From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ding Tianhong Subject: Re: [PATCH v4 net-next 01/12] bonding: remove bond->lock from bond_arp_rcv Date: Tue, 18 Feb 2014 15:07:46 +0800 Message-ID: <530306C2.1050003@huawei.com> References: <1392648088-21336-1-git-send-email-vfalico@redhat.com> <1392648088-21336-2-git-send-email-vfalico@redhat.com> <5302DB61.4070505@huawei.com> <20140218061222.GA18373@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: , Jay Vosburgh , Andy Gospodarek To: Veaceslav Falico Return-path: Received: from szxga01-in.huawei.com ([119.145.14.64]:29077 "EHLO szxga01-in.huawei.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753605AbaBRHIJ (ORCPT ); Tue, 18 Feb 2014 02:08:09 -0500 In-Reply-To: <20140218061222.GA18373@redhat.com> Sender: netdev-owner@vger.kernel.org List-ID: On 2014/2/18 14:12, Veaceslav Falico wrote: > On Tue, Feb 18, 2014 at 12:02:41PM +0800, Ding Tianhong wrote: >> On 2014/2/17 22:41, Veaceslav Falico wrote: >>> We're always called with rcu_read_lock() held (bond_arp_rcv() is on= ly >>> called from bond_handle_frame(), which is rx_handler and always cal= led >>> under rcu from __netif_receive_skb_core() ). >>> >>> The slave active/passive and/or bonding params can change in-flight= , however >>> we don't really care about that - we only modify the last time pack= et was >>> received, which is harmless. >>> >>> CC: Jay Vosburgh >>> CC: Andy Gospodarek >>> Signed-off-by: Veaceslav Falico >>> --- >>> drivers/net/bonding/bond_main.c | 3 --- >>> 1 file changed, 3 deletions(-) >>> >>> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/= bond_main.c >>> index 3bce855..3c50bec 100644 >>> --- a/drivers/net/bonding/bond_main.c >>> +++ b/drivers/net/bonding/bond_main.c >>> @@ -2260,8 +2260,6 @@ int bond_arp_rcv(const struct sk_buff *skb, s= truct bonding *bond, >>> if (skb->protocol !=3D __cpu_to_be16(ETH_P_ARP)) >>> return RX_HANDLER_ANOTHER; >>> >>> - read_lock(&bond->lock); >>> - >>> if (!slave_do_arp_validate(bond, slave)) >>> goto out_unlock; >>> >>> @@ -2318,7 +2316,6 @@ int bond_arp_rcv(const struct sk_buff *skb, s= truct bonding *bond, >>> bond_validate_arp(bond, slave, tip, sip); >>> >>> out_unlock: >>> - read_unlock(&bond->lock); >>> if (arp !=3D (struct arphdr *)skb->data) >>> kfree(arp); >>> return RX_HANDLER_ANOTHER; >>> >> >> I think it is not enough, you should add rcu_dereference for bond->c= urr_active_slave, it may be changed during >> the recv processing. >=20 > bond->lock has absolutely nothing to du with bond->curr_active_slave = =2E >=20 Yep, this problem is introduced by commit aeea64ac7=EF=BC=8C there is n= o way to protect the curr_active_slave, so I think you could fix it in this patch together. =20 else if (bond->curr_active_slave && time_after(slave_last_rx(bond, bond->curr_active_slave), bond->curr_active_slave->jiffies)) >> >> Regards >> Ding >> >> >=20 >=20