RE: [PATCH] RDMA/siw: Fix potential siw_mem refcnt leak in nr_add_node

From: Bernard Metzler
Date: Wed Apr 15 2020 - 10:17:18 EST


-----"Jason Gunthorpe" <jgg@xxxxxxxx> wrote: -----

>To: "Xiyu Yang" <xiyuyang19@xxxxxxxxxxxx>
>From: "Jason Gunthorpe" <jgg@xxxxxxxx>
>Date: 04/15/2020 04:09PM
>Cc: "Bernard Metzler" <bmt@xxxxxxxxxxxxxx>, "Doug Ledford"
><dledford@xxxxxxxxxx>, linux-rdma@xxxxxxxxxxxxxxx,
>linux-kernel@xxxxxxxxxxxxxxx, yuanxzhang@xxxxxxxxxxxx, kjlu@xxxxxxx,
>"Xin Tan" <tanxin.ctf@xxxxxxxxx>
>Subject: [EXTERNAL] Re: [PATCH] RDMA/siw: Fix potential siw_mem
>refcnt leak in nr_add_node
>
>On Wed, Apr 15, 2020 at 04:39:08PM +0800, Xiyu Yang wrote:
>> siw_fastreg_mr() invokes siw_mem_id2obj(), which returns a local
>> reference of the siw_mem object to "mem" with increased refcnt.
>> When siw_fastreg_mr() returns, "mem" becomes invalid, so the
>refcount
>> should be decreased to keep refcount balanced.
>>
>> The issue happens in one error path of siw_fastreg_mr(). When
>"base_mr"
>> equals to NULL but "mem" is not NULL, the function forgets to
>decrease
>> the refcnt increased by siw_mem_id2obj() and causes a refcnt leak.
>>
>> Fix this issue by calling siw_mem_put() on this error path when mem
>is
>> not NULL.
>>
>> Signed-off-by: Xiyu Yang <xiyuyang19@xxxxxxxxxxxx>
>> Signed-off-by: Xin Tan <tanxin.ctf@xxxxxxxxx>
>> drivers/infiniband/sw/siw/siw_qp_tx.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/infiniband/sw/siw/siw_qp_tx.c
>b/drivers/infiniband/sw/siw/siw_qp_tx.c
>> index ae92c8080967..86044a44b83b 100644
>> +++ b/drivers/infiniband/sw/siw/siw_qp_tx.c
>> @@ -926,6 +926,8 @@ static int siw_fastreg_mr(struct ib_pd *pd,
>struct siw_sqe *sqe)
>> siw_dbg_pd(pd, "STag 0x%08x\n", sqe->rkey);
>>
>> if (unlikely(!mem || !base_mr)) {
>> + if (mem)
>> + siw_mem_put(mem);
>> pr_warn("siw: fastreg: STag 0x%08x unknown\n", sqe->rkey);
>> return -EINVAL;
>> }
>
>I think I prefer this version, which is what I'll use if nobody has
>concerns:
>
>diff --git a/drivers/infiniband/sw/siw/siw_qp_tx.c
>b/drivers/infiniband/sw/siw/siw_qp_tx.c
>index ae92c8080967c5..0580bbf535ceb7 100644
>--- a/drivers/infiniband/sw/siw/siw_qp_tx.c
>+++ b/drivers/infiniband/sw/siw/siw_qp_tx.c
>@@ -920,20 +920,28 @@ static int siw_fastreg_mr(struct ib_pd *pd,
>struct siw_sqe *sqe)
> {
> struct ib_mr *base_mr = (struct ib_mr *)(uintptr_t)sqe->base_mr;
> struct siw_device *sdev = to_siw_dev(pd->device);
>- struct siw_mem *mem = siw_mem_id2obj(sdev, sqe->rkey >> 8);
>+ struct siw_mem *mem;
> int rv = 0;
>
> siw_dbg_pd(pd, "STag 0x%08x\n", sqe->rkey);
>
>- if (unlikely(!mem || !base_mr)) {
>+ if (unlikely(!base_mr)) {
> pr_warn("siw: fastreg: STag 0x%08x unknown\n", sqe->rkey);
> return -EINVAL;
> }
>+
> if (unlikely(base_mr->rkey >> 8 != sqe->rkey >> 8)) {
> pr_warn("siw: fastreg: STag 0x%08x: bad MR\n", sqe->rkey);
>+ return -EINVAL;
>+ }
>+
>+ mem = siw_mem_id2obj(sdev, sqe->rkey >> 8);
>+ if (unlikely(!mem)) {
>+ pr_warn("siw: fastreg: STag 0x%08x unknown\n", sqe->rkey);
> rv = -EINVAL;
> goto out;
> }
>+

Fine with me in principle, but we would have to return
directly here as well - since we do not have a valid mem
to be put back.

Thanks
Bernard.