Thread (4 messages) 4 messages, 2 authors, 2023-01-14

Re: [PATCH v2] inet: fix fast path in __inet_hash_connect()

From: Eric Dumazet <edumazet@google.com>
Date: 2023-01-13 12:22:38
Also in: lkml

On Fri, Jan 13, 2023 at 12:40 PM Pietro Borrello
[off-list ref] wrote:
quoted hunk ↗ jump to hunk
__inet_hash_connect() has a fast path taken if sk_head(&tb->owners) is
equal to the sk parameter.
sk_head() returns the hlist_entry() with respect to the sk_node field.
However entries in the tb->owners list are inserted with respect to the
sk_bind_node field with sk_add_bind_node().
Thus the check would never pass and the fast path never execute.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reviewed-by: Kuniyuki Iwashima <redacted>
Signed-off-by: Pietro Borrello <redacted>
---
Changes in v2:
- nit: s/list_entry/hlist_entry/
- Link to v1: https://lore.kernel.org/r/20230112-inet_hash_connect_bind_head-v1-1-7e3c770157c8@diag.uniroma1.it (local)
---
 include/net/sock.h         | 10 ++++++++++
 net/ipv4/inet_hashtables.c |  2 +-
 2 files changed, 11 insertions(+), 1 deletion(-)
diff --git a/include/net/sock.h b/include/net/sock.h
index dcd72e6285b2..23fc403284db 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -860,6 +860,16 @@ static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_hea
        __sk_nulls_add_node_rcu(sk, list);
 }

+static inline struct sock *__sk_bind_head(const struct hlist_head *head)
+{
+       return hlist_entry(head->first, struct sock, sk_bind_node);
+}
+
+static inline struct sock *sk_bind_head(const struct hlist_head *head)
+{
+       return hlist_empty(head) ? NULL : __sk_bind_head(head);
+}
+
 static inline void __sk_del_bind_node(struct sock *sk)
 {
        __hlist_del(&sk->sk_bind_node);
diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index d039b4e732a3..a805e086fb48 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -998,7 +998,7 @@ int __inet_hash_connect(struct inet_timewait_death_row *death_row,
                                                  hinfo->bhash_size)];
                tb = inet_csk(sk)->icsk_bind_hash;
                spin_lock_bh(&head->lock);
-               if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
+               if (sk_bind_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
                        inet_ehash_nolisten(sk, NULL, NULL);
1) Given this path was never really used, we have no coverage.

2) Given that we do not check inet_ehash_nolisten() return code here.

I would recommend _not_ adding the Fixes: tag, and target net-next tree

In fact, I would remove this dead code, and reduce complexity.

I doubt the difference is going to be noticed.
(We have to access the ehash bucket anyway)

                        spin_unlock_bh(&head->lock);
                        return 0;

---
base-commit: 1b929c02afd37871d5afb9d498426f83432e71c2
change-id: 20230112-inet_hash_connect_bind_head-8f2dc98f08b1

Best regards,
--
Pietro Borrello [off-list ref]
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help