dccp: don't restart ccid2_hc_tx_rto_expire() if sk in closed state
authorAlexey Kodanev <alexey.kodanev@oracle.com>
Fri, 26 Jan 2018 12:14:16 +0000 (15:14 +0300)
committerBen Hutchings <ben@decadent.org.uk>
Sat, 3 Mar 2018 15:51:03 +0000 (15:51 +0000)
commit dd5684ecae3bd8e44b644f50e2c12c7e57fdfef5 upstream.

ccid2_hc_tx_rto_expire() timer callback always restarts the timer
again and can run indefinitely (unless it is stopped outside), and after
commit 120e9dabaf55 ("dccp: defer ccid_hc_tx_delete() at dismantle time"),
which moved ccid_hc_tx_delete() (also includes sk_stop_timer()) from
dccp_destroy_sock() to sk_destruct(), this started to happen quite often.
The timer prevents releasing the socket, as a result, sk_destruct() won't
be called.

Found with LTP/dccp_ipsec tests running on the bonding device,
which later couldn't be unloaded after the tests were completed:

  unregister_netdevice: waiting for bond0 to become free. Usage count = 148

Fixes: 2a91aa396739 ("[DCCP] CCID2: Initial CCID2 (TCP-Like) implementation")
Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

index 67164bb..eaa9d3e 100644 (file)
@@ -140,6 +140,9 @@ static void ccid2_hc_tx_rto_expire(unsigned long data)
+       if (sk->sk_state == DCCP_CLOSED)
+               goto out;
        /* back-off timer */
        hc->tx_rto <<= 1;
        if (hc->tx_rto > DCCP_RTO_MAX)