Skip to contentSkip to conversation

rpmsg: fix possible refcount leak in rpmsg_register_device_override()

PATCH Superseded Hangyu Hua 2 participants 4 messages Jun 23, 2022 Remote processor messaging (rpmsg)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Hangyu Hua

author
#0 lore raw

PATCH 1/1 rpmsg: fix possible refcount leak in rpmsg_register_device_override()

[1] commit 1680939e9ecf ("rpmsg: virtio: Fix possible double free in rpmsg_virtio_add_ctrl_dev()")
[2] commit c2eecefec5df ("rpmsg: virtio: Fix possible double free in rpmsg_probe()")
[3] commit bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")

The above three patches merged at the same time introduced a new bug.
[1] and [2] make rpmsg_ns_register_device and rpmsg_ctrldev_register_device need to call the callback function internally to free vch when it fails.
[3] has an error return path not handled vch.

Fix this by adding a put_device() to the error path.

Fixes bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")
---
 drivers/rpmsg/rpmsg_core.c | 1 +
 1 file changed, 1 insertion(+)
drivers/rpmsg/rpmsg_core.c +1 -0
index 290c1f02da10..5a47cad89fdc 100644
@@ -618,6 +618,7 @@ int rpmsg_register_device_override(struct rpmsg_device *rpdev,
 					  strlen(driver_override));
 		if (ret) {
 			dev_err(dev, "device_set_override failed: %d\n", ret);
+			put_device(dev);
 			return ret;
 		}
 	}
Signature · 1 line
2.25.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 6 lines · Hangyu Hua

[1] commit 1680939e9ecf ("rpmsg: virtio: Fix possible double free in rpmsg_virtio_add_ctrl_dev()")
[2] commit c2eecefec5df ("rpmsg: virtio: Fix possible double free in rpmsg_probe()")
[3] commit bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")

I think only the last [3] introduced it, because it's the commit missing put_device in first error path.

Quoted 10 lines

The above three patches merged at the same time introduced a new bug.
[1] and [2] make rpmsg_ns_register_device and rpmsg_ctrldev_register_device need to call the callback function internally to free vch when it fails.
[3] has an error return path not handled vch.

Fix this by adding a put_device() to the error path.

Fixes bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")

Best regards,
Krzysztof

author
#2 lore raw
Quoted 11 lines · Krzysztof Kozlowski
Quoted 6 lines · Hangyu Hua

[1] commit 1680939e9ecf ("rpmsg: virtio: Fix possible double free in rpmsg_virtio_add_ctrl_dev()")
[2] commit c2eecefec5df ("rpmsg: virtio: Fix possible double free in rpmsg_probe()")
[3] commit bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")

I think only the last [3] introduced it, because it's the commit missing put_device in first error path.

I see. Do i need to change the commit log and then send a v2?

Quoted 16 lines
Quoted 10 lines

The above three patches merged at the same time introduced a new bug.
[1] and [2] make rpmsg_ns_register_device and rpmsg_ctrldev_register_device need to call the callback function internally to free vch when it fails.
[3] has an error return path not handled vch.

Fix this by adding a put_device() to the error path.

Fixes bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")

Best regards,
Krzysztof

#3 lore raw
Quoted 9 lines · Hangyu Hua
Quoted 8 lines · Krzysztof Kozlowski
Quoted 6 lines · Hangyu Hua

[1] commit 1680939e9ecf ("rpmsg: virtio: Fix possible double free in rpmsg_virtio_add_ctrl_dev()")
[2] commit c2eecefec5df ("rpmsg: virtio: Fix possible double free in rpmsg_probe()")
[3] commit bb17d110cbf2 ("rpmsg: Fix calling device_lock() on non-initialized device")

Quoted 5 lines
Quoted 3 lines

I think only the last [3] introduced it, because it's the commit missing put_device in first error path.

I see. Do i need to change the commit log and then send a v2?

Yes, please. With my Reviewed-by tag.

Best regards,
Krzysztof