Donald Sharp [Fri, 8 Jan 2021 21:52:41 +0000 (16:52 -0500)]
tests: Wait for bgp convergence *after* we re-add the interfaces
In test_bgp_mutli_vrf_topo2.py it's clear that we remove then
re-add the vrf interfaces. Then the test was immediately
checking to ensure that the routes were available.
BGP needs time to reconverge. Let's ensure that first.
Donald Sharp [Fri, 8 Jan 2021 12:44:42 +0000 (07:44 -0500)]
tools: Cleanup some zebra section of the support bundle
a) Add some useful commands
b) Remove `show error all` this just dumps the error codes. If
we know the version we don't need this. Additionally this is
rather large.
Donald Sharp [Fri, 8 Jan 2021 12:42:08 +0000 (07:42 -0500)]
tools: Cleanup bgp commands in the support bundle
Add some missing commands ( I am sure that there are more useful ones to )
Cleanup to use the modern non-deprecated syntax in case anyone runs across
this.
Donald Sharp [Fri, 8 Jan 2021 00:24:13 +0000 (19:24 -0500)]
bgpd: Move afi indepedent code outside of checks for it
In bgp_zebra_announce we do work to apply the table map.
This is the same for both v4 and v6 but we have the code
duplicated in both v4 and v6 if statements. Move outside
to reduce the duplications.
Wesley Coakley [Tue, 5 Jan 2021 09:22:57 +0000 (04:22 -0500)]
bgpd: separate lcommunity validation from tokenizer
`lcommunity_gettoken` expects a space-delimeted list of 0 or more large
communities. `lcommunity_list_valid` can perform this check.
`lcommunity_list_valid` now validates large community lists more
accurately based on the following condition: Each quantity in a standard bgp
large community must:
1. Contain at least one digit
2. Fit within 4 octets
3. Contain only digits unless the lcommunity is "expanded"
4. Contain a valid regex if the lcommunity is "expanded"
Moreover we validate that each large community list contains exactly 3
such values separated by a single colon each.
One quirk of our validation which is worth documenting is:
The first line will throw an error complaining about a "malformed community-list
value". The second line will be accepted because the each value is each treated as
a regex when matching large communities, it simply will never match anything so
it's rather useless.
Pat Ruddy [Tue, 5 Jan 2021 17:18:39 +0000 (17:18 +0000)]
tests: add test for labelmanager plugin
An external label manager plugin may want to use the following
functions:
- create_label_chunk
- assign_label_chunk
- delete_label_chunk
- release_label_chunk
This test ensures that they are externally visible.
Olivier Dugeon [Tue, 26 May 2020 12:51:02 +0000 (14:51 +0200)]
lib: Add Link State Database
Define new models for Link State Database a.k.a TED
and functions to manipulate the new database as well as exchange Link State
information through ZAPI Opaque message.
Pat Ruddy [Wed, 16 Dec 2020 17:02:53 +0000 (17:02 +0000)]
tests: add BGPLU label sanity test
This test checks the interactions between the BGP label requesting
code and the labelpool code to ensure the correct number of labels
and label chunks are requested and those labels are freed back into
the pool when the corresponding prefix is removed.
Pat Ruddy [Wed, 16 Dec 2020 15:06:30 +0000 (15:06 +0000)]
bgpd: fix zclient return code damage in labelpool
The check for the return code for zclient_send_get_label_chunk is
reversed and therefore the pending count does not get incremented
for each successful label chunk request.
This has the effect of requesting a 50 label chunk per label request
from BGP i.e we request 50 times the labels we require.
Pat Ruddy [Thu, 17 Dec 2020 10:49:30 +0000 (10:49 +0000)]
bgpd: release label to labelpool on dereg
when we de-register for a label associated with a prefix then release
it back to the labelpool to clean up the stale reference held in the
ledger and inuse lists.
Pat Ruddy [Thu, 17 Dec 2020 10:41:07 +0000 (10:41 +0000)]
bgpd: refactor label allocation code
To prepare for fixing an issue where labels do not get released back
to the labelpool when the route is deleted some refactoring is
necessary. There are 2 parts to this.
1. restructure the code to remove the circular nature of label
allocations via the labelpool and decouple the label type decision
from the notification fo the FEC.
The code to notify the FEC association to zebra has been split out
into a separate function so that it can be called from the synchronous
path (for registration of index-based labels and de-registration of all
labels), and from the asynchronous path where we need to wait for a
callback from the labelpool code with a label allocation.
The decision about whether we are using an index-based label or an
allocated label is reflected in the state of the BGP_NODE_LABEL_REQUESTED
flag so the checks on the path_info in the labelpool callback code are
no longer required.
2. change the owned of a labelpool allocated label from the path info
structure to the bgp_dest structure. This allows labels to be released
(in a subsequent commit) when the owner (bgp_dest) goes away.
Pat Ruddy [Wed, 9 Dec 2020 10:25:38 +0000 (10:25 +0000)]
zebra: labelmanager could return reserved labels
when checking if there is a "hole" behind the current reservation
marker the calculation of whether the hole is big enough to satisfy
the requested chunk is out by 1. This could result in returning a label
which has already been allocated.
Pat Ruddy [Wed, 9 Dec 2020 10:23:44 +0000 (10:23 +0000)]
zebra: label manager should never return a reserved block
if the requested chunk size was less than 16 then a chunk
within the reserved block would be returned. Make sure that
we never return labels that are below MPLS_LABEL_UNRESERVED_MIN
Pat Ruddy [Fri, 4 Dec 2020 12:16:32 +0000 (12:16 +0000)]
bgpd: add missing unlock of path_info
when the path info information is queued on the work queue it
is protected by a lock to avoid the rug being pulled whilst it
resides on the queue add an unlock in the error case where we do
no queue the reference to the workqueue.
G. Paul Ziemba [Sat, 19 Dec 2020 19:41:05 +0000 (11:41 -0800)]
tests: topotests/lib/lutil.py: add timestamp to log of test start
Timestamps in test logs are needed for correlation with messages in
routing protocol log files. Vox populi indicates preference for
timestamp at beginning of line.
Test case 5.10 sends leave message to unicast address, the leave
packet is accepted and a query message is sent in response to this.
No validation for address is present in the function
Add check for addresses as per RFC. Leave messages are allowed only
sent to either ALL-ROUTERS (224.0.0.2) or group address.
Duncan Eastoe [Tue, 22 Dec 2020 21:23:43 +0000 (21:23 +0000)]
zebra: zebra2proto() handle kernel/connect type
When dplane_fpm_nl is used the "Please add this protocol(n) to proper
rt_netlink.c handling" debug message is emitted for any route of type
kernel or connected.
This severely reduces performance of dplane_fpm_nl when large numbers
of these routes are present in the RIB.
The messages are not observed when using the original fpm module since
this uses a custom function, netlink_proto_from_route_type().
zebra2proto() now returns RTPROT_KERNEL for ZEBRA_ROUTE_CONNECT and
ZEBRA_ROUTE_KERNEL. This should only impact dplane_fpm_nl's use of
the common netlink routines since these routes generally ignored via
checking of RSYSTEM_ROUTE().
Duncan Eastoe [Tue, 22 Dec 2020 19:47:17 +0000 (19:47 +0000)]
zebra: fpm_nl_process() reschedule dp thread
fpm_nl_process() now ensures that the dataplane thread is rescheduled
if it hits the work limit while processing its incoming work queue.
This would probably already occur due to some other event, such as
fpm_process_queue() enqueuing completed work to the output queue,
however it does no harm to add this explicit reschedule.
Duncan Eastoe [Tue, 22 Dec 2020 18:35:40 +0000 (18:35 +0000)]
zebra: resched dp thread if output queue limit hit
If the dataplane thread hits the work limit while processing the
output queue for any given provider, we now explicitly reschedule
the thread.
Otherwise, if the number of items in the output queue is greater than
the work limit, draining of that output queue is dependent on new
dataplane work.
Routes which are not drained from the output queue are stuck with
the 'q' flag, so this is a similar issue to that observed in 164d8e86081fdf33992b6c45af446bac6103e20c.
zebra: skip EVI setup if an ES is applied to a pseudo interface
zebra maintains pseudo interface for hanging off user config after
the interface is deleted in the kernel. If an user tried to config
an ES against such an interface zebra would crash with the following
call stack -
at zebra/zebra_evpn_mh.c:2095
sysmac=sysmac@entry=0x55cfbadd3160) at zebra/zebra_evpn_mh.c:2258
at zebra/zebra_evpn_mh.c:3222
argv=<optimized out>, es_lid_str=<optimized out>, es_lid=1, no=0x0, vty=0x55cfbaf4c7b0)
at zebra/zebra_evpn_mh.c:3222
argv=<optimized out>) at ./zebra/zebra_evpn_mh_clippy.c:202
vty=vty@entry=0x55cfbaf4c7b0, cmd=cmd@entry=0x0, filter=FILTER_RELAXED)
at lib/command.c:1073
zebra: accept bgp remote mac-ip update if the higher-seq-local mac is not bgp-ready
If a local-MAC or local-neigh is not active locally it is not sent to BGP.
At this point if BGP rxes a remote route it accepts it and installs in
zebra. Zebra was rejecting BGP's update if it had a higher seq local (inactive)
entry. This would result in bgp and zebra falling out of sync.
In some cases zebra would delete the local-inactive entries in sometime (as
a part of the dplane/kernel garbage collection). This would leave zebra
with missing remote entries (which were still present in bgpd).
This change allows lower-seq BGP updates to overwrite zebra's local entry if
that entry happens to be local-inactive.
Note: This logic was already in use for sync-mac-ip updates. Extended the
same logic to remote-mac-ip updates.
zebra: clean zevpn references in the access bd database when the VNI is deleted
When an VNI was deleted as a part of FRR/zebra shutdown the zevpn entry
was being freed without removing its reference in the access vlan
entry (i.e. without clearing the VLAN->VNI mapping) used by MH.
If a netlink/dp notification is rxed for a neigh without the peer-sync
flag FRR re-installs the entry with the right flags. This change is
needed to handle cases where the dataplane and FRR may fall out of
sync because of neigh learning on the network ports (i.e. via
the VxLAN).
Ticket: CM-30693
The problem was found during VM mobility "torture" tests where 100s
of extended VM moves were done.
zebra: fix a problem with local MAC pointing to a remote ES
If a remote MAC update is rxed from BGP with a lower sequence number than
the local one zebra ignores the MAC update. This typically happens if
there is a race condition (where updates are in flight from zebra to BGP).
There was a bug in zebra because of which the dest ES was being updated
before this check. This left the local MAC pointing to a remote ES.
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
Relevant Dumps:
===============
root@leaf21:mgmt:~# net show evpn mac vni 101101 mac 00:93:00:00:00:01
MAC: 00:93:00:00:00:01
ESI: 03:00:00:00:77:01:03:00:00:0d
Intf: - VLAN: 101
Sync-info: neigh#: 1 peer-proxy
Local Seq: 3 Remote Seq: 0
Neighbors:
21.1.13.1 Active
root@leaf21:mgmt:~# net sho evpn es
Type: L local, R remote, N non-DF
ESI Type ES-IF VTEPs
03:00:00:00:77:01:02:00:00:0c R - 6.0.0.10,6.0.0.11
03:00:00:00:77:01:03:00:00:0d R - 6.0.0.10,6.0.0.11,6.0.0.12
03:00:00:00:77:01:04:00:00:0e R - 6.0.0.10,6.0.0.11,6.0.0.12,6.0.0.13
03:00:00:00:77:02:02:00:00:16 LR bondP2-H2 6.0.0.15
03:00:00:00:77:02:03:00:00:17 LR bondP2-H3 6.0.0.15,6.0.0.16
03:00:00:00:77:02:04:00:00:18 LR bondP2-H4 6.0.0.15,6.0.0.16,6.0.0.17
root@leaf21:mgmt:~#
Relevant logs:
===============
2020/07/29 15:41:27.110846 ZEBRA: Recv MACIP ADD VNI 101101 MAC 00:93:00:00:00:01 IP 21.1.13.1 flags 0x0 seq 2 VTEP 0.0.0.0 ESI 03:00:00:00:77:01:03:00:00:0d from bgp
2020/07/29 15:41:27.110867 ZEBRA: Ignore remote MACIP ADD VNI 101101 MAC 00:93:00:00:00:01 IP 21.1.13.1 as existing MAC has higher seq 3 flags 0x401
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
zebra: advertise stale neighs if EVPN-MH is not enabled
With EVPN-MH, Type-2 routes are also used for MAC-IP syncing between
ES peers so a change was done to only treat REACHABLE local neigh
entries as local-active and advertise them as Type-2 routes i.e. STALE
neigh entries are no longer advertised as Type-2s.
This however exposed some unexpected problems with MLAG where a
secondary reboot followed by a primary reboot left a lot of neighs
in STALE state (on the primary) resulting in them not being
advertised. And remote routed traffic to those hosts being
blackholed in a sym-IRB setup.
This commit is a workaround to fix the regression (it doesn't fix
the underlying problems with entries not becoming REACHABLE; which
maybe a day-1 problem). The workaround is to continue advertising
STALE neighbors if EVPN-MH is not enabled.
projects / mirror_frr.git / log