Re: commit b9ef5513c99b breaks ping to ipv6 linklocal addresses on debian buster

[David Ahern <dsahern@gmail.com>]

On 1/6/20 4:04 AM, Tetsuo Handa wrote:
>>
>> This change seems more appropriate to me (and fixes the failure):
>>
>> diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
>> index ecea41ce919b..ce5e3be7c111 100644
>> --- a/security/smack/smack_lsm.c
>> +++ b/security/smack/smack_lsm.c
>> @@ -2854,7 +2854,7 @@ static int smack_socket_connect(struct socket
>> *sock, struct sockaddr *sap,
>>                 rc = smack_netlabel_send(sock->sk, (struct sockaddr_in
>> *)sap);
>>                 break;
>>         case PF_INET6:
>> -               if (addrlen < SIN6_LEN_RFC2133 || sap->sa_family !=
>> AF_INET6)
>> +               if (addrlen < SIN6_LEN_RFC2133)
>>                         return -EINVAL;
> 
> This is called upon connect(), isn't it? Then, it is possible that a socket's
> protocol family is PF_INET6 but address given is AF_INET, isn't it? For example,
> __ip6_datagram_connect() checks for AF_INET before checking addrlen is at least
> SIN6_LEN_RFC2133 bytes. Thus, I think that we need to return 0 if address given
> is AF_INET even if socket is PF_INET6.

In this case, the sockaddr is AF_UNSPEC. From the first message:

$ strace -e connect ping6 -c1 -w1 ff02::1%eth1
connect(4, {sa_family=AF_UNSPEC,
sa_data="\4\1\0\0\0\0\377\2\0\0\0\0\0\0\0\0\0\0\0\0\0\1\3\0\0\0"}, 28) = 0

That's why I was suggesting if the socket is PF_INET6 and the address
length is at least SIN6_LEN_RFC2133, then don't worry about the sa_family.


> 
>>  #ifdef SMACK_IPV6_SECMARK_LABELING
>>                 rsp = smack_ipv6host_label(sip);
>>
>>
>> ie., if the socket family is AF_INET6 the address length should be an
>> IPv6 address. The family in the sockaddr is not as important.
>>
> 
> Commit b9ef5513c99b was wrong, but we need to also fix commit c673944347ed ?
> 

not sure. I have not seen a problem related to it yet.