Re: [PATCH net-next v1 2/3] inet: Avoid ehash lookup race in inet_ehash_insert()

[luoxuanqiang <xuanqiang.luo@linux.dev>]

在 2025/9/16 07:00, Kuniyuki Iwashima 写道:
> On Mon, Sep 15, 2025 at 12:04 AM <xuanqiang.luo@linux.dev> wrote:
>> From: Xuanqiang Luo <luoxuanqiang@kylinos.cn>
>>
>> Since ehash lookups are lockless, if one CPU performs a lookup while
>> another concurrently deletes and inserts (removing reqsk and inserting sk),
>> the lookup may fail to find the socket, an RST may be sent.
>>
>> The call trace map is drawn as follows:
>>     CPU 0                           CPU 1
>>     -----                           -----
>>                                  inet_ehash_insert()
>>                                  spin_lock()
>>                                  sk_nulls_del_node_init_rcu(osk)
>> __inet_lookup_established()
>>          (lookup failed)
>>                                  __sk_nulls_add_node_rcu(sk, list)
>>                                  spin_unlock()
>>
>> As both deletion and insertion operate on the same ehash chain, this patch
>> introduces two new sk_nulls_replace_* helper functions to implement atomic
>> replacement.
>>
>> If sk_nulls_replace_node_init_rcu() fails, it indicates osk is either
>> hlist_unhashed or hlist_nulls_unhashed. The former returns false; the
>> latter performs insertion without deletion.
>>
>> Fixes: 5e0724d027f0 ("tcp/dccp: fix hashdance race for passive sessions")
>> Signed-off-by: Xuanqiang Luo <luoxuanqiang@kylinos.cn>
>> ---
>>   include/net/sock.h         | 23 +++++++++++++++++++++++
>>   net/ipv4/inet_hashtables.c |  7 +++++++
>>   2 files changed, 30 insertions(+)
>>
>> diff --git a/include/net/sock.h b/include/net/sock.h
>> index 896bec2d2176..26dacf7bc93e 100644
>> --- a/include/net/sock.h
>> +++ b/include/net/sock.h
>> @@ -859,6 +859,29 @@ static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
>>          return rc;
>>   }
>>
>> +static inline bool __sk_nulls_replace_node_init_rcu(struct sock *old,
>> +                                                   struct sock *new)
>> +{
>> +       if (sk_hashed(old) &&
>> +           hlist_nulls_replace_init_rcu(&old->sk_nulls_node,
>> +                                        &new->sk_nulls_node))
>> +               return true;
>> +
>> +       return false;
>> +}
>> +
>> +static inline bool sk_nulls_replace_node_init_rcu(struct sock *old,
>> +                                                 struct sock *new)
>> +{
>> +       bool rc = __sk_nulls_replace_node_init_rcu(old, new);
>> +
>> +       if (rc) {
>> +               WARN_ON(refcount_read(&old->sk_refcnt) == 1);
>> +               __sock_put(old);
>> +       }
>> +       return rc;
>> +}
>> +
>>   static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
>>   {
>>          hlist_add_head(&sk->sk_node, list);
>> diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
>> index ef4ccfd46ff6..7803fd3cc8e9 100644
>> --- a/net/ipv4/inet_hashtables.c
>> +++ b/net/ipv4/inet_hashtables.c
>> @@ -685,6 +685,12 @@ bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
>>          spin_lock(lock);
>>          if (osk) {
>>                  WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
>> +               /* Since osk and sk should be in the same ehash bucket, try
>> +                * direct replacement to avoid lookup gaps. On failure, no
>> +                * changes. sk_nulls_del_node_init_rcu() will handle the rest.
> Both sk_nulls_replace_node_init_rcu() and
> sk_nulls_del_node_init_rcu() return true only when
> sk_hashed(osk) == true.
>
> Only thing sk_nulls_del_node_init_rcu() does is to
> set ret to false.
>
>
>> +                */
>> +               if (sk_nulls_replace_node_init_rcu(osk, sk))
>> +                       goto unlock;
>>                  ret = sk_nulls_del_node_init_rcu(osk);
> So, should we simply do
>
> ret = sk_nulls_replace_node_init_rcu(osk, sk);
> goto unlock;
>
> ?

sk_nulls_replace_node_init_rcu() only returns true if both 
sk_hashed(osk) == true and hlist_nulls_unhashed(old) == false.
However, in the original sk_nulls_del_node_init_rcu() logic, when 
sk_hashed(osk) == true, it always performs __sock_put(sk) regardless of 
the hlist_nulls_unhashed(old) check. Therefore, if 
sk_nulls_replace_node_init_rcu() fails, we can safely let ret or 
__sock_put(sk) be handled by the subsequent 
sk_nulls_del_node_init_rcu(osk) call. Thanks Xuanqiang.

>
>>          } else if (found_dup_sk) {
>>                  *found_dup_sk = inet_ehash_lookup_by_sk(sk, list);
>> @@ -695,6 +701,7 @@ bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
>>          if (ret)
>>                  __sk_nulls_add_node_rcu(sk, list);
>>
>> +unlock:
>>          spin_unlock(lock);
>>
>>          return ret;
>> --
>> 2.27.0
>>