I have some late transport comments on this ID. The update seems to
retain a lot of thinking that is really historical and I'd really
encourage people to look again to making the document uptodate.
Detailed comments follow.
Best wishes,
Gorry
----
The following text strikes me as a little odd in an update:
" Moreover, TCP implementations that follow the "slow-
start" congestion-avoidance algorithm [CONG] typically calculate and
cache several other values derived from the PMTU. It may be simpler
to receive asynchronous notification when the PMTU changes, so that
these variables may be updated.”
- A modern TCP caches at least some path information in the TCB, why
start with this clause at all:
"Moreover, TCP implementations that follow the "slow start"
congestion-avoidance algorithm [CONG] typically calculate and”
and simply replace this with something like:
"TCP implementations”?
—---
The following text also seems to not reflect a modern TCP stack:
" It is sufficient
to treat this as any other dropped segment, and wait until the
retransmission timer expires to cause retransmission of the segment.”
(and following 3 paras).
Could this be replaced by text that does not exclude modern
retransmission methods:
" It is sufficient
to treat this in the same way as any other dropped segment, and
will be recovered by normal retransmission methods."
—
There is a block of text that describes retransmission triggered by ICMPv6.
Has this code been implemented in modern releases of TCP?:
" Alternatively, the retransmission could be done in immediate response
to a notification that the Path MTU has changed, but only for the
specific connection specified by the Packet Too Big message.”
- It seems to expose a number of attack vectors that really should not
be exposed!!
---
The discussion of NFS may still be a reasonable historic example, but to
be current it should really refer also to NFSv4/TCP as utlising the MTU
discovery provided by TCP, since UDP-based NFS is no longer a key
application.
---
There is no mention that paths including tunnels can eat ICMPv6 PTB
messages on the tunnel segment, blackholing them, which prevents
reaching the destination.
---
I think the security consideration is naive!
This statement in particular seems to open DOS vulnerability:
"
When a node receives a Packet Too Big message, it MUST reduce its
estimate of the PMTU for the relevant path, based on the value of the
MTU field in the message."
- Introdueces a significant vulnerability. A rogue PTB message that
reduces the PMTU to a minimum, can result in a path too small to carry
an encapsulated packet. (Recently noted by Fernando Gont).
Moreover, other layers view ICMP messages with suspicion and have long
noted the need to check ICMP payload and match only packets that relate
to actual 5-tuples in use (effectively reducing vulnerability to
off-path attacks). For example, the Guidelines for UDP, rfc5405bis, state:
" Applications SHOULD appropriately validate the payload of ICMP
messages to ensure these are received in response to transmitted
traffic (i.e., a reported error condition that corresponds to a UDP
datagram actually sent by the application). …“
- clearly handling this in IP-layer tunnels can be troublesome, but
that's a problem that should be described, not obscured.
——
I’d finally like to add my concerns about the understatement of the
value of PLPMTUD, which seems to not reflect the recommendations to use
this method:
“ It defines a method for Packetization Layer Path
MTU Discovery (PLPMTUD) designed for use over paths where delivery of
ICMP messages to a host is not assured.”
This seems under-stating the value and recommendations to deploy
PLMTUD, compared with current transport-area recommendations, for
instance, the UDP Guidelines provide much more on this important design
consideration:
" Packetization Layer Path MTU Discovery (PLPMTUD) [RFC4821] does not
rely upon network support for ICMP messages and is therefore
considered more robust than standard PMTUD. It is not susceptible to
"black holing" of ICMP message. To operate, PLPMTUD requires changes
to the way the transport is used, both to transmit probe packets, and
to account for the loss or success of these probes. This updates not
only the PMTU algorithm, it also impacts loss recovery, congestion
control, etc. These updated mechanisms can be implemented within a
connection-oriented transport (e.g., TCP, SCTP, DCCP), but are not a
part of UDP, but this type of feedback is not typically present for
unidirectional applications."
----
The examples used in the definition of "upper layer" and "link" also
makes this document appear as historic, rather than a new RFC!