Re: [PATCH net v6 09/10] ip6mr: Lock RCU before ip6mr_get_table() call in ip6mr_rtm_getroute()

[Stefan Wiehler]

>> When IPV6_MROUTE_MULTIPLE_TABLES is enabled, multicast routing tables
>> must be read under RCU or RTNL lock.
>>
>> Fixes: d1db275dd3f6 ("ipv6: ip6mr: support multiple tables")
>> Signed-off-by: Stefan Wiehler <stefan.wiehler@xxxxxxxxx>
>> ---
>>  net/ipv6/ip6mr.c | 10 +++++++---
>>  1 file changed, 7 insertions(+), 3 deletions(-)
>>
>> diff --git a/net/ipv6/ip6mr.c b/net/ipv6/ip6mr.c
>> index b169b27de7e1..39aac81a30f1 100644
>> --- a/net/ipv6/ip6mr.c
>> +++ b/net/ipv6/ip6mr.c
>> @@ -2633,27 +2633,31 @@ static int ip6mr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
>>               grp = nla_get_in6_addr(tb[RTA_DST]);
>>       tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
>>
>> +     rcu_read_lock();
> 
> AFAICS ip6mr_rtm_getroute() runs with RTNL held, so I don't see
> why this patch is needed.

That's true, but it's called neither with RCU nor RTNL lock when
RTNL_FLAG_DOIT_UNLOCKED is set in rtnetlink_rcv_msg():

> if (flags & RTNL_FLAG_DOIT_UNLOCKED) {
> 	doit = link->doit;
> 	rcu_read_unlock();
> 	if (doit)
> 		err = doit(skb, nlh, extack);
> 	module_put(owner);
> 	return err;
> }
> rcu_read_unlock();

I realized now that I completely misunderstood how ip6mr_rtm_dumproute() gets
called - it should be still safe though because mpls_netconf_dump_devconf() and
getaddr_dumpit() hold the RCU lock while mpls_dump_routes() asserts that the
RTNL lock is held. Is that observation correct?