diff mbox series

ath11k: Fix potential RCU dereference issue in ath11k_debugfs_htt_ext_stats_handler

Message ID tencent_A64CA96B962349E369B349EA01EBC53C3505@qq.com (mailing list archive)
State Rejected
Delegated to: Kalle Valo
Headers show
Series ath11k: Fix potential RCU dereference issue in ath11k_debugfs_htt_ext_stats_handler | expand

Commit Message

Jiawei Ye Aug. 30, 2024, 12:02 p.m. UTC
In the `ath11k_debugfs_htt_ext_stats_handler` function, the `ar` pointer
obtained via RCU lock is accessed after the RCU read-side critical
section might be unlocked. According to RCU usage rules, this is illegal.
Reusing this pointer can lead to unpredictable behavior, including
accessing memory that has been updated or causing use-after-free issues.
The `ath12k_debugfs_htt_ext_stats_handler` function in the
`drivers/net/wireless/ath/ath12k/debugfs_htt_stats.c` file provides a good
example to follow for addressing this issue.

This possible bug was identified using a static analysis tool developed
by myself, specifically designed to detect RCU-related issues.

To address this issue, the RCU read lock is now kept until all accesses
to the `ar` pointer are completed. A `goto exit` statement is introduced
to ensure that the RCU read unlock is called appropriately, regardless of
the function's exit path.

Signed-off-by: Jiawei Ye <jiawei.ye@foxmail.com>
---
 drivers/net/wireless/ath/ath11k/debugfs_htt_stats.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Comments

Jeff Johnson Aug. 30, 2024, 4:01 p.m. UTC | #1
On 8/30/2024 5:02 AM, Jiawei Ye wrote:
> In the `ath11k_debugfs_htt_ext_stats_handler` function, the `ar` pointer
> obtained via RCU lock is accessed after the RCU read-side critical
> section might be unlocked. According to RCU usage rules, this is illegal.
> Reusing this pointer can lead to unpredictable behavior, including
> accessing memory that has been updated or causing use-after-free issues.
> The `ath12k_debugfs_htt_ext_stats_handler` function in the
> `drivers/net/wireless/ath/ath12k/debugfs_htt_stats.c` file provides a good
> example to follow for addressing this issue.
> 
> This possible bug was identified using a static analysis tool developed
> by myself, specifically designed to detect RCU-related issues.
> 
> To address this issue, the RCU read lock is now kept until all accesses
> to the `ar` pointer are completed. A `goto exit` statement is introduced
> to ensure that the RCU read unlock is called appropriately, regardless of
> the function's exit path.

This analysis is incorrect since ar is not an RCU-protected structure

The rcu_read_lock() is required internally within
ath11k_mac_get_ar_by_pdev_id() when retrieving the RCU-protected pdev pointer.

So NAK this patch.
diff mbox series

Patch

diff --git a/drivers/net/wireless/ath/ath11k/debugfs_htt_stats.c b/drivers/net/wireless/ath/ath11k/debugfs_htt_stats.c
index 870e86a31bf8..325377e00818 100644
--- a/drivers/net/wireless/ath/ath11k/debugfs_htt_stats.c
+++ b/drivers/net/wireless/ath/ath11k/debugfs_htt_stats.c
@@ -4572,15 +4572,14 @@  void ath11k_debugfs_htt_ext_stats_handler(struct ath11k_base *ab,
 	pdev_id = FIELD_GET(HTT_STATS_COOKIE_LSB, cookie);
 	rcu_read_lock();
 	ar = ath11k_mac_get_ar_by_pdev_id(ab, pdev_id);
-	rcu_read_unlock();
 	if (!ar) {
 		ath11k_warn(ab, "failed to get ar for pdev_id %d\n", pdev_id);
-		return;
+		goto exit;
 	}
 
 	stats_req = ar->debug.htt_stats.stats_req;
 	if (!stats_req)
-		return;
+		goto exit;
 
 	spin_lock_bh(&ar->debug.htt_stats.lock);
 
@@ -4599,6 +4598,8 @@  void ath11k_debugfs_htt_ext_stats_handler(struct ath11k_base *ab,
 
 	if (send_completion)
 		complete(&stats_req->cmpln);
+exit:
+	rcu_read_unlock();
 }
 
 static ssize_t ath11k_read_htt_stats_type(struct file *file,