netdev - Re: [PATCH net 1/3] bonding: access curr_active_slave with rtnl

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Sat, 10 Dec 2022 18:41:01 +0100
From:   Eric Dumazet <edumazet@...gle.com>
To:     Hangbin Liu <liuhangbin@...il.com>
Cc:     netdev@...r.kernel.org, Jay Vosburgh <j.vosburgh@...il.com>,
        "David S . Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Jonathan Toppins <jtoppins@...hat.com>,
        Paolo Abeni <pabeni@...hat.com>, liali <liali@...hat.com>
Subject: Re: [PATCH net 1/3] bonding: access curr_active_slave with rtnl_dereference

On Sat, Dec 10, 2022 at 1:28 PM Hangbin Liu <liuhangbin@...il.com> wrote:
>
> On Sat, Dec 10, 2022 at 12:58:59AM +0100, Eric Dumazet wrote:
> > On Fri, Dec 9, 2022 at 11:13 AM Hangbin Liu <liuhangbin@...il.com> wrote:
> > >
> > > Looks commit 4740d6382790 ("bonding: add proper __rcu annotation for
> > > curr_active_slave") missed rtnl_dereference for curr_active_slave
> > > in bond_miimon_commit().
> > >
> > > Fixes: 4740d6382790 ("bonding: add proper __rcu annotation for curr_active_slave")
> >
> >
> > > Signed-off-by: Hangbin Liu <liuhangbin@...il.com>
> > > ---
> > >  drivers/net/bonding/bond_main.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> > > index b9a882f182d2..2b6cc4dbb70e 100644
> > > --- a/drivers/net/bonding/bond_main.c
> > > +++ b/drivers/net/bonding/bond_main.c
> > > @@ -2689,7 +2689,7 @@ static void bond_miimon_commit(struct bonding *bond)
> > >
> > >                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
> > >
> > > -                       if (!bond->curr_active_slave || slave == primary)
> > > +                       if (!rtnl_dereference(bond->curr_active_slave) || slave == primary)
> >
> > We do not dereference the pointer here.
> >
> > If this is fixing a sparse issue, then use the correct RCU helper for this.
> >
> > ( rcu_access_pointer())
>
> Hmm... I saw in 4740d6382790 ("bonding: add proper __rcu annotation for
>  curr_active_slave") there are also some dereference like that. Should I also
> fix them at the same time? e.g.

There is no 'fix' really. I do not see any reason to change this part.

It is merely a matter of repeating or not the fact that RTNL (or
another lock) is held.

>
> @@ -2607,8 +2612,8 @@ static void bond_ab_arp_commit(struct bonding *bond)
>
>                 case BOND_LINK_UP:
>                         trans_start = dev_trans_start(slave->dev);
> -                       if (bond->curr_active_slave != slave ||
> -                           (!bond->curr_active_slave &&
> +                       if (rtnl_dereference(bond->curr_active_slave) != slave ||
> +                           (!rtnl_dereference(bond->curr_active_slave) &&
>

At the time of commit  4740d6382790 we wanted to make sure the
bond->curr_slave_lock
was taken, because it was the assertion at that time.

Then later, bond_deref_active_protected() has been removed, because
curr_slave_lock has been removed.

$ git log --oneline --reverse
b25bd2515ea32cf5ddd5fd5a2a93b8c9dd875e4f..8c0bc550288d81e9ad8a2ed9136a72140b9ef507
86e749866d7c6b0ee1f9377cf7142f2690596a05 bonding: 3ad: clean up
curr_slave_lock usage
62c5f5185397f4bd8e5defe6fcb86420deeb2b38 bonding: alb: remove curr_slave_lock
1c72cfdc96e63bf975cab514c4ca4d8a661ba0e6 bonding: clean curr_slave_lock use
b743562819bd97cc7c282e870896bae8016b64b5 bonding: convert
curr_slave_lock to a spinlock and rename it
4bab16d7c97498e91564231b922d49f52efaf7d4 bonding: alb: convert to
bond->mode_lock
e470259fa1bd7ce5a375b16c5ec97cc0e83b058d bonding: 3ad: convert to
bond->mode_lock
8c0bc550288d81e9ad8a2ed9136a72140b9ef507 bonding: adjust locking comments


Now, you post a patch for bond_miimon_commit() which already has :

if (slave == rcu_access_pointer(bond->curr_active_slave))
      goto do_failover;

So really it is a matter of consistency in _this_ function, which is
run under RTNL for sure.

It is also a patch for net-next tree, because it fixes no bug.

I would not add a Fixes: tag to avoid dealing with useless backports.