WARNING: suspicious RCU usage in bond_ipsec_add_sa

[syzbot]

Hello,

syzbot found the following crash on:

HEAD commit: e44f65fd xen-netfront: remove redundant assignment to vari..
git tree: net-next
console output: https://syzkaller.appspot.com/x/log.txt?x=16148f87100000
kernel config: https://syzkaller.appspot.com/x/.config?x=829871134ca5e230
dashboard link: https://syzkaller.appspot.com/bug?extid=582c98032903dcc04816
compiler: gcc (GCC) 10.1.0-syz 20200507

Unfortunately, I don't have any reproducer for this crash yet.

IMPORTANT: if you fix the bug, please add the following tag to the commit:
Reported-by: syzbot+582c98...@syzkaller.appspotmail.com

=============================
WARNING: suspicious RCU usage
5.8.0-rc2-syzkaller #0 Not tainted
-----------------------------
drivers/net/bonding/bond_main.c:387 suspicious rcu_dereference_protected() usage!

other info that might help us debug this:


rcu_scheduler_active = 2, debug_locks = 1
1 lock held by syz-executor.0/5186:
#0: ffff888089791a28 (&net->xfrm.xfrm_cfg_mutex){+.+.}-{3:3}, at: xfrm_netlink_rcv+0x5c/0x90 net/xfrm/xfrm_user.c:2687

stack backtrace:
CPU: 1 PID: 5186 Comm: syz-executor.0 Not tainted 5.8.0-rc2-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
__dump_stack lib/dump_stack.c:77 [inline]
dump_stack+0x18f/0x20d lib/dump_stack.c:118
bond_ipsec_add_sa+0x1c8/0x220 drivers/net/bonding/bond_main.c:387
xfrm_dev_state_add+0x2da/0x7b0 net/xfrm/xfrm_device.c:268
xfrm_state_construct net/xfrm/xfrm_user.c:655 [inline]
xfrm_add_sa+0x2166/0x34f0 net/xfrm/xfrm_user.c:684
xfrm_user_rcv_msg+0x414/0x700 net/xfrm/xfrm_user.c:2680
netlink_rcv_skb+0x15a/0x430 net/netlink/af_netlink.c:2469
xfrm_netlink_rcv+0x6b/0x90 net/xfrm/xfrm_user.c:2688
netlink_unicast_kernel net/netlink/af_netlink.c:1303 [inline]
netlink_unicast+0x533/0x7d0 net/netlink/af_netlink.c:1329
netlink_sendmsg+0x856/0xd90 net/netlink/af_netlink.c:1918
sock_sendmsg_nosec net/socket.c:652 [inline]
sock_sendmsg+0xcf/0x120 net/socket.c:672
____sys_sendmsg+0x6e8/0x810 net/socket.c:2352
___sys_sendmsg+0xf3/0x170 net/socket.c:2406
__sys_sendmsg+0xe5/0x1b0 net/socket.c:2439
do_syscall_64+0x60/0xe0 arch/x86/entry/common.c:359
entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x45cb29
Code: Bad RIP value.
RSP: 002b:00007ff7e9a92c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
RAX: ffffffffffffffda RBX: 00000000005027e0 RCX: 000000000045cb29
RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003
RBP: 000000000078bf00 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff
R13: 0000000000000a45 R14: 00000000004cd2c9 R15: 00007ff7e9a936d4
bond0: (slave bond_slave_0): Slave does not support ipsec offload


---
This bug is generated by a bot. It may contain errors.
See https://goo.gl/tpsmEJ for more information about syzbot.
syzbot engineers can be reached at syzk...@googlegroups.com.

syzbot will keep track of this bug report. See:
https://goo.gl/tpsmEJ#status for how to communicate with syzbot.

[Jarod Wilson]

On Mon, Jul 6, 2020 at 11:44 AM syzbot
<syzbot+582c98...@syzkaller.appspotmail.com> wrote:
>
> Hello,
>
> syzbot found the following crash on:
>
> HEAD commit: e44f65fd xen-netfront: remove redundant assignment to vari..
> git tree: net-next
> console output: https://syzkaller.appspot.com/x/log.txt?x=16148f87100000
> kernel config: https://syzkaller.appspot.com/x/.config?x=829871134ca5e230
> dashboard link: https://syzkaller.appspot.com/bug?extid=582c98032903dcc04816
> compiler: gcc (GCC) 10.1.0-syz 20200507
>
> Unfortunately, I don't have any reproducer for this crash yet.
>
> IMPORTANT: if you fix the bug, please add the following tag to the commit:
> Reported-by: syzbot+582c98...@syzkaller.appspotmail.com
>
> =============================
> WARNING: suspicious RCU usage
> 5.8.0-rc2-syzkaller #0 Not tainted
> -----------------------------
> drivers/net/bonding/bond_main.c:387 suspicious rcu_dereference_protected() usage!

Hm. Access to curr_active_slave in the bonding driver is kind of all
over the place, between rtnl_dereference, rcu_deference,
rcu_access_pointer and just reading it without any protections. It
does look like this is a case where bond_ipsec_add_sa() gets called
without RTNL being held, so perhaps we should be using rcu_dereference
here, since we do need to dereference the acquired pointer, but
probably don't need to be holding RTNL here.

--
Jarod Wilson
ja...@redhat.com