Re: [PATCH net-next] hinic: fix bug of wq out of bound access

[Xunlei Pang]

On 2022/4/28 PM8:46, maqiao wrote:
> cc Paolo Abeni, Guangbin Huang, Kees Cook, Gustavo A. R. Silva
>
> On 2022/4/28 PM8:30, Qiao Ma wrote:
> > If wq has only one page, we need to check wqe rolling over page by
> > compare end_idx and curr_idx, and then copy wqe to shadow wqe to
> > avoid out of bound access.
> > This work has been done in hinic_get_wqe, but missed for hinic_read_wqe.
> > This patch fixes it, and removes unnecessary MASKED_WQE_IDX().
> >
> > Fixes: 7dd29ee12865 ("hinic: add sriov feature support")
> > Signed-off-by: Qiao Ma <mqaio@xxxxxxxxxxxxxxxxx>
> > ---
> >   drivers/net/ethernet/huawei/hinic/hinic_hw_wq.c | 7 +++++--
> >   1 file changed, 5 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/huawei/hinic/hinic_hw_wq.c 
> > b/drivers/net/ethernet/huawei/hinic/hinic_hw_wq.c
> > index 5dc3743f8091..f04ac00e3e70 100644
> > --- a/drivers/net/ethernet/huawei/hinic/hinic_hw_wq.c
> > +++ b/drivers/net/ethernet/huawei/hinic/hinic_hw_wq.c
> > @@ -771,7 +771,7 @@ struct hinic_hw_wqe *hinic_get_wqe(struct hinic_wq 
> > *wq, unsigned int wqe_size,
> >       /* If we only have one page, still need to get shadown wqe when
> >        * wqe rolling-over page
> >        */
> > -    if (curr_pg != end_pg || MASKED_WQE_IDX(wq, end_prod_idx) < 
> > *prod_idx) {
> > +    if (curr_pg != end_pg || end_prod_idx < *prod_idx) {
> >           void *shadow_addr = &wq->shadow_wqe[curr_pg * 
> > wq->max_wqe_size];
> >           copy_wqe_to_shadow(wq, shadow_addr, num_wqebbs, *prod_idx);
> > @@ -841,7 +841,10 @@ struct hinic_hw_wqe *hinic_read_wqe(struct 
> > hinic_wq *wq, unsigned int wqe_size,
> >       *cons_idx = curr_cons_idx;
> > -    if (curr_pg != end_pg) {
> > +    /* If we only have one page, still need to get shadown wqe when
> > +     * wqe rolling-over page
> > +     */
> > +    if (curr_pg != end_pg || end_cons_idx < curr_cons_idx) {
> >           void *shadow_addr = &wq->shadow_wqe[curr_pg * 
> > wq->max_wqe_size];
> >           copy_wqe_to_shadow(wq, shadow_addr, num_wqebbs, *cons_idx);

This is a fundamental problem, and caused kernel panic as follows:

 Unable to handle kernel paging request at virtual address ffff800041371000
 Call trace:
  hinic_sq_get_sges+0x50/0x84 [hinic]
  free_tx_poll+0x84/0x2fc [hinic]
  napi_poll+0xcc/0x270
  net_rx_action+0xd8/0x280
  __do_softirq+0x120/0x37c
  __irq_exit_rcu+0x108/0x140
  irq_exit+0x14/0x20
  __handle_domain_irq+0x84/0xe0
  gic_handle_irq+0x80/0x108

Reviewed-by: Xunlei Pang <xlpang@xxxxxxxxxxxxxxxxx>