On 21/05/2019 20:55, Jason Gunthorpe wrote: > The core code should not pass a udata to the driver destroy function that > contains the input from the create command. Otherwise the driver will > attempt to interpret the create udata as destroy udata, and at least > in the case of EFA, will leak resources. > > Zero this stuff out before invoking destroy. > > Reported-by: Leon Romanovsky <leonro@xxxxxxxxxxxx> > Fixes: c4367a26357b ("IB: Pass uverbs_attr_bundle down ib_x destroy path") > Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxxxx> Reviewed-by: Gal Pressman <galpress@xxxxxxxxxx> I will send a followup patch that handles the flows that pass NULL udata.