Re: [PATCH v2 1/2] usb: gadget: f_hid: don't call copy_from_user() under get_report_spinlock
From: Greg KH
Date: Fri Nov 20 2026 - 05:42:57 EST
On Sat, Sep 19, 2026 at 11:00:49AM +0000, Hui Peng wrote:
> f_hidg_get_report() already copies the incoming struct usb_hidg_report
> from userspace into the freshly allocated 'entry' before acquiring
> hidg->get_report_spinlock.
>
> When a report with the same report_id is already present in
> hidg->report_list, the update path copies from userspace a second time,
> this time directly into ptr->report_data while the spinlock is held with
> interrupts disabled:
>
> spin_lock_irqsave(&hidg->get_report_spinlock, flags);
> ptr = f_hidg_search_for_report(hidg, report_id);
> if (ptr) {
> if (copy_from_user(&ptr->report_data, buffer,
> sizeof(struct usb_hidg_report))) {
>
> copy_from_user() may fault and sleep, so this is a sleeping function
> called from atomic context, reported by CONFIG_DEBUG_ATOMIC_SLEEP as
> "BUG: sleeping function called from invalid context". Userspace can
> reach it at will by issuing GADGET_HID_WRITE_GET_REPORT twice with the
> same report_id on /dev/hidgN.
>
> The second copy is also redundant: the same user buffer has already been
> copied into entry->report_data a few lines above, and report_id was
> derived from that copy. Assign from the already copied data instead,
> which removes the fault from the critical section and makes the update
> atomic with respect to the list lookup.
>
> Fixes: a139c98f760e ("USB: gadget: f_hid: Add GET_REPORT via userspace IOCTL")
> Assisted-by: LLM
> Signed-off-by: Hui Peng <benquike@xxxxxxxxx>
> ---
> v2: Add the Fixes: tag Greg asked for.
>
> f_hidg_get_report(), the spinlock and the copy_from_user() call under
> it were all added together by a139c98f760e ("USB: gadget: f_hid: Add
> GET_REPORT via userspace IOCTL"), first released in v6.12, so that is
> where this starts. git blame on the removed lines agrees.
>
> Found by inspection while investigating the use-after-free fixed in
> patch 2/2, and build tested only; I do not have HID gadget hardware,
> but the path is reachable from /dev/hidgN with dummy_hcd.
>
> drivers/usb/gadget/function/f_hid.c | 8 +-------
> 1 file changed, 1 insertion(+), 7 deletions(-)
>
> --- a/drivers/usb/gadget/function/f_hid.c
> +++ b/drivers/usb/gadget/function/f_hid.c
> @@ -668,13 +668,7 @@ static int f_hidg_get_report(struct file
>
> if (ptr) {
> /* Report already exists in list - update it */
> - if (copy_from_user(&ptr->report_data, buffer,
> - sizeof(struct usb_hidg_report))) {
> - spin_unlock_irqrestore(&hidg->get_report_spinlock, flags);
> - ERROR(cdev, "copy_from_user error\n");
> - kfree(entry);
> - return -EINVAL;
> - }
> + ptr->report_data = entry->report_data;
> kfree(entry);
> } else {
> /* Report does not exist in list - add it */
> --
> 2.43.0
>
Series does not apply anymore, can you please rebase and resend?
thanks,
greg k-h