Leak in rdma_get_cm_event on pthread_cancel?

Leak in rdma_get_cm_event on pthread_cancel?

[Flavio Baronti]

When I want to stop the thread reading rdma events, I kill it with a pthread_cancel(), since it is blocked in a 
rdma_get_cm_event(). This causes a leak of the evt variable in the rdma_get_cm_event() function. Not a big deal, of 
course, but a couple of pthread_cleanup_push(free, evt) and pthread_cleanup_pop() should solve it.

Flavio
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: Leak in rdma_get_cm_event on pthread_cancel?

[Hefty, Sean]

> When I want to stop the thread reading rdma events, I kill it with a
> pthread_cancel(), since it is blocked in a
> rdma_get_cm_event(). This causes a leak of the evt variable in the
> rdma_get_cm_event() function. Not a big deal, of
> course, but a couple of pthread_cleanup_push(free, evt) and
> pthread_cleanup_pop() should solve it.

What version of the librdmacm are you using?
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Leak in rdma_get_cm_event on pthread_cancel?

[Flavio Baronti]

Il 1/13/2012 17:43 PM, Hefty, Sean ha scritto:
>> When I want to stop the thread reading rdma events, I kill it with a
>> pthread_cancel(), since it is blocked in a
>> rdma_get_cm_event(). This causes a leak of the evt variable in the
>> rdma_get_cm_event() function. Not a big deal, of
>> course, but a couple of pthread_cleanup_push(free, evt) and
>> pthread_cleanup_pop() should solve it.
>
> What version of the librdmacm are you using?

I'm using 1.0.13; I checked in 1.0.15 too, and the code appears to have the same structure.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Leak in rdma_get_cm_event on pthread_cancel?

[Jason Gunthorpe]

On Fri, Jan 13, 2012 at 12:46:00PM +0100, Flavio Baronti wrote:
> When I want to stop the thread reading rdma events, I kill it with a
> pthread_cancel(), since it is blocked in a rdma_get_cm_event(). This
> causes a leak of the evt variable in the rdma_get_cm_event()
> function. Not a big deal, of course, but a couple of
> pthread_cleanup_push(free, evt) and pthread_cleanup_pop() should
> solve it.

None of the IB libraries have been audited for posix async cancellation
safety.

To do this work someone would have to review every call to determine
if it is a cancelation point, and document this as part of the API.

Then review if any of the internal cancelation points in the
cancelable functions require a resource clean up, and add it.

IHMO, cancelation should not be used with threads that call into the
ib libraries.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Leak in rdma_get_cm_event on pthread_cancel?

[Flavio Baronti]

Il 1/13/2012 18:59 PM, Jason Gunthorpe ha scritto:
> On Fri, Jan 13, 2012 at 12:46:00PM +0100, Flavio Baronti wrote:
>> When I want to stop the thread reading rdma events, I kill it with a
>> pthread_cancel(), since it is blocked in a rdma_get_cm_event(). This
>> causes a leak of the evt variable in the rdma_get_cm_event()
>> function. Not a big deal, of course, but a couple of
>> pthread_cleanup_push(free, evt) and pthread_cleanup_pop() should
>> solve it.
>
> None of the IB libraries have been audited for posix async cancellation
> safety.
>
> To do this work someone would have to review every call to determine
> if it is a cancelation point, and document this as part of the API.
>
> Then review if any of the internal cancelation points in the
> cancelable functions require a resource clean up, and add it.
>
> IHMO, cancelation should not be used with threads that call into the
> ib libraries.
>
> Jason
>

Ok, but what is then the proper way to stop a thread blocked on rdma_get_cm_event() (or on ibv_get_cq_event())?

Flavio
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Leak in rdma_get_cm_event on pthread_cancel?

[Or Gerlitz]

Flavio Baronti <f.baronti-ngIpsMLAhaq41k5uCYKmRQ@public.gmane.org> wrote:
> what is then the proper way to stop a thread blocked on
> rdma_get_cm_event() (or on ibv_get_cq_event())?

Oh, this is simple - don't get there... have your code

1. get the FDs associated with IB - which you can retrieve from
librdmacm and from libibverbs

2. make them non blocking

3. add them to you select/poll  FD group

Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Leak in rdma_get_cm_event on pthread_cancel?

[Jason Gunthorpe]

On Tue, Jan 17, 2012 at 12:15:23AM +0200, Or Gerlitz wrote:
> Flavio Baronti <f.baronti-ngIpsMLAhaq41k5uCYKmRQ@public.gmane.org> wrote:
> > what is then the proper way to stop a thread blocked on
> > rdma_get_cm_event() (or on ibv_get_cq_event())?
> 
> Oh, this is simple - don't get there... have your code
> 
> 1. get the FDs associated with IB - which you can retrieve from
> librdmacm and from libibverbs
> 
> 2. make them non blocking
> 
> 3. add them to you select/poll  FD group

4. Use an eventfd or otherwise to break out of the select/poll loop
and then exit your thread.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Leak in rdma_get_cm_event on pthread_cancel?

[Flavio Baronti]

Il 1/16/2012 23:35 PM, Jason Gunthorpe ha scritto:
> On Tue, Jan 17, 2012 at 12:15:23AM +0200, Or Gerlitz wrote:
>> Flavio Baronti<f.baronti-ngIpsMLAhaq41k5uCYKmRQ@public.gmane.org>  wrote:
>>> what is then the proper way to stop a thread blocked on
>>> rdma_get_cm_event() (or on ibv_get_cq_event())?
>>
>> Oh, this is simple - don't get there... have your code
>>
>> 1. get the FDs associated with IB - which you can retrieve from
>> librdmacm and from libibverbs
>>
>> 2. make them non blocking
>>
>> 3. add them to you select/poll  FD group
>
> 4. Use an eventfd or otherwise to break out of the select/poll loop
> and then exit your thread.
>
> Jason

Thanks, I did not think about eventfd+select.

Flavio
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html