Merge pull request #861 from dwalton76/bgpd-attr-unchanged

bgpd: attribute-unchanged issues with peer-groups

Merge pull request #859 from donaldsharp/bgp_ll_global

Bgp -> Allow comparison of correct v6 nexthop

Merge pull request #862 from dwalton76/bgpd-update-too-large

bgpd: UPDATE may be larger than 4096 if addpath is used

Merge pull request #863 from dwalton76/zebra-static-reject

zebra: static route cleanup

Merge pull request #864 from dwalton76/zebra-debug-packet-detail

Zebra debug packet detail

Merge pull request #869 from donaldsharp/hash_a_roni

lib: Remove expansion of hash table

Merge pull request #868 from LabNConsulting/working/master/patch-set/rfapi-lib+

Restore rfapi and encap attribute, broken by some recent changes

bgpd rfapi: use afi_t

Signed-off-by: Lou Berger <lberger@labn.net>

bgpd rfapi: fix breakage introduced in refactoring

Signed-off-by: Lou Berger <lberger@labn.net>

lib: Remove expansion of hash table

The hash code has the idea of stopping expanding
the hash table when certain criteria are set.
With the recent addition of `show hashtable`
we can now see that when we have a full internet
feed we've stopped expanding the table at 1k
buckets.  This results in some serious performance
issues at scale.

Since we now have the ability to see the statistics
on a hash table, let's allow it to expand.  Doing
so on a full feed showed this:

before:
   Hash table          |  Buckets   Entries   Empty   LF        SD      FLF       SD
 ----------------------+----------------------------------------------------------------
   route table hash    |  1024      1187579   0%      1159.75   34.06   1159.75   35.08
   route table hash    |  32768     76208     10%     2.33      2.80    2.58      4.03
   route table hash    |  1024      1187572   0%      1159.74   34.06   1159.74   35.08
   route table hash    |  2048      76205     0%      37.21     6.13    37.21     7.29

Showing hash table statistics for BGP
-------------------------------------

   Hash table         |  Buckets   Entries   Empty   LF       SD      FLF      SD
 ---------------------+--------------------------------------------------------------
   BGP Attributes     |  131072    251229    15%     1.92     2.48    2.25     3.33
   route table hash   |  4096      1187572   0%      289.93   17.03   289.93   17.87
   route table hash   |  32768     76205     10%     2.33     2.90    2.58     4.21

After:

   Hash table          |  Buckets   Entries   Empty   LF     SD     FLF    SD
 ----------------------+--------------------------------------------------------
   route table hash    |  1048576   1187349   32%     1.13   2.57   1.67   3.16
   route table hash    |  32768     76195     10%     2.33   2.81   2.58   4.03
   route table hash    |  1048576   1187342   32%     1.13   2.58   1.67   3.16
   route table hash    |  32768     76192     10%     2.33   2.68   2.58   3.81

Showing hash table statistics for BGP
-------------------------------------

   Hash table         |  Buckets   Entries   Empty   LF     SD     FLF    SD
 ---------------------+--------------------------------------------------------
   BGP Attributes     |  131072    251222    15%     1.92   2.64   2.25   3.58
   route table hash   |  1048576   1187342   32%     1.13   2.52   1.67   3.07
   route table hash   |  32768     76192     10%     2.33   2.86   2.58   4.12

We should see some significant performance improvements across the board
for full feeds.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

bgp rfapi: use route_table_init and _finish (fixes crash due to recent lib change)

Signed-off-by: Lou Berger <lberger@labn.net>

bgpd: Fix nexthop comparison for v6

When we have both a LL and a Global address,
use what the attribute wants for comparison
instead of assuming Global than LL.

This was causing BGP to install v6 routes
that used the LL as the nexthop, where
the global address was different and
being used as the basis for comparison.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

bgpd: Set the ifindex to DELETED after we notify zebra

The code path for a deleted interface was calling
zebra with a IFINDEX_DELETED, which caused zebra
to bitch and moan about the issue.  Since the
only thing this function does is call zebra
to deregister the RA stuff, don't set the
ifindex to DELETED till afterwords.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

Merge branch 'master' of https://github.com/dwalton76/frr into zebra-debug-packet-detail

Conflicts:
zebra/debug.c

bgpd: attribute-unchanged issues with peer-groups

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>

bgpd: UPDATE may be larger than 4096 if addpath is used

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>

Merge pull request #842 from qlyoung/update-style-guide

Update COMMUNITY.md

zebra: static route cleanup

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>
- The 'reject' keyword was being lost
- Hide the "ip route A.B.C.D A.B.C.D" config options since these will be
  displayed as "ip route A.B.C.D/X"

COMMUNITY.md: rewrap paragraphs, update PR section

* Wrap paragraphs at 80 lines
* Update Github PR section to remove mention of develop branch

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

COMMUNITY.md: minor grammatical fixes

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

COMMUNITY.md: document documentation

Add guidelines for documentation to COMMUNITY.md

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

COMMUNITY.md: miscellaneous documentation

Also:
- Add documentation on mailing list
- Update mentions of unit tests
- Update process description

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

COMMUNITY.md: s/PROJECT/FRRouting

Also:
* Make headers consistently capitalized
* Some extra backticks where needed

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

COMMUNITY.md: update style guide

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

zebra: debug cleanup

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>

Merge pull request #848 from opensourcerouting/rb-tree-fix

Fix corrupted RB trees

Merge pull request #847 from donaldsharp/cmd_success

Cmd success

Merge pull request #838 from dwalton76/lib-copy-nexthops

lib: copy_nexthops() only copies the first nexthop

Merge pull request #837 from qlyoung/fix-eigrp-routemap-cli

eigrpd, vtysh: add dummy route-map cli

Merge pull request #836 from dwalton76/zebra-debug-packet-send-detail

zebra: "debug zebra packet" config display

lib: fix corrupted RB trees

Commit 8f942af90 introduced a bug while silencing a clang warning. Silence
the warning in a different way to fix our red-black tree implementation.

Fixes #841.

Signed-off-by: Renato Westphal <renato@opensourcerouting.org>

lib: revert reindent of files imported from OpenBSD

We should preserve the original indentation to make it easier to keep
these files in sync with the upstream.

Signed-off-by: Renato Westphal <renato@opensourcerouting.org>

ripd: Fix debug config failures

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

ripngd: Fix failure returned with some debug commands

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

*: fix git-reindent-branch.py reversing order

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Revert "*: reindent pt. 2"

This reverts commit c14777c6bfd0a446c85243d3a9835054a259c276.

clang 5 is not widely available enough for people to indent with.  This
is particularly problematic when rebasing/adjusting branches.

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

lib: copy_nexthops() only copies the first nexthop

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>
Before the fix NHT had each path resolving via swp1

cel-redxp-10# show ip route 20.0.11.253
Routing entry for 20.0.11.0/24
  Known via "bgp", distance 20, metric 0, best
  Last update 00:00:20 ago
  * 169.254.0.1, via swp1
  * 169.254.0.17, via swp2

cel-redxp-10#

cel-redxp-10# show ip nht
[snip]
20.0.11.253
 resolved via bgp
 via 169.254.0.1, swp1
 via 169.254.0.1, swp1
 Client list: pim(fd 19)

After the fix

cel-redxp-10# show ip nht
[snip]
20.0.11.253
 resolved via bgp
 via 169.254.0.1, swp1
 via 169.254.0.17, swp2
 Client list: pim(fd 19)

*: add git-reindent-branch.py

This tries to heave a branch over the reindentation step.  Somewhat
tested, don't trust it blindly...

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

eigrpd, vtysh: add dummy route-map cli

eigrpd will successfully accept `(conf)# route-map foo ...` because it
is not sent to eigrpd from vtysh, but of course, this is the classic
node sync syndrome.

Since eigrpd apparently doesn't support proper routemaps yet just add
the cli so we can suppress the vtysh errors.

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

Merge pull request #834 from dwalton76/ospf6d-show-crash

ospf6d: crash in ospf6_lsdb_show

Merge pull request #835 from dwalton76/lib-route-node-lookup

lib: route_node_lookup() needs to apply_mask() to prefix

zebra: "debug zebra packet" config display

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>

lib: route_node_lookup() needs to apply_mask() to prefix

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>
Reviewed-by: <qlyoung@cumulusnetworks.com>

Merge pull request #828 from tigranmartirosyan/master

#752 ARP is not updating if mac address changed on remote side

Add 1 more identation to correspond to kernel style multi-line comment

Merge pull request #833 from dslicenc/cm16779-bgp-startup-ll

bgpd: fix peer startup for labeled-unicast if linklocal address not f…

Merge pull request #827 from devicenull/patch-1

Add note about bridge limitations

Merge pull request #805 from Orange-OpenSource/master

Add OSPF API and FRR Hacking documents

ospf6d: crash in ospf6_lsdb_show

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>

bgpd: fix peer startup for labeled-unicast if linklocal address not found

Problem found in testing where ipv6 labeled-unicast prefixes were not received
on the peers if a "service networking restart" was issued.  Same problem would
happen with an ifdown/ifup on the link to the peer.  Found the problem to be
that peers would establish for labeled-unicast even if a link-local address was
not yet available on the interface toward the peer, causing updates to be sent
without a nexthop value. These were then rejected by the peer. Fix is to delay
peer establishment until after the link-local addresses are available.

Ticket: CM-16779
Signed-off-by: Don Slice <dslice@cumulusnetworks.com>
Reviewed By: Donald Sharp <sharpd@cumulusnetworks.com>
Testing Done: Manual testing successful.  Bgp-smoke completed with no new failures

replace space to tabs, add kernel styles multiline, remove trailing whitespaces.

*: reindent pt. 2

w/ clang 5

* reflow comments
* struct members go 1 per line
* binpack algo was adjusted

Add note about bridge limitations

Add note that Zebra will never be able to detect the master interface for any interface due to IFLA_INFO_SLAVE_KIND not being defined

Merge pull request #826 from qlyoung/rm-vtyoutln

eigrpd: remove last vty_outln

eigrpd: remove last vty_outln

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>

merge with upstream

*: reindent

indent.py `git ls-files | pcregrep '\.[ch]$' | pcregrep -v '^(ldpd|babeld|nhrpd)/'`

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: add indent control files

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge pull request #818 from donaldsharp/eigrp_afi

Eigrp afi

Merge pull request #820 from opensourcerouting/gcc7

fix new warnings emitted by GCC 7

Merge pull request #794 from opensourcerouting/table-hash-ospf6-lsdb-refactor

use hash in lib/table, also refactor ospf6 lsdb code

lib: parser: remove incorrect assert

A {foo|bar|baz} graph node will have more than 1/2 incoming links even
if no other references are left to it (which is what the assert was
previously trying to ensure.)

I don't see a good way to "fix" the assert so I'm just removing it.

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge pull request #806 from dwalton76/bgpd-set-ipv4-vpn-nexthop

bgpd: "set ipv4 vpn next-hop 1.1.1.1" is saved/displayed as "set ip vpn next-hop 1.1.1.1"

tests/ospf6d/test_lsdb: remove vty_outln

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge remote-tracking branch 'frr/master' into table-hash-ospf6-lsdb-refactor

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

nhrpd: enable -Werror

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: fix GCC 7 warnings/issues

The label initializer & nhrpd variable are just to shut up GCC 7,
the other two are actual bugs.

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: fix GCC 7 switch/case fallthrough warnings

Need a comment on these.

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge pull request #803 from opensourcerouting/redhat-doc-fixes-master

Redhat doc fixes for master branch

Merge pull request #809 from donaldsharp/evpn_plus_struct_attr

Evpn plus struct attr

bgpd, zebra: Cleanup warnings from new code

1) Clean up VTY_NEWLINE -> \n
2) Remove usages of VTY_GET_INTEGER

Signed-off-by: Donald Sharp

Merge remote-tracking branch 'origin/master' into evpn_plus_struct_attr

Merge remote-tracking branch 'origin/master' into evpn_plus_struct_attr

bgpd: Disable new EVPN CLI

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

Merge pull request #813 from opensourcerouting/newline-redux

newline redux

Merge pull request #762 from bingen/mutiple_nh_recursive_levels

Allow for more than 1 NH recursion level

Merge pull request #804 from opensourcerouting/fixes/tests/prng

tests: update prng to return better pseudo random numbers

lib, vtysh: remove now-useless newline arg

It's always \n now, hooray.

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge remote-tracking branch 'frr/master' into newline-redux

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge pull request #799 from dwalton76/bgpd-show-labeled-unicast

bgpd: show bgp ipv4 label-unicast does not provide any output

Merge pull request #808 from qlyoung/vtysh-termcols

lib, vtysh: pretty-print variable autocompletions

*: remove vty_outln again

(PRs merged to master added another few vty_outln() calls)

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge remote-tracking branch 'frr/master' into newline-redux

Lots of conflicts from CMD_WARNING_CONFIG_FAILED...

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

Merge pull request #800 from qlyoung/fix-old-vpn-commands

bgpd: fix old-style vpn commands

Merge pull request #815 from dwalton76/CMD_WARNING_take2

vtysh: return non-zero for configuration failures

Merge pull request #817 from donaldsharp/babel_redist

babeld: Allow babel to redistribute v6 routes

Merge pull request #814 from qlyoung/vtysh-password-commands

vtysh: fixup password commands

*: remove VTYNL, part 6 of 6

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: remove VTYNL, part 5 of 6

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: remove VTYNL, part 4 of 6

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: remove VTYNL, part 3 of 6

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: remove VTYNL, part 2 of 6

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: remove VTYNL, part 1 of 6

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

ospf6d: VNL -> VTYNL

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

*: ditch vty_outln(), part 2 of 2

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

eigrpd: Fix up some redistribution issues

Get EIGRP to start attempting to redistribute routes
as well as to display the fact it thinks it is redistributed.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

eigrpd: Let eigrp setup it's router-id in some situations

If the router-id is passed up to eigrp before we have
started the instance, allow it to set it up properly
as we start the instance.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

eigrpd: Fixed wrong type used

Recent additions to prefix_copy to check afi type exposed
an issue with eigrp setting the wrong afi type for a `struct prefix`.

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

babeld: Allow babel to redistribute v6 routes

Turn on the ability in babel to receive v6 routes when requested.

robot# show babel route
192.168.201.0/24 metric 0 (exported)
2606:a000:111d:803e::/64 metric 0 (exported)
robot# conf t
robot(config)# int enp3s0
robot(config-if)# ipv6 addr 2404:4000:991d:804c:c32:ee94:742c:4d5/73
robot(config-if)# end
robot# show babel route
192.168.201.0/24 metric 0 (exported)
2606:a000:111d:803e::/64 metric 0 (exported)
2404:4000:991d:804c:c00::/73 metric 0 (exported)
robot#

Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>

vtysh: return non-zero for configuration failures

Signed-off-by: Daniel Walton <dwalton@cumulusnetworks.com>
This allows frr-reload.py (or anything else that scripts via vtysh)
to know if the vtysh command worked or hit an error.