Issues Meeting – SIP Bar BOF Tuesday, March 20, 2001
Notes taken by David Shrader
Issues to Discuss
- Overlap
- Record
Route
- Early
Media
- Re-INVITE
Glare
- Pre-loaded
Route headers
- SRV
Randomization issues
- 409
- Replaces
Header
- Request-URI
parmaeters
Closed Issues from Previous Meetings
- Re-INVITE
without SDP – closed
- The
re-INVITE without SDP means “no change, keep whatever you already
have”
- Refer-to
issue
Jonathan R indicated that he would re-issue the bis
draft as soon as possible after the IETF meeting.
Issues Work
Record Route
- What
methods do you Record-Route in?
- JR
thought it meant only for session initiation methods, something that
installs state
- RM
Any transaction that has subsequent transactions in the same call leg.
- There
is also a request to not shove garbage into SIP messages because we are
using UDP. We would like to avoid fragmentation because there are problems
with fragments.
- Go
ahead and add RR in other messages that don’t need it, such as Options.
- There
is also a comment about the difficulty of implementing exception cases.
- RS
You should put RR in things that are useful, but MAY put it anywhere.
- JR
added comment that the UA will ignore RR in methods that don’t use it.
- Note:
when RR comes in subsequent messages. The receipt of it MUST NOT update the
route set. “If you have a route set for this call leg and you receive a
RR, don’t create it.” “If you don’t have a route set for a call leg
and you receive a RR, install it.”
- “A
RR in a 200 OK from a PRACK and COMET are both ignored.” “A RR that
occurs inside an INVITE is ignored.” This is a little more difficult in
that the RR within an INVITE may have a limited lifetime. This is added as
issue “RR w/in pending INVITE”.
- JR
– probably add text in bis about the overlap transaction limitations. In
particular, the INVITE is the only transaction that allows overlap.
- Basic
RR issue closed.
Re-INVITE Glare
- If
you have an invite transaction in progress and you receive an INVITE, you
should reject it with a 500 and a Retry-After granularity of seconds. There
are concerns that this was not fast enough with proposals to make faster.
- Is
this even a problem? Is the SDP a one-sided view or a two-sided view? Are
there are other state items in the re-INVITE that cause this to be a
problem?
- SDP
negotiation may be a future problem.
- This
item is still open.
Pre-loaded Route headers
- Clearly
recognized as a need. There are good reasons for it.
- A
request that already has a route in it, still needs to be record-routed.
- Are
there potential security threats to this? These considerations need to be
added. Using pre-loaded headers and modified Route headers in general should
be discussed explicitly. Also modified and pre-loaded Via headers.
- It
is up to the policy of the proxy server to know whether or not to accept
Route headers.
- This
should be clarified in the spec (new RR text) to indicate that pre-loaded
headers might exist.
- Partial
Route processing should already be covered in the text.
- Issue
closed.
409 and the Action Parameter
- If A
sends Registrar with an action parameter of “proxy” and B sends same To
field with an action parameter of “redirect”.
- Conservative
solution: keep action parameter as is, since we don’t know what this
means, we reject the conflicting request
- Liberal
solution: there are cases where you can reasonably determine what happens,
such as method dependent. Allow different actions to be submitted.
- Third
solution: recognize a mistake; Action parameter was prior to CPL and better
mechanisms are now available. This parameter should not be used any longer
in new implementations. Thus, the registrar could “ignore” this or
provide the 409 for a conflict.
- HS
– Current model using 409 is confusing because two user agents could be
conflicting and could have registration problems. You have to look really
hard to notice that UA is not registered.
- Is
this action parameter useful for new applications? Is it realistic for
new/small implementations to support CPL? Is there even a reasonable
expectation of the registrar currently implementing the hint for behavior?
What happens if there is a conflict between the action parameter and the
CPL?
- Application
policy is really what should determine this. We don’t need to have text
indicating when the application should respond with an explicit error code.
- Action
parameter is deprecated. Reg. MAY send a 409.
- Is
this parameter useful or not?
- Simplest
proposal is to change MUST send 409 to a MAY send 409.
- Action
is useful as a hint. There is some language to use it so people want to.
- There
is a comparison with unknown parameters that implies we don’t need to have
it standardized.
- HS
and JR do not want to have a specified item whose meaning is not
identifiable.
- Rough
consensus to remove action from the spec with a reference that it has
historic reference in the RFC.
- Issue
closed.
Request-URI Parameters
- Whatever
was in the URI to begin with, stays there in the message. Nothing gets
stripped out or added.
- Issue
closed.
Unknown URI Parameters
- These
are allowed, in particular for record-route headers
- Text
to be added to the spec for this
- RR
text should be clear that these unknown parameters are reflected in the
Request-URI
- A
proxy that receives unknown parameters in Request-URI just ignores them.
- Issue
closed.
Precedence of Route and Request-URI
- What
if the domain is mine, but parts of the header are not?
- If
we are an outbound proxy, do we take the Request-URI or the Route header to
route the call. Does this imply that all outgoing proxies need to
record-route?
- Relates
to outbound proxies, record routing of outbound proxies.
- This
issue discussion fell apart requiring pictures and more thinking.
Attendees
-
Rohan Mahy
-
Robert Sparks
-
Steve Donovan
-
Ben Campbell
-
Alan Johnston
-
James Undery
-
Neal Deeason
-
Dean Willis
-
Itamar
-
Bill Marshall
-
Don Stanwyck
-
Jonathan Lennox
-
Jayshree Bharatia
-
Mary Barnes
-
James Yu
-
Orit Levin
-
Francois Audet
-
Dan Petrie
-
Tom Taylor
-
Robert Freillich
-
Sean Olsen
-
Adam Roach
-
Gonzalo Caramillo
-
Bob Penfield
-
Joerg Ott
-
Brian Rosen
-
Jonathan Rosenberg
-
Jon Peterson
-
David Shrader
-
Henning Schulzrinne
-
Keith Drage