Re: RFC 8201 Packet Too Big Processing

[Timothy Carlin <[e-mail]>]

On Fri, Apr 10, 2020 at 2:33 PM Timothy Carlin <[e-mail]> wrote:

> On Fri, Apr 10, 2020 at 2:20 PM Fernando Gont <[e-mail]>
> wrote:
>
>> On 10/4/20 14:36, Timothy Carlin wrote:
>> > On Fri, Apr 10, 2020 at 1:27 PM Fernando Gont <[e-mail]
>> > <mailto:[e-mail]>> wrote:
>> >
>> >     On 10/4/20 14:19, Timothy Carlin wrote:
>> >      > Hi Fernando,
>> >      >
>> >      > On Fri, Apr 10, 2020 at 1:14 PM Fernando Gont
>> >     <[e-mail] <mailto:[e-mail]>
>> >      > <mailto:[e-mail] <mailto:[e-mail]>>>
>> wrote:
>> >      >
>> >      >     Hello, Tim,
>> >      >
>> >      >     On 10/4/20 14:07, Timothy Carlin wrote:
>> >      >      > Hello,
>> >      >      >
>> >      >      > We've noticed during testing for RFC 8200 and 8201 that,
>> >     for packets
>> >      >      > larger than 1280, the Linux kernel is processing invalid
>> >     Packet
>> >      >     Too Big
>> >      >      > messages that indicate an MTU less than 1280, and
>> subsequently
>> >      >      > fragmenting packets to a size of 1280. We've seen this
>> >     with 4.15
>> >      >     and 4.18.
>> >      >      >
>> >      >      > This is from Section 4 of RFC 8201:
>> >      >      >
>> >      >      >  >   If a node receives a Packet Too Big message
>> reporting a
>> >      >     next-hop MTU
>> >      >      >  >   that is less than the IPv6 minimum link MTU, it must
>> >     discard it.
>> >      >      >
>> >      >      > Have others noticed this issue with Linux or other
>> OSes?  I'll
>> >      >     also note
>> >      >      > that it correctly does not generate an atomic fragment if
>> the
>> >      >     packet is
>> >      >      > less than 1280 bytes.
>> >      >
>> >      >     I'm trying to understand the scenario...
>> >      >
>> >      >     Host sends a packet of size > 1280
>> >      >     It receives an ICMPv6 PTB < 1280
>> >      >     And it retransmit the packet as a fragmented packet, where
>> >     none of the
>> >      >     fragments is larger than 1280 bytes?
>> >      >
>> >      >
>> >      > This is correct.  Since the ICMPv6 PTB < 1280, and invalid, we
>> would
>> >      > expect the PTB to be discarded, and subsequent packets (for that
>> >      > destination) to remain unfragmented.
>> >
>> >     Agreed. Unless I'm missing something, there's no point in doing that
>> >     (at
>> >     the end of the day, if the offending MTU was < 1280, fragmenting
>> >     packets
>> >     at 1280 will be of no use).
>> >
>> >     Can you provide the exact kernel version, so I may try to take a
>> >     look at
>> >     the kernel code and figure out what's going on?
>> >
>> >
>> > 4.15.0-96-generic and 4.18.0-147 both seem to have this issue.
>>
>> Have you tried with newer kernels? e.g., I'm running 5.3.0-42-generic
>> here.
>
>
>  I have not.  These were from two relatively new distributions, but
> apparently are lagging on the kernel version.  I'll try something newer.
>

Indications are that this remains broken as of 5.3.0-050300-generic.  Let
me know if you want me to try another version.