Re: [Lse-tech] [RFC] fork historic module

Re: [Lse-tech] [RFC] fork historic module

[Manfred Spraul]

Andi wrote:

>> +/* IOCTL numbers */
>> +/* If you add a new IOCTL number don't forget to update FH_MAXNR */
>> +#define FH_MAGIC	0x35
>> +#define FH_REGISTER	_IO(FH_MAGIC,0)
>> +#define FH_UNREGISTER	_IO(FH_MAGIC,1)
>
>Is this really unique? 32bit emulation currently needs unique ioctl numbers.
>  
>
Are there plans to fix that? Perhaps move the emulation callback into 
struct file?

--
    Manfred

Re: [Lse-tech] [RFC] fork historic module

[Andi Kleen]

On Fri, Dec 17, 2004 at 05:52:38PM +0100, Manfred Spraul wrote:
> Andi wrote:
> 
> >>+/* IOCTL numbers */
> >>+/* If you add a new IOCTL number don't forget to update FH_MAXNR */
> >>+#define FH_MAGIC	0x35
> >>+#define FH_REGISTER	_IO(FH_MAGIC,0)
> >>+#define FH_UNREGISTER	_IO(FH_MAGIC,1)
> >
> >Is this really unique? 32bit emulation currently needs unique ioctl 
> >numbers.
> > 
> >
> Are there plans to fix that? Perhaps move the emulation callback into 
> struct file?

Yes. See the thread on l-k i started two days ago.

Actually the main reason is that the current conversion setup has a 
unfixable module unload race.

-Andi

Re: [Lse-tech] [RFC] fork historic module

[Guillaume Thouvenin]

On Fri, 2004-12-17 at 16:11 +0100, Andi Kleen wrote:
> > +/* IOCTL numbers */
> > +/* If you add a new IOCTL number don't forget to update FH_MAXNR */
> > +#define FH_MAGIC	0x35
> > +#define FH_REGISTER	_IO(FH_MAGIC,0)
> > +#define FH_UNREGISTER	_IO(FH_MAGIC,1)
> 
> Is this really unique? 32bit emulation currently needs unique ioctl numbers.

I read the Documentation/ioctl-number.txt file and 0x35 is not mentioned
in this file. But I made a grep on 'MAGIC' and '35' in the Linux source
tree and you're right, this number is already used by POR_MAGIC_2. 

Is a grep in the Linux source tree is enough to know if the value is
unique?

Thanks,
Guillaume  

Re: [Lse-tech] [RFC] fork historic module

[Erich Focht]

On Friday 17 December 2004 15:58, Guillaume Thouvenin wrote:
>   Here is the fork historic module. This module keeps the historic about
> the creation of processes and it gives this information to a user space
> application if one is registered. Currently, only one application can
> use it. If needed it could be extended to several applications.

This sounds very useful for a lot of interesting stuff. Though in the
current form maybe the name is not appropriate: the module doesn't
keep track of the fork history, this would be done in user-space. The
module is more like a relay for fork monitoring.

If you plan to extend this to multiple registered user space processes
it would be great to have per process reporting policies, for example:
 - all forks,
 - only forks of descendants (children, etc).

Erich

Re: [Lse-tech] [RFC] fork historic module

[Andi Kleen]

> +/* IOCTL numbers */
> +/* If you add a new IOCTL number don't forget to update FH_MAXNR */
> +#define FH_MAGIC	0x35
> +#define FH_REGISTER	_IO(FH_MAGIC,0)
> +#define FH_UNREGISTER	_IO(FH_MAGIC,1)

Is this really unique? 32bit emulation currently needs unique ioctl numbers.

-Andi