On Thu, Nov 17, 2022 at 08:41:52PM +0800, Zhang Changzhong wrote:
>
>
> On 2022/11/17 19:36, Leon Romanovsky wrote:
> > On Thu, Nov 17, 2022 at 03:50:09PM +0800, Zhang Changzhong wrote:
> >> The __ef100_hard_start_xmit() returns NETDEV_TX_OK without freeing skb
> >> in error handling case, add dev_kfree_skb_any() to fix it.
> >>
> >> Fixes: 51b35a454efd ("sfc: skeleton EF100 PF driver")
> >> Signed-off-by: Zhang Changzhong <[email protected]>
> >> ---
> >> drivers/net/ethernet/sfc/ef100_netdev.c | 1 +
> >> 1 file changed, 1 insertion(+)
> >>
> >> diff --git a/drivers/net/ethernet/sfc/ef100_netdev.c b/drivers/net/ethernet/sfc/ef100_netdev.c
> >> index 88fa295..ddcc325 100644
> >> --- a/drivers/net/ethernet/sfc/ef100_netdev.c
> >> +++ b/drivers/net/ethernet/sfc/ef100_netdev.c
> >> @@ -218,6 +218,7 @@ netdev_tx_t __ef100_hard_start_xmit(struct sk_buff *skb,
> >> skb->len, skb->data_len, channel->channel);
> >> if (!efx->n_channels || !efx->n_tx_channels || !channel) {
> >> netif_stop_queue(net_dev);
> >> + dev_kfree_skb_any(skb);
> >> goto err;
> >> }
> >
> > ef100 doesn't release in __ef100_enqueue_skb() either. SKB shouldn't be
> > NULL or ERR at this stage.
>
> SKB shouldn't be NULL or ERR, so it can be freed. But this code looks weird.
Please take a look __ef100_enqueue_skb() and see if it frees SKB on
error or not. If not, please fix it.
Thanks
>
> >
> > diff --git a/drivers/net/ethernet/sfc/ef100_tx.c b/drivers/net/ethernet/sfc/ef100_tx.c
> > index 29ffaf35559d..426706b91d02 100644
> > --- a/drivers/net/ethernet/sfc/ef100_tx.c
> > +++ b/drivers/net/ethernet/sfc/ef100_tx.c
> > @@ -497,7 +497,7 @@ int __ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
> >
> > err:
> > efx_enqueue_unwind(tx_queue, old_insert_count);
> > - if (!IS_ERR_OR_NULL(skb))
> > + if (rc)
> > dev_kfree_skb_any(skb);
> >
> > /* If we're not expecting another transmit and we had something to push
> >
> >
> >>
> >> --
> >> 2.9.5
> >>
> > .
> >
On Thu, Nov 17, 2022 at 03:05:27PM +0200, Leon Romanovsky wrote:
> On Thu, Nov 17, 2022 at 08:41:52PM +0800, Zhang Changzhong wrote:
> >
> >
> > On 2022/11/17 19:36, Leon Romanovsky wrote:
> > > On Thu, Nov 17, 2022 at 03:50:09PM +0800, Zhang Changzhong wrote:
> > >> The __ef100_hard_start_xmit() returns NETDEV_TX_OK without freeing skb
> > >> in error handling case, add dev_kfree_skb_any() to fix it.
> > >>
> > >> Fixes: 51b35a454efd ("sfc: skeleton EF100 PF driver")
> > >> Signed-off-by: Zhang Changzhong <[email protected]>
> > >> ---
> > >> drivers/net/ethernet/sfc/ef100_netdev.c | 1 +
> > >> 1 file changed, 1 insertion(+)
> > >>
> > >> diff --git a/drivers/net/ethernet/sfc/ef100_netdev.c b/drivers/net/ethernet/sfc/ef100_netdev.c
> > >> index 88fa295..ddcc325 100644
> > >> --- a/drivers/net/ethernet/sfc/ef100_netdev.c
> > >> +++ b/drivers/net/ethernet/sfc/ef100_netdev.c
> > >> @@ -218,6 +218,7 @@ netdev_tx_t __ef100_hard_start_xmit(struct sk_buff *skb,
> > >> skb->len, skb->data_len, channel->channel);
> > >> if (!efx->n_channels || !efx->n_tx_channels || !channel) {
> > >> netif_stop_queue(net_dev);
> > >> + dev_kfree_skb_any(skb);
> > >> goto err;
> > >> }
> > >
> > > ef100 doesn't release in __ef100_enqueue_skb() either. SKB shouldn't be
> > > NULL or ERR at this stage.
> >
> > SKB shouldn't be NULL or ERR, so it can be freed. But this code looks weird.
>
> Please take a look __ef100_enqueue_skb() and see if it frees SKB on
> error or not. If not, please fix it.
That function looks ok to me, but I appreciate the extra eyes on it.
Martin
> Thanks
>
> >
> > >
> > > diff --git a/drivers/net/ethernet/sfc/ef100_tx.c b/drivers/net/ethernet/sfc/ef100_tx.c
> > > index 29ffaf35559d..426706b91d02 100644
> > > --- a/drivers/net/ethernet/sfc/ef100_tx.c
> > > +++ b/drivers/net/ethernet/sfc/ef100_tx.c
> > > @@ -497,7 +497,7 @@ int __ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
> > >
> > > err:
> > > efx_enqueue_unwind(tx_queue, old_insert_count);
> > > - if (!IS_ERR_OR_NULL(skb))
> > > + if (rc)
> > > dev_kfree_skb_any(skb);
> > >
> > > /* If we're not expecting another transmit and we had something to push
> > >
> > >
> > >>
> > >> --
> > >> 2.9.5
> > >>
> > > .
> > >
On Fri, Nov 18, 2022 at 09:15:43AM +0000, Martin Habets wrote:
> On Thu, Nov 17, 2022 at 03:05:27PM +0200, Leon Romanovsky wrote:
> > On Thu, Nov 17, 2022 at 08:41:52PM +0800, Zhang Changzhong wrote:
> > >
> > >
> > > On 2022/11/17 19:36, Leon Romanovsky wrote:
> > > > On Thu, Nov 17, 2022 at 03:50:09PM +0800, Zhang Changzhong wrote:
> > > >> The __ef100_hard_start_xmit() returns NETDEV_TX_OK without freeing skb
> > > >> in error handling case, add dev_kfree_skb_any() to fix it.
> > > >>
> > > >> Fixes: 51b35a454efd ("sfc: skeleton EF100 PF driver")
> > > >> Signed-off-by: Zhang Changzhong <[email protected]>
> > > >> ---
> > > >> drivers/net/ethernet/sfc/ef100_netdev.c | 1 +
> > > >> 1 file changed, 1 insertion(+)
> > > >>
> > > >> diff --git a/drivers/net/ethernet/sfc/ef100_netdev.c b/drivers/net/ethernet/sfc/ef100_netdev.c
> > > >> index 88fa295..ddcc325 100644
> > > >> --- a/drivers/net/ethernet/sfc/ef100_netdev.c
> > > >> +++ b/drivers/net/ethernet/sfc/ef100_netdev.c
> > > >> @@ -218,6 +218,7 @@ netdev_tx_t __ef100_hard_start_xmit(struct sk_buff *skb,
> > > >> skb->len, skb->data_len, channel->channel);
> > > >> if (!efx->n_channels || !efx->n_tx_channels || !channel) {
> > > >> netif_stop_queue(net_dev);
> > > >> + dev_kfree_skb_any(skb);
> > > >> goto err;
> > > >> }
> > > >
> > > > ef100 doesn't release in __ef100_enqueue_skb() either. SKB shouldn't be
> > > > NULL or ERR at this stage.
> > >
> > > SKB shouldn't be NULL or ERR, so it can be freed. But this code looks weird.
> >
> > Please take a look __ef100_enqueue_skb() and see if it frees SKB on
> > error or not. If not, please fix it.
>
> That function looks ok to me, but I appreciate the extra eyes on it.
__ef100_enqueue_skb() has the following check in error path:
498 err:
499 efx_enqueue_unwind(tx_queue, old_insert_count);
500 if (!IS_ERR_OR_NULL(skb))
501 dev_kfree_skb_any(skb);
502
The issue is that skb is never error or null here and this "if" is
actually always true and can be deleted.
Thanks
Hello,
On Fri, 2022-11-18 at 19:11 +0200, Leon Romanovsky wrote:
> On Fri, Nov 18, 2022 at 09:15:43AM +0000, Martin Habets wrote:
> > On Thu, Nov 17, 2022 at 03:05:27PM +0200, Leon Romanovsky wrote:
> > > Please take a look __ef100_enqueue_skb() and see if it frees SKB on
> > > error or not. If not, please fix it.
> >
> > That function looks ok to me, but I appreciate the extra eyes on it.
>
> __ef100_enqueue_skb() has the following check in error path:
>
> 498 err:
> 499 efx_enqueue_unwind(tx_queue, old_insert_count);
> 500 if (!IS_ERR_OR_NULL(skb))
> 501 dev_kfree_skb_any(skb);
> 502
>
> The issue is that skb is never error or null here and this "if" is
> actually always true and can be deleted.
I think that such additional change could be suite for a different net-
next patch, while this -net patch could land as is, @Leon: do you
agree?
Thanks!
Paolo
On Tue, Nov 22, 2022 at 09:28:42AM +0100, Paolo Abeni wrote:
> Hello,
>
> On Fri, 2022-11-18 at 19:11 +0200, Leon Romanovsky wrote:
> > On Fri, Nov 18, 2022 at 09:15:43AM +0000, Martin Habets wrote:
> > > On Thu, Nov 17, 2022 at 03:05:27PM +0200, Leon Romanovsky wrote:
> > > > Please take a look __ef100_enqueue_skb() and see if it frees SKB on
> > > > error or not. If not, please fix it.
> > >
> > > That function looks ok to me, but I appreciate the extra eyes on it.
> >
> > __ef100_enqueue_skb() has the following check in error path:
> >
> > 498 err:
> > 499 efx_enqueue_unwind(tx_queue, old_insert_count);
> > 500 if (!IS_ERR_OR_NULL(skb))
> > 501 dev_kfree_skb_any(skb);
> > 502
> >
> > The issue is that skb is never error or null here and this "if" is
> > actually always true and can be deleted.
>
> I think that such additional change could be suite for a different net-
> next patch, while this -net patch could land as is, @Leon: do you
> agree?
>
Sure, thanks,
Reviewed-by: Leon Romanovsky <[email protected]>