svn commit: r358167 - head/sys/netinet6

[Kristof Provost]

On 24 Feb 2020, at 15:21, Hans Petter Selasky wrote:
> On 2020-02-20 11:56, Bjoern A. Zeeb wrote:
>> +
>> +		unfragpartlen = optlen + sizeof(struct ip6_hdr);
>> +
>
> Hi Bjoren,
>
> This change introduces a slight regression when a host replies to IPv6 
> ping fragmented packets. The problem is the "unfragpartlen" must also 
> be set in the else case of "if (opt)", else the payload offset 
> computation for IPv6 fragments goes wrong by the size of the IPv6 
> header!
>
Confirmed, because the pf fragmentation:v6 test also fails on this.

> After r358167:
>
>> ping6 -s 3000 fe80::ee0d:9aff:fed4:2c8c%mce2
>> PING6(3048=40+8+3000 bytes) fe80::ee0d:9aff:fed4:2c94%mce2 --> 
>> fe80::ee0d:9aff:fed4:2c8c%mce2
>> ^C
>> --- fe80::ee0d:9aff:fed4:2c8c%mce2 ping6 statistics ---
>> 2 packets transmitted, 0 packets received, 100.0% packet loss
>
> With the patch mentioned in the end of this e-mail:
>
>> ping6 -s 3000 fe80::ee0d:9aff:fed4:2c8c%mce2
>> PING6(3048=40+8+3000 bytes) fe80::ee0d:9aff:fed4:2c8c%mce2 --> 
>> fe80::ee0d:9aff:fed4:2c8c%mce2
>> 3008 bytes from fe80::ee0d:9aff:fed4:2c8c%mce2, icmp_seq=0 hlim=64 
>> time=0.499 ms
>> 3008 bytes from fe80::ee0d:9aff:fed4:2c8c%mce2, icmp_seq=1 hlim=64 
>> time=0.405 ms
>> 3008 bytes from fe80::ee0d:9aff:fed4:2c8c%mce2, icmp_seq=2 hlim=64 
>> time=0.097 ms
>> ^C
>> --- fe80::ee0d:9aff:fed4:2c8c%mce2 ping6 statistics ---
>> 3 packets transmitted, 3 packets received, 0.0% packet loss
>> round-trip min/avg/max/std-dev = 0.097/0.334/0.499/0.172 ms
>
> Patch goes like this:
>
>> diff --git a/sys/netinet6/ip6_output.c b/sys/netinet6/ip6_output.c
>> index 06c57bcec48..a6c8d148833 100644
>> --- a/sys/netinet6/ip6_output.c
>> +++ b/sys/netinet6/ip6_output.c
>> @@ -459,7 +459,6 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts 
>> *opt,
>>          */
>>         bzero(&exthdrs, sizeof(exthdrs));
>>         optlen = 0;
>> -       unfragpartlen = 0;
>>         if (opt) {
>>                 /* Hop-by-Hop options header. */
>>                 MAKE_EXTHDR(opt->ip6po_hbh, &exthdrs.ip6e_hbh, 
>> optlen);
>> @@ -497,8 +496,6 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts 
>> *opt,
>>                 /* Routing header. */
>>                 MAKE_EXTHDR(opt->ip6po_rthdr, &exthdrs.ip6e_rthdr, 
>> optlen);
>> -               unfragpartlen = optlen + sizeof(struct ip6_hdr);
>> -
>>                 /*
>>                  * NOTE: we don't add AH/ESP length here (done in
>>                  * ip6_ipsec_output()).
>> @@ -508,6 +505,8 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts 
>> *opt,
>>                 MAKE_EXTHDR(opt->ip6po_dest2, &exthdrs.ip6e_dest2, 
>> optlen);
>>         }
>> +       unfragpartlen = optlen + sizeof(struct ip6_hdr);
>> +
>>         /*
>>          * If there is at least one extension header,
>>          * separate IP6 header from the payload.
>
And with this patch the test passes again.

Regards,
Kristof