Re: [PATCH] sys_getppid oopses on debug kernel

Re: [PATCH] sys_getppid oopses on debug kernel

[Andi Kleen]

Kirill Korotaev <dev@sw.ru> writes:

[adding linux-arch]

> > Accessing freed memory is a bug, always, not just *only* when slab
> > debugging is on, right?  Doesn't this mean we could get junk, or that
> > the reader could potentially run off a bad pointer?
> no, read the comment in sys_getppid.
> It is a valid optimization. _safe_ and alowing to bypass taking the lock.
> BUT! This optimization relies on the fact that kernel memory (DMA + normal zone)
> is always mapped into virtual address space.
> Which is invalid for debug kernels only.

In x86 arch code we would use __get_user for this (and we do in a couple 
of places). But it wouldn't be portable because sometimes _user is 
in a different address space.

Maybe it would be time to make a similar facility (read/write_kernel_safe() or similar)
with error return available to generic code? 

It should be easy to implement - iirc near all architectures already
use the exception handling frame work and it is a simple extension 
of that. x86 could just define it to __put/get_user

-Andi

Re: [PATCH] sys_getppid oopses on debug kernel

[Andrew Morton]

On 09 Aug 2006 05:09:11 +0200
Andi Kleen <ak@suse.de> wrote:

> Kirill Korotaev <dev@sw.ru> writes:
> 
> [adding linux-arch]
> 
> > > Accessing freed memory is a bug, always, not just *only* when slab
> > > debugging is on, right?  Doesn't this mean we could get junk, or that
> > > the reader could potentially run off a bad pointer?
> > no, read the comment in sys_getppid.
> > It is a valid optimization. _safe_ and alowing to bypass taking the lock.
> > BUT! This optimization relies on the fact that kernel memory (DMA + normal zone)
> > is always mapped into virtual address space.
> > Which is invalid for debug kernels only.
> 
> In x86 arch code we would use __get_user for this (and we do in a couple 
> of places). But it wouldn't be portable because sometimes _user is 
> in a different address space.
> 
> Maybe it would be time to make a similar facility (read/write_kernel_safe() or similar)
> with error return available to generic code? 
> 
> It should be easy to implement - iirc near all architectures already
> use the exception handling frame work and it is a simple extension 
> of that. x86 could just define it to __put/get_user
> 

I just did something like that:

Similar to ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.18-rc3/2.6.18-rc3-mm2/broken-out/add-probe_kernel_address.patch

Although I'm not sure it's needed for this problem. A getppid() which does

asmlinkage long sys_getppid(void)
{
	int pid;

	read_lock(&tasklist_lock);
	pid = current->group_leader->real_parent->tgid;
	read_unlock(&tasklist_lock);

	return pid;
}

seems like a fine implementation to me ;)