cxl: fix possible null-ptr-deref in cxl_guest_init_afu|adapter()

If device_register() fails in cxl_register_afu|adapter(), the device
is not added, device_unregister() can not be called in the error path,
otherwise it will cause a null-ptr-deref because of removing not added
device.

As comment of device_register() says, it should use put_device() to give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register fails.

Fixes: 14baf4d9c739 ("cxl: Add guest-specific code")
Signed-off-by: Yang Yingliang <redacted>
---
 drivers/misc/cxl/guest.c | 24 ++++++++++++++----------
 1 file changed, 14 insertions(+), 10 deletions(-)

diff --git a/drivers/misc/cxl/guest.c b/drivers/misc/cxl/guest.c
index 375f692ae9d6..fb95a2d5cef4 100644
--- a/drivers/misc/cxl/guest.c
+++ b/drivers/misc/cxl/guest.c

@@ -965,10 +965,10 @@ int cxl_guest_init_afu(struct cxl *adapter, int slice, struct device_node *afu_n
 	 * if it returns an error!
 	 */
 	if ((rc = cxl_register_afu(afu)))
-		goto err_put1;
+		goto err_put_dev;
 
 	if ((rc = cxl_sysfs_afu_add(afu)))
-		goto err_put1;
+		goto err_del_dev;
 
 	/*
 	 * pHyp doesn't expose the programming models supported by the

@@ -984,7 +984,7 @@ int cxl_guest_init_afu(struct cxl *adapter, int slice, struct device_node *afu_n
 		afu->modes_supported = CXL_MODE_DIRECTED;
 
 	if ((rc = cxl_afu_select_best_mode(afu)))
-		goto err_put2;
+		goto err_remove_sysfs;
 
 	adapter->afu[afu->slice] = afu;
 

@@ -1004,10 +1004,12 @@ int cxl_guest_init_afu(struct cxl *adapter, int slice, struct device_node *afu_n
 
 	return 0;
 
-err_put2:
+err_remove_sysfs:
 	cxl_sysfs_afu_remove(afu);
-err_put1:
-	device_unregister(&afu->dev);
+err_del_dev:
+	device_del(&afu->dev);
+err_put_dev:
+	put_device(&afu->dev);
 	free = false;
 	guest_release_serr_irq(afu);
 err2:

@@ -1141,18 +1143,20 @@ struct cxl *cxl_guest_init_adapter(struct device_node *np, struct platform_devic
 	 * even if it returns an error!
 	 */
 	if ((rc = cxl_register_adapter(adapter)))
-		goto err_put1;
+		goto err_put_dev;
 
 	if ((rc = cxl_sysfs_adapter_add(adapter)))
-		goto err_put1;
+		goto err_del_dev;
 
 	/* release the context lock as the adapter is configured */
 	cxl_adapter_context_unlock(adapter);
 
 	return adapter;
 
-err_put1:
-	device_unregister(&adapter->dev);
+err_del_dev:
+	device_del(&adapter->dev);
+err_put_dev:
+	put_device(&adapter->dev);
 	free = false;
 	cxl_guest_remove_chardev(adapter);
 err1:

-- 
2.25.1

If device_register() fails in cxl_pci_afu|adapter(), the device
is not added, device_unregister() can not be called in the error
path, otherwise it will cause a null-ptr-deref because of removing
not added device.

As comment of device_register() says, it should use put_device() to give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register fails.

Fixes: f204e0b8cedd ("cxl: Driver code for powernv PCIe based cards for userspace access")
Signed-off-by: Yang Yingliang <redacted>
---
 drivers/misc/cxl/pci.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/drivers/misc/cxl/pci.c b/drivers/misc/cxl/pci.c
index 3de0aea62ade..6d495d641c95 100644
--- a/drivers/misc/cxl/pci.c
+++ b/drivers/misc/cxl/pci.c

@@ -1164,10 +1164,10 @@ static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
 	 * if it returns an error!
 	 */
 	if ((rc = cxl_register_afu(afu)))
-		goto err_put1;
+		goto err_put_dev;
 
 	if ((rc = cxl_sysfs_afu_add(afu)))
-		goto err_put1;
+		goto err_del_dev;
 
 	adapter->afu[afu->slice] = afu;
 

@@ -1176,10 +1176,12 @@ static int pci_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
 
 	return 0;
 
-err_put1:
+err_del_dev:
+	device_del(&afu->dev);
+err_put_dev:
 	pci_deconfigure_afu(afu);
 	cxl_debugfs_afu_remove(afu);
-	device_unregister(&afu->dev);
+	put_device(&afu->dev);
 	return rc;
 
 err_free_native:

@@ -1667,23 +1669,25 @@ static struct cxl *cxl_pci_init_adapter(struct pci_dev *dev)
 	 * even if it returns an error!
 	 */
 	if ((rc = cxl_register_adapter(adapter)))
-		goto err_put1;
+		goto err_put_dev;
 
 	if ((rc = cxl_sysfs_adapter_add(adapter)))
-		goto err_put1;
+		goto err_del_dev;
 
 	/* Release the context lock as adapter is configured */
 	cxl_adapter_context_unlock(adapter);
 
 	return adapter;
 
-err_put1:
+err_del_dev:
+	device_del(&adapter->dev);
+err_put_dev:
 	/* This should mirror cxl_remove_adapter, except without the
 	 * sysfs parts
 	 */
 	cxl_debugfs_adapter_remove(adapter);
 	cxl_deconfigure_adapter(adapter);
-	device_unregister(&adapter->dev);
+	put_device(&adapter->dev);
 	return ERR_PTR(rc);
 
 err_release:

-- 
2.25.1

On 11/11/2022 15:54, Yang Yingliang wrote:

If device_register() fails in cxl_register_afu|adapter(), the device
is not added, device_unregister() can not be called in the error path,
otherwise it will cause a null-ptr-deref because of removing not added
device.

As comment of device_register() says, it should use put_device() to give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register fails.

Fixes: 14baf4d9c739 ("cxl: Add guest-specific code")
Signed-off-by: Yang Yingliang <redacted>
---

Thanks for fixing it!
At first, I was slightly uneasy about calling device_del() and put_device() directly, i.e. it would be better not to worry about what's under the hood of device_unregister(). But 1) I don't see how else to fix it and 2) more importantly, I looked at the history of device_unregister() to see how frequently it changed. I can now rest easy :-)

Acked-by: Frederic Barrat <redacted>

  Fred

30 more quoted lines

On 11/11/2022 15:54, Yang Yingliang wrote:

If device_register() fails in cxl_pci_afu|adapter(), the device
is not added, device_unregister() can not be called in the error
path, otherwise it will cause a null-ptr-deref because of removing
not added device.

As comment of device_register() says, it should use put_device() to give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register fails.

Fixes: f204e0b8cedd ("cxl: Driver code for powernv PCIe based cards for userspace access")
Signed-off-by: Yang Yingliang <redacted>
---

Acked-by: Frederic Barrat <redacted>

  Fred

25 more quoted lines

On Fri, 2022-11-11 at 22:54 +0800, Yang Yingliang wrote:

If device_register() fails in cxl_register_afu|adapter(), the device
is not added, device_unregister() can not be called in the error
path,
otherwise it will cause a null-ptr-deref because of removing not
added
device.

As comment of device_register() says, it should use put_device() to
give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register
fails.

The comment around device_add() is relevant too:

  Rule of thumb is: if device_add() succeeds, you should call
  device_del() when you want to get rid of it. If device_add() has
  *not* succeeded, use *only* put_device() to drop the reference count.

Fixes: 14baf4d9c739 ("cxl: Add guest-specific code")
Signed-off-by: Yang Yingliang <redacted>

Thanks for the fix!

Acked-by: Andrew Donnellan <redacted>

36 more quoted lines

-- 
Andrew Donnellan    OzLabs, ADL Canberra
ajd@linux.ibm.com   IBM Australia Limited

On Fri, 2022-11-11 at 22:54 +0800, Yang Yingliang wrote:

If device_register() fails in cxl_pci_afu|adapter(), the device
is not added, device_unregister() can not be called in the error
path, otherwise it will cause a null-ptr-deref because of removing
not added device.

As comment of device_register() says, it should use put_device() to
give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register
fails.

Fixes: f204e0b8cedd ("cxl: Driver code for powernv PCIe based cards
for userspace access")
Signed-off-by: Yang Yingliang <redacted>

See reply to patch #1.

This patch puts a few things between the device_del() call and the
put_device() call, but that looks fine - we don't touch afu->dev in the
middle.

Thanks for the fix!

Acked-by: Andrew Donnellan <redacted>

29 more quoted lines

-- 
Andrew Donnellan    OzLabs, ADL Canberra
ajd@linux.ibm.com   IBM Australia Limited

On Fri, 11 Nov 2022 22:54:39 +0800, Yang Yingliang wrote:

If device_register() fails in cxl_register_afu|adapter(), the device
is not added, device_unregister() can not be called in the error path,
otherwise it will cause a null-ptr-deref because of removing not added
device.

As comment of device_register() says, it should use put_device() to give
up the reference in the error path. So split device_unregister() into
device_del() and put_device(), then goes to put dev when register fails.

[...]

Applied to powerpc/next.

[1/2] cxl: fix possible null-ptr-deref in cxl_guest_init_afu|adapter()
      https://git.kernel.org/powerpc/c/f949ccee1dde970bc77dc871b4f0b5e651577344
[2/2] cxl: fix possible null-ptr-deref in cxl_pci_init_afu|adapter()
      https://git.kernel.org/powerpc/c/8bf03f557d6c6e108cf47bea32f4a68e276e1157

cheers