|
Hello,
On Tue, 9 Oct 2001, Roberto Nibali wrote:
> Hi Julian,
>
> > The code was present before this change. It seems it is used
> > to avoid checksum problems in the hardware, for example, when CHECKSUM_HW
> > or CHECKSUM_UNNECESSARY are returned.
... by this way the masq code can be debugged without worrying
for wrong calcs in skb->csum (CHECKSUM_HW), the check is performed
from scratch.
> I fail to see this. It looks bogus to me. Can you tell me a
> configuration
> or setup when CONFIG_IP_MASQ_DEBUG is set? There should also be a
> &sysctl_ip_masq_debug, sizeof(int), 0644, NULL, &proc_dointvec} proc-fs
> entry, or not? I just cannot see where this is triggered. #ifdef's
> everywhere but no #define.
There is no #define but it is added from the authors in
include/linux/config.h when needed :) But the sysctl var is present
in net/ipv4/sysctl_net_ipv4.c
> > Yes, it seems 2.2.20 is full with net bugs. But we can port
>
> I hope you mean 2.2.19. If 2.2.20 is still full of bugs, I'll definitely
> change to OpenBSD or back to 2.0.39. :)
Oh, yes, bug fixes. Better to use latest 2.0.40pre :)
> > 1.0.9 to 2.2.20 without any problems after this change (it is the only
> > one for 1.0.9 for now) is tested with 2.2.19. For now we need this change
> > tested from many aspects before 1.0.9.
>
> Ok, I've a LVS workshop an thursday. I can let the people test this
> patch.
> I'm sure they will love to see a kernel crashing and blame me for it :)
What is the life without fun :) What I know is that this
patch works for:
- LVS-DR
- LVS-NAT FTP with ip_masq_ftp
- Normal LVS-NAT without FTP
- Masquerade
I still didn't found where it crashes but may be you'll be
lucky enough :) The other danger is when the packets leave the box
with wrong checksum. You have time to make new kernel before your
workshop tomorrow :) And tomorrow I'll run testlvs on my NICs.
> Best regards,
> Roberto Nibali, ratz
Regards
--
Julian Anastasov <ja@xxxxxx>
|
