[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANn89iJ09MAg5Zf0b7LPby--1YVi9vcwKQDDGH=E5_Bvw1P89Q@mail.gmail.com>
Date: Tue, 5 Nov 2024 09:46:29 +0100
From: Eric Dumazet <edumazet@...gle.com>
To: Kuniyuki Iwashima <kuniyu@...zon.com>
Cc: courmisch@...il.com, davem@...emloft.net, eric.dumazet@...il.com,
kuba@...nel.org, netdev@...r.kernel.org, pabeni@...hat.com,
syzkaller@...glegroups.com
Subject: Re: [PATCH net-next] phonet: do not call synchronize_rcu() from phonet_route_del()
On Mon, Nov 4, 2024 at 9:40 PM Kuniyuki Iwashima <kuniyu@...zon.com> wrote:
>
> From: Eric Dumazet <edumazet@...gle.com>
> Date: Mon, 4 Nov 2024 15:26:22 +0000
> > Calling synchronize_rcu() while holding rcu_read_lock() is not
> > permitted [1]
>
> Thanks for catching this !
>
> >
> > Move the synchronize_rcu() to route_doit().
> >
> > [1]
> > WARNING: suspicious RCU usage
> > 6.12.0-rc5-syzkaller-01056-gf07a6e6ceb05 #0 Not tainted
> > -----------------------------
> > kernel/rcu/tree.c:4092 Illegal synchronize_rcu() in RCU read-side critical section!
> >
> > other info that might help us debug this:
> >
> > rcu_scheduler_active = 2, debug_locks = 1
> > 1 lock held by syz-executor427/5840:
> > #0: ffffffff8e937da0 (rcu_read_lock){....}-{1:2}, at: rcu_lock_acquire include/linux/rcupdate.h:337 [inline]
> > #0: ffffffff8e937da0 (rcu_read_lock){....}-{1:2}, at: rcu_read_lock include/linux/rcupdate.h:849 [inline]
> > #0: ffffffff8e937da0 (rcu_read_lock){....}-{1:2}, at: route_doit+0x3d6/0x640 net/phonet/pn_netlink.c:264
> >
> > stack backtrace:
> > CPU: 1 UID: 0 PID: 5840 Comm: syz-executor427 Not tainted 6.12.0-rc5-syzkaller-01056-gf07a6e6ceb05 #0
> > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 09/13/2024
> > Call Trace:
> > <TASK>
> > __dump_stack lib/dump_stack.c:94 [inline]
> > dump_stack_lvl+0x241/0x360 lib/dump_stack.c:120
> > lockdep_rcu_suspicious+0x226/0x340 kernel/locking/lockdep.c:6821
> > synchronize_rcu+0xea/0x360 kernel/rcu/tree.c:4089
> > phonet_route_del+0xc6/0x140 net/phonet/pn_dev.c:409
> > route_doit+0x514/0x640 net/phonet/pn_netlink.c:275
> > rtnetlink_rcv_msg+0x791/0xcf0 net/core/rtnetlink.c:6790
> > netlink_rcv_skb+0x1e3/0x430 net/netlink/af_netlink.c:2551
> > netlink_unicast_kernel net/netlink/af_netlink.c:1331 [inline]
> > netlink_unicast+0x7f6/0x990 net/netlink/af_netlink.c:1357
> > netlink_sendmsg+0x8e4/0xcb0 net/netlink/af_netlink.c:1901
> > sock_sendmsg_nosec net/socket.c:729 [inline]
> > __sock_sendmsg+0x221/0x270 net/socket.c:744
> > sock_write_iter+0x2d7/0x3f0 net/socket.c:1165
> > new_sync_write fs/read_write.c:590 [inline]
> > vfs_write+0xaeb/0xd30 fs/read_write.c:683
> > ksys_write+0x183/0x2b0 fs/read_write.c:736
> > do_syscall_x64 arch/x86/entry/common.c:52 [inline]
> > do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
> > entry_SYSCALL_64_after_hwframe+0x77/0x7f
> >
> > Fixes: 17a1ac0018ae ("phonet: Don't hold RTNL for route_doit().")
> > Reported-by: syzbot <syzkaller@...glegroups.com>
> > Signed-off-by: Eric Dumazet <edumazet@...gle.com>
> > Cc: Kuniyuki Iwashima <kuniyu@...zon.com>
> > Cc: Remi Denis-Courmont <courmisch@...il.com>
> > ---
> > net/phonet/pn_dev.c | 4 +++-
> > net/phonet/pn_netlink.c | 10 ++++++++--
> > 2 files changed, 11 insertions(+), 3 deletions(-)
> >
> > diff --git a/net/phonet/pn_dev.c b/net/phonet/pn_dev.c
> > index 19234d664c4fb537eba0267266efbb226cf103c3..578d935f2b11694fd1004c5f854ec344b846eeb2 100644
> > --- a/net/phonet/pn_dev.c
> > +++ b/net/phonet/pn_dev.c
> > @@ -406,7 +406,9 @@ int phonet_route_del(struct net_device *dev, u8 daddr)
> >
> > if (!dev)
> > return -ENOENT;
> > - synchronize_rcu();
> > +
> > + /* Note : our caller must call synchronize_rcu() */
> > +
> > dev_put(dev);
>
> Are these synchronize_rcu() + dev_put() paired with rcu_read_lock()
> and dev_hold() in phonet_route_output() ?
This dev_put() was before your patch.
It is paired the dev_hold() in phonet_route_add()
>
> If so, we need to move dev_put() too or maybe we can remove
> synchronize_rcu() here and replace rcu_read_lock() with
> spin_lock(&routes->lock) in phonet_route_output().
Not at all, let's not make phonet slower than needed :)
Powered by blists - more mailing lists