Re: [PATCH net-next v2 1/3] net: ipv6: Allow shorthand delete of all nexthops in multipath route

[David Miller <davem@davemloft.net>]

From: David Ahern <dsa@cumulusnetworks.com>
Date: Mon, 16 Jan 2017 18:27:36 -0700

> On 1/16/17 5:51 PM, David Miller wrote:
>> From: David Ahern <dsa@cumulusnetworks.com>
>> Date: Sun, 15 Jan 2017 12:07:04 -0800
>> 
>>> @@ -2143,6 +2143,26 @@ int ip6_del_rt(struct rt6_info *rt)
>>>  	return __ip6_del_rt(rt, &info);
>>>  }
>>>  
>>> +/* called with table lock held */
>>  ...
>>> @@ -2176,10 +2196,9 @@ static int ip6_route_del(struct fib6_config *cfg)
>>>  				continue;
>>>  			if (cfg->fc_protocol && cfg->fc_protocol != rt->rt6i_protocol)
>>>  				continue;
>>> -			dst_hold(&rt->dst);
>>> -			read_unlock_bh(&table->tb6_lock);
>>>  
>>> -			return __ip6_del_rt(rt, &cfg->fc_nlinfo);
>>> +			err = __ip6_route_del(rt, cfg);
>>> +			break;
>>>  		}
>> 
>> fib6_del() (invoked by __ip6_route_del()) has to be invoked with the
>> table lock held a sa writer, but here you are only holding it as a
>> reader.
> 
> That table lock is still held. If you look up 2 lines I remove the line that releases the lock.

It's held in this function as a reader, it needs to be held as a writer.

That's why the lock is dropped in the current code and the existing
wrapper around fib6_del() takes it as a writer.

Is it clear now?

	read_lock_bh(&table->lock);
	fib6_del();

is invalid.

	write_lock_bh(&table->lock);
	fib6_del();

is required.