[Patch net] net: explain the lockdep annotations for dev_uc_unsync()

[Patch net] net: explain the lockdep annotations for dev_uc_unsync()

[Cong Wang]

The lockdep annotations for dev_uc_unsync() and dev_mc_unsync()
are not easy to understand, so add some comments to explain
why they are correct.

Similar for the rest netif_addr_lock_bh() cases, they don't
need nested version.

Cc: Taehee Yoo <ap420073@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
---
 net/core/dev_addr_lists.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/net/core/dev_addr_lists.c b/net/core/dev_addr_lists.c
index 6393ba930097..54cd568e7c2f 100644
--- a/net/core/dev_addr_lists.c
+++ b/net/core/dev_addr_lists.c
@@ -690,6 +690,15 @@ void dev_uc_unsync(struct net_device *to, struct net_device *from)
 	if (to->addr_len != from->addr_len)
 		return;
 
+	/* netif_addr_lock_bh() uses lockdep subclass 0, this is okay for two
+	 * reasons:
+	 * 1) This is always called without any addr_list_lock, so as the
+	 *    outermost one here, it must be 0.
+	 * 2) This is called by some callers after unlinking the upper device,
+	 *    so the dev->lower_level becomes 1 again.
+	 * Therefore, the subclass for 'from' is 0, for 'to' is either 1 or
+	 * larger.
+	 */
 	netif_addr_lock_bh(from);
 	netif_addr_lock_nested(to);
 	__hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
@@ -911,6 +920,7 @@ void dev_mc_unsync(struct net_device *to, struct net_device *from)
 	if (to->addr_len != from->addr_len)
 		return;
 
+	/* See the above comments inside dev_uc_unsync(). */
 	netif_addr_lock_bh(from);
 	netif_addr_lock_nested(to);
 	__hw_addr_unsync(&to->mc, &from->mc, to->addr_len);
-- 
2.27.0

Re: [Patch net] net: explain the lockdep annotations for dev_uc_unsync()

[David Miller]

From: Cong Wang <xiyou.wangcong@gmail.com>
Date: Fri, 26 Jun 2020 11:25:27 -0700

> The lockdep annotations for dev_uc_unsync() and dev_mc_unsync()
> are not easy to understand, so add some comments to explain
> why they are correct.
> 
> Similar for the rest netif_addr_lock_bh() cases, they don't
> need nested version.
> 
> Cc: Taehee Yoo <ap420073@gmail.com>
> Cc: Dmitry Vyukov <dvyukov@google.com>
> Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>

Applied.