unloading, trying to resolve this
> > > one is a valiant try, but not realistic.
> > 
> > Miroslav, your input / help here would be valuable. I did the
> > generalization work because you said it would be worthy for you too...
> 
> Yes, we have the option to revert and remove the existing live patch from 
> the system. I am not sure how (if) it is used in practice.
> 
> At least at SUSE we do not support the option. But we are only one of the 
> many downstream users. So yes, there is the option.

Same for Red Hat.  Unloading livepatch modules seems to work fine, but
isn't officially supported.

That said, if rmmod is just considered a development aid, and we're
going to be ignoring bugs, we should make it official with a new
TAINT_RMMOD.

-- 
Josh

Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

[Peter Zijlstra]

On Tue, Apr 06, 2021 at 10:54:23AM -0500, Josh Poimboeuf wrote:

> Same for Red Hat.  Unloading livepatch modules seems to work fine, but
> isn't officially supported.
> 
> That said, if rmmod is just considered a development aid, and we're
> going to be ignoring bugs, we should make it official with a new
> TAINT_RMMOD.

Another option would be to have live-patch modules leak a module
reference by default, except when some debug sysctl is set or something.
Then only those LP modules loaded while the sysctl is set to 'YOLO' can
be unloaded.

Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

[Josh Poimboeuf]

On Wed, Apr 07, 2021 at 04:09:44PM +0200, Peter Zijlstra wrote:
> On Tue, Apr 06, 2021 at 10:54:23AM -0500, Josh Poimboeuf wrote:
> 
> > Same for Red Hat.  Unloading livepatch modules seems to work fine, but
> > isn't officially supported.
> > 
> > That said, if rmmod is just considered a development aid, and we're
> > going to be ignoring bugs, we should make it official with a new
> > TAINT_RMMOD.
> 
> Another option would be to have live-patch modules leak a module
> reference by default, except when some debug sysctl is set or something.
> Then only those LP modules loaded while the sysctl is set to 'YOLO' can
> be unloaded.

The issue is broader than just live patching.

My suggestion was that if we aren't going to fix bugs in kernel module
unloading, then unloading modules shouldn't be supported, and should
taint the kernel.

-- 
Josh

Re: [PATCH 1/2] zram: fix crashes due to use of cpu hotplug multistate

[Peter Zijlstra]

On Wed, Apr 07, 2021 at 10:30:31AM -0500, Josh Poimboeuf wrote:
> On Wed, Apr 07, 2021 at 04:09:44PM +0200, Peter Zijlstra wrote:
> > On Tue, Apr 06, 2021 at 10:54:23AM -0500, Josh Poimboeuf wrote:
> > 
> > > Same for Red Hat.  Unloading livepatch modules seems to work fine, but
> > > isn't officially supported.
> > > 
> > > That said, if rmmod is just considered a development aid, and we're
> > > going to be ignoring bugs, we should make it official with a new
> > > TAINT_RMMOD.
> > 
> > Another option would be to have live-patch modules leak a module
> > reference by default, except when some debug sysctl is set or something.
> > Then only those LP modules loaded while the sysctl is set to 'YOLO' can
> > be unloaded.
> 
> The issue is broader than just live patching.
> 
> My suggestion was that if we aren't going to fix bugs in kernel module
> unloading, then unloading modules shouldn't be supported, and should
> taint the kernel.

Hold up, what? However much I dislike modules (and that is lots), if you
don't want to support rmmod, you have to leak a reference to self in
init. Barring that you get to fix any and all unload bugs.