bonding: avoid lockdep confusion in bond_get_stats()
authorEric Dumazet <edumazet@google.com>
Tue, 31 Jul 2018 13:30:54 +0000 (06:30 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 1 Aug 2018 16:34:07 +0000 (09:34 -0700)
syzbot found that the following sequence produces a LOCKDEP splat [1]

ip link add bond10 type bond
ip link add bond11 type bond
ip link set bond11 master bond10

To fix this, we can use the already provided nest_level.

This patch also provides correct nesting for dev->addr_list_lock

[1]
WARNING: possible recursive locking detected
4.18.0-rc6+ #167 Not tainted
--------------------------------------------
syz-executor751/4439 is trying to acquire lock:
(____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline]
(____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426

but task is already holding lock:
(____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline]
(____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426

other info that might help us debug this:
 Possible unsafe locking scenario:

       CPU0
       ----
  lock(&(&bond->stats_lock)->rlock);
  lock(&(&bond->stats_lock)->rlock);

 *** DEADLOCK ***

 May be due to missing lock nesting notation

3 locks held by syz-executor751/4439:
 #0: (____ptrval____) (rtnl_mutex){+.+.}, at: rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77
 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline]
 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426
 #2: (____ptrval____) (rcu_read_lock){....}, at: bond_get_stats+0x0/0x560 include/linux/compiler.h:215

stack backtrace:
CPU: 0 PID: 4439 Comm: syz-executor751 Not tainted 4.18.0-rc6+ #167
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+0x1c9/0x2b4 lib/dump_stack.c:113
 print_deadlock_bug kernel/locking/lockdep.c:1765 [inline]
 check_deadlock kernel/locking/lockdep.c:1809 [inline]
 validate_chain kernel/locking/lockdep.c:2405 [inline]
 __lock_acquire.cold.64+0x1fb/0x486 kernel/locking/lockdep.c:3435
 lock_acquire+0x1e4/0x540 kernel/locking/lockdep.c:3924
 __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline]
 _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:144
 spin_lock include/linux/spinlock.h:310 [inline]
 bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426
 dev_get_stats+0x10f/0x470 net/core/dev.c:8316
 bond_get_stats+0x232/0x560 drivers/net/bonding/bond_main.c:3432
 dev_get_stats+0x10f/0x470 net/core/dev.c:8316
 rtnl_fill_stats+0x4d/0xac0 net/core/rtnetlink.c:1169
 rtnl_fill_ifinfo+0x1aa6/0x3fb0 net/core/rtnetlink.c:1611
 rtmsg_ifinfo_build_skb+0xc8/0x190 net/core/rtnetlink.c:3268
 rtmsg_ifinfo_event.part.30+0x45/0xe0 net/core/rtnetlink.c:3300
 rtmsg_ifinfo_event net/core/rtnetlink.c:3297 [inline]
 rtnetlink_event+0x144/0x170 net/core/rtnetlink.c:4716
 notifier_call_chain+0x180/0x390 kernel/notifier.c:93
 __raw_notifier_call_chain kernel/notifier.c:394 [inline]
 raw_notifier_call_chain+0x2d/0x40 kernel/notifier.c:401
 call_netdevice_notifiers_info+0x3f/0x90 net/core/dev.c:1735
 call_netdevice_notifiers net/core/dev.c:1753 [inline]
 netdev_features_change net/core/dev.c:1321 [inline]
 netdev_change_features+0xb3/0x110 net/core/dev.c:7759
 bond_compute_features.isra.47+0x585/0xa50 drivers/net/bonding/bond_main.c:1120
 bond_enslave+0x1b25/0x5da0 drivers/net/bonding/bond_main.c:1755
 bond_do_ioctl+0x7cb/0xae0 drivers/net/bonding/bond_main.c:3528
 dev_ifsioc+0x43c/0xb30 net/core/dev_ioctl.c:327
 dev_ioctl+0x1b5/0xcc0 net/core/dev_ioctl.c:493
 sock_do_ioctl+0x1d3/0x3e0 net/socket.c:992
 sock_ioctl+0x30d/0x680 net/socket.c:1093
 vfs_ioctl fs/ioctl.c:46 [inline]
 file_ioctl fs/ioctl.c:500 [inline]
 do_vfs_ioctl+0x1de/0x1720 fs/ioctl.c:684
 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701
 __do_sys_ioctl fs/ioctl.c:708 [inline]
 __se_sys_ioctl fs/ioctl.c:706 [inline]
 __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706
 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x440859
Code: e8 2c af 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 3b 10 fc ff c3 66 2e 0f 1f 84 00 00 00 00
RSP: 002b:00007ffc51a92878 EFLAGS: 00000213 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000440859
RDX: 0000000020000040 RSI: 0000000000008990 RDI: 0000000000000003
RBP: 0000000000000000 R08: 00000000004002c8 R09: 00000000004002c8
R10: 00000000022d5880 R11: 0000000000000213 R12: 0000000000007390
R13: 0000000000401db0 R14: 0000000000000000 R15: 0000000000000000

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jay Vosburgh <j.vosburgh@gmail.com>
Cc: Veaceslav Falico <vfalico@gmail.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/bonding/bond_main.c

index 63e3844c5becf5e973e10fa2aa533f668ac8e30b..217b790d22edc2fdd90d4f3c634cb2fa38bc095f 100644 (file)
@@ -1717,6 +1717,8 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
                goto err_upper_unlink;
        }
 
+       bond->nest_level = dev_get_nest_level(bond_dev) + 1;
+
        /* If the mode uses primary, then the following is handled by
         * bond_change_active_slave().
         */
@@ -1764,7 +1766,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
        if (bond_mode_can_use_xmit_hash(bond))
                bond_update_slave_arr(bond, NULL);
 
-       bond->nest_level = dev_get_nest_level(bond_dev);
 
        netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
                    slave_dev->name,
@@ -3415,6 +3416,13 @@ static void bond_fold_stats(struct rtnl_link_stats64 *_res,
        }
 }
 
+static int bond_get_nest_level(struct net_device *bond_dev)
+{
+       struct bonding *bond = netdev_priv(bond_dev);
+
+       return bond->nest_level;
+}
+
 static void bond_get_stats(struct net_device *bond_dev,
                           struct rtnl_link_stats64 *stats)
 {
@@ -3423,7 +3431,7 @@ static void bond_get_stats(struct net_device *bond_dev,
        struct list_head *iter;
        struct slave *slave;
 
-       spin_lock(&bond->stats_lock);
+       spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
        memcpy(stats, &bond->bond_stats, sizeof(*stats));
 
        rcu_read_lock();
@@ -4227,6 +4235,7 @@ static const struct net_device_ops bond_netdev_ops = {
        .ndo_neigh_setup        = bond_neigh_setup,
        .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
        .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
+       .ndo_get_lock_subclass  = bond_get_nest_level,
 #ifdef CONFIG_NET_POLL_CONTROLLER
        .ndo_netpoll_setup      = bond_netpoll_setup,
        .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
@@ -4725,6 +4734,7 @@ static int bond_init(struct net_device *bond_dev)
        if (!bond->wq)
                return -ENOMEM;
 
+       bond->nest_level = SINGLE_DEPTH_NESTING;
        netdev_lockdep_set_classes(bond_dev);
 
        list_add_tail(&bond->bond_list, &bn->dev_list);