Re: [PATCH -mm] ehca: avoid crash on kthread_create() failure

[Akinobu Mita <akinobu.mita@gmail.com>]

On Mon, Dec 25, 2006 at 11:41:32AM +0200, Muli Ben-Yehuda wrote:
> On Mon, Dec 25, 2006 at 06:35:57PM +0900, Akinobu Mita wrote:
> > On Mon, Dec 25, 2006 at 10:55:51AM +0200, Muli Ben-Yehuda wrote:
> > > This is correct because cct is allocated via alloc_percpu, which in
> > > turn calls kzalloc, which means cct->task is NULL by default, but it's
> > > a little too obscure for me. How about making it explicit?
> > > 
> > > task = kthread_create(...)
> > > if (!IS_ERR(task))
> > >    cct->task = task;
> > > else
> > >   cct->task = NULL;
> > > 
> > > return cct->task;
> > 
> > Subject: [PATCH -mm] ehca: avoid crash on kthread_create() failure (v3)
> > 
> > This patch disallows invalid task_struct pointer returned by
> > kthread_create() to be written to percpu data to avoid crash.
> > 
> > Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
> > Cc: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
> > Cc: Christoph Raisch <raisch@de.ibm.com>
> > Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
> > 
> > ---
> >  drivers/infiniband/hw/ehca/ehca_irq.c |   11 ++++++++---
> >  1 file changed, 8 insertions(+), 3 deletions(-)
> > 
> > Index: 2.6-mm/drivers/infiniband/hw/ehca/ehca_irq.c
> > ===================================================================
> > --- 2.6-mm.orig/drivers/infiniband/hw/ehca/ehca_irq.c
> > +++ 2.6-mm/drivers/infiniband/hw/ehca/ehca_irq.c
> > @@ -606,15 +606,20 @@ static int comp_task(void *__cct)
> >  static struct task_struct *create_comp_task(struct ehca_comp_pool *pool,
> >  					    int cpu)
> >  {
> > +	struct task_struct *task;
> >  	struct ehca_cpu_comp_task *cct;
> >  
> >  	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
> >  	spin_lock_init(&cct->task_lock);
> >  	INIT_LIST_HEAD(&cct->cq_list);
> >  	init_waitqueue_head(&cct->wait_queue);
> > -	cct->task = kthread_create(comp_task, cct, "ehca_comp/%d", cpu);
> > +	task = kthread_create(comp_task, cct, "ehca_comp/%d", cpu);
> > +	if (!IS_ERR(task))
> > +		cct->task = task;
> > +	else
> > +		cct->task = NULL;
> >  
> > -	return cct->task;
> > +	return task;
> 
> This should be return cct->task, since we later test the return value
> of create_comp_task against NULL, e.g., in comp_poll_callback and
> ehca_create_comp_pool.

Yeah, I already sent the patch:
http://lkml.org/lkml/2006/12/19/56

Maybe I should reorganize these ehca fixes later.
(make ehca_create_task() return NULL on failure)