Re: [PATCH] uacce: fix concurrency of fops_open and uacce_remove

[Zhangfei Gao]

On 2022/6/16 下午4:14, Jean-Philippe Brucker wrote:
> On Thu, Jun 16, 2022 at 12:10:18PM +0800, Zhangfei Gao wrote:
> > > > diff --git a/drivers/misc/uacce/uacce.c b/drivers/misc/uacce/uacce.c
> > > > index 281c54003edc..b6219c6bfb48 100644
> > > > --- a/drivers/misc/uacce/uacce.c
> > > > +++ b/drivers/misc/uacce/uacce.c
> > > > @@ -136,9 +136,16 @@ static int uacce_fops_open(struct inode *inode, struct file *filep)
> > > >    	if (!q)
> > > >    		return -ENOMEM;
> > > > +	mutex_lock(&uacce->queues_lock);
> > > > +
> > > > +	if (!uacce->parent->driver) {
> > > I don't think this is useful, because the core clears parent->driver after
> > > having run uacce_remove():
> > >
> > >     rmmod hisi_zip		open()
> > >      ...				 uacce_fops_open()
> > >      __device_release_driver()	  ...
> > >       pci_device_remove()
> > >        hisi_zip_remove()
> > >         hisi_qm_uninit()
> > >          uacce_remove()
> > >           ...			  ...
> > >      				  mutex_lock(uacce->queues_lock)
> > >       ...				  if (!uacce->parent->driver)
> > >       device_unbind_cleanup()	  /* driver still valid, proceed */
> > >        dev->driver = NULL
> > The check  if (!uacce->parent->driver) is required, otherwise NULL pointer
> > may happen.
> I agree we need something, what I mean is that this check is not
> sufficient.
>
> > iommu_sva_bind_device
> > const struct iommu_ops *ops = dev_iommu_ops(dev);  ->
> > dev->iommu->iommu_dev->ops
> >
> > rmmod has no issue, but remove parent pci device has the issue.
> Ah right, relying on the return value of bind() wouldn't be enough even if
> we mandated SVA.
>
> [...]
> > > I think we need the global uacce_mutex to serialize uacce_remove() and
> > > uacce_fops_open(). uacce_remove() would do everything, including
> > > xa_erase(), while holding that mutex. And uacce_fops_open() would try to
> > > obtain the uacce object from the xarray while holding the mutex, which
> > > fails if the uacce object is being removed.
> > Since fops_open get char device refcount, uacce_release will not happen
> > until open returns.
> The refcount only ensures that the uacce_device object is not freed as
> long as there are open fds. But uacce_remove() can run while there are
> open fds, or fds in the process of being opened. And atfer uacce_remove()
> runs, the uacce_device object still exists but is mostly unusable. For
> example once the module is freed, uacce->ops is not valid anymore. But
> currently uacce_fops_open() may dereference the ops in this case:
>
> 	uacce_fops_open()
> 	 if (!uacce->parent->driver)
> 	 /* Still valid, keep going */		
> 	 ...					rmmod
> 						 uacce_remove()
> 	 ...					 free_module()
> 	 uacce->ops->get_queue() /* BUG */

uacce_remove should wait for uacce->queues_lock, until fops_open release 
the lock.
If open happen just after the uacce_remove: unlock, uacce_bind_queue in 
open should fail.

> Accessing uacce->ops after free_module() is a use-after-free. We need all
you men parent release the resources.
> the fops to synchronize with uacce_remove() to ensure they don't use any
> resource of the parent after it's been freed.
After fops_open, currently we are counting on parent driver stop all dma 
first, then call uacce_remove, which is assumption.
Like drivers/crypto/hisilicon/zip/zip_main.c: hisi_qm_wait_task_finish, 
which will wait uacce_release.
If comments this , there may other issue,
Unable to handle kernel paging request at virtual address ffff80000b700204
pc : hisi_qm_cache_wb.part.0+0x2c/0xa0

> I see uacce_fops_poll() may have the same problem, and should be inside
> uacce_mutex.
Do we need consider this, uacce_remove can happen anytime but not 
waiting dma stop?

Not sure uacce_mutex can do this.
Currently the sequence is
mutex_lock(&uacce->queues_lock);
mutex_lock(&uacce_mutex);

Or we set all the callbacks of uacce_ops to NULL?
Module_get/put only works for module, but not for removing device.

Thanks

> Thanks,
> Jean