| Message ID | 1668416136-33530-1-git-send-email-zhangchangzhong@huawei.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net] net: nixge: fix potential memory leak in nixge_start_xmit() | expand |
Zhang Changzhong <zhangchangzhong@huawei.com> : > The nixge_start_xmit() returns NETDEV_TX_OK but does not free skb on two > error handling cases, which can lead to memory leak. > > To fix this, return NETDEV_TX_BUSY in case of nixge_check_tx_bd_space() > fails and add dev_kfree_skb_any() in case of dma_map_single() fails. This patch merge two unrelated changes. Please split. > Fixes: 492caffa8a1a ("net: ethernet: nixge: Add support for National Instruments XGE netdev") > Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> > --- > drivers/net/ethernet/ni/nixge.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/ethernet/ni/nixge.c b/drivers/net/ethernet/ni/nixge.c > index 19d043b593cc..b9091f9bbc77 100644 > --- a/drivers/net/ethernet/ni/nixge.c > +++ b/drivers/net/ethernet/ni/nixge.c > @@ -521,13 +521,15 @@ static netdev_tx_t nixge_start_xmit(struct sk_buff *skb, > if (nixge_check_tx_bd_space(priv, num_frag)) { > if (!netif_queue_stopped(ndev)) > netif_stop_queue(ndev); > - return NETDEV_TX_OK; > + return NETDEV_TX_BUSY; > } The driver should probably check the available room before returning from hard_start_xmit and turn the check above unlikely(). Btw there is no lock and the Tx completion is irq driven: the driver is racy. :o(
On 2022/11/14 18:40, Francois Romieu wrote: > Zhang Changzhong <zhangchangzhong@huawei.com> : >> The nixge_start_xmit() returns NETDEV_TX_OK but does not free skb on two >> error handling cases, which can lead to memory leak. >> >> To fix this, return NETDEV_TX_BUSY in case of nixge_check_tx_bd_space() >> fails and add dev_kfree_skb_any() in case of dma_map_single() fails. > > This patch merge two unrelated changes. Please split. > >> Fixes: 492caffa8a1a ("net: ethernet: nixge: Add support for National Instruments XGE netdev") >> Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> >> --- >> drivers/net/ethernet/ni/nixge.c | 6 ++++-- >> 1 file changed, 4 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/net/ethernet/ni/nixge.c b/drivers/net/ethernet/ni/nixge.c >> index 19d043b593cc..b9091f9bbc77 100644 >> --- a/drivers/net/ethernet/ni/nixge.c >> +++ b/drivers/net/ethernet/ni/nixge.c >> @@ -521,13 +521,15 @@ static netdev_tx_t nixge_start_xmit(struct sk_buff *skb, >> if (nixge_check_tx_bd_space(priv, num_frag)) { >> if (!netif_queue_stopped(ndev)) >> netif_stop_queue(ndev); >> - return NETDEV_TX_OK; >> + return NETDEV_TX_BUSY; >> } > > The driver should probably check the available room before returning > from hard_start_xmit and turn the check above unlikely(). > > Btw there is no lock and the Tx completion is irq driven: the driver > is racy. :o( > Hi Francois, Thanks for you review. I'll make v2 according to your suggestion. Changzhong
On 15 Nov 21:20, Zhang Changzhong wrote: >On 2022/11/14 18:40, Francois Romieu wrote: >> Zhang Changzhong <zhangchangzhong@huawei.com> : >>> The nixge_start_xmit() returns NETDEV_TX_OK but does not free skb on two >>> error handling cases, which can lead to memory leak. >>> >>> To fix this, return NETDEV_TX_BUSY in case of nixge_check_tx_bd_space() >>> fails and add dev_kfree_skb_any() in case of dma_map_single() fails. >> >> This patch merge two unrelated changes. Please split. >> >>> Fixes: 492caffa8a1a ("net: ethernet: nixge: Add support for National Instruments XGE netdev") >>> Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> >>> --- >>> drivers/net/ethernet/ni/nixge.c | 6 ++++-- >>> 1 file changed, 4 insertions(+), 2 deletions(-) >>> >>> diff --git a/drivers/net/ethernet/ni/nixge.c b/drivers/net/ethernet/ni/nixge.c >>> index 19d043b593cc..b9091f9bbc77 100644 >>> --- a/drivers/net/ethernet/ni/nixge.c >>> +++ b/drivers/net/ethernet/ni/nixge.c >>> @@ -521,13 +521,15 @@ static netdev_tx_t nixge_start_xmit(struct sk_buff *skb, >>> if (nixge_check_tx_bd_space(priv, num_frag)) { >>> if (!netif_queue_stopped(ndev)) >>> netif_stop_queue(ndev); >>> - return NETDEV_TX_OK; >>> + return NETDEV_TX_BUSY; >>> } >> >> The driver should probably check the available room before returning >> from hard_start_xmit and turn the check above unlikely(). >> >> Btw there is no lock and the Tx completion is irq driven: the driver >> is racy. :o( >> > >Hi Francois, > >Thanks for you review. I'll make v2 according to your suggestion. > you will probably need to check out: Transmit path guidelines: https://www.kernel.org/doc/Documentation/networking/driver.rst
On 2022/11/17 6:36, Saeed Mahameed wrote: > On 15 Nov 21:20, Zhang Changzhong wrote: >> On 2022/11/14 18:40, Francois Romieu wrote: >>> Zhang Changzhong <zhangchangzhong@huawei.com> : >>>> The nixge_start_xmit() returns NETDEV_TX_OK but does not free skb on two >>>> error handling cases, which can lead to memory leak. >>>> >>>> To fix this, return NETDEV_TX_BUSY in case of nixge_check_tx_bd_space() >>>> fails and add dev_kfree_skb_any() in case of dma_map_single() fails. >>> >>> This patch merge two unrelated changes. Please split. >>> >>>> Fixes: 492caffa8a1a ("net: ethernet: nixge: Add support for National Instruments XGE netdev") >>>> Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> >>>> --- >>>> drivers/net/ethernet/ni/nixge.c | 6 ++++-- >>>> 1 file changed, 4 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/drivers/net/ethernet/ni/nixge.c b/drivers/net/ethernet/ni/nixge.c >>>> index 19d043b593cc..b9091f9bbc77 100644 >>>> --- a/drivers/net/ethernet/ni/nixge.c >>>> +++ b/drivers/net/ethernet/ni/nixge.c >>>> @@ -521,13 +521,15 @@ static netdev_tx_t nixge_start_xmit(struct sk_buff *skb, >>>> if (nixge_check_tx_bd_space(priv, num_frag)) { >>>> if (!netif_queue_stopped(ndev)) >>>> netif_stop_queue(ndev); >>>> - return NETDEV_TX_OK; >>>> + return NETDEV_TX_BUSY; >>>> } >>> >>> The driver should probably check the available room before returning >>> from hard_start_xmit and turn the check above unlikely(). >>> >>> Btw there is no lock and the Tx completion is irq driven: the driver >>> is racy. :o( >>> >> >> Hi Francois, >> >> Thanks for you review. I'll make v2 according to your suggestion. >> > > you will probably need to check out: Transmit path guidelines: > https://www.kernel.org/doc/Documentation/networking/driver.rst > Thank! This document is very helpful. > .
diff --git a/drivers/net/ethernet/ni/nixge.c b/drivers/net/ethernet/ni/nixge.c index 19d043b593cc..b9091f9bbc77 100644 --- a/drivers/net/ethernet/ni/nixge.c +++ b/drivers/net/ethernet/ni/nixge.c @@ -521,13 +521,15 @@ static netdev_tx_t nixge_start_xmit(struct sk_buff *skb, if (nixge_check_tx_bd_space(priv, num_frag)) { if (!netif_queue_stopped(ndev)) netif_stop_queue(ndev); - return NETDEV_TX_OK; + return NETDEV_TX_BUSY; } cur_phys = dma_map_single(ndev->dev.parent, skb->data, skb_headlen(skb), DMA_TO_DEVICE); - if (dma_mapping_error(ndev->dev.parent, cur_phys)) + if (dma_mapping_error(ndev->dev.parent, cur_phys)) { + dev_kfree_skb_any(skb); goto drop; + } nixge_hw_dma_bd_set_phys(cur_p, cur_phys); cur_p->cntrl = skb_headlen(skb) | XAXIDMA_BD_CTRL_TXSOF_MASK;
The nixge_start_xmit() returns NETDEV_TX_OK but does not free skb on two error handling cases, which can lead to memory leak. To fix this, return NETDEV_TX_BUSY in case of nixge_check_tx_bd_space() fails and add dev_kfree_skb_any() in case of dma_map_single() fails. Fixes: 492caffa8a1a ("net: ethernet: nixge: Add support for National Instruments XGE netdev") Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> --- drivers/net/ethernet/ni/nixge.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)