Re: net/ipv4/fib_trie.c:1550 suspicious rcu_dereference_check() usage

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

On 03/10/2015 09:20 AM, Fabio Estevam wrote:
> On Tue, Mar 10, 2015 at 1:19 AM, Fabio Estevam <festevam@gmail.com> wrote:
>> Hi,
>>
>> On a imx6sl-warp board running linux-next 20150306:
>>
>> root@imx6sl-warp:~# reboot
>>
>> The system is going down for reboot NOW!(ttymxc0) (Wed Feb 25 18:41:59 2015):
>> INIT: Switching to runlevel: 6
>> INIT: Sending processes the TERM signal
>> INIT: Sending processe Stopping syslogd/klogd: stopped syslogd (pid 301)
>> stopped klogd (pid 304)
>> done
>> Deconfiguring network interfaces... [   36.083246]
>> [   36.084753] ===============================
>> [   36.089026] [ INFO: suspicious RCU usage. ]
>> [   36.093217] 4.0.0-rc2-next-20150306-00003-g62c7037-dirty #2699 Not tainted
> It seems that the issue is gone in linux-next 20150310.

I found the bug.  It was in "fib_trie: Only resize tnodes once instead 
of on each leaf removal in fib_table_flush", commit 7289e6dd.  I had 
incorrectly used a rcu_dereference where I should have used an 
rtnl_dereference.  The offending code was removed and replaced 4 days 
ago when "fib_trie: Add key vector to root, return parent key_vector in 
resize", commit 88bae714, was applied.

- Alex