Received-SPF: pass (google.com: domain of linux-kernel+bounces-142393-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
From: Junxian Huang <huangjunxian6@hisilicon.com>
To: <jgg@ziepe.ca>, <leon@kernel.org>
CC: <linux-rdma@vger.kernel.org>, <linuxarm@huawei.com>,
	<linux-kernel@vger.kernel.org>, <huangjunxian6@hisilicon.com>
Subject: [PATCH for-next 05/10] RDMA/hns: Fix UAF for cq async event
Date: Fri, 12 Apr 2024 17:16:11 +0800
Message-ID: <20240412091616.370789-6-huangjunxian6@hisilicon.com>
In-Reply-To: <20240412091616.370789-1-huangjunxian6@hisilicon.com>
References: <20240412091616.370789-1-huangjunxian6@hisilicon.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain

From: Chengchang Tang <tangchengchang@huawei.com>

The refcount of CQ is not protected by locks. When CQ asynchronous
events and CQ destruction are concurrent, CQ may have been released,
which will cause UAF.

Use the xa_lock() to protect the CQ refcount.

Fixes: 9a4435375cd1 ("IB/hns: Add driver files for hns RoCE driver")
Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
---
 drivers/infiniband/hw/hns/hns_roce_cq.c | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_cq.c b/drivers/infiniband/hw/hns/hns_roce_cq.c
index 7250d0643b5c..68e22f368d43 100644
--- a/drivers/infiniband/hw/hns/hns_roce_cq.c
+++ b/drivers/infiniband/hw/hns/hns_roce_cq.c
@@ -149,7 +149,7 @@ static int alloc_cqc(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq)
 		return ret;
 	}
 
-	ret = xa_err(xa_store(&cq_table->array, hr_cq->cqn, hr_cq, GFP_KERNEL));
+	ret = xa_err(xa_store_irq(&cq_table->array, hr_cq->cqn, hr_cq, GFP_KERNEL));
 	if (ret) {
 		ibdev_err(ibdev, "failed to xa_store CQ, ret = %d.\n", ret);
 		goto err_put;
@@ -163,7 +163,7 @@ static int alloc_cqc(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq)
 	return 0;
 
 err_xa:
-	xa_erase(&cq_table->array, hr_cq->cqn);
+	xa_erase_irq(&cq_table->array, hr_cq->cqn);
 err_put:
 	hns_roce_table_put(hr_dev, &cq_table->table, hr_cq->cqn);
 
@@ -182,7 +182,7 @@ static void free_cqc(struct hns_roce_dev *hr_dev, struct hns_roce_cq *hr_cq)
 		dev_err(dev, "DESTROY_CQ failed (%d) for CQN %06lx\n", ret,
 			hr_cq->cqn);
 
-	xa_erase(&cq_table->array, hr_cq->cqn);
+	xa_erase_irq(&cq_table->array, hr_cq->cqn);
 
 	/* Waiting interrupt process procedure carried out */
 	synchronize_irq(hr_dev->eq_table.eq[hr_cq->vector].irq);
@@ -476,13 +476,6 @@ void hns_roce_cq_event(struct hns_roce_dev *hr_dev, u32 cqn, int event_type)
 	struct ib_event event;
 	struct ib_cq *ibcq;
 
-	hr_cq = xa_load(&hr_dev->cq_table.array,
-			cqn & (hr_dev->caps.num_cqs - 1));
-	if (!hr_cq) {
-		dev_warn(dev, "async event for bogus CQ 0x%06x\n", cqn);
-		return;
-	}
-
 	if (event_type != HNS_ROCE_EVENT_TYPE_CQ_ID_INVALID &&
 	    event_type != HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR &&
 	    event_type != HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW) {
@@ -491,7 +484,16 @@ void hns_roce_cq_event(struct hns_roce_dev *hr_dev, u32 cqn, int event_type)
 		return;
 	}
 
-	refcount_inc(&hr_cq->refcount);
+	xa_lock(&hr_dev->cq_table.array);
+	hr_cq = xa_load(&hr_dev->cq_table.array,
+			cqn & (hr_dev->caps.num_cqs - 1));
+	if (hr_cq)
+		refcount_inc(&hr_cq->refcount);
+	xa_unlock(&hr_dev->cq_table.array);
+	if (!hr_cq) {
+		dev_warn(dev, "async event for bogus CQ 0x%06x\n", cqn);
+		return;
+	}
 
 	ibcq = &hr_cq->ib_cq;
 	if (ibcq->event_handler) {
-- 
2.30.0