Re: suspicious RCU usage in fib_table_flush_external

[Alexander Duyck <alexander.h.duyck@redhat.com>]

On 03/10/2015 11:35 AM, Sabrina Dubroca wrote:
> I ran into this while deleting a netns with some routes:
>
> ===============================
> [ INFO: suspicious RCU usage. ]
> 4.0.0-rc2-00722-g8ac467e837a2-dirty #674 Not tainted
> -------------------------------
> net/ipv4/fib_trie.c:1606 suspicious rcu_dereference_protected() usage!
> other info that might help us debug this:
>
> rcu_scheduler_active = 1, debug_locks = 1
> 4 locks held by kworker/u4:1/50:
>   #0:  ("%s""netns"){.+.+.+}, at: [<ffffffff81086525>] process_one_work+0x155/0x880
>   #1:  (net_cleanup_work){+.+.+.}, at: [<ffffffff81086525>] process_one_work+0x155/0x880
>   #2:  (net_mutex){+.+.+.}, at: [<ffffffff81647cfd>] cleanup_net+0x7d/0x230
>   #3:  (&(&net->rules_mod_lock)->rlock){+.+...}, at: [<ffffffff81674ce4>] fib_rules_unregister+0x34/0x100
>   stack backtrace:
> CPU: 1 PID: 50 Comm: kworker/u4:1 Not tainted 4.0.0-rc2-00722-g8ac467e837a2-dirty #674
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140617_173321-var-lib-archbuild-testing-x86_64-tobias 04/01/2014
> Workqueue: netns cleanup_net
>   0000000000000001 ffff88001dffbba8 ffffffff81835bd0 0000000000000011
>   ffff88001dd95690 ffff88001dffbbd8 ffffffff810b5505 0000000000000001
>   0000000000000001 ffff88001da86270 ffff88001d528008 ffff88001dffbc08
> Call Trace:
>   [<ffffffff81835bd0>] dump_stack+0x4c/0x6e
>   [<ffffffff810b5505>] lockdep_rcu_suspicious+0xd5/0x110
>   [<ffffffff81710421>] fib_table_flush_external+0xe1/0x130
>   [<ffffffff81708808>] fib_flush_external+0x38/0x60
>   [<ffffffff81719dee>] fib4_rule_delete+0x2e/0x30
>   [<ffffffff81674d4b>] fib_rules_unregister+0x9b/0x100
>   [<ffffffff8171a2a5>] fib4_rules_exit+0x15/0x20
>   [<ffffffff817078ab>] ip_fib_net_exit+0x1b/0x120
>   [<ffffffff817079e5>] fib_net_exit+0x35/0x40
>
>
>
> since fib_table_flush_external wants RTNL, should we just move
> rtnl_lock() before fib4_rules_exit in ip_fib_net_exit?
>
>
> diff --git a/net/ipv4/fib_frontend.c b/net/ipv4/fib_frontend.c
> index e067770235bf..ed94bc80eb78 100644
> --- a/net/ipv4/fib_frontend.c
> +++ b/net/ipv4/fib_frontend.c
> @@ -1126,11 +1126,12 @@ static void ip_fib_net_exit(struct net *net)
>   {
>   	unsigned int i;
>   
> +	rtnl_lock();
> +
>   #ifdef CONFIG_IP_MULTIPLE_TABLES
>   	fib4_rules_exit(net);
>   #endif
>   
> -	rtnl_lock();
>   	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
>   		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
>   		struct hlist_node *tmp;
>

This fix looks like it is probably correct.  I'd recommend including a 
note that it fixes 104616e74 ("switchdev: don't support custom ip rules, 
for now"), and make sure to include a signed-off-by.

- Alex