| Message ID | 20211206120216.mgo6qibl5fmzdcrp@linutronix.de (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net,v2] tcp: Don't acquire inet_listen_hashbucket::lock with disabled BH. | expand |
On Mon, Dec 06, 2021 at 01:02:16PM +0100, Sebastian Andrzej Siewior wrote: > Commit > 9652dc2eb9e40 ("tcp: relax listening_hash operations") > > removed the need to disable bottom half while acquiring > listening_hash.lock. There are still two callers left which disable > bottom half before the lock is acquired. > > On PREEMPT_RT the softirqs are preemptible and local_bh_disable() acts > as a lock to ensure that resources, that are protected by disabling > bottom halves, remain protected. > This leads to a circular locking dependency if the lock acquired with > disabled bottom halves is also acquired with enabled bottom halves > followed by disabling bottom halves. This is the reverse locking order. > It has been observed with inet_listen_hashbucket::lock: > > local_bh_disable() + spin_lock(&ilb->lock): > inet_listen() > inet_csk_listen_start() > sk->sk_prot->hash() := inet_hash() > local_bh_disable() > __inet_hash() > spin_lock(&ilb->lock); > acquire(&ilb->lock); > > Reverse order: spin_lock(&ilb->lock) + local_bh_disable(): > tcp_seq_next() > listening_get_next() > spin_lock(&ilb->lock); The net tree has already been using ilb2 instead of ilb. It does not change the problem though but updating the commit log will be useful to avoid future confusion. iiuc, established_get_next() does not hit this because it calls spin_lock_bh() which then keeps the local_bh_disable() => spin_lock() ordering? The patch lgtm. Acked-by: Martin KaFai Lau <kafai@fb.com> > acquire(&ilb->lock); > > tcp4_seq_show() > get_tcp4_sock() > sock_i_ino() > read_lock_bh(&sk->sk_callback_lock); > acquire(softirq_ctrl) // <---- whoops > acquire(&sk->sk_callback_lock) > > Drop local_bh_disable() around __inet_hash() which acquires > listening_hash->lock. Split inet_unhash() and acquire the > listen_hashbucket lock without disabling bottom halves; the inet_ehash > lock with disabled bottom halves.
On 2021-12-09 12:06:32 [-0800], Martin KaFai Lau wrote: > > local_bh_disable() + spin_lock(&ilb->lock): > > inet_listen() > > inet_csk_listen_start() > > sk->sk_prot->hash() := inet_hash() > > local_bh_disable() > > __inet_hash() > > spin_lock(&ilb->lock); > > acquire(&ilb->lock); > > > > Reverse order: spin_lock(&ilb->lock) + local_bh_disable(): > > tcp_seq_next() > > listening_get_next() > > spin_lock(&ilb->lock); > The net tree has already been using ilb2 instead of ilb. > It does not change the problem though but updating > the commit log will be useful to avoid future confusion. You think so? But having ilb2 and ilb might suggest that these two are different locks while they are the same. I could repost it early next week if you this actually confuses more… > iiuc, established_get_next() does not hit this because > it calls spin_lock_bh() which then keeps the > local_bh_disable() => spin_lock() ordering? Yes. BH was disabled before the spin_lock was acquired so it is the same ordering. The important part is not to disable BH after the spin_lock was acquired and in the reverse order. > The patch lgtm. > Acked-by: Martin KaFai Lau <kafai@fb.com> Sebastian
On Fri, Dec 10, 2021 at 05:21:13PM +0100, Sebastian Andrzej Siewior wrote: > On 2021-12-09 12:06:32 [-0800], Martin KaFai Lau wrote: > > > local_bh_disable() + spin_lock(&ilb->lock): > > > inet_listen() > > > inet_csk_listen_start() > > > sk->sk_prot->hash() := inet_hash() > > > local_bh_disable() > > > __inet_hash() > > > spin_lock(&ilb->lock); > > > acquire(&ilb->lock); > > > > > > Reverse order: spin_lock(&ilb->lock) + local_bh_disable(): > > > tcp_seq_next() > > > listening_get_next() > > > spin_lock(&ilb->lock); > > The net tree has already been using ilb2 instead of ilb. > > It does not change the problem though but updating > > the commit log will be useful to avoid future confusion. > > You think so? But having ilb2 and ilb might suggest that these two are > different locks while they are the same. I could repost it early next > week if you this actually confuses more… Yes, they are different locks. ilb2->lock is also taken in the inet_listen() path. ilb->lock is not even taken in the listening_get_next() side.
diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c index 75737267746f8..7bd1e10086f0a 100644 --- a/net/ipv4/inet_hashtables.c +++ b/net/ipv4/inet_hashtables.c @@ -637,7 +637,9 @@ int __inet_hash(struct sock *sk, struct sock *osk) int err = 0; if (sk->sk_state != TCP_LISTEN) { + local_bh_disable(); inet_ehash_nolisten(sk, osk, NULL); + local_bh_enable(); return 0; } WARN_ON(!sk_unhashed(sk)); @@ -669,45 +671,54 @@ int inet_hash(struct sock *sk) { int err = 0; - if (sk->sk_state != TCP_CLOSE) { - local_bh_disable(); + if (sk->sk_state != TCP_CLOSE) err = __inet_hash(sk, NULL); - local_bh_enable(); - } return err; } EXPORT_SYMBOL_GPL(inet_hash); -void inet_unhash(struct sock *sk) +static void __inet_unhash(struct sock *sk, struct inet_listen_hashbucket *ilb) { - struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; - struct inet_listen_hashbucket *ilb = NULL; - spinlock_t *lock; - if (sk_unhashed(sk)) return; - if (sk->sk_state == TCP_LISTEN) { - ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; - lock = &ilb->lock; - } else { - lock = inet_ehash_lockp(hashinfo, sk->sk_hash); - } - spin_lock_bh(lock); - if (sk_unhashed(sk)) - goto unlock; - if (rcu_access_pointer(sk->sk_reuseport_cb)) reuseport_stop_listen_sock(sk); if (ilb) { + struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; + inet_unhash2(hashinfo, sk); ilb->count--; } __sk_nulls_del_node_init_rcu(sk); sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); -unlock: - spin_unlock_bh(lock); +} + +void inet_unhash(struct sock *sk) +{ + struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; + + if (sk_unhashed(sk)) + return; + + if (sk->sk_state == TCP_LISTEN) { + struct inet_listen_hashbucket *ilb; + + ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)]; + /* Don't disable bottom halves while acquiring the lock to + * avoid circular locking dependency on PREEMPT_RT. + */ + spin_lock(&ilb->lock); + __inet_unhash(sk, ilb); + spin_unlock(&ilb->lock); + } else { + spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash); + + spin_lock_bh(lock); + __inet_unhash(sk, NULL); + spin_unlock_bh(lock); + } } EXPORT_SYMBOL_GPL(inet_unhash); diff --git a/net/ipv6/inet6_hashtables.c b/net/ipv6/inet6_hashtables.c index 67c9114835c84..0a2e7f2283911 100644 --- a/net/ipv6/inet6_hashtables.c +++ b/net/ipv6/inet6_hashtables.c @@ -333,11 +333,8 @@ int inet6_hash(struct sock *sk) { int err = 0; - if (sk->sk_state != TCP_CLOSE) { - local_bh_disable(); + if (sk->sk_state != TCP_CLOSE) err = __inet_hash(sk, NULL); - local_bh_enable(); - } return err; }
Commit 9652dc2eb9e40 ("tcp: relax listening_hash operations") removed the need to disable bottom half while acquiring listening_hash.lock. There are still two callers left which disable bottom half before the lock is acquired. On PREEMPT_RT the softirqs are preemptible and local_bh_disable() acts as a lock to ensure that resources, that are protected by disabling bottom halves, remain protected. This leads to a circular locking dependency if the lock acquired with disabled bottom halves is also acquired with enabled bottom halves followed by disabling bottom halves. This is the reverse locking order. It has been observed with inet_listen_hashbucket::lock: local_bh_disable() + spin_lock(&ilb->lock): inet_listen() inet_csk_listen_start() sk->sk_prot->hash() := inet_hash() local_bh_disable() __inet_hash() spin_lock(&ilb->lock); acquire(&ilb->lock); Reverse order: spin_lock(&ilb->lock) + local_bh_disable(): tcp_seq_next() listening_get_next() spin_lock(&ilb->lock); acquire(&ilb->lock); tcp4_seq_show() get_tcp4_sock() sock_i_ino() read_lock_bh(&sk->sk_callback_lock); acquire(softirq_ctrl) // <---- whoops acquire(&sk->sk_callback_lock) Drop local_bh_disable() around __inet_hash() which acquires listening_hash->lock. Split inet_unhash() and acquire the listen_hashbucket lock without disabling bottom halves; the inet_ehash lock with disabled bottom halves. Reported-by: Mike Galbraith <efault@gmx.de> Fixes: 9652dc2eb9e40 ("tcp: relax listening_hash operations") Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: https://lkml.kernel.org/r/12d6f9879a97cd56c09fb53dee343cbb14f7f1f7.camel@gmx.de Link: https://lkml.kernel.org/r/X9CheYjuXWc75Spa@hirez.programming.kicks-ass.net Acked-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> --- On 2021-12-03 18:23:20 [-0800], Jakub Kicinski wrote: > On Fri, 3 Dec 2021 10:39:34 +0900 Kuniyuki Iwashima wrote: > > I think this patch is for the net tree and needs a Fixes: tag of the commit > > mentioned in the description. > > The patch itself looks good to me. > > > > Acked-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> > > Makes sense, please repost for net with the Fixes tag and CC Eric. v1…v2: Reposted with fixes and net-tree as requested. Please keep in mind that this only effects the PREEMPT_RT preemption model and I'm posting this as part of the merging efforts. Therefore I didn't add the Fixes: tag and used net-next as I didn't expect any -stable backports (but then Greg sometimes backports RT-only patches since "it makes the life of some folks easier" as he puts it). net/ipv4/inet_hashtables.c | 53 ++++++++++++++++++++++--------------- net/ipv6/inet6_hashtables.c | 5 +--- 2 files changed, 33 insertions(+), 25 deletions(-)