Re: kern/52353: [netbsd-8] A crash in icmpv6 code (?)

[Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>]

Hi,

On 2017/07/06 22:17, Dominik Bialy wrote:
> On Tue, Jul 04, 2017 at 02:25:25PM +0900, Kengo NAKAHARA wrote:
>> Hi,
>>
>> Thank you for your quick and detailed reply.
>>
>> On 2017/07/03 12:30, Dominik Bialy wrote:
>>> On Sun, Jul 02, 2017 at 11:35:01PM +0000, Kengo NAKAHARA wrote:
>>>> The following reply was made to PR kern/52353; it has been noted by GNATS.
>>>>
>>>> From: Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>
>>>> To: gnats-bugs%NetBSD.org@localhost, kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost,
>>>>         netbsd-bugs%netbsd.org@localhost
>>>> Cc: 
>>>> Subject: Re: kern/52353: [netbsd-8] A crash in icmpv6 code (?)
>>>> Date: Mon, 3 Jul 2017 08:33:32 +0900
>>>>
>>>>  Hi,
>>>>  
>>>>  On 2017/07/01 22:40, dmb%yenn.ulegend.net@localhost wrote:
>>>>  > System: NetBSD yenn 8.0_BETA NetBSD 8.0_BETA (YENN) #6: Mon Jun 26 08:49:07 UTC 2017 builds@yenn:/var/obj/sys/arch/amd64/compile/YENN amd64
>>>>  
>>>>  It seems you use other than GENERIC kernel config.Could you show your
>>>>  kernel config?
>>>>  
>>> It is almost GENERIC with: GATEWAY, and altq* options added.  The setup
>>> is somewhat unusual, since I'm using ipf+pf+the old altq
>>>
>>> Here's the config:
>>>
>>> http://yenn.ulegend.net/~dmb/YENN
>>>
>>> Here's dmesg.boot:
>>>
>>> http://yenn.ulegend.net/~dmb/dmesg.boot
>>>
>>> (The panic in this dmesg is probably unrelated.)
>>>
>>> In altq I'm using WFQ over a 0.75 Mbps uplink.
>>>
>>>>  > Architecture: x86_64
>>>>  > Machine: amd64
>>>>  >> Description:
>>>>  > yenn# crash -M netbsd.3.core -N netbsd.3
>>>>  > Crash version 8.0_BETA, image version 8.0_BETA.
>>>>  > System panicked: in6_cksum: mbuf too short for IPv6 header
>>>>  > Backtrace from time of crash is available.
>>>>  > crash> bt
>>>>  > _KERNEL_OPT_NARCNET() at 0
>>>>  > ?() at fffffe80a11bcc00
>>>>  > vpanic() at vpanic+0x149
>>>>  > snprintf() at snprintf
>>>>  > in6_cksum() at in6_cksum+0x1a2
>>>>  > _icmp6_input() at _icmp6_input+0xb4
>>>>  > wqinput_work() at wqinput_work+0x88
>>>>  > workqueue_worker() at workqueue_worker+0xbc
>>>>  > 
>>>>  > yes, IPv6 is via gif(4), but _before_ the patch for MP-fy
>>>>  >> How-To-Repeat:
>>>>  > 	possibly ping the machine for some time (?) with IPv6 on a gif(4) (?)
>>>>  
>>>>  Which do you use IPv6 over IPv6 or IPv6 over IPv4?
>>>>  
>>> IPv6 over IPv4 -- he.net tunnelbroker
>>
>> Hmm, I guess there may be the issue in combination ALTQ and gif(4).
>> So, I try to reproduce it in my simple environment. That is,
>>     - use two NetBSD-8 machine and connect directly their ethernet ports
>>     - create IPv6 over IPv4 gif(4) between the two machines
>>     - apply below WFQ to the gif(4) psrc, pdst ethernet
>>       ====================
>>       interface wm2 bandwidth 750000 wfq
>>       ====================
>>     - ping6 over gif(4) each other
>> However, I cannot reproduce it yet...
>>
>> By the way, ozaki-r@n.o help me to research this issue. He also implements
>> a patch. Could you try below patch?
>> ====================
>> diff --git a/sys/netinet6/icmp6.c b/sys/netinet6/icmp6.c
>> index f740932036d..3b45ba8d785 100644
>> --- a/sys/netinet6/icmp6.c
>> +++ b/sys/netinet6/icmp6.c
>> @@ -494,6 +494,15 @@ _icmp6_input(struct mbuf *m, int off, int proto)
>>  		goto freeit;
>>  	}
>>  
>> +	if (m->m_len < sizeof(struct ip6_hdr)) {
>> +		m = m_pullup(m, sizeof(struct ip6_hdr));
>> +		if (m == NULL) {
>> +			ICMP6_STATINC(ICMP6_STAT_TOOSHORT);
>> +			icmp6_ifstat_inc(rcvif, ifs6_in_error);
>> +			goto freeit;
>> +		}
>> +	}
>> +
>>  	ip6 = mtod(m, struct ip6_hdr *);
>>  	IP6_EXTHDR_GET(icmp6, struct icmp6_hdr *, m, off, sizeof(*icmp6));
>>  	if (icmp6 == NULL) {
>> ====================
>>
>> If the issue is reproduced after applying above patch, could you tell
>> me your detailed network configuration and ipf/pf setting? 
>> Of course, within the range you can show with no problem.
>>
> OK, then I applied the patch, and I'm pinging the machine
> for quite some time.  The uptime is almost 2 days.

Thank you for your testing! I committed above patch to -current branch,
and sent pullup request to -8 branch. It will merge to -8 branch soon. 

> Close the PR?  When it'd be needed it could be reopened again, no?

Ok, I close this PR.


Thanks,

-- 
//////////////////////////////////////////////////////////////////////
Internet Initiative Japan Inc.

Device Engineering Section,
IoT Platform Development Department,
Network Division,
Technology Unit

Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>