Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 15 Apr 2020 11:09:10 -0300
From:   Jason Gunthorpe <jgg@ziepe.ca>
To:     Xiyu Yang <xiyuyang19@fudan.edu.cn>
Cc:     Bernard Metzler <bmt@zurich.ibm.com>,
        Doug Ledford <dledford@redhat.com>, linux-rdma@vger.kernel.org,
        linux-kernel@vger.kernel.org, yuanxzhang@fudan.edu.cn,
        kjlu@umn.edu, Xin Tan <tanxin.ctf@gmail.com>
Subject: Re: [PATCH] RDMA/siw: Fix potential siw_mem refcnt leak in
 nr_add_node
Message-ID: <20200415140910.GN5100@ziepe.ca>
References: <1586939949-69856-1-git-send-email-xiyuyang19@fudan.edu.cn>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1586939949-69856-1-git-send-email-xiyuyang19@fudan.edu.cn>
User-Agent: Mutt/1.9.4 (2018-02-28)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

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@fudan.edu.cn>
> Signed-off-by: Xin Tan <tanxin.ctf@gmail.com>
>  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;
 	}
+
 	if (unlikely(mem->pd != pd)) {
 		pr_warn("siw: fastreg: PD mismatch\n");
 		rv = -EINVAL;