2020-12-20 08:15:22

by Dinghao Liu

[permalink] [raw]
Subject: [PATCH] IB/sa: Fix memleak in ib_nl_make_request

When rdma_nl_multicast() fails, skb should be freed
just like when ibnl_put_msg() fails.

Signed-off-by: Dinghao Liu <[email protected]>
---
drivers/infiniband/core/sa_query.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index 89a831fa1885..8bd23b5cc913 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -873,8 +873,10 @@ static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
spin_lock_irqsave(&ib_nl_request_lock, flags);
ret = rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, gfp_flag);

- if (ret)
+ if (ret) {
+ nlmsg_free(skb);
goto out;
+ }

/* Put the request on the list.*/
delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
--
2.17.1


2020-12-27 07:18:52

by Leon Romanovsky

[permalink] [raw]
Subject: Re: [PATCH] IB/sa: Fix memleak in ib_nl_make_request

On Sun, Dec 20, 2020 at 04:13:14PM +0800, Dinghao Liu wrote:
> When rdma_nl_multicast() fails, skb should be freed
> just like when ibnl_put_msg() fails.

It is not so simple as you wrote in the description.

There are no other places in the linux kernel that free
SKBs after netlink_multicast() failure.

Thanks

>
> Signed-off-by: Dinghao Liu <[email protected]>
> ---
> drivers/infiniband/core/sa_query.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
> index 89a831fa1885..8bd23b5cc913 100644
> --- a/drivers/infiniband/core/sa_query.c
> +++ b/drivers/infiniband/core/sa_query.c
> @@ -873,8 +873,10 @@ static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
> spin_lock_irqsave(&ib_nl_request_lock, flags);
> ret = rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, gfp_flag);
>
> - if (ret)
> + if (ret) {
> + nlmsg_free(skb);
> goto out;
> + }
>
> /* Put the request on the list.*/
> delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
> --
> 2.17.1
>

2020-12-28 03:10:36

by Dinghao Liu

[permalink] [raw]
Subject: Re: Re: [PATCH] IB/sa: Fix memleak in ib_nl_make_request

> On Sun, Dec 20, 2020 at 04:13:14PM +0800, Dinghao Liu wrote:
> > When rdma_nl_multicast() fails, skb should be freed
> > just like when ibnl_put_msg() fails.
>
> It is not so simple as you wrote in the description.
>
> There are no other places in the linux kernel that free
> SKBs after netlink_multicast() failure.
>

It's clear for me, thanks.

Regards,
Dinghao