On Mon, Jun 4, 2018 at 4:34 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> On Tue, May 29, 2018 at 7:45 PM, Xin Long <lucien.xin@xxxxxxxxx> wrote:
>> On Wed, May 30, 2018 at 1:06 AM, Marcelo Ricardo Leitner
>> <marcelo.leitner@xxxxxxxxx> wrote:
>>> On Tue, May 29, 2018 at 12:03:46PM -0400, Neal Cardwell wrote:
>>>> On Tue, May 29, 2018 at 11:45 AM Marcelo Ricardo Leitner <
>>>> marcelo.leitner@xxxxxxxxx> wrote:
>>>> > - patch2 - fix rtx attack vector
>>>> > - Add the floor value to rto_min to HZ/20 (which fits the values
>>>> > that Michael shared on the other email)
>>>>
>>>> I would encourage allowing minimum RTO values down to 5ms, if the ACK
>>>> policy in the receiver makes this feasible. Our experience is that in
>>>> datacenter environments it can be advantageous to allow timer-based loss
>>>> recoveries using timeout values as low as 5ms, e.g.:
>>>
>>> Thanks Neal. On Xin's tests, the hearbeat timer becomes an issue at
>>> ~25ms already. Xin, can you share more details on the hw, which CPU
>>> was used?
>
> Hi,
>
> Did we reach any decision on this? This continues to produce bug
> reports on syzbot.
I will post a patch later today for the suggestion:
- patch1 - fix issue at hand
- Use the max_t above
to fix this.
As for patch2 and patch3:
- patch2 - fix rtx attack vector
- Add the floor value to rto_min to HZ/20 (which fits the values
that Michael shared on the other email)
- patch3 - speed up initial HB again
- change sctp_cmd_hb_timers_start() so hb timers are kickstarted
when the association is established. AFAICT RFC doesn't specify
when these initial ones should be sent, and I see no issues with
speeding them up.
They are more like improvements, we will do it in the future after
getting more information.
>
> I am not sure whom you are asking, because Xin is you unless I am
> missing something :)
> But if you mean syzbot hardware, then it's GCE VMs with modern Intel
> CPUs but an important aspect is a heavy-debug config (which you can
> take from here https://syzkaller.appspot.com/bug?extid=3dcd59a1f907245f891f)
> and systematic bug reporting. So if it's any flaky in your testing, it
> will produce dozens of bug emails on syzbot.
>
>
>> It was on a KVM guest, "-smp 2,cores=1,threads=1,sockets=2"
>> # lscpu
>> Architecture: x86_64
>> CPU op-mode(s): 32-bit, 64-bit
>> Byte Order: Little Endian
>> CPU(s): 2
>> On-line CPU(s) list: 0,1
>> Thread(s) per core: 1
>> Core(s) per socket: 1
>> Socket(s): 2
>> NUMA node(s): 1
>> Vendor ID: GenuineIntel
>> CPU family: 6
>> Model: 13
>> Model name: QEMU Virtual CPU version 1.5.3
>> Stepping: 3
>> CPU MHz: 2397.222
>> BogoMIPS: 4794.44
>> Hypervisor vendor: KVM
>> Virtualization type: full
>> L1d cache: 32K
>> L1i cache: 32K
>> L2 cache: 4096K
>> NUMA node0 CPU(s): 0,1
>> Flags: fpu de pse tsc msr pae mce cx8 apic sep mtrr
>> pge mca cmov pse36 clflush mmx fxsr sse sse2 syscall nx lm rep_good
>> nopl cpuid pni cx16 hypervisor lahf_lm abm pti
>>
>> If we're counting on max_t to fix this CPU stuck. It should not that
>> matter if min rto < the value causing that stuck.
>>
>>>
>>> Anyway, what about we add a floor to rto_max too, so that RTO can
>>> actually grow into something bigger that don't hog the CPU? Like:
>>> rto_min floor = 5ms
>>> rto_max floor = 50ms
>>>
>>> Marcelo
--
To unsubscribe from this list: send the line "unsubscribe linux-sctp" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
