In msi2500_probe, it invokes spi_alloc_master to allocate master
controller and spi_register_master to register master controller.
Then in msi2500_disconnect, it calls spi_unregister_master to unregister
the master controller. And in spi_unregister_master, it calls put_device
to decrease the refcount and the device object will be freed then. As a
result, the dereference of dev->lock will cause a use-after-free bug.
Fix it by changing spi_alloc_master to devm_spi_alloc_master, and
removing spi_master_put in the error handling code.
Note that this patch can prevent UAF occurring any more
Fixes: fd8b5f502929 ("msi2500: move msi3101 out of staging and rename")
Signed-off-by: Dongliang Mu <[email protected]>
---
drivers/media/usb/msi2500/msi2500.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/media/usb/msi2500/msi2500.c b/drivers/media/usb/msi2500/msi2500.c
index 71de6b4c4e4c..64c3ec9f1d0c 100644
--- a/drivers/media/usb/msi2500/msi2500.c
+++ b/drivers/media/usb/msi2500/msi2500.c
@@ -1220,7 +1220,7 @@ static int msi2500_probe(struct usb_interface *intf,
}
/* SPI master adapter */
- master = spi_alloc_master(dev->dev, 0);
+ master = devm_spi_alloc_master(dev->dev, 0);
if (master == NULL) {
ret = -ENOMEM;
goto err_unregister_v4l2_dev;
@@ -1233,7 +1233,6 @@ static int msi2500_probe(struct usb_interface *intf,
spi_master_set_devdata(master, dev);
ret = spi_register_master(master);
if (ret) {
- spi_master_put(master);
goto err_unregister_v4l2_dev;
}
--
2.25.1
Em Wed, 17 Nov 2021 00:32:07 +0800
Dongliang Mu <[email protected]> escreveu:
> In msi2500_probe, it invokes spi_alloc_master to allocate master
> controller and spi_register_master to register master controller.
> Then in msi2500_disconnect, it calls spi_unregister_master to unregister
> the master controller. And in spi_unregister_master, it calls put_device
> to decrease the refcount and the device object will be freed then. As a
> result, the dereference of dev->lock will cause a use-after-free bug.
>
> Fix it by changing spi_alloc_master to devm_spi_alloc_master, and
> removing spi_master_put in the error handling code.
>
> Note that this patch can prevent UAF occurring any more
>
> Fixes: fd8b5f502929 ("msi2500: move msi3101 out of staging and rename")
> Signed-off-by: Dongliang Mu <[email protected]>
> ---
> drivers/media/usb/msi2500/msi2500.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/drivers/media/usb/msi2500/msi2500.c b/drivers/media/usb/msi2500/msi2500.c
> index 71de6b4c4e4c..64c3ec9f1d0c 100644
> --- a/drivers/media/usb/msi2500/msi2500.c
> +++ b/drivers/media/usb/msi2500/msi2500.c
> @@ -1220,7 +1220,7 @@ static int msi2500_probe(struct usb_interface *intf,
> }
>
> /* SPI master adapter */
> - master = spi_alloc_master(dev->dev, 0);
> + master = devm_spi_alloc_master(dev->dev, 0);
We had some bad past experiences on using devm_* on USB media devices,
as there are almost always multiple drivers loaded and multiple
interfaces.
Could you please fix the free logic instead of using devm_*?
Regards,
Mauro
> if (master == NULL) {
> ret = -ENOMEM;
> goto err_unregister_v4l2_dev;
> @@ -1233,7 +1233,6 @@ static int msi2500_probe(struct usb_interface *intf,
> spi_master_set_devdata(master, dev);
> ret = spi_register_master(master);
> if (ret) {
> - spi_master_put(master);
> goto err_unregister_v4l2_dev;
> }
>
Thanks,
Mauro
On Tue, Dec 14, 2021 at 9:32 PM Mauro Carvalho Chehab
<[email protected]> wrote:
>
> Em Wed, 17 Nov 2021 00:32:07 +0800
> Dongliang Mu <[email protected]> escreveu:
>
> > In msi2500_probe, it invokes spi_alloc_master to allocate master
> > controller and spi_register_master to register master controller.
> > Then in msi2500_disconnect, it calls spi_unregister_master to unregister
> > the master controller. And in spi_unregister_master, it calls put_device
> > to decrease the refcount and the device object will be freed then. As a
> > result, the dereference of dev->lock will cause a use-after-free bug.
> >
> > Fix it by changing spi_alloc_master to devm_spi_alloc_master, and
> > removing spi_master_put in the error handling code.
> >
> > Note that this patch can prevent UAF occurring any more
> >
> > Fixes: fd8b5f502929 ("msi2500: move msi3101 out of staging and rename")
> > Signed-off-by: Dongliang Mu <[email protected]>
> > ---
> > drivers/media/usb/msi2500/msi2500.c | 3 +--
> > 1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/drivers/media/usb/msi2500/msi2500.c b/drivers/media/usb/msi2500/msi2500.c
> > index 71de6b4c4e4c..64c3ec9f1d0c 100644
> > --- a/drivers/media/usb/msi2500/msi2500.c
> > +++ b/drivers/media/usb/msi2500/msi2500.c
> > @@ -1220,7 +1220,7 @@ static int msi2500_probe(struct usb_interface *intf,
> > }
> >
> > /* SPI master adapter */
> > - master = spi_alloc_master(dev->dev, 0);
> > + master = devm_spi_alloc_master(dev->dev, 0);
>
> We had some bad past experiences on using devm_* on USB media devices,
> as there are almost always multiple drivers loaded and multiple
> interfaces.
>
> Could you please fix the free logic instead of using devm_*?
Hi Mauro,
When I resume working on this patch, I found this UAF bug is already
fixed in the spi part and merged to the upstream.
The correct fix is spi: fix use-after-free of the add_lock mutex [1].
So let's forget this patch.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6c53b45c71b4920b5e62f0ea8079a1da382b9434
>
> Regards,
> Mauro
>
> > if (master == NULL) {
> > ret = -ENOMEM;
> > goto err_unregister_v4l2_dev;
> > @@ -1233,7 +1233,6 @@ static int msi2500_probe(struct usb_interface *intf,
> > spi_master_set_devdata(master, dev);
> > ret = spi_register_master(master);
> > if (ret) {
> > - spi_master_put(master);
> > goto err_unregister_v4l2_dev;
> > }
> >
>
>
>
> Thanks,
> Mauro