RFC 3973
Protocol Independent Multicast - Dense Mode (PIM-DM): Protocol Specification (Revised)
4.4. PIM-DM Prune, Join, and Graft Messages
This section describes the generation and processing of PIM-DM Join, Prune, and Graft messages. Prune messages are sent toward the upstream neighbor for S to indicate that traffic from S addressed to group G is not desired. In the case of downstream routers A and B, where A wishes to continue receiving data and B does not, A will send a Join in response to B's Prune to override the Prune. This is the only situation in PIM-DM in which a Join message is used. Finally, a Graft message is used to re-join a previously pruned branch to the delivery tree.
4.4.1. Upstream Prune, Join, and Graft Messages
The Upstream(S,G) state machine for sending Prune, Graft, and Join messages is given below. There are three states. Forwarding (F) This is the starting state of the Upsteam(S,G) state machine. The state machine is in this state if it just started or if oiflist(S,G) != NULL. Pruned (P) The set, olist(S,G), is empty. The router will not forward data from S addressed to group G. AckPending (AP) The router was in the Pruned(P) state, but a transition has occurred in the Downstream(S,G) state machine for one of this (S,G) entry's outgoing interfaces, indicating that traffic from S addressed to G should again be forwarded. A Graft message has been sent to RPF'(S), but a Graft Ack message has not yet been received. In addition, there are three state-machine-specific timers: GraftRetry Timer (GRT(S,G)) This timer is set when a Graft is sent upstream. If a corresponding GraftAck is not received before the timer expires, then another Graft is sent, and the GraftRetry Timer is reset. The timer is stopped when a Graft Ack message is received. This timer is normally set to Graft_Retry_Period (see 4.8). Override Timer (OT(S,G)) This timer is set when a Prune(S,G) is received on the upstream interface where olist(S,G) != NULL. When the timer expires, a Join(S,G) message is sent on the upstream interface. This timer is normally set to t_override (see 4.8). Prune Limit Timer (PLT(S,G)) This timer is used to rate-limit Prunes on a LAN. It is only used when the Upstream(S,G) state machine is in the Pruned state. A Prune cannot be sent if this timer is running. This timer is normally set to t_limit (see 4.8).
+-------------+ +-------------+ | | olist == NULL | | | Forward |----------------------->| Pruned | | | | | +-------------+ +-------------+ ^ | ^ | | | | | | |RPF`(S) Changes olist == NULL| | | | | | | | +-------------+ | | | +-------->| |----------+ | | | AckPending | | +-------------| |<-------------+ Rcv GraftAck OR +-------------+ olist != NULL Rcv State Refresh With (P==0) OR S Directly Connect Figure 1: Upstream Interface State Machine In tabular form, the state machine is defined as follows: +-------------------------------+--------------------------------------+ | | Previous State | | +------------+------------+------------+ | Event | Forwarding | Pruned | AckPending | +-------------------------------+------------+------------+------------+ | Data packet arrives on | ->P Send | ->P Send | N/A | | RPF_Interface(S) AND | Prune(S,G) | Prune(S,G) | | | olist(S,G) == NULL AND |Set PLT(S,G)|Set PLT(S,G)| | | PLT(S,G) not running | | | | +-------------------------------+------------+------------+------------+ | State Refresh(S,G) received | ->F Set | ->P Reset |->AP Set | | from RPF`(S) AND | OT(S,G) | PLT(S,G) | OT(S,G) | | Prune Indicator == 1 | | | | +-------------------------------+------------+------------+------------+ | State Refresh(S,G) received | ->F | ->P Send |->F Cancel | | from RPF`(S) AND | | Prune(S,G) | GRT(S,G) | | Prune Indicator == 0 AND | |Set PLT(S,G)| | | PLT(S,G) not running | | | | +-------------------------------+------------+------------+------------+
+-------------------------------+--------------------------------------+ | | Previous State | + +------------+------------+------------+ | Event | Forwarding | Pruned | AckPending | +-------------------------------+------------+------------+------------+ | See Join(S,G) to RPF'(S) | ->F Cancel | ->P |->AP Cancel | | | OT(S,G) | | OT(S,G) | +-------------------------------+------------+------------+------------+ | See Prune(S,G) | ->F Set | ->P |->AP Set | | | OT(S,G) | | OT(S,G) | +-------------------------------+------------+------------+------------+ | OT(S,G) Expires | ->F Send | N/A |->AP Send | | | Join(S,G) | | Join(S,G) | +-------------------------------+------------+------------+------------+ | olist(S,G)->NULL | ->P Send | N/A |->P Send | | | Prune(S,G) | | Prune(S,G) | | |Set PLT(S,G)| |Set PLT(S,G)| | | | | Cancel | | | | | GRT(S,G) | +-------------------------------+------------+------------+------------+ | olist(S,G)->non-NULL | N/A | ->AP Send | N/A | | | | Graft(S,G) | | | | |Set GRT(S,G)| | +-------------------------------+------------+------------+------------+ | RPF'(S) Changes AND | ->AP Send | ->AP Send |->AP Send | | olist(S,G) != NULL | Graft(S,G) | Graft(S,G) | Graft(S,G) | | |Set GRT(S,G)|Set GRT(S,G)|Set GRT(S,G)| +-------------------------------+------------+------------+------------+ | RPF'(S) Changes AND | ->P | ->P Cancel |->P Cancel | | olist(S,G) == NULL | | PLT(S,G) | GRT(S,G) | +-------------------------------+------------+------------+------------+ | S becomes directly connected | ->F | ->P |->F Cancel | | | | | GRT(S,G) | +-------------------------------+------------+------------+------------+ | GRT(S,G) Expires | N/A | N/A |->AP Send | | | | | Graft(S,G) | | | | |Set GRT(S,G)| +-------------------------------+------------+------------+------------+ | Receive GraftAck(S,G) from | ->F | ->P |->F Cancel | | RPF'(S) | | | GRT(S,G) | +-------------------------------+------------+------------+------------+ The transition event "RcvGraftAck(S,G)" implies receiving a Graft Ack message targeted to this router's address on the incoming interface for the (S,G) entry. If the destination address is not correct, the state transitions in this state machine must not occur.
4.4.1.1. Transitions from the Forwarding (F) State
When the Upstream(S,G) state machine is in the Forwarding (F) state, the following events may trigger a transition: Data Packet arrives on RPF_Interface(S) AND olist(S,G) == NULL AND S NOT directly connected The Upstream(S,G) state machine MUST transition to the Pruned (P) state, send a Prune(S,G) to RPF'(S), and set PLT(S,G) to t_limit seconds. State Refresh(S,G) Received from RPF'(S) The Upstream(S,G) state machine remains in a Forwarding state. If the received State Refresh has the Prune Indicator bit set to one, this router must override the upstream router's Prune state after a short random interval. If OT(S,G) is not running and the Prune Indicator bit equals one, the router MUST set OT(S,G) to t_override seconds. See Join(S,G) to RPF'(S) This event is only relevant if RPF_interface(S) is a shared medium. This router sees another router on RPF_interface(S) send a Join(S,G) to RPF'(S,G). If the OT(S,G) is running, then it means that the router had scheduled a Join to override a previously received Prune. Another router has responded more quickly with a Join, so the local router SHOULD cancel its OT(S,G), if it is running. The Upstream(S,G) state machine remains in the Forwarding (F) state. See Prune(S,G) AND S NOT directly connected This event is only relevant if RPF_interface(S) is a shared medium. This router sees another router on RPF_interface(S) send a Prune(S,G). As this router is in Forwarding state, it must override the Prune after a short random interval. If OT(S,G) is not running, the router MUST set OT(S,G) to t_override seconds. The Upstream(S,G) state machine remains in Forwarding (F) state. OT(S,G) Expires AND S NOT directly connected The OverrideTimer (OT(S,G)) expires. The router MUST send a Join(S,G) to RPF'(S) to override a previously detected prune. The Upstream(S,G) state machine remains in the Forwarding (F) state. olist(S,G) -> NULL AND S NOT directly connected The Upstream(S,G) state machine MUST transition to the Pruned (P) state, send a Prune(S,G) to RPF'(S), and set PLT(S,G) to t_limit seconds.
RPF'(S) Changes AND olist(S,G) is non-NULL AND S NOT directly
connected
Unicast routing or Assert state causes RPF'(S) to change,
including changes to RPF_Interface(S). The Upstream(S,G) state
machine MUST transition to the AckPending (AP) state, unicast a
Graft to the new RPF'(S), and set the GraftRetry Timer (GRT(S,G))
to Graft_Retry_Period.
RPF'(S) Changes AND olist(S,G) is NULL
Unicast routing or Assert state causes RPF'(S) to change,
including changes to RPF_Interface(S). The Upstream(S,G) state
machine MUST transition to the Pruned (P) state.
4.4.1.2. Transitions from the Pruned (P) State
When the Upstream(S,G) state machine is in the Pruned (P) state, the
following events may trigger a transition:
Data arrives on RPF_interface(S) AND PLT(S,G) not running AND S NOT
directly connected
Either another router on the LAN desires traffic from S addressed
to G or a previous Prune was lost. To prevent generating a
Prune(S,G) in response to every data packet, the PruneLimit Timer
(PLT(S,G)) is used. Once the PLT(S,G) expires, the router needs
to send another prune in response to a data packet not received
directly from the source. A Prune(S,G) MUST be sent to RPF'(S),
and the PLT(S,G) MUST be set to t_limit.
State Refresh(S,G) Received from RPF'(S)
The Upstream(S,G) state machine remains in a Pruned state. If
the State Refresh has its Prune Indicator bit set to zero and
PLT(S,G) is not running, a Prune(S,G) MUST be sent to RPF'(S),
and the PLT(S,G) MUST be set to t_limit. If the State Refresh
has its Prune Indicator bit set to one, the router MUST reset
PLT(S,G) to t_limit.
See Prune(S,G) to RPF'(S)
A Prune(S,G) is seen on RPF_interface(S) to RPF'(S). The
Upstream(S,G) state machine stays in the Pruned (P) state. The
router MAY reset its PLT(S,G) to the value in the Holdtime field
of the received message if it is greater than the current value
of the PLT(S,G).
olist(S,G)->non-NULL AND S NOT directly connected
The set of interfaces defined by the olist(S,G) macro becomes
non-empty, indicating that traffic from S addressed to group G
must be forwarded. The Upstream(S,G) state machine MUST cancel
PLT(S,G), transition to the AckPending (AP) state and unicast a
Graft message to RPF'(S). The Graft Retry Timer (GRT(S,G)) MUST
be set to Graft_Retry_Period.
RPF'(S) Changes AND olist(S,G) == non-NULL AND S NOT directly
connected
Unicast routing or Assert state causes RPF'(S) to change,
including changes to RPF_Interface(S). The Upstream(S,G) state
machine MUST cancel PLT(S,G), transition to the AckPending (AP)
state, send a Graft unicast to the new RPF'(S), and set the
GraftRetry Timer (GRT(S,G)) to Graft_Retry_Period.
RPF'(S) Changes AND olist(S,G) == NULL AND S NOT directly connected
Unicast routing or Assert state causes RPF'(S) to change,
including changes to RPF_Interface(S). The Upstream(S,G) state
machine stays in the Pruned (P) state and MUST cancel the
PLT(S,G) timer.
S becomes directly connected
Unicast routing changed so that S is directly connected. The
Upstream(S,G) state machine remains in the Pruned (P) state.
4.4.1.3. Transitions from the AckPending (AP) State
When the Upstream(S,G) state machine is in the AckPending (AP) state,
the following events may trigger a transition:
State Refresh(S,G) Received from RPF'(S) with Prune Indicator == 1
The Upstream(S,G) state machine remains in an AckPending state.
The router must override the upstream router's Prune state after
a short random interval. If OT(S,G) is not running and the Prune
Indicator bit equals one, the router MUST set OT(S,G) to
t_override seconds.
State Refresh(S,G) Received from RPF'(S) with Prune Indicator == 0
The router MUST cancel its GraftRetry Timer (GRT(S,G)) and
transition to the Forwarding (F) state.
See Join(S,G) to RPF'(S,G)
This event is only relevant if RPF_interface(S) is a shared
medium. This router sees another router on RPF_interface(S) send
a Join(S,G) to RPF'(S,G). If the OT(S,G) is running, then it
means that the router had scheduled a Join to override a
previously received Prune. Another router has responded more
quickly with a Join, so the local router SHOULD cancel its
OT(S,G), if it is running. The Upstream(S,G) state machine
remains in the AckPending (AP) state.
See Prune(S,G)
This event is only relevant if RPF_interface(S) is a shared
medium. This router sees another router on RPF_interface(S) send
a Prune(S,G). As this router is in AckPending (AP) state, it
must override the Prune after a short random interval. If
OT(S,G) is not running, the router MUST set OT(S,G) to t_override
seconds. The Upstream(S,G) state machine remains in AckPending
(AP) state.
OT(S,G) Expires
The OverrideTimer (OT(S,G)) expires. The router MUST send a
Join(S,G) to RPF'(S). The Upstream(S,G) state machine remains in
the AckPending (AP) state.
olist(S,G) -> NULL
The set of interfaces defined by the olist(S,G) macro becomes
null, indicating that traffic from S addressed to group G should
no longer be forwarded. The Upstream(S,G) state machine MUST
transition to the Pruned (P) state. A Prune(S,G) MUST be
multicast to the RPF_interface(S), with RPF'(S) named in the
upstream neighbor field. The GraftRetry Timer (GRT(S,G)) MUST be
cancelled, and PLT(S,G) MUST be set to t_limit seconds.
RPF'(S) Changes AND olist(S,G) does not become NULL AND S NOT
directly connected
Unicast routing or Assert state causes RPF'(S) to change,
including changes to RPF_Interface(S). The Upstream(S,G) state
machine stays in the AckPending (AP) state. A Graft MUST be
unicast to the new RPF'(S) and the GraftRetry Timer (GRT(S,G))
reset to Graft_Retry_Period.
RPF'(S) Changes AND olist(S,G) == NULL AND S NOT directly connected
Unicast routing or Assert state causes RPF'(S) to change,
including changes to RPF_Interface(S). The Upstream(S,G) state
machine MUST transition to the Pruned (P) state. The GraftRetry
Timer (GRT(S,G)) MUST be cancelled.
S becomes directly connected
Unicast routing has changed so that S is directly connected. The
GraftRetry Timer MUST be cancelled, and the Upstream(S,G) state
machine MUST transition to the Forwarding(F) state.
GRT(S,G) Expires
The GraftRetry Timer (GRT(S,G)) expires for this (S,G) entry.
The Upstream(S,G) state machine stays in the AckPending (AP)
state. Another Graft message for (S,G) SHOULD be unicast to
RPF'(S) and the GraftRetry Timer (GRT(S,G)) reset to
Graft_Retry_Period. It is RECOMMENDED that the router retry a
configured number of times before ceasing retries.
See GraftAck(S,G) from RPF'(S)
A GraftAck is received from RPF'(S). The GraftRetry Timer MUST
be cancelled, and the Upstream(S,G) state machine MUST transition
to the Forwarding(F) state.
4.4.2. Downstream Prune, Join, and Graft Messages
The Prune(S,G) Downstream state machine for receiving Prune, Join and
Graft messages on interface I is given below. This state machine
MUST always be in the NoInfo state on the upstream interface. It
contains three states.
NoInfo(NI)
The interface has no (S,G) Prune state, and neither the Prune
timer (PT(S,G,I)) nor the PrunePending timer ((PPT(S,G,I)) is
running.
PrunePending(PP)
The router has received a Prune(S,G) on this interface from a
downstream neighbor and is waiting to see whether the prune will
be overridden by another downstream router. For forwarding
purposes, the PrunePending state functions exactly like the
NoInfo state.
Pruned(P)
The router has received a Prune(S,G) on this interface from a
downstream neighbor, and the Prune was not overridden. Data from
S addressed to group G is no longer being forwarded on this
interface.
In addition, there are two timers:
PrunePending Timer (PPT(S,G,I))
This timer is set when a valid Prune(S,G) is received. Expiry of
the PrunePending Timer (PPT(S,G,I)) causes the interface to
transition to the Pruned state.
Prune Timer (PT(S,G,I))
This timer is set when the PrunePending Timer (PT(S,G,I))
expires. Expiry of the Prune Timer (PT(S,G,I)) causes the
interface to transition to the NoInfo (NI) state, thereby
allowing data from S addressed to group G to be forwarded on the
interface.
+-------------+ +-------------+
| | PPT Expires | |
|PrunePending |----------------------->| Pruned |
| | | |
+-------------+ +-------------+
| ^ |
| | |
| |Rcv Prune |
| | |
| | +-------------+ |
| +---------| | |
| | NoInfo |<-------------+
+------------>| | Rcv Join/Graft OR
Rcv Join/Graft OR +-------------+ PT Expires OR
RPF_Interface(S)->I RPF_Interface(S)->I
Figure 2: Downstream Interface State Machine
In tabular form, the state machine is as follows: +-------------------------------+--------------------------------------+ | | Previous State | + +------------+------------+------------+ | Event | No Info | PrunePend | Pruned | +-------------------------------+------------+------------+------------+ | Receive Prune(S,G) |->PP Set |->PP |->P Reset | | | PPT(S,G,I) | | PT(S,G,I) | +-------------------------------+------------+------------+------------+ | Receive Join(S,G) |->NI |->NI Cancel |->NI Cancel | | | | PPT(S,G,I) | PT(S,G,I) | +-------------------------------+------------+------------+------------+ | Receive Graft(S,G) |->NI Send |->NI Send |->NI Send | | | GraftAck | GraftAck | GraftAck | | | | Cancel | Cancel | | | | PPT(S,G,I) | PT(S,G,I) | +-------------------------------+------------+------------+------------+ | PPT(S,G) Expires | N/A |->P Set | N/A | | | | PT(S,G,I) | | +-------------------------------+------------+------------+------------+ | PT(S,G) Expires | N/A | N/A |->NI | +-------------------------------+------------+------------+------------+ | RPF_Interface(S) becomes I |->NI |->NI Cancel |->NI Cancel | | | | PPT(S,G,I) | PT(S,G,I) | +-------------------------------+------------+------------+------------+ | Send State Refresh(S,G) out I |->NI |->PP |->P Reset | | | | | PT(S,G,I) | +-------------------------------+------------+------------+------------+ The transition events "Receive Graft(S,G)", "Receive Prune(S,G)", and "Receive Join(S,G)" denote receiving a Graft, Prune, or Join message in which this router's address on I is contained in the message's upstream neighbor field. If the upstream neighbor field does not match this router's address on I, then these state transitions in this state machine must not occur.4.4.2.1. Transitions from the NoInfo State
When the Prune(S,G) Downstream state machine is in the NoInfo (NI) state, the following events may trigger a transition: Receive Prune(S,G) A Prune(S,G) is received on interface I with the upstream neighbor field set to the router's address on I. The Prune(S,G) Downstream state machine on interface I MUST transition to the PrunePending (PP) state. The PrunePending Timer (PPT(S,G,I)) MUST be set to J/P_Override_Interval if the router has more than
one neighbor on I. If the router has only one neighbor on
interface I, then it SHOULD set the PPT(S,G,I) to zero,
effectively transitioning immediately to the Pruned (P) state.
Receive Graft(S,G)
A Graft(S,G) is received on the interface I with the upstream
neighbor field set to the router's address on I. The Prune(S,G)
Downstream state machine on interface I stays in the NoInfo (NI)
state. A GraftAck(S,G) MUST be unicast to the originator of the
Graft(S,G) message.
4.4.2.2. Transitions from the PrunePending (PP) State
When the Prune(S,G) downstream state machine is in the PrunePending
(PP) state, the following events may trigger a transition.
Receive Join(S,G)
A Join(S,G) is received on interface I with the upstream neighbor
field set to the router's address on I. The Prune(S,G)
Downstream state machine on interface I MUST transition to the
NoInfo (NI) state. The PrunePending Timer (PPT(S,G,I)) MUST be
cancelled.
Receive Graft(S,G)
A Graft(S,G) is received on interface I with the upstream
neighbor field set to the router's address on I. The Prune(S,G)
Downstream state machine on interface I MUST transition to the
NoInfo (NI) state and MUST unicast a Graft Ack message to the
Graft originator. The PrunePending Timer (PPT(S,G,I)) MUST be
cancelled.
PPT(S,G,I) Expires
The PrunePending Timer (PPT(S,G,I)) expires, indicating that no
neighbors have overridden the previous Prune(S,G) message. The
Prune(S,G) Downstream state machine on interface I MUST
transition to the Pruned (P) state. The Prune Timer (PT(S,G,I))
is started and MUST be initialized to the received
Prune_Hold_Time minus J/P_Override_Interval. A PruneEcho(S,G)
MUST be sent on I if I has more than one PIM neighbor. A
PruneEcho(S,G) is simply a Prune(S,G) message multicast by the
upstream router to a LAN, with itself as the Upstream Neighbor.
Its purpose is to add additional reliability so that if a Join
that should have overridden the Prune is lost locally on the LAN,
the PruneEcho(S,G) may be received and trigger a new Join
message. A PruneEcho(S,G) is OPTIONAL on an interface with only
one PIM neighbor. In addition, the router MUST evaluate any
possible transitions in the Upstream(S,G) state machine.
RPF_Interface(S) becomes interface I
The upstream interface for S has changed. The Prune(S,G)
Downstream state machine on interface I MUST transition to the
NoInfo (NI) state. The PrunePending Timer (PPT(S,G,I)) MUST be
cancelled.
4.4.2.3. Transitions from the Prune (P) State
When the Prune(S,G) Downstream state machine is in the Pruned (P)
state, the following events may trigger a transition.
Receive Prune(S,G)
A Prune(S,G) is received on the interface I with the upstream
neighbor field set to the router's address on I. The Prune(S,G)
Downstream state machine on interface I remains in the Pruned (P)
state. The Prune Timer (PT(S,G,I)) SHOULD be reset to the
holdtime contained in the Prune(S,G) message if it is greater
than the current value.
Receive Join(S,G)
A Join(S,G) is received on the interface I with the upstream
neighbor field set to the router's address on I. The Prune(S,G)
downstream state machine on interface I MUST transition to the
NoInfo (NI) state. The Prune Timer (PT(S,G,I)) MUST be
cancelled. The router MUST evaluate any possible transitions in
the Upstream(S,G) state machine.
Receive Graft(S,G)
A Graft(S,G) is received on interface I with the upstream
neighbor field set to the router's address on I. The Prune(S,G)
Downstream state machine on interface I MUST transition to the
NoInfo (NI) state and send a Graft Ack back to the Graft's
source. The Prune Timer (PT(S,G,I)) MUST be cancelled. The
router MUST evaluate any possible transitions in the
Upstream(S,G) state machine.
PT(S,G,I) Expires
The Prune Timer (PT(S,G,I)) expires, indicating that it is again
time to flood data from S addressed to group G onto interface I.
The Prune(S,G) Downstream state machine on interface I MUST
transition to the NoInfo (NI) state. The router MUST evaluate
any possible transitions in the Upstream(S,G) state machine.
RPF_Interface(S) becomes interface I
The upstream interface for S has changed. The Prune(S,G)
Downstream state machine on interface I MUST transition to the
NoInfo (NI) state. The PruneTimer (PT(S,G,I)) MUST be cancelled.
Send State Refresh(S,G) out interface I
The router has refreshed the Prune(S,G) state on interface I.
The router MUST reset the Prune Timer (PT(S,G,I)) to the Holdtime
from an active Prune received on interface I. The Holdtime used
SHOULD be the largest active one but MAY be the most recently
received active Prune Holdtime.
4.5. State Refresh
This section describes the major portions of the state refresh
mechanism.
4.5.1. Forwarding of State Refresh Messages
When a State Refresh message, SRM, is received, it is forwarded
according to the following pseudo-code.
if (iif != RPF_interface(S))
return;
if (RPF'(S) != srcaddr(SRM))
return;
if (StateRefreshRateLimit(S,G) == TRUE)
return;
for each interface I in pim_nbrs {
if (TTL(SRM) == 0 OR (TTL(SRM) - 1) < Threshold(I))
continue; /* Out of TTL, skip this interface */
if (boundary(I,G))
continue; /* This interface is scope boundary, skip it */
if (I == iif)
continue; /* This is the incoming interface, skip it */
if (lost_assert(S,G,I) == TRUE)
continue; /* Let the Assert Winner do State Refresh */
Copy SRM to SRM'; /* Make a copy of SRM to forward */
if (I contained in prunes(S,G)) {
set Prune Indicator bit of SRM' to 1;
if StateRefreshCapable(I) == TRUE
set PT(S,G) to largest active holdtime read from a Prune
message accepted on I;
} else {
set Prune Indicator bit of SRM' to 0;
}
set srcaddr(SRM') to my_addr(I);
set TTL of SRM' to TTL(SRM) - 1;
set metric of SRM' to metric of unicast route used to reach S;
set pref of SRM' to preference of unicast route used to reach S;
set mask of SRM' to mask of route used to reach S;
if (AssertState == NoInfo) {
set Assert Override of SRM' to 1;
} else {
set Assert Override of SRM' to 0;
}
transmit SRM' on I;
}
The pseudocode above employs the following macro definitions.
Boundary(I,G) is TRUE if an administratively scoped boundary for
group G is configured on interface I.
StateRefreshCapable(I) is TRUE if all neighbors on an interface use
the State Refresh option.
StateRefreshRateLimit(S,G) is TRUE if the time elapsed since the last
received StateRefresh(S,G) is less than the configured
RefreshLimitInterval.
TTL(SRM) returns the TTL contained in the State Refresh Message, SRM.
This is different from the TTL contained in the IP header.
Threshold(I) returns the minimum TTL that a packet must have before
it can be transmitted on interface I.
srcaddr(SRM) returns the source address contained in the network
protocol (e.g., IPv4) header of the State Refresh Message, SRM.
my_addr(I) returns this node's network (e.g., IPv4) address on
interface I.
4.5.2. State Refresh Message Origination
This section describes the origination of State Refresh messages. These messages are generated periodically by the PIM-DM router directly connected to a source. One Origination(S,G) state machine exists per (S,G) entry in a PIM-DM router. The Origination(S,G) state machine has the following states: NotOriginator(NO) This is the starting state of the Origination(S,G) state machine. While in this state, a router will not originate State Refresh messages for the (S,G) pair. Originator(O) When in this state the router will periodically originate State Refresh messages. Only routers directly connected to S may transition to this state. In addition, there are two state machine specific timers: State Refresh Timer (SRT(S,G)) This timer controls when State Refresh messages are generated. The timer is initially set when that Origination(S,G) state machine transitions to the O state. It is cancelled when the Origination(S,G) state machine transitions to the NO state. This timer is normally set to StateRefreshInterval (see 4.8). Source Active Timer (SAT(S,G)) This timer is first set when the Origination(S,G) state machine transitions to the O state and is reset on the receipt of every data packet from S addressed to group G. When it expires, the Origination(S,G) state machine transitions to the NO state. This timer is normally set to SourceLifetime (see 4.8). +-------------+ Rcv Directly From S +-------------+ | |----------------------->| | |NotOriginator| | Originator | | |<-----------------------| | +-------------+ SAT Expires OR +-------------+ S NOT Direct Connect Figure 3: State Refresh State Machine
In tabular form, the state machine is defined as follows: +----------------------------------------------------------------------+ | | Previous State | | +---------------+-------------------+ | Event | NotOriginator | Originator | +----------------------------------+---------------+-------------------+ | Receive Data from S AND | ->O | ->O Reset | | S directly connected | Set SRT(S,G) | SAT(S,G) | | | Set SAT(S,G) | | +----------------------------------+---------------+-------------------+ | SRT(S,G) Expires | N/A | ->O Send | | | | StateRefresh(S,G) | | | | Reset SRT(S,G) | +----------------------------------+---------------+-------------------+ | SAT(S,G) Expires | N/A | ->NO Cancel | | | | SRT(S,G) | +----------------------------------+---------------+-------------------+ | S no longer directly connected | ->NO | ->NO | | | | Cancel SRT(S,G) | | | | Cancel SAT(S,G) | +----------------------------------+---------------+-------------------+4.5.2.1. Transitions from the NotOriginator (NO) State
When the Originating(S,G) state machine is in the NotOriginator (NO) state, the following event may trigger a transition: Data Packet received from directly connected Source S addressed to group G The router MUST transition to an Originator (O) state, set SAT(S,G) to SourceLifetime, and set SRT(S,G) to StateRefreshInterval. The router SHOULD record the TTL of the packet for use in State Refresh messages.4.5.2.2. Transitions from the Originator (O) State
When the Originating(S,G) state machine is in the Originator (O) state, the following events may trigger a transition: Receive Data Packet from S addressed to G The router remains in the Originator (O) state and MUST reset SAT(S,G) to SourceLifetime. The router SHOULD increase its recorded TTL to match the TTL of the packet, if the packet's TTL is larger than the previously recorded TTL. A router MAY record the TTL based on an implementation specific sampling policy to avoid examining the TTL of every multicast packet it handles.
SRT(S,G) Expires
The router remains in the Originator (O) state and MUST reset
SRT(S,G) to StateRefreshInterval. The router MUST also generate
State Refresh messages for transmission, as described in the
State Refresh Forwarding rules (Section 4.5.1), except for the
TTL. If the TTL of data packets from S to G are being recorded,
then the TTL of each State Refresh message is set to the highest
recorded TTL. Otherwise, the TTL is set to the configured State
Refresh TTL. Let I denote the interface over which a State
Refresh message is being sent. If the Prune(S,G) Downstream
state machine is in the Pruned (P) state, then the Prune-
Indicator bit MUST be set to 1 in the State Refresh message being
sent over I. Otherwise, the Prune-Indicator bit MUST be set to 0.
SAT(S,G) Expires
The router MUST cancel the SRT(S,G) timer and transition to the
NotOriginator (NO) state.
S is no longer directly connected
The router MUST transition to the NotOriginator (NO) state and
cancel both the SAT(S,G) and SRT(S,G).
4.6. PIM Assert Messages
4.6.1. Assert Metrics
Assert metrics are defined as follows:
struct assert_metric {
metric_preference;
route_metric;
ip_address;
};
When assert_metrics are compared, the metric_preference and
route_metric field are compared in order, where the first lower value
wins. If all fields are equal, the IP address of the router that
sourced the Assert message is used as a tie-breaker, with the highest
IP address winning.
An Assert metric for (S,G) to include in (or compare against) an
Assert message sent on interface I should be computed by using the
following pseudocode:
assert_metric
my_assert_metric(S,G,I) {
if (CouldAssert(S,G,I) == TRUE) {
return spt_assert_metric(S,G,I)
} else {
return infinite_assert_metric()
}
}
spt_assert_metric(S,I) gives the Assert metric we use if we're
sending an Assert based on active (S,G) forwarding state:
assert_metric
spt_assert_metric(S,I) {
return {0,MRIB.pref(S),MRIB.metric(S),my_addr(I)}
}
MRIB.pref(X) and MRIB.metric(X) are the routing preference and
routing metrics associated with the route to a particular (unicast)
destination X, as determined by the MRIB. my_addr(I) is simply the
router's network (e.g., IP) address associated with the local
interface I.
infinite_assert_metric() gives the Assert metric we need to send an
Assert but doesn't match (S,G) forwarding state:
assert_metric
infinite_assert_metric() {
return {1,infinity,infinity,0}
}
4.6.2. AssertCancel Messages
An AssertCancel(S,G) message is simply an Assert message for (S,G)
with infinite metric. The Assert winner sends this message when it
changes its upstream interface to this interface. Other routers will
see this metric, causing those with forwarding state to send their
own Asserts and re-establish an Assert winner.
AssertCancel messages are simply an optimization. The original
Assert timeout mechanism will eventually allow a subnet to become
consistent; the AssertCancel mechanism simply causes faster
convergence. No special processing is required for an AssertCancel
message, as it is simply an Assert message from the current winner.
4.6.3. Assert State Macros
The macro lost_assert(S,G,I), is used in the olist computations of Section 4.1.3, and is defined as follows: bool lost_assert(S,G,I) { if ( RPF_interface(S) == I ) { return FALSE } else { return (AssertWinner(S,G,I) != me AND (AssertWinnerMetric(S,G,I) is better than spt_assert_metric(S,G,I))) } } AssertWinner(S,G,I) defaults to NULL, and AssertWinnerMetric(S,G,I) defaults to Infinity when in the NoInfo state.4.6.4. (S,G) Assert Message State Machine
The (S,G) Assert state machine for interface I is shown in Figure 4. There are three states: NoInfo (NI) This router has no (S,G) Assert state on interface I. I am Assert Winner (W) This router has won an (S,G) Assert on interface I. It is now responsible for forwarding traffic from S destined for G via interface I. I am Assert Loser (L) This router has lost an (S,G) Assert on interface I. It must not forward packets from S destined for G onto interface I. In addition, an Assert Timer (AT(S,G,I)) is used to time out the Assert state.
+-------------+ +-------------+ | | Rcv Pref Assert or SR | | | Winner |----------------------->| Loser | | | | | +-------------+ +-------------+ ^ | ^ | | | Rcv Pref Assert or| | | |AT Expires OR State Refresh| | | |CouldAssert->FALSE | | | | | | | | +-------------+ | | | +-------->| |----------+ | | | No Info | | +-------------| |<-------------+ Rcv Data from dnstrm +-------------+ Rcv Inf Assert from Win OR OR Rcv Inferior Assert Rcv Inf SR from Winner OR OR Rcv Inferior SR AT Expires OR CouldAssert Changes OR Winner's NLT Expires Figure 4: Assert State Machine In tabular form, the state machine is defined as follows: +-------------------------------+--------------------------------------+ | | Previous State | | +------------+------------+------------+ | Event | No Info | Winner | Loser | +-------------------------------+------------+------------+------------+ | An (S,G) Data packet received | ->W Send | ->W Send | ->L | | on downstream interface | Assert(S,G)| Assert(S,G)| | | | Set | Set | | | | AT(S,G,I) | AT(S,G,I) | | +-------------------------------+--------------------------------------+ | Receive Inferior (Assert OR | N/A | N/A |->NI Cancel | | State Refresh) from Assert | | | AT(S,G,I) | | Winner | | | | +-------------------------------+--------------------------------------+ | Receive Inferior (Assert OR | ->W Send | ->W Send | ->L | | State Refresh) from non-Assert| Assert(S,G)| Assert(S,G)| | | Winner AND CouldAssert==TRUE | Set | Set | | | | AT(S,G,I) | AT(S,G,I) | | +-------------------------------+--------------------------------------+
+-------------------------------+--------------------------------------+ | | Previous State | | +------------+------------+------------+ | Event | No Info | Winner | Loser | +-------------------------------+------------+------------+------------+ | Receive Preferred Assert OR | ->L Send | ->L Send | ->L Set | | State Refresh | Prune(S,G) | Prune(S,G) | AT(S,G,I) | | | Set | Set | | | | AT(S,G,I) | AT(S,G,I) | | +-------------------------------+--------------------------------------+ | Send State Refresh | ->NI | ->W Reset | N/A | | | | AT(S,G,I) | | +-------------------------------+--------------------------------------+ | AT(S,G) Expires | N/A | ->NI | ->NI | +-------------------------------+--------------------------------------+ | CouldAssert -> FALSE | ->NI |->NI Cancel |->NI Cancel | | | | AT(S,G,I) | AT(S,G,I) | +-------------------------------+--------------------------------------+ | CouldAssert -> TRUE | ->NI | N/A |->NI Cancel | | | | | AT(S,G,I) | +-------------------------------+--------------------------------------+ | Winner's NLT(N,I) Expires | N/A | N/A |->NI Cancel | | | | | AT(S,G,I) | +-------------------------------+--------------------------------------+ | Receive Prune(S,G), Join(S,G) | ->NI | ->W | ->L Send | | or Graft(S,G) | | | Assert(S,G)| +-------------------------------+--------------------------------------+ Terminology: A "preferred assert" is one with a better metric than the current winner. An "inferior assert" is one with a worse metric than my_assert_metric(S,G,I). The state machine uses the following macro: CouldAssert(S,G,I) = (RPF_interface(S) != I)4.6.4.1. Transitions from NoInfo State
In the NoInfo state, the following events may trigger transitions: An (S,G) data packet arrives on downstream interface I An (S,G) data packet arrived on a downstream interface. It is optimistically assumed that this router will be the Assert winner for this (S,G). The Assert state machine MUST transition to the "I am Assert Winner" state, send an Assert(S,G) to interface I, store its own address and metric as the Assert Winner, and set the Assert_Timer (AT(S,G,I) to Assert_Time, thereby initiating the Assert negotiation for (S,G).
Receive Inferior (Assert OR State Refresh) AND
CouldAssert(S,G,I)==TRUE
An Assert or State Refresh is received for (S,G) that is inferior
to our own assert metric on interface I. The Assert state machine
MUST transition to the "I am Assert Winner" state, send an
Assert(S,G) to interface I, store its own address and metric as
the Assert Winner, and set the Assert Timer (AT(S,G,I)) to
Assert_Time.
Receive Preferred Assert or State Refresh
The received Assert or State Refresh has a better metric than
this router's, and therefore the Assert state machine MUST
transition to the "I am Assert Loser" state and store the Assert
Winner's address and metric. If the metric was received in an
Assert, the router MUST set the Assert Timer (AT(S,G,I)) to
Assert_Time. If the metric was received in a State Refresh, the
router MUST set the Assert Timer (AT(S,G,I)) to three times the
received State Refresh Interval. If CouldAssert(S,G,I) == TRUE,
the router MUST also multicast a Prune(S,G) to the Assert winner
with a Prune Hold Time equal to the Assert Timer and evaluate any
changes in its Upstream(S,G) state machine.
4.6.4.2. Transitions from Winner State
When in "I am Assert Winner" state, the following events trigger
transitions:
An (S,G) data packet arrives on downstream interface I
An (S,G) data packet arrived on a downstream interface. The
Assert state machine remains in the "I am Assert Winner" state.
The router MUST send an Assert(S,G) to interface I and set the
Assert Timer (AT(S,G,I) to Assert_Time.
Receive Inferior Assert or State Refresh
An (S,G) Assert is received containing a metric for S that is
worse than this router's metric for S. Whoever sent the Assert
is in error. The router MUST send an Assert(S,G) to interface I
and reset the Assert Timer (AT(S,G,I)) to Assert_Time.
Receive Preferred Assert or State Refresh
An (S,G) Assert or State Refresh is received that has a better
metric than this router's metric for S on interface I. The
Assert state machine MUST transition to "I am Assert Loser" state
and store the new Assert Winner's address and metric. If the
metric was received in an Assert, the router MUST set the Assert
Timer (AT(S,G,I)) to Assert_Time. If the metric was received in
a State Refresh, the router MUST set the Assert Timer (AT(S,G,I))
to three times the State Refresh Interval. The router MUST also
multicast a Prune(S,G) to the Assert winner, with a Prune Hold
Time equal to the Assert Timer, and evaluate any changes in its
Upstream(S,G) state machine.
Send State Refresh
The router is sending a State Refresh(S,G) message on interface
I. The router MUST set the Assert Timer (AT(S,G,I)) to three
times the State Refresh Interval contained in the State
Refresh(S,G) message.
AT(S,G,I) Expires
The (S,G) Assert Timer (AT(S,G,I)) expires. The Assert state
machine MUST transition to the NoInfo (NI) state.
CouldAssert(S,G,I) -> FALSE
This router's RPF interface changed, making CouldAssert(S,G,I)
false. This router can no longer perform the actions of the
Assert winner, so the Assert state machine MUST transition to
NoInfo (NI) state, send an AssertCancel(S,G) to interface I,
cancel the Assert Timer (AT(S,G,I)), and remove itself as the
Assert Winner.
4.6.4.3. Transitions from Loser State
When in "I am Assert Loser" state, the following transitions can
occur:
Receive Inferior Assert or State Refresh from Current Winner
An Assert or State Refresh is received from the current Assert
winner that is worse than this router's metric for S (typically,
the winner's metric became worse). The Assert state machine MUST
transition to NoInfo (NI) state and cancel AT(S,G,I). The router
MUST delete the previous Assert Winner's address and metric and
evaluate any possible transitions to its Upstream(S,G) state
machine. Usually this router will eventually re-assert and win
when data packets from S have started flowing again.
Receive Preferred Assert or State Refresh
An Assert or State Refresh is received that has a metric better
than or equal to that of the current Assert winner. The Assert
state machine remains in Loser (L) state. If the metric was
received in an Assert, the router MUST set the Assert Timer
(AT(S,G,I)) to Assert_Time. If the metric was received in a
State Refresh, the router MUST set the Assert Timer (AT(S,G,I))
to three times the received State Refresh Interval. If the
metric is better than the current Assert Winner, the router MUST
store the address and metric of the new Assert Winner, and if
CouldAssert(S,G,I) == TRUE, the router MUST multicast a
Prune(S,G) to the new Assert winner.
AT(S,G,I) Expires
The (S,G) Assert Timer (AT(S,G,I)) expires. The Assert state
machine MUST transition to NoInfo (NI) state. The router MUST
delete the Assert Winner's address and metric. If CouldAssert ==
TRUE, the router MUST evaluate any possible transitions to its
Upstream(S,G) state machine.
CouldAssert -> FALSE
CouldAssert has become FALSE because interface I has become the
RPF interface for S. The Assert state machine MUST transition to
NoInfo (NI) state, cancel AT(S,G,I), and delete information
concerning the Assert Winner on I.
CouldAssert -> TRUE
CouldAssert has become TRUE because interface I used to be the
RPF interface for S, and now it is not. The Assert state machine
MUST transition to NoInfo (NI) state, cancel AT(S,G,I), and
delete information concerning the Assert Winner on I.
Current Assert Winner's NeighborLiveness Timer Expires
The current Assert winner's NeighborLiveness Timer (NLT(N,I)) has
expired. The Assert state machine MUST transition to the NoInfo
(NI) state, delete the Assert Winner's address and metric, and
evaluate any possible transitions to its Upstream(S,G) state
machine.
Receive Prune(S,G), Join(S,G), or Graft(S,G)
A Prune(S,G), Join(S,G), or Graft(S,G) message was received on
interface I with its upstream neighbor address set to the
router's address on I. The router MUST send an Assert(S,G) on
the receiving interface I to initiate an Assert negotiation. The
Assert state machine remains in the Assert Loser(L) state. If a
Graft(S,G) was received, the router MUST respond with a
GraftAck(S,G).
4.6.5. Rationale for Assert Rules
The following is a summary of the rules for generating and processing Assert messages. It is not intended to be definitive (the state machines and pseudocode provide the definitive behavior). Instead, it provides some rationale for the behavior. 1. The Assert winner for (S,G) must act as the local forwarder for (S,G) on behalf of all downstream members. 2. PIM messages are directed to the RPF' neighbor and not to the regular RPF neighbor. 3. An Assert loser that receives a Prune(S,G), Join(S,G), or Graft(S,G) directed to it initiates a new Assert negotiation so that the downstream router can correct its RPF'(S). 4. An Assert winner for (S,G) sends a cancelling assert when it is about to stop forwarding on an (S,G) entry. Example: If a router is being taken down, then a canceling assert is sent.
(page 38 continued on part 3)
