Problem Statement:
=================
In scale setup BGP sessions start flapping.
RCA:
====
In virtualized environment there are multiple places where
MTU need to be set. If there are some places were MTU is not set
properly then there is chances that BGP packets get fragmented,
in scale setup this will lead to BGP session flap.
Fix:
====
A new tcp option is provided as part of this implementation,
which can be configured per neighbor and helps to set the TCP
max segment size. User need to derive the path MTU between the BGP
neighbors and set that value as part of tcp-mss setting.
1. CLI Configuration:
[no] neighbor <A.B.C.D|X:X::X:X|WORD> tcp-mss (1-65535)
2. Running config
frr# show running-config
router bgp 100
neighbor 198.51.100.2 tcp-mss 150 => new entry
neighbor 2001:DB8::2 tcp-mss 400 => new entry
3. Show command
frr# show bgp neighbors 198.51.100.2
BGP neighbor is 198.51.100.2, remote AS 100, local AS 100, internal link
Hostname: frr
Configured tcp-mss is 150, synced tcp-mss is 138 => new display
4. Show command json output
frr# show bgp neighbors 2001:DB8::2 json
{
"2001:DB8::2":{
"remoteAs":100,
"bgpTimerKeepAliveIntervalMsecs":60000,
"bgpTcpMssConfigured":400, => new entry
"bgpTcpMssSynced":388, => new entry
Risk:
=====
Low - This is a config driven feature and it sets the max segment
size for the TCP session between BGP peers.
Tests Executed:
===============
Have done manual testing with three router topology.
1. Executed basic config and un config scenarios
2. Verified if the config is updated in running config
during config and no config operation
3. Verified the show command output in both CLI format and
JSON format.
4. Verified if TCP SYN messages carry the max segment size
in their initial packets.
5. Verified the behaviour during clear bgp session.
6. done packet capture to see if the new segment size
takes effect.
Signed-off-by: Abhinay Ramesh <rabhinay@vmware.com>
Current code has an inconsistent behavior with redistribute routes.
Suppose you have a kernel route that is being read w/ a distance
of 255:
eva# show ip route kernel
Codes: K - kernel route, C - connected, S - static, R - RIP,
O - OSPF, I - IS-IS, B - BGP, E - EIGRP, N - NHRP,
T - Table, v - VNC, V - VNC-Direct, A - Babel, D - SHARP,
F - PBR, f - OpenFabric,
> - selected route, * - FIB route, q - queued, r - rejected, b - backup
t - trapped, o - offload failure
K>* 0.0.0.0/0 [0/100] via 192.168.161.1, enp39s0, 00:06:39
K>* 4.4.4.4/32 [255/8192] via 192.168.161.1, enp39s0, 00:01:26
eva#
If you have redistribution already turned on for kernel routes
you will be notified of the 4.4.4.4/32 route. If you turn
on kernel route redistribution watching after the 4.4.4.4/32 route
has been read by zebra you will never learn of it.
There is no need to look for infinite distance in the redistribution
code. Either we are selected or not. In other words non kernel routes
with an 255 distance are never installed so the checks were pointless.
So let's just remove the distance checking and tell interested parties
about the 255 kernel route if it exists.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Not having scapy in the docker image leads to very obtuse failures in
the pim bsm tests (obtuse, as in, it just fails without any hint as to
why...)
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
pimd was the only user of this function, and that has gone away now.
So just kill the function.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Just some cleanup before I touch this code; switching to typesafe list
macros & putting the data directly inline.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
... yes we may need it later, but if and when that happens we can put it
back there. No point carrying around unused things.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
The .pb-c.c files pick up our assert() override, but that needs config.h
to be included too, and that needs to go at the very top of the file...
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Display the MSDP peer configuration in `show running-config` so it can
be saved on configuration write.
Signed-off-by: Rafael Zalamena <rzalamena@opensourcerouting.org>
The current implementation of TI-LFA computes link-protecting
repair paths (even when node protection is enabled) to have repair
paths to all destinations when no node-protecting repair has been
found. This may be desired or not. E.g. the link-protecting paths
may use the protected node and be, therefore, useless if the node
fails. Also, computing link-protecting repairs incurs extra
calculations.
With this patch, when node protection is enabled, link protecting
repair paths are only computed if "link-fallback" is specified in
the configuration, on a per interface and IS-IS level.
Signed-off-by: Fredi Raspall <fredi@voltanet.io>
Currently FRR reads the kernel for interface state and FRR
creates a connected route per address on an interface. If
you are in a situation where you have multiple addresses
on an interface just create 1 connected route for them:
sharpd@eva:/tmp/topotests$ vtysh -c "show int dummy302"
Interface dummy302 is up, line protocol is up
Link ups: 0 last: (never)
Link downs: 0 last: (never)
vrf: default
index 3279 metric 0 mtu 1500 speed 0
flags: <UP,BROADCAST,RUNNING,NOARP>
Type: Ethernet
HWaddr: aa:4a:ed:95:9f:18
inet 10.4.1.1/24
inet 10.4.1.2/24 secondary
inet 10.4.1.3/24 secondary
inet 10.4.1.4/24 secondary
inet 10.4.1.5/24 secondary
inet6 fe80::a84a:edff:fe95:9f18/64
Interface Type Other
Interface Slave Type None
protodown: off
sharpd@eva:/tmp/topotests$ vtysh -c "show ip route connected"
Codes: K - kernel route, C - connected, S - static, R - RIP,
O - OSPF, I - IS-IS, B - BGP, E - EIGRP, N - NHRP,
T - Table, v - VNC, V - VNC-Direct, A - Babel, D - SHARP,
F - PBR, f - OpenFabric,
> - selected route, * - FIB route, q - queued, r - rejected, b - backup
t - trapped, o - offload failure
C>* 10.4.1.0/24 is directly connected, dummy302, 00:10:03
C>* 192.168.161.0/24 is directly connected, enp39s0, 00:10:03
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
If LDP is miss configured in a setup and the router has LSPs with no remote
label, this code installs the LSP with a pop instruction of the top-level
label so the packet can be forwarded using IP. This is a best-effort
attempt to deliver labeled IP packets to their final destination instead of
dropping them. If this config is turned off the code will only install
LSPs that have a valid remote label.
Signed-off-by: Lynne Morrison <lynne@voltanet.io>
As per the ospfv3 conformance test 24.3
SETUP: Configure DIface-0 with priority set to <hprty>.
ANVL: Establish full adjacency with DUT for neighbor Rtr-0-A on DIface-0.
DUT: Exchange all the <OSPF-DD> packets, during adjacency establish- ment.
ANVL: Verify that the received <OSPF-DD> packets contain: • one header of Link-LSA, originated by DUT.
ANVL: Send <OSPF-LSR> packet from neighbor Rtr-0-A to DIface-0 con- taining:
• One Request Tuple for Link-LSA originated by DUT.
ANVL: Listen (for upto 2 * <RxmtInterval> seconds) on DIface-0. DUT: Send <OSPF-LSU> packet.
ANVL: Verify that the received <OSPF-LSU> packet contains:
• •
one Link-LSA, originated by DUT, contains: Rtr Pri field set to <hprty>.
----------
When interface priority is changed Link LSAs should be tranmitted
with the priority set.
When the link priorty chanages, the drbdr algorithm is called, which
can change the state of the interface. But if the state does not
changes then LINK LSAs are not transmitted.
This PR fixes this issue. If the state is changed, then LINK LSAs
will anyways be tranmitted. But in case the state is not changed,
even in that case Link LSAs are tranmitted.
Signed-off-by: Yash Ranjan <ranjany@vmware.com>
These hoops to get warnings for mis-printing `uint64_t` are apparently
breaking some C++ bits...
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
The previous method, using zassert.h and hoping nothing includes
assert.h (which, on glibc at least, just does "#undef assert" and puts
its own definition in...) was fragile - and actually broke undetected.
Just provide our own assert.h and control overriding by putting it in a
separate directory to add to the include path (or not.)
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Since _rnode_zlog was wrapping zlog(), these messages weren't getting an
unique ID assigned through the xref mechanism. Replace macro with a
small extension that prints (almost) the same thing.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
When an operator encounters a situation where the number
of FD's open is greater than what we have been configured
to legitimately handle via uname or the `--limit-fds` command
line, abort with a message that they should be able to
debug and figure out what is going on.
Fixes: #8596
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Delay setting local data about a remote peer until after BGP
has decided to allow an open connection to proceed.
Modifying local peer data structures based upon what is
received from a peer should not be done until after BGP
has decided that the open is allowed to proceed.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
FRR in thread.c clears the passed in double pointer when
we pull it off the ready queue and pass it back to
the calling function via thread_fetch().
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Initially the reading of the speed of an interface happened
upon interface creation and happened until the speed of a link
settled down to a single value. The speed of an interface
can also change as that a new optic can be inserted that
changes the speed, in which case FRR would see a interface
down (optic removal) and then a interface up (optic insertion).
In this case FRR would not treat this as an event that changed
the speed. Let's expand the checking a bit more.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
When enabling 'debug isis lfa', the option was correctly enabled
but not displayed by 'show debugging' command.
Signed-off-by: Fredi Raspall <fredi@voltanet.io>
When enabling TI-LFA the forward SPF for neighbors adjacent to the
PLR is computed. Later, when computing the PQ spaces, the reverse
SPF trees for those adjacent neighbors affected by the protected
interface are computed.
When node protection is enabled, TI-LFA link protection is run
immediately afterwards to compute repairs in case no
node-protecting backup path exists. In this second run, the
existing code tries to compute the reverse SPF tree for the same
node, without freeing the SPF tree of the prior run.
This patch fixes this by not computing the reverse SPF again, thus
avoiding a memory leak and an unnecessary SPF run.
Signed-off-by: Fredi Raspall <fredi@voltanet.io>
Individual tests must not depend on each other. In particular, a test
can't be sure that the previous test config is applied or cleared.
It is definitely not true when a single test is executed, for example:
`test_bgp_auth.py::test_prefix_peer_remove_passwords`.
This commit makes all tests independent of each other. It also adds a
call to check_all_peers_established at the start of "remove_passwords"
tests to make sure that we not only block new peers with an incorrect
password, but also clean the existing peers.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
