CVE-2024-56545

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> HID: hyperv: streamline driver probe to avoid devres issues<br /> <br /> It was found that unloading &amp;#39;hid_hyperv&amp;#39; module results in a devres<br /> complaint:<br /> <br /> ...<br /> hv_vmbus: unregistering driver hid_hyperv<br /> ------------[ cut here ]------------<br /> WARNING: CPU: 2 PID: 3983 at drivers/base/devres.c:691 devres_release_group+0x1f2/0x2c0<br /> ...<br /> Call Trace:<br /> <br /> ? devres_release_group+0x1f2/0x2c0<br /> ? __warn+0xd1/0x1c0<br /> ? devres_release_group+0x1f2/0x2c0<br /> ? report_bug+0x32a/0x3c0<br /> ? handle_bug+0x53/0xa0<br /> ? exc_invalid_op+0x18/0x50<br /> ? asm_exc_invalid_op+0x1a/0x20<br /> ? devres_release_group+0x1f2/0x2c0<br /> ? devres_release_group+0x90/0x2c0<br /> ? rcu_is_watching+0x15/0xb0<br /> ? __pfx_devres_release_group+0x10/0x10<br /> hid_device_remove+0xf5/0x220<br /> device_release_driver_internal+0x371/0x540<br /> ? klist_put+0xf3/0x170<br /> bus_remove_device+0x1f1/0x3f0<br /> device_del+0x33f/0x8c0<br /> ? __pfx_device_del+0x10/0x10<br /> ? cleanup_srcu_struct+0x337/0x500<br /> hid_destroy_device+0xc8/0x130<br /> mousevsc_remove+0xd2/0x1d0 [hid_hyperv]<br /> device_release_driver_internal+0x371/0x540<br /> driver_detach+0xc5/0x180<br /> bus_remove_driver+0x11e/0x2a0<br /> ? __mutex_unlock_slowpath+0x160/0x5e0<br /> vmbus_driver_unregister+0x62/0x2b0 [hv_vmbus]<br /> ...<br /> <br /> And the issue seems to be that the corresponding devres group is not<br /> allocated. Normally, devres_open_group() is called from<br /> __hid_device_probe() but Hyper-V HID driver overrides &amp;#39;hid_dev-&gt;driver&amp;#39;<br /> with &amp;#39;mousevsc_hid_driver&amp;#39; stub and basically re-implements<br /> __hid_device_probe() by calling hid_parse() and hid_hw_start() but not<br /> devres_open_group(). hid_device_probe() does not call __hid_device_probe()<br /> for it. Later, when the driver is removed, hid_device_remove() calls<br /> devres_release_group() as it doesn&amp;#39;t check whether hdev-&gt;driver was<br /> initially overridden or not.<br /> <br /> The issue seems to be related to the commit 62c68e7cee33 ("HID: ensure<br /> timely release of driver-allocated resources") but the commit itself seems<br /> to be correct.<br /> <br /> Fix the issue by dropping the &amp;#39;hid_dev-&gt;driver&amp;#39; override and using<br /> hid_register_driver()/hid_unregister_driver() instead. Alternatively, it<br /> would have been possible to rely on the default handling but<br /> HID_CONNECT_DEFAULT implies HID_CONNECT_HIDRAW and it doesn&amp;#39;t seem to work<br /> for mousevsc as-is.