tcp: Don't acquire inet_listen_hashbucket::lock with disabled BH.
authorSebastian Andrzej Siewior <bigeasy@linutronix.de>
Wed, 9 Feb 2022 18:56:57 +0000 (19:56 +0100)
committerJakub Kicinski <kuba@kernel.org>
Thu, 10 Feb 2022 05:28:36 +0000 (21:28 -0800)
commit4f9bf2a2f5aacf988e6d5e56b961ba45c5a25248
tree25f6d2c15f9b599dfef7d55201d0a6077d2597dd
parent1127170d457eb9bcc839ef7f2064634f92fe83e2
tcp: Don't acquire inet_listen_hashbucket::lock with disabled BH.

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(&ilb2->lock) + local_bh_disable():
  tcp_seq_next()
    listening_get_next()
      spin_lock(&ilb2->lock);
acquire(&ilb2->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>
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
Link: https://lore.kernel.org/r/YgQOebeZ10eNx1W6@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/ipv4/inet_hashtables.c
net/ipv6/inet6_hashtables.c