Bluetooth: bfusb: Fix use-after-free and memory leak in device lifecycle

The driver stores a reference to the usb_device structure (udev) in its private data (data->udev), which can persist beyond the immediate context of the bfusb_probe() function.

Without proper reference count management, this can lead to two issues:

1. A use-after-free scenario if udev is accessed after its main reference count drops to zero (e.g., if the device is disconnected and the data structure is still active).
2. A memory leak if udev's reference count is not properly decremented during driver disconnect, preventing the usb_device object from being freed.

To correctly manage the udev lifetime, explicitly increment its reference count with usb_get_dev(udev) when storing it in the driver's private data. Correspondingly, decrement the reference count with usb_put_dev(data->udev) in the bfusb_disconnect() callback.

This ensures udev remains valid while referenced by the driver's private data and is properly released when no longer needed.

---
Changes in v3:
- Add tag Cc

Changes in v2:
- Add tags Fixes and Cc

 drivers/bluetooth/bfusb.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

index 8df310983bf6..f966bd8361b0 100644
@@ -622,7 +622,7 @@ static int bfusb_probe(struct usb_interface *intf, const struct usb_device_id *i
 	if (!data)
 		return -ENOMEM;
 
-	data->udev = udev;
+	data->udev = usb_get_dev(udev);
 	data->bulk_in_ep    = bulk_in_ep->desc.bEndpointAddress;
 	data->bulk_out_ep   = bulk_out_ep->desc.bEndpointAddress;
 	data->bulk_pkt_size = le16_to_cpu(bulk_out_ep->desc.wMaxPacketSize);

@@ -701,6 +701,8 @@ static void bfusb_disconnect(struct usb_interface *intf)
 
 	usb_set_intfdata(intf, NULL);
 
+	usb_put_dev(data->udev);
+
 	bfusb_close(hdev);
 
 	hci_unregister_dev(hdev);

2.43.0

This is automated email and please do not reply to this email!

Dear submitter,

Thank you for submitting the patches to the linux bluetooth mailing list. This is a CI test results with your patch series:
PW Link:https://patchwork.kernel.org/project/bluetooth/list/?series=987211

---Test result---

Test Summary:

CheckPatch                    PENDING   0.26 seconds
GitLint                       PENDING   0.21 seconds
SubjectPrefix                 PASS      0.13 seconds
BuildKernel                   PASS      24.58 seconds
CheckAllWarning               PASS      27.10 seconds
CheckSparse                   PASS      30.69 seconds
BuildKernel32                 PASS      24.34 seconds
TestRunnerSetup               PASS      485.22 seconds
TestRunner_l2cap-tester       PASS      24.93 seconds
TestRunner_iso-tester         PASS      44.50 seconds
TestRunner_bnep-tester        PASS      6.09 seconds
TestRunner_mgmt-tester        FAIL      127.01 seconds
TestRunner_rfcomm-tester      PASS      10.20 seconds
TestRunner_sco-tester         PASS      14.79 seconds
TestRunner_ioctl-tester       PASS      10.29 seconds
TestRunner_mesh-tester        FAIL      11.56 seconds
TestRunner_smp-tester         PASS      8.67 seconds
TestRunner_userchan-tester    PASS      6.30 seconds
IncrementalBuild              PENDING   0.44 seconds

Details
##############################
Test: CheckPatch - PENDING
Desc: Run checkpatch.pl script Output:

##############################
Test: GitLint - PENDING
Desc: Run gitlint
Output:

##############################
Test: TestRunner_mgmt-tester - FAIL
Desc: Run mgmt-tester with test-runner Output:
Total: 490, Passed: 485 (99.0%), Failed: 1, Not Run: 4

Failed Test Cases
LL Privacy - Add Device 3 (AL is full) Failed 0.220 seconds
##############################
Test: TestRunner_mesh-tester - FAIL
Desc: Run mesh-tester with test-runner Output:
Total: 10, Passed: 8 (80.0%), Failed: 2, Not Run: 0

Failed Test Cases

Mesh - Send cancel - 1                               Timed out    2.071 seconds
Mesh - Send cancel - 2                               Timed out    1.997 seconds

##############################
Test: IncrementalBuild - PENDING
Desc: Incremental build with the patches in the series Output:

---
Regards,
Linux Bluetooth

How can that happen as during the probe/remove cycle, the reference count is always properly incremetned.

There is no leak here at all, sorry.

How was this tested?

I'm not saying the change is wrong, just that I don't think it's actually a leak, or fix of anything real.

Or do you have a workload that shows this is needed? If so, what is the crash reported?

thanks,

greg k-h

Hello Greg,

Thanks for your feedback.

I understand your concern about the existence of a memory leak or use-after-free scenario in the driver's current context.

My intention with this patch is to ensure the driver adheres to best practices for managing usb_device structure references, as outlined in the kernel's documentation. The usb_get_dev() function is explicitly designed for use when a driver stores a reference to a usb_device structure in its private data, which is the case here with data->udev.

As the documentation for usb_get_dev() states:

``Each live reference to a device should be refcounted. Drivers for USB interfaces should normally record such references in their probe() methods, when they bind to an interface, and release them by calling usb_put_dev(), in their disconnect() methods.``

By following this recommendation, adding usb_get_dev(udev) in bfusb_probe() and usb_put_dev(data->udev) in bfusb_disconnect() ensures the udev structure's lifetime is explicitly managed by the driver as long as it's being referenced. This proactively prevents potential issues that could arise in future scenarios, even if a specific problem hasn't been observed or reported yet.

While I don't have a specific workload that reproduces a current crash or memory leak, this patch aims to enhance the driver's robustness by aligning its behavior with the established conventions for managing usb_device object references. It's a preventive measure to ensure the driver correctly handles the lifetime of the usb_device object it references, even in scenarios of unexpected disconnection or re-enumeration that might otherwise have unforeseen consequences.

Please let me know if you have any further questions.

Regards,

Salah Triki

Yes, I agree with the documentation, I wrote it :)

But, I am saying, you are NOT actually fixing anything here. It's a "best practice" but due to the fact that the dev pointer is only being reference counted by your change across the probe/release function, it is a pointless change.

It's also a "dangerous" change in that you are trying to say "this fixes a security issue!" when it does not do anything like that at all.

Please test this to see if it actually makes any difference in the code before making claims that it fixes a real bug.

thanks,

greg k-h