OSDN Git Service

tipc: fix lockdep warning during node delete
authorJon Maloy <donmalo99@gmail.com>
Mon, 26 Nov 2018 17:26:14 +0000 (12:26 -0500)
committerDavid S. Miller <davem@davemloft.net>
Wed, 28 Nov 2018 00:30:39 +0000 (16:30 -0800)
commitec835f891232d7763dea9da0358f31e24ca6dfb7
tree3e7c7e6caee06163b89789bccf52f3a1f9d2e21a
parentcc5922054131f9abefdc0622ae64fc55e6b2671d
tipc: fix lockdep warning during node delete

We see the following lockdep warning:

[ 2284.078521] ======================================================
[ 2284.078604] WARNING: possible circular locking dependency detected
[ 2284.078604] 4.19.0+ #42 Tainted: G            E
[ 2284.078604] ------------------------------------------------------
[ 2284.078604] rmmod/254 is trying to acquire lock:
[ 2284.078604] 00000000acd94e28 ((&n->timer)#2){+.-.}, at: del_timer_sync+0x5/0xa0
[ 2284.078604]
[ 2284.078604] but task is already holding lock:
[ 2284.078604] 00000000f997afc0 (&(&tn->node_list_lock)->rlock){+.-.}, at: tipc_node_stop+0xac/0x190 [tipc]
[ 2284.078604]
[ 2284.078604] which lock already depends on the new lock.
[ 2284.078604]
[ 2284.078604]
[ 2284.078604] the existing dependency chain (in reverse order) is:
[ 2284.078604]
[ 2284.078604] -> #1 (&(&tn->node_list_lock)->rlock){+.-.}:
[ 2284.078604]        tipc_node_timeout+0x20a/0x330 [tipc]
[ 2284.078604]        call_timer_fn+0xa1/0x280
[ 2284.078604]        run_timer_softirq+0x1f2/0x4d0
[ 2284.078604]        __do_softirq+0xfc/0x413
[ 2284.078604]        irq_exit+0xb5/0xc0
[ 2284.078604]        smp_apic_timer_interrupt+0xac/0x210
[ 2284.078604]        apic_timer_interrupt+0xf/0x20
[ 2284.078604]        default_idle+0x1c/0x140
[ 2284.078604]        do_idle+0x1bc/0x280
[ 2284.078604]        cpu_startup_entry+0x19/0x20
[ 2284.078604]        start_secondary+0x187/0x1c0
[ 2284.078604]        secondary_startup_64+0xa4/0xb0
[ 2284.078604]
[ 2284.078604] -> #0 ((&n->timer)#2){+.-.}:
[ 2284.078604]        del_timer_sync+0x34/0xa0
[ 2284.078604]        tipc_node_delete+0x1a/0x40 [tipc]
[ 2284.078604]        tipc_node_stop+0xcb/0x190 [tipc]
[ 2284.078604]        tipc_net_stop+0x154/0x170 [tipc]
[ 2284.078604]        tipc_exit_net+0x16/0x30 [tipc]
[ 2284.078604]        ops_exit_list.isra.8+0x36/0x70
[ 2284.078604]        unregister_pernet_operations+0x87/0xd0
[ 2284.078604]        unregister_pernet_subsys+0x1d/0x30
[ 2284.078604]        tipc_exit+0x11/0x6f2 [tipc]
[ 2284.078604]        __x64_sys_delete_module+0x1df/0x240
[ 2284.078604]        do_syscall_64+0x66/0x460
[ 2284.078604]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[ 2284.078604]
[ 2284.078604] other info that might help us debug this:
[ 2284.078604]
[ 2284.078604]  Possible unsafe locking scenario:
[ 2284.078604]
[ 2284.078604]        CPU0                    CPU1
[ 2284.078604]        ----                    ----
[ 2284.078604]   lock(&(&tn->node_list_lock)->rlock);
[ 2284.078604]                                lock((&n->timer)#2);
[ 2284.078604]                                lock(&(&tn->node_list_lock)->rlock);
[ 2284.078604]   lock((&n->timer)#2);
[ 2284.078604]
[ 2284.078604]  *** DEADLOCK ***
[ 2284.078604]
[ 2284.078604] 3 locks held by rmmod/254:
[ 2284.078604]  #0: 000000003368be9b (pernet_ops_rwsem){+.+.}, at: unregister_pernet_subsys+0x15/0x30
[ 2284.078604]  #1: 0000000046ed9c86 (rtnl_mutex){+.+.}, at: tipc_net_stop+0x144/0x170 [tipc]
[ 2284.078604]  #2: 00000000f997afc0 (&(&tn->node_list_lock)->rlock){+.-.}, at: tipc_node_stop+0xac/0x19
[...}

The reason is that the node timer handler sometimes needs to delete a
node which has been disconnected for too long. To do this, it grabs
the lock 'node_list_lock', which may at the same time be held by the
generic node cleanup function, tipc_node_stop(), during module removal.
Since the latter is calling del_timer_sync() inside the same lock, we
have a potential deadlock.

We fix this letting the timer cleanup function use spin_trylock()
instead of just spin_lock(), and when it fails to grab the lock it
just returns so that the timer handler can terminate its execution.
This is safe to do, since tipc_node_stop() anyway is about to
delete both the timer and the node instance.

Fixes: 6a939f365bdb ("tipc: Auto removal of peer down node instance")
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tipc/node.c