A BGP update-group is dynamically created to group together a set of peers
such that any BGP updates can be formed just once for the entire group and
only the next hop attribute may need to be modified when the update is sent
out to each peer in the group. The update formation code attempts to
determine as much as possible if the next hop will be set to our own IP
address for every peer in the group. This helps to avoid additional checks
at the point of sending the update (which happens on a per-peer basis) and
also because some other attributes may/could vary depending on whether the
next hop is set to our own IP or not. Resetting the next hop to our own IP
address is the most common behavior for EBGP peerings in the absence of
other user-configured or internal (e.g., for l2vpn/evpn) settings and
peerings on a shared subnet.
The code had a flaw in the multiaccess check to see if there are peers in
the update group which are on a shared subnet as the next hop of the path
being announced - the source peer could itself be in the same update group
and cause the check to give an incorrect result. Modify the check to skip
the source peer so that the check is more accurate.
Signed-off-by: Vivek Venkatraman <vivek@cumulusnetworks.com>
Reviewed-by: Donald Sharp <sharpd@cumulusnetworks.com>
Reviewed-by: Don Slice <dslice@cumulusnetworks.com>
The bgp reason code was being reset in bgp_best_selection
by rerunning bgp_path_info_cmp multiple times under certain
receiving patterns of data from peers.
This is the debugs that show this issue:
2020/03/16 19:17:22.523780 BGP: 2001:20:1:1::6 rcvd UPDATE w/ attr: nexthop 20.1.1.6, origin i, metric 600, community 1000:1006, path 20
2020/03/16 19:17:22.523819 BGP: 2001:20:1:1::6 rcvd 20.10.0.6/32 IPv4 unicast
2020/03/16 19:17:22.556168 BGP: 20.1.1.6 rcvd UPDATE w/ attr: nexthop 20.1.1.6, origin i, metric 500, community 1000:1006, path 20
2020/03/16 19:17:22.556209 BGP: 20.1.1.6 rcvd 20.10.0.6/32 IPv4 unicast
2020/03/16 19:17:22.572358 BGP: bgp_process_main_one: p=20.10.0.6/32 afi=IPv4, safi=unicast start
2020/03/16 19:17:22.572408 BGP: 20.10.0.6/32: Comparing path 2001:20:1:1::6 flags 0x410 with path 20.1.1.6 flags 0x410
2020/03/16 19:17:22.572415 BGP: 20.10.0.6/32: path 2001:20:1:1::6 loses to path 20.1.1.6 due to MED 600 > 500
2020/03/16 19:17:22.572422 BGP: 20.10.0.6/32: path 20.1.1.6 is the bestpath from AS 20
2020/03/16 19:17:22.572429 BGP: 20.10.0.6/32: path 20.1.1.6 is the initial bestpath
2020/03/16 19:17:22.572435 BGP: bgp_best_selection: pi 0x5627187c66c0 dmed
2020/03/16 19:17:22.572441 BGP: 20.10.0.6/32: After path selection, newbest is path 20.1.1.6 oldbest was NONE
2020/03/16 19:17:22.572447 BGP: 20.10.0.6/32: path 20.1.1.6 is the bestpath, add to the multipath list
2020/03/16 19:17:22.572453 BGP: 20.10.0.6/32: path 2001:20:1:1::6 has the same nexthop as the bestpath, skip it
2020/03/16 19:17:22.572460 BGP: 20.10.0.6/32: starting mpath update, newbest 20.1.1.6 num candidates 1 old-mpath-count 0 old-cum-bw u0
2020/03/16 19:17:22.572466 BGP: 20.10.0.6/32: comparing candidate 20.1.1.6 with existing mpath NONE
2020/03/16 19:17:22.572473 BGP: 20.10.0.6/32: New mpath count (incl newbest) 1 mpath-change NO all_paths_lb 0 cum_bw u0
Effectively if BGP receives 2 paths it could end up running bgp_path_info_cmp multiple times
and in some situations overwrite the reason selected the first time through.
In this example path selection is run and the MED is the reason for the choice.
Then in bgp_best_selection is run again this time clearing new_select
to NULL before calling path selection for the first time. This second
call into path selection resets the reason, since it is only passing in one
path. So save the last reason selected and restore in this case.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Override ORIGIN attribute if defined.
E.g.: Cisco and Juniper set ORIGIN for aggregated address
to IGP which is not what rfc4271 says.
This enables the same behavior, optionally.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
The act of peer_sort() being called always set this value
even when we are just looking it up. We need to seperate
out the idea of lookup from set.
For those places that this is immediately obvious that
this is a lookup switch over to using this function.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
The coverity SA believes that the regex value can possibly
be NULL. Not possible so let's make it happy.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
* While the Deferral timer is running, signal route update pending
(ZEBRA_CLIENT_ROUTE_UPDATE_PENDING) from BGPD to Zebra.
* After expiry of the Deferral timer, the deferred routes are processed.
When the deferred route_list becomes empty, End-of-Rib is send to the
peer and route processing complete message (ZEBRA_CLIENT_ROUTE_UPDATE_COMPLETE)
is sent to Zebra. So that Zebra would delete any stale routes still
present in the rib.
Signed-off-by: Biswajit Sadhu <sadhub@vmware.com>
*After a restarting router comes up and the bgp session is
successfully established with the peer. If the restarting
router doesn’t have any route to send, it send EOR to
the peer immediately before receiving updates from its peers.
*Instead the restarting router should send EOR, if the
selection deferral timer is not running OR count of eor received
and eor required are matches then send EOR.
Signed-off-by: Biswajit Sadhu <sadhub@vmware.com>
& GR is enabled.
When GR with deferral is enabled and connected routes are
distributed then in one race condition route node gets added
in to both deferred queue and work queue. If deferred queue
gets processed first then it ends up delete only flag while
leaving the entry in the work queue as it is. When a new update
comes for the same route node next time from peer then it hits
assert. Assert check is added to ensure we don’t add to work queue
again while it is already present.
So, check before adding in to deferred queue if it is already present
in work queue and bail if so.
Signed-off-by: Biswajit Sadhu <sadhub@vmware.com>
* Selection Deferral Timer for Graceful Restart.
* Added selection deferral timer handling function.
* Route marking as selection defer when update message is received.
* Staggered processing of routes which are pending best selection.
* Fix for multi-path test case.
Signed-off-by: Biswajit Sadhu <sadhub@vmware.com>
bgpd already supports BGP Prefix-SID path attribute and
there are some sub-types of Prefix-SID path attribute.
This commits makes bgpd to support additional sub-types.
sub-Type-4 and sub-Type-5 for construct the VPNv4 SRv6 backend
with vpnv4-unicast address family.
This path attributes is already supported by Ciscos IOS-XR and NX-OS.
Prefix-SID sub-Type-4 and sub-Type-5 is defined on following
IETF-drafts.
Supports(A-part-of):
- https://tools.ietf.org/html/draft-dawra-idr-srv6-vpn-04
- https://tools.ietf.org/html/draft-dawra-idr-srv6-vpn-05
Signed-off-by: Hiroki Shirokura <slank.dev@gmail.com>
uint8_t * cannot be cast to uint32_t * unless the
pointed-to address is aligned according to uint32_t's
alignment rules. And it usually is not.
Signed-off-by: Santosh P K <sapk@vmware.com>
uint8_t * cannot be cast to uint32_t * unless the pointed-to address is
aligned according to uint32_t's alignment rules. And it usually is not.
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
Before it was:
```
exit1-debian-9# show ip bgp regexp ^200a
Invalid character in as-path access-list ^200a
```
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
This should keep backward compatibility when bgp show-hostname is
enabled/disabled.
Also show the real originator IP instead of showing fqdn of the route
reflector.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
