Re: FreeBSD 10-RC4: Got crash in igb driver

Le 10 janv. 2014 à 02:21, Yonghyeon PYUN <pyunyh_at_gmail.com> a écrit :

> On Thu, Jan 09, 2014 at 04:06:09PM +0100, Alexandre Martins wrote:
>> Dear,
>> 
>> I experience some troubles with the igb device driver on FreeBSD 10-RC4.
>> 
>> The kernel make a pagefault in the igb_tx_ctx_setup function when accessing to 
>> a IPv6 header.
>> 
>> The network configuration is the following:
>> - box acting as an IPv6 router
>> - one interface with an IPv6 (igb0)
>> - another interface with a vlan, and IPv6 on it (vlan0 on igb1)
>> 
>> Vlan Hardware tagging is set on both interfaces.
>> 
>> The packet that cause the crash come from igb0 and go to vlan0.
>> 
>> After investigation, i see that the mbuf is split in two. The first one carry 
>> the ethernet header, the second, the IPv6 header and data payload.
>> 
>> The split is due to the "m_copy" done in ip6_forward, that make the mbuf not 
>> writable and the "M_PREPEND" in ether_output that insert the new mbuf before 
>> the original one.
>> 
>> The kernel crashes only if the newly allocated mbuf is at the end of a memory 
>> page, and no page is available after this one. So, it's extremly rare.
>> 
>> I inserted a "KASSERT" into the function (see attached patch) to check this 
>> behavior, and it raises on every IPv6 forwarded packet to the vlan. The 
>> problem disapear if i remove hardware tagging.
>> 
>> In the commit 256200, i see that pullups has been removed. May it be related ?
>> 
> 
> I think I introduced the header parsing code to meet controller
> requirement in em(4) and Jack borrowed that code in the past but it
> seems it was removed in r256200.  It seems igb_tx_ctx_setup()
> assumes it can access ethernet/IP/TCP/UDP headers in the first mbuf
> of the chain.
> This looks wrong to me.

Instead of patching each driver with pullup code we can add a generic pullup code ?
- get the contiguous protocol requirement (L2, L3, L4) from underlying driver.
- do the pullup

> 
>> Can you confirm the problem ?
>> 
> 
> Probably Jack can tell more about change made in r256200.  It's not
> easy for me to verify correctness of igb(4) at this moment.
> 
>> Best regards
>> 
>> -- 
>> Alexandre Martins
>> NETASQ -- We secure IT