ikvswitch (1.0.8~bpo12+1) UNRELEASED; urgency=medium

  [ Thomas Goirand ]
  * Added the --map-redistribute-vlan-subnets option to the start
    command, accepting all (the default), none, or a comma separated
    list of subnet last-octets (eg: 117,118), controlling whether the
    VLAN-enabled leaf switch redistributes the per-VLAN subnets as
    connected routes into BGP. These subnets are for the OVS mode:
    its network nodes are L2-attached to these VLANs and source their
    SNAT traffic and BGP next-hops from these subnets, so the default
    of all keeps the SNAT return path working. For the OVN mode it's
    the opposite: reachability is provided by the /32 host route
    announced by the compute node hosting each floating IP, and a
    covering /24 would shadow it on the compute nodes, so fabrics
    without OVS network nodes can pass none or a restricted list.
  * Originate the cluster management subnets (CLUSTER_MGMT_SUBNETS) from
    every leaf switch with a blackhole static plus network statement,
    untagged so they reach the computes: BGP-2-the-host compute VRFs
    reject their own management subnets as AS loops, so without these
    covering prefixes the OVN chassis router can't route floating IP
    replies back to management-source clients. Also add the missing
    cl3 management /26 (192.168.123.0/26) to the default value.
  * Extend the VLAN subnets to 7 (VLAN 11 to 17, 192.168.113 to
    192.168.119), so that a 7th pool subnet and VLAN are available for
    a second external network, eg: an ext-net1 alongside ext-floating1.
  * Rehome the IPMI bridge network within the /19: HOST_IPMI_IP moves
    from 192.168.200.1 to 192.168.109.1. Also derive DHCP_RELAY_DESTINATION
    and CLUSTER_MGMT_SUBNETS from HOST_VIRTUAL_SUBNET_START_PREFIX, so
    that the whole address plan follows a single variable, and fix the
    hardcoded 113.201 BGP peer address of the VLAN-enabled leaf's FRR
    template, which was wrong whenever VLAN_SBUNETS_START_IP wasn't 113.
  * Bump the default VLAN_SUBNETS_NUM to 8 in the example configuration,
    so that the 7th and 8th pool subnets are really created by default
    (VLAN 11 to 18, 192.168.113 to 192.168.120): the 8th pool carries
    the OCI PoC's cl3 Octavia transport network.
  * Extend the hardcoded BGP speaker peers of the VLAN-enabled leaf from
    2 to 5: cl1's two network nodes on the legacy speaker VLAN
    (192.168.113.200 and .201), plus cl3's three controllers on its
    transport VLAN (192.168.118.200 to .202), so that controller-hosted
    BGP speakers can peer and announce routed networks.
  * Write the CLUSTER_MGMT_SUBNETS blackhole statics before the "router
    bgp" block of the leaf FRR template: bgpd stops reading frr.conf
    when it meets a command it doesn't own inside the BGP context, which
    silently dropped the leaf peer-group members, both address-families,
    the network statements and the route-maps, leaving every leaf unable
    to peer with the spines ("Configured AFI/SAFIs do not overlap with
    received MP capabilities").
  * Fix the VLAN pool count: create exactly VLAN_SUBNETS_NUM bridges
    (VLAN 11 to 18 for the default of 8, paired with 192.168.113 to
    192.168.120) instead of NUM+1, and align the --map-redistribute-
    vlan-subnets bound accordingly: the extra 9th bridge sat on
    192.168.121, where the cluster management /26s live.
  * Skip the BGP speaker peers of an undefined VLAN_SBUNETn_PREFIX in
    the VLAN-enabled leaf's FRR template, so that configuration files
    predating the cl3 transport VLAN don't get malformed "neighbor .200"
    statements.
  * Run the switch VMs with the q35 machine type instead of the legacy
    i440fx "pc" one, and pin the PCI address of every device (the SCSI
    controller at 0x3, then one slot per NIC from 0x4 on): the guests
    name their interfaces enp0s4, enp0s5, ... and the generated network
    and FRR configurations follow.
  * Build the switch template with root auto-login on both the VGA (VNC)
    and the serial (IPMI SOL) consoles, which are only reachable from
    the host: pass --tty-autologin yes and --add-serial-getty ttyS0 to
    build-openstack-debian-image. Rebuilding the cached template (rm
    /var/lib/ikvswitch/templates/ikvswitch-template.raw) is needed to
    apply it.
  * Update the README examples to the enp0sN guest interface names,
    following the q35 switch.
  * Disable reverse path filtering on every switch VM
    (net.ipv4.conf.all.rp_filter=0 and
    net.ipv4.conf.default.rp_filter=0, written to /etc/sysctl.d/99-rp-
    filter.conf), as neutron's own BGP CI (tempest-multinode-with-bgp)
    does: the kernel's reverse path validation, loose mode included,
    drops the asymmetric return traffic of BGP-2-the-host floating IPs
    whenever the chassis advertising the /32 host route differs from the
    one egressing the reply. Named 99- so that it applies after Debian's
    10-network-security.conf, whose default.rp_filter=2 would otherwise
    win the key.

  [ Jenkins ]
  * Rebuilt by bop.

 -- Jenkins <jenkins@bookworm-caracal.infomaniak.ch>  Sun, 15 Nov 2026 11:55:36 +0000

ikvswitch (1.0.7) unstable; urgency=medium

  * Cleanup the mounted switch partition and the loop device when
    ikvswitch-setup fails or gets interrupted, as the leaked mounts and
    busy loop devices were making every subsequent run fail with kpartx
    errors. Comes with a unit test simulating a mount failure.

 -- Thomas Goirand <zigo@debian.org>  Sun, 01 Nov 2026 17:29:58 +0100

ikvswitch (1.0.6) unstable; urgency=medium

  * Add DHCP_SERVER_DAEMON option, so that dnsmasq can be used as the DHCP
    server of the odd leaf switches, instead of isc-dhcp which isn't
    maintained upstream anymore (dnsmasq is the new default).
  * Also add DHCP_* options (lease times, pool bounds, domain name, DNS
    servers, PXE bootfiles and iPXE boot URL), as the DHCP configuration
    of the leaf switches is now generated from ikvswitch.conf, without
    any hardcoded value.
  * Add DHCP_HTTP_BOOT_URL option, so that native UEFI HTTP Boot clients
    (client-arch 16) are handed a real EFI loader to fetch over HTTP,
    with no PXE nor iPXE involved, defaulting to the shim EFI binary of
    the openstack-cluster-installer PXE server, so that Secure Boot is
    supported.
  * Guard the dnsmasq client-arch boot entries with tag:!ipxe, as an
    iPXE client also sends a value in option 93, and dnsmasq matches
    dhcp-boot entries in the reverse order of the config file: without
    this, iPXE clients could be handed a TFTP bootfile instead of the
    iPXE boot URL.
  * Add a shell script unit test suite under tests/, run at build time
    through run_tests, which runs the real ikvswitch-setup against
    mock binaries, and checks the DHCP configuration generated for
    both daemons, the HTTP Boot matching, and the configuration file
    validation.
  * Route to the internet VM using the configured prefix, instead of
    the hardcoded 192.168.96.2, so that overriding
    HOST_VIRTUAL_SUBNET_START_PREFIX in ikvswitch.conf keeps working.
  * Make ikvswitch-host-networking re-entrant: running start on an
    already or partially configured host now skips existing bridges,
    TAPs and dummies, and replaces the route and the IPMI bridge
    address, instead of failing.
  * Add a unit test suite for ikvswitch-host-networking: it checks the
    full creation inventory of start, the create/teardown symmetry of
    stop, that repeated start/stop cycles are re-entrant and never
    crash (including on a partially configured host), and that a
    customized HOST_VIRTUAL_SUBNET_START_PREFIX propagates everywhere.
    It runs against stateful mock binaries, deriving the kernel state
    from a shared log, with a sandboxed copy of the script.
  * Use pkgos-run-tests from openstack-pkg-tools (>= 157~) to run the
    unit tests in parallel, instead of the in-tree run_tests script,
    so that the test runner is shared between packages.
  * Fix the lintian bash-term-in-posix-shell tags on ikvswitch-setup:
    rename the HOSTNAME local variable of ikvswitch_customize_images
    to NAME, as HOSTNAME is a bash-only variable which isn't set under
    a POSIX shell. The "source /etc/network/interfaces.d/*" strings
    appended to the /etc/network/interfaces of the generated switch VM
    images are ifupdown directives, not shell code, and must stay
    verbatim, as the order in which the sourced files matters: they
    are covered by a package wide lintian override.
  * Added man pages for ikvswitch-setup and ikvswitch-host-networking,
    documenting the start and stop commands, the layout and networking
    variables they consume from /etc/ikvswitch/ikvswitch.conf, and the
    files they read and write, which also fixes the lintian
    no-manual-page warnings.
  * Standards-Version: 4.7.4.

 -- Thomas Goirand <zigo@debian.org>  Sun, 01 Nov 2026 12:16:39 +0100

ikvswitch (1.0.5) unstable; urgency=medium

  * Write in /etc/sysctl.d/00-forward-internet.conf as sysctl.conf is gone in
    Trixie.
  * Use Trixie as default distro for the setup.
  * Add || true when doing "ip link set down dev" if ipmi bridge.

 -- Thomas Goirand <zigo@debian.org>  Sun, 03 Aug 2025 22:54:49 +0200

ikvswitch (1.0.4) unstable; urgency=medium

  * Allow cl1-network-2 to adverize BGP.

 -- Thomas Goirand <zigo@debian.org>  Fri, 11 Apr 2025 14:07:58 +0200

ikvswitch (1.0.3) unstable; urgency=medium

  * Use --vlan-setup for leaf2 virtual switch, not leaf1.

 -- Thomas Goirand <zigo@debian.org>  Tue, 25 Mar 2025 14:15:10 +0100

ikvswitch (1.0.2) unstable; urgency=medium

  * Also do ip link set down dev ${HOST_IPMI_BRIDGE_NAME} when teardown.

 -- Thomas Goirand <zigo@debian.org>  Wed, 12 Feb 2025 11:08:55 +0100

ikvswitch (1.0.1) unstable; urgency=medium

  * Add ip link set up dev ${HOST_IPMI_BRIDGE_NAME}, otherwise the IPMI bridge
    cannot forward traffic to the host.

 -- Thomas Goirand <zigo@debian.org>  Thu, 06 Feb 2025 11:06:14 +0100

ikvswitch (1.0.0) unstable; urgency=medium

  * Switch default from 22 to 23u rack.
  * Release as 1.0, as it proves very stable.

 -- Thomas Goirand <zigo@debian.org>  Fri, 15 Nov 2024 14:48:40 +0100

ikvswitch (0.0.8) unstable; urgency=medium

  * Depends on qemu-kvm | qemu-system-any, not just qemu-system which installs
    qemu for all archs.

 -- Thomas Goirand <zigo@debian.org>  Thu, 27 Jun 2024 11:45:51 +0200

ikvswitch (0.0.7) unstable; urgency=medium

  * [a6235d2] Fancier prints.
  * [97860bf] Increase VNC port num from 50 to 80 to avoid conflict with oci-poc.
  * [ce38315] Increase default NUM_U to 19.
  * [16c01f3] Allow BGP advertizing from leaf1 switch VLANs.
  * [1a7aaff] Add IPMI bridge.
  * [a78f3e3] Create tuntap NIC and plug it to IPMI bridge.

 -- Thomas Goirand <zigo@debian.org>  Tue, 19 Mar 2024 14:50:30 +0100

ikvswitch (0.0.6) unstable; urgency=medium

  * [c450c4e] ikvswitch-setup: fix BGP accept list interface in leafs with
    arbitrary number of U.

 -- Thomas Goirand <zigo@debian.org>  Wed, 28 Feb 2024 14:39:26 +0100

ikvswitch (0.0.5) unstable; urgency=medium

  * [26a17e6] Make the list of iface in the dhcp0 bridge, be dynamic and depend
    on the number of U.

 -- Thomas Goirand <zigo@debian.org>  Wed, 28 Feb 2024 11:54:00 +0100

ikvswitch (0.0.4) unstable; urgency=medium

  * Increase the default number of ports from 13 to 16 per rack.
  * Update d/copyright.

 -- Thomas Goirand <zigo@debian.org>  Wed, 28 Feb 2024 11:21:07 +0100

ikvswitch (0.0.3) unstable; urgency=medium

  * [1963b5b] and [bc9e7cc] More docs.
  * [93f8ec6] Add 6x connected VLANs from 192.168.113/24 to 168.168.118.0/24,
    and accept annoucements from network nodes.

 -- Thomas Goirand <zigo@debian.org>  Tue, 16 Jan 2024 15:58:10 +0100

ikvswitch (0.0.2) unstable; urgency=medium

  * Add another "internet" virtual switch, and setup DHCPd on odd switches.

 -- Thomas Goirand <zigo@debian.org>  Tue, 09 Jan 2024 20:02:50 +0100

ikvswitch (0.0.1) unstable; urgency=medium

  * Initial release. (Closes: #1059614)

 -- Thomas Goirand <zigo@debian.org>  Fri, 29 Dec 2023 11:14:49 +0100
