Bluetooth: hci_core: Fix UAF in hci_unregister_dev()
PATCH 1/1 Bluetooth: hci_core: Fix UAF in hci_unregister_dev()
commit eec3deaeaafe upstream.
[manual backport: 6.6.y uses cancel_*_work_sync() instead of
disable_*_work_sync() which was introduced in a later cycle]
hci_unregister_dev() does not cancel cmd_timer and ncmd_timer before the hci_dev structure is freed. If a timeout fires during device teardown, the callback dereferences freed memory (including the hdev->reset function pointer), leading to a use-after-free.
Add cancel_delayed_work_sync() calls alongside the existing cancel_work_sync() calls to ensure both timers are fully quiesced before teardown proceeds.
--- net/bluetooth/hci_core.c | 2 ++ 1 file changed, 2 insertions(+)
net/bluetooth/hci_core.c +2 -0
@@ -2701,6 +2701,8 @@ void hci_unregister_dev(struct hci_dev *hdev)cancel_work_sync (&hdev->tx_work);cancel_work_sync (&hdev->power_on);cancel_work_sync (&hdev->error_reset); +cancel_delayed_work_sync (&hdev->cmd_timer); +cancel_delayed_work_sync (&hdev->ncmd_timer);hci_cmd_sync_clear (hdev);
Signature · 1 line
2.43.0
Sasha Levin
maintainerQuoted 1 line
[PATCH 6.6.y] Bluetooth:
hci_core: Fix UAF inhci_unregister_dev()
Thanks for the patch, but I can't take this as-is -- a v2 is needed:
1) The "commit ... upstream" line cites a hash that doesn't exist in mainline, and there's no upstream commit titled "Fix UAF in hci_unregister_dev()".
2) The diff doesn't match that fix: upstream uses disable_work_sync() / disable_delayed_work_sync() (guarded by HCI_UNREGISTER), not an added cancel_delayed_work_sync() in hci_unregister_dev(). cancel_* doesn't prevent re-queue, and hci_unregister_dev() later calls hci_dev_close_sync() which can re-touch those timers, so this may not actually close the UAF.
Signature · 2 lines
Thanks, Sasha