| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
COMMAND:
table-map <route-map-name>
DESCRIPTION:
This feature is used to apply a route-map on route updates from BGP to Zebra.
All the applicable match operations are allowed, such as match on prefix,
next-hop, communities, etc. Set operations for this attach-point are limited
to metric and next-hop only. Any operation of this feature does not affect
BGPs internal RIB.
Supported for ipv4 and ipv6 address families. It works on multi-paths as well,
however, metric setting is based on the best-path only.
IMPLEMENTATION NOTES:
The route-map application at this point is not supposed to modify any of BGP
route's attributes (anything in bgp_info for that matter). To achieve that,
creating a copy of the bgp_attr was inevitable. Implementation tries to keep
the memory footprint low, code comments do point out the rationale behind a
few choices made.
bgp_zebra_announce() was already a big routine, adding this feature would
extend it further. Patch has created a few smaller routines/macros whereever
possible to keep the size of the routine in check without compromising on the
readability of the code/flow inside this routine.
For updating a partially filtered route (with its nexthops), BGP to Zebra
replacement semantic of the next-hops serves the purpose well. However, with
this patch there could be some redundant withdraws each time BGP announces a
route thats (all the nexthops) gets denied by the route-map application.
Handling of this case could be optimized by keeping state with the prefix and
the nexthops in BGP. The patch doesn't optimizing that case, as even with the
redundant withdraws the total number of updates to zebra are still be capped
by the total number of routes in the table.
Signed-off-by: Vipin Kumar <vipin@cumulusnetworks.com>
Reviewed-by: Pradosh Mohapatra <pmohapat@cumulusnetworks.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A fat tree topology running IBGP gets into two issues with anycast address
routing. Consider the following topology:
R9 R10
x x
R3 R4 R7 R8
x x
R1 R2 R5 R6
| | | |
10/8 10/8 10/8 S
Let's remind ourselves of BGP decision process steps:
1. Highest Local Preference
2. Shortest AS Path Length
3. Lowest Origin Type
4. Lowest MED (Multi-Exit Discriminator)
5. Prefer External to Internal
6. Closest Egress (Lowest IGP Distance)
7. Tie Breaking (Lowest-Router-ID)
8. Tie Breaking (Lowest-cluster-list length)
9. Tie Breaking (Lowest-neighbor-address)
Without any policies, steps 1-6 will almost always evaluate identically for
all paths received on any router in the above topology. Let's assume that
the router-ids follow the following inequality: R1 < R2 < R5 < R6. Owing to
the 7th step above, all routers will now choose R1's path as the best. This
is undesirable. As an example, traffic from S to 10/8 will follow the path
S -> R6 -> R7 -> R9 -> R4 -> R2 -> 10/8 instead of S -> R6 -> R7 -> R5 -> 10/8.
Furthermore, once R7 (& R8) chooses R1's path as the best, it would withdraw
its path learned through (R5, R6) from (R9, R10). This leads to inefficient
load balancing - e.g. R9 can't do ECMP across all available egresses -
(R1, R2, R5).
The patch addresses these issues by noting that that cluster list is always
carried along with the routes and its length is a good indicator of IBGP
hops. It thus makes sense to compare that as an extension to metric after
step 6. That automatically ensures correct multipath computation.
Unfortunately a partial deployment of this in a generic topology (note:
fat-tree/clos topologies work fine) may lead to potential loops. It needs
to be looked into.
Signed-off-by: Pradosh Mohapatra <pmohapat@cumulusnetworks.com>
Reviewed-by: Dinesh G Dutt <ddutt@cumulusnetworks.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
advertised is based on the bestpath attribute set, but the
following attributes are aggregated from the attribute sets
of the multipath constituents:
- AS_PATH
- ORIGIN
- COMMUNITIES
- EXTENDED COMMUNITIES
In addition the route is advertised with the NEXT_HOP set
to the router's interface IP address, instead of the NEXT_HOP
of the best path. This is to ensure that traffic will go to this
router so it can be fanned out via the multipath route.
* bgpd/ecommunity.c
* ecommunity_uniq_sort(): Make this function externally accessible
* bgpd/ecommunity.h
* Add external declaration for ecommunity_uniq_sort()
* bgpd/bgp_mpath.c
* bgp_info_nexthop_cmp(): Replace calls to bgp_attr_extra_get()
to avoid unwanted memory allocation
* bgp_info_mpath_free(): Free aggregate attribute for multipath
* bgp_info_mpath_attr(): Lookup aggregate attribute of a multipath route
* bgp_info_mpath_attr_set(): Set aggregate attribute of a multipath route
* bgp_info_mpath_aggregate_update(): Update the aggregate attribute
of a multipath route
* bgpd/bgp_mpath.h
* bgp_info_mpath: Add pointer to hold aggregate attribute of a multipath
* Add external declarations for new functions
* bgpd/bgp_route.c
* bgp_announce_check(): Use aggregate attribute when announcing multipath
route
* bgp_announce_check_rsclient(): Use aggregate attribute when announcing
multipath route
* bgp_best_selection(): After updating multipath set, update the
multipath aggregate attribute
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
first stage of the best path calculation. The second stage then
selects a winner from each peer AS's best path. In the second stage we
clear multipath set of the non-selected best paths via
bgp_mp_dmed_deselect(). Since the multipath set is already marked up
for the winning path, we don't call bgp_info_mpath_update() after the
second stage calculation.
* bgpd/bgp_mpath.c
* bgp_mp_dmed_deselect(): New function to cleanup the multipath
markup if a DMED selected path loses in stage 2 of the best path
calculation
* bgpd/bgp_mpath.h
* Add external declaration of bgp_mp_dmed_deselect()
* bgpd/bgp_route.c
* bgp_best_selection(): If multipath is enabled, build up the mp_list
for the current peer AS, and do the RIB markup the best path from
that AS. In the second stage, clear the RIB markup for the DMED
selected path if it is not selected as best. Only call
bgp_info_mpath_update() in the second stage when not doing
deterministic MED.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
information based on the multipath list (mp_list) generated during
the best path calculation. Display "multipath" for paths that are
multipath and also on bestpath if the route is multipath. Flag a
best path with the BGP_INFO_MULTIPATH_CHG if the multipath
set has changed since the last update. This can be used to trigger
updates to zebra and peers.
The multipath markup is a lazily allocated bgp_info_mpath structure
that is added to the best path and any multipaths. The mpath structures
are linked together with the best path element at the head and the
other elements ordered by nexthop and then by peer address. This
markup scheme is updated by calling bgp_info_mpath_update() and passing
in a new mp_list the the current multipath set. There are additional
API's for walking the multipath set, querying the count of multipaths,
and for cleaning up the multipath markup information when freeing path
information.
* bgpd/bgp_mpath.c
* bgp_info_mpath_new(): Allocation of new mpath element
* bgp_info_mpath_free(): Release memory for mpath element
* bgp_info_mpath_get(): Access mpath element of path. Allocate memory
on-demand
* bgp_info_mpath_enqueue(): Enqueue a path onto the multipath list
* bgp_info_mpath_dequeue(): Remove a path from the multipath list
* bgp_info_mpath_first(): Return first path on the multipath list
* bgp_info_mpath_next(): Return next path on the multipath list
* bgp_info_mpath_count(): Return the number of paths on the multipath list
* bgp_info_mpath_count_set(): Set the number of paths on the multipath list
* bgp_info_mpath_update(): Update multipath markup on bgp route table entry
and flag any changes. Emit 'debug bgp event' output on any multipath
change.
* bgpd/bgp_mpath.h
* struct bgp_info_mpath: Information added to a bgp_info path to record
multipath information
* External declarations for new functions in bgp_mpath.c
* bgpd/bgp_route.c
* bgp_info_free(): Free mpath memory when freeing path information
* bgp_info_reap(): Dequeue path from multipath queue before deleting it
* bgp_best_selection(): Calls bgp_info_mpath_update() with latest
mp_list to mark-up rib table entry
* bgp_vty_out_detail(): Add display of multipath flag for a path. Also
display 'multipath' for bestpath if it is a multipath route
* bgpd/bgp_route.h
* struct bgp_info: Add pointer to bgp_info_mpath information
* Add flags to mark a path as multipath (BGP_INFO_MULTIPATH) and
to mark bestpath if multipath information has changed
(BGP_INFO_MULTIPATH_CHG)
* lib/memtypes.c
* Add MTYPE_BGP_MPATH_INFO for allocating memory for bgp_info_mpath
* tests/bgp_mpath_test.c
* Add test case for bgp_info_mpath_update() and supporting functions
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
equal to the best path are accumulated onto an ordered list (mp_list)
if maximum-paths is configured. A future commit will add the
multipath markup to the BGP rib table based on the mp_list. Add
unit test for the added mp_list functions.
Deterministic MED is not supported in this commit, it will be
added later.
* bgpd/bgp_aspath.c
* Make aspath_cmp() an external symbol so it can be used in
equivalent paths check
* bgpd/bgp_aspath.h
* Add extern declaration of aspath_cmp()
* bgpd/bgp_mpath.c
* bgp_info_nexthop_cmp(): Compares nexthops of two paths
* bgp_info_mpath_cmp(): Compare function to order multipaths by
nexthop and then by peer address
* bgp_mp_list_init(): Initialize a list with the multipath order function
* bgp_mp_list_clear(): Clear out the mp_list
* bgp_mp_list_add(): Add a multipath to mp_list
* bgpd/bgp_mpath.h
* External declarations for above added functions in bgp_mpath.c
* bgpd/bgp_route.c
* bgp_info_cmp(): Add equivalent paths result (paths_eq). If eBGP
paths are equal down to IGP metric check, flag as equal if peer AS
matches. Similarly for iBGP paths but compare full AS_PATH.
* bgp_best_selection(): If multipath is enabled, accumulate equivalent paths
in mp_list. Add debug bgp event output to see result (will be filtered
later to display only when change occurs)
* bgp_process_rsclient(): Pass multipath config to bgp_best_selection()
* bgp_process_main(): Pass multipath config to bgp_best_selection()
* tests/bgp_mpath_test.c
* Add unit test case for bgp_mp_list functions
|
|
|
There is support to configure this for each (AFI,SAFI), but
currently this configuration is only present for IPv4 unicast:
maximum-paths [ibgp] <1-255>
no maximum-paths [ibgp] [<1-255>]
* bgpd/Makefile.am
* Add bgp_mpath.h and bgp_mpath.c to build
* bgpd/bgp_mpath.h
* New file for bgp multipath declarations
* define BGP_DEFAULT_MAXPATHS
* bgpd/bgp_mpath.c
* bgp_maximum_paths_set(): Configure maximum paths for the given
afi, safi and bgp instance
* bgp_maximum_paths_unset(): Return maximum paths configuration to
the default setting for the given afi, safi and bgp instance
* bgpd/bgp_vty.c
* Define command strings for above CLI
* bgp_config_write_maxpaths(): Outputs configuration for the given
afi, safi and bgp instance
* Install command elements for IPv4 unicast
* bgpd/bgp_zebra.h
* bgp_config_write_maxpaths(): External declaration
* bgpd/bgpd.c
* bgp_create(): Initialize bgp instance to default maximum paths setting
* bgp_config_write_family(): Output maximum paths configuration
for the given address family
* bgp_config_write(): Output maximum paths configuration for
IPv4 unicast address family
* bgpd/bgpd.h
* struct bgp: Add storage for maximum paths configuration for
each afi, safi
|
