On Thu, Feb 23, 2017 at 12:20 PM, Jason Gunthorpe
<jgunthorpe(a)obsidianresearch.com> wrote:
On Thu, Feb 23, 2017 at 11:22:03AM -0800, Dan Williams wrote:
> If device_add() fails, cleanup the cdev. Otherwise, we leak a kobj_map()
> with a stale device number.
>
> Fixes: ba09c01d2fa8 ("dax: convert to the cdev api")
> Cc: <stable(a)vger.kernel.org>
> Cc: Logan Gunthorpe <logang(a)deltatee.com>
> Reported-by: Jason Gunthorpe <jgunthorpe(a)obsidianresearch.com>
> Signed-off-by: Dan Williams <dan.j.williams(a)intel.com>
> drivers/dax/dax.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/dax/dax.c b/drivers/dax/dax.c
> index ed758b74ddf0..0f8008dd0b0c 100644
> +++ b/drivers/dax/dax.c
> @@ -724,6 +724,7 @@ struct dax_dev *devm_create_dax_dev(struct dax_region
*dax_region,
> dev_set_name(dev, "dax%d.%d", dax_region->id, dax_dev->id);
> rc = device_add(dev);
> if (rc) {
> + cdev_del(&dax_dev->cdev);
This probably should call into unregister_dax_dev and just skip the
device_unregister part.
Once cdev_add returns it is possible for a mmap to have been created,
so cleanup after that point has to go through all the other
unregister_dax_dev steps.
Ah true. I was thinking the device node is not auto-created if the
device_add() fails, but there's theoretically nothing stopping
userspace from trying to hit this race with its own self-created
device-node.