Re: [PATCH] Correct return codes during racy destruction.

[Paul Durrant <xadimgnik@xxxxxxxxx>]

On 08/11/2022 10:49, Martin Harvey wrote:

Can we have an explanatory commit comment? How did you discover the 
races (since the HLK/WLK has not complained before AFAIK)? What is the 
fall-out of the race? BSOD?
  Paul

> Signed-off-by: Martin Harvey <martin.harvey@xxxxxxxxxx>
> ---
>   src/xenvif/driver.c | 12 ++++++++++++
>   1 file changed, 12 insertions(+)
>
> diff --git a/src/xenvif/driver.c b/src/xenvif/driver.c
> index 8bc097a..cac0cee 100644
> --- a/src/xenvif/driver.c
> +++ b/src/xenvif/driver.c
> @@ -350,8 +350,20 @@ Dispatch(
>       ASSERT3P(Dx->DeviceObject, ==, DeviceObject);
>   
>       if (Dx->DevicePnpState == Deleted) {
> +        PIO_STACK_LOCATION StackLocation = IoGetCurrentIrpStackLocation(Irp);
> +        UCHAR MajorFunction = StackLocation->MajorFunction;
> +        UCHAR MinorFunction = StackLocation->MinorFunction;
>           status = STATUS_NO_SUCH_DEVICE;
>   
> +        if (MajorFunction == IRP_MJ_PNP) {
> +            if ((MinorFunction == IRP_MN_SURPRISE_REMOVAL) ||
> +               (MinorFunction == IRP_MN_REMOVE_DEVICE)) {
> +                /* FDO and PDO deletions can block after being marked deleted, 
> but before IoDeleteDevice */
> +                status = STATUS_SUCCESS;
> +            }
> +            ASSERT((MinorFunction != IRP_MN_CANCEL_REMOVE_DEVICE) && 
> (MinorFunction != IRP_MN_CANCEL_STOP_DEVICE));
> +        }
> +
>           Irp->IoStatus.Status = status;
>           IoCompleteRequest(Irp, IO_NO_INCREMENT);
>           goto done;