** Description changed:

  [Impact]
  In Xenial/16.04LTS, one can't generate network interface name from 
"phys_port_name" attribute.
  
  "phys_port_name" indicates the interface physical port name within the
  NIC.
  
  [Test Case]
  
  Check that udev (systemd-udevd) provides the phys_port_name property
  Tests should be done on kernel versions: v4.15 (HWE)
  
  # cat /sys/class/net/<INTERFACE>/phys_port_name
  yyy
  
  Look if interface name contains the 'phys_port_name':
  
  $ ip link show
  ....
  3: ens3nyyy: <BROADCAST,MULTICAST,UP,LOWER_UP> ...
  ....
  
  [Regression Potential]
  
  * This piece of code is already in place in Bionic (systemd) and late.
  AFAICT, nothing has been reported since then with regards to this feature.
  
  * phys_port_name kernel support has been introduced in v4.1, but none of
  the current v4.4 Xenial kernel drivers uses it (minus rocker which is a
  test bed software switch for devel work on switchdev, it has no real
  function)
  
  # Xenial - kernel v4.4
  config-4.4.0-142-generic:# CONFIG_NET_SWITCHDEV is not set
  
  No drivers uses the phys_port_name method at all + NET_SWITCHDEV is not
  even turned on.
  
  # Xenial HWE - kernel v4.15
  config-4.15.0-99-generic:CONFIG_NET_SWITCHDEV=y
  
  Meaning if a regression arise, it will be limited to the HWE kernel (v4.15) 
and to the "Ethernet switch device driver model (switchdev)":
  mlx5
  mlxsw
  bnxt
  sfc (solarflar)
  
  --
  drivers/net/ethernet:
  --
  broadcom/bnxt/bnxt_vfr.c:     .ndo_get_phys_port_name = 
bnxt_vf_rep_get_phys_port_name
  broadcom/bnxt/bnxt.c: .ndo_get_phys_port_name = bnxt_get_phys_port_name
  cavium/liquidio/lio_vf_rep.c: .ndo_get_phys_port_name = 
lio_vf_rep_phys_port_name,
  mellanox/mlx5/core/en_rep.c:  .ndo_get_phys_port_name  = 
mlx5e_rep_get_phys_port_name,
  mellanox/mlxsw/switchx2.c:    .ndo_get_phys_port_name = 
mlxsw_sx_port_get_phys_port_name,
  mellanox/mlxsw/spectrum.c:    .ndo_get_phys_port_name = 
mlxsw_sp_port_get_phys_port_name,
  netronome/nfp/nfp_net_repr.c: .ndo_get_phys_port_name = 
nfp_port_get_phys_port_name,
  netronome/nfp/nfp_net_common.c:       .ndo_get_phys_port_name = 
nfp_port_get_phys_port_name,
  rocker/rocker_main.c: .ndo_get_phys_port_name         = 
rocker_port_get_phys_port_name,
  sfc/efx.c:    .ndo_get_phys_port_name = efx_get_phys_port_name,
  --
  
  # On other more specific kernels the regression potential can be
  expanded to virtio-net driver as well.
  
  # cloud-init may taint the result as it renames switchdev(s) after
  systemd at first initialization (even without the phys_port_name
  support) as follows:
  
  (Testing on a sfc card / server deployed by MAAS)
  
  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 36.199674] sfc 0000:08:00.1 
ens1f1: renamed from eth1 ## systemd
  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 37.128236] sfc 0000:08:00.0 
ens1f0: renamed from eth0 ## systemd
  
  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 84.653460] sfc 0000:08:00.0 
ens1f0np0: renamed from ens1f0 ## cloud-init
  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 84.697177] sfc 0000:08:00.1 
ens1f1np1: renamed from ens1f1 ## cloud-init
  
  cloud-init.log:2020-05-20 22:04:53,980 - util.py[DEBUG]: Running command 
['ip', 'link', 'set', 'ens1f0', 'name', 'ens1f0np0'] with allowed return codes 
[0] (shell=False, capture=True)
  cloud-init.log:2020-05-20 22:04:54,016 - util.py[DEBUG]: Running command 
['ip', 'link', 'set', 'ens1f1', 'name', 'ens1f1np1'] with allowed return codes 
[0] (shell=False, capture=True)
  
+ systemd:
+ ... OBFUSCATED_HOST systemd[1]: sys-subsystem-net-devices-ens1f1.device: Dev 
sys-subsystem-net-devices-ens1f1.device appeared twice with different sysfs 
paths /sys/devices/pci0000:00/0000:00:03.0/0000:08:00.1/net/ens1f1 and 
/sys/devices/pci0000:00/0000:00:03.0/0000:08:00.1/net/ens1f1np1
+ 
  # 1 concern was raise here by ddstreet:
  https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1875927/comments/1
  
  Here's the result of the test I have performed:
  https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1875927/comments/5
  
  This change indeed rename existing switchdev interfaces to add the
  'phys_port_name' (if any) into it. I'm afraid this will be a non-
  acceptable behaviour change in stable release and won't be SRU'able.
  
  More discussion on this topic to come ....
  
  [Other informations]
  
https://github.com/systemd/systemd/commit/4887b656c22af059d4e833de7b56544f24951184
  https://github.com/systemd/systemd/pull/4506
  
  [Original Description]
  
  It has been brought to my attention that systemd in Xenial/16.04LTS
  doesn't have support for phys_port_name[0] attribute.
  
  The support has been first introduced in systemd version "232" via:
  
https://github.com/systemd/systemd/commit/4887b656c22af059d4e833de7b56544f24951184
  https://github.com/systemd/systemd/pull/4506
  
  Bionic and late have the necessary bits ( systemd >232), but not Xenial
  (229)[1]
  
  Support for "phys_port_name" has been first introduced in the kernel
  with v4.1[2]
  
  [0]
  - https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-class-net
  - 
https://www.freedesktop.org/software/systemd/man/systemd.net-naming-scheme.html
  - https://www.kernel.org/doc/Documentation/networking/switchdev.txt
  
  [1]
  # git systemd/systemd
  git describe --contains 4887b656c22af059d4e833de7b56544f24951184
  v232~15
  
  # rmadison
   => systemd | 229-4ubuntu21.27 | xenial-updates
   systemd | 237-3ubuntu10.39 | bionic-updates
   systemd | 240-6ubuntu5.8   | disco-updates
   systemd | 242-7ubuntu3.7   | eoan-updates
   systemd | 245.4-4ubuntu3   | focal
   systemd | 245.4-4ubuntu3   | groovy
  
  [2]
  https://github.com/torvalds/linux/commit/db24a9044ee1
  
  $ git describe --contains db24a9044ee1
  v4.1-rc1

-- 
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to systemd in Ubuntu.
https://bugs.launchpad.net/bugs/1875927

Title:
  add support for phys_port_name attribute in Xenial/16.04LTS

Status in systemd package in Ubuntu:
  Fix Released
Status in systemd source package in Xenial:
  In Progress

Bug description:
  [Impact]
  In Xenial/16.04LTS, one can't generate network interface name from 
"phys_port_name" attribute.

  "phys_port_name" indicates the interface physical port name within the
  NIC.

  [Test Case]

  Check that udev (systemd-udevd) provides the phys_port_name property
  Tests should be done on kernel versions: v4.15 (HWE)

  # cat /sys/class/net/<INTERFACE>/phys_port_name
  yyy

  Look if interface name contains the 'phys_port_name':

  $ ip link show
  ....
  3: ens3nyyy: <BROADCAST,MULTICAST,UP,LOWER_UP> ...
  ....

  [Regression Potential]

  * This piece of code is already in place in Bionic (systemd) and late.
  AFAICT, nothing has been reported since then with regards to this feature.

  * phys_port_name kernel support has been introduced in v4.1, but none
  of the current v4.4 Xenial kernel drivers uses it (minus rocker which
  is a test bed software switch for devel work on switchdev, it has no
  real function)

  # Xenial - kernel v4.4
  config-4.4.0-142-generic:# CONFIG_NET_SWITCHDEV is not set

  No drivers uses the phys_port_name method at all + NET_SWITCHDEV is
  not even turned on.

  # Xenial HWE - kernel v4.15
  config-4.15.0-99-generic:CONFIG_NET_SWITCHDEV=y

  Meaning if a regression arise, it will be limited to the HWE kernel (v4.15) 
and to the "Ethernet switch device driver model (switchdev)":
  mlx5
  mlxsw
  bnxt
  sfc (solarflar)

  --
  drivers/net/ethernet:
  --
  broadcom/bnxt/bnxt_vfr.c:     .ndo_get_phys_port_name = 
bnxt_vf_rep_get_phys_port_name
  broadcom/bnxt/bnxt.c: .ndo_get_phys_port_name = bnxt_get_phys_port_name
  cavium/liquidio/lio_vf_rep.c: .ndo_get_phys_port_name = 
lio_vf_rep_phys_port_name,
  mellanox/mlx5/core/en_rep.c:  .ndo_get_phys_port_name  = 
mlx5e_rep_get_phys_port_name,
  mellanox/mlxsw/switchx2.c:    .ndo_get_phys_port_name = 
mlxsw_sx_port_get_phys_port_name,
  mellanox/mlxsw/spectrum.c:    .ndo_get_phys_port_name = 
mlxsw_sp_port_get_phys_port_name,
  netronome/nfp/nfp_net_repr.c: .ndo_get_phys_port_name = 
nfp_port_get_phys_port_name,
  netronome/nfp/nfp_net_common.c:       .ndo_get_phys_port_name = 
nfp_port_get_phys_port_name,
  rocker/rocker_main.c: .ndo_get_phys_port_name         = 
rocker_port_get_phys_port_name,
  sfc/efx.c:    .ndo_get_phys_port_name = efx_get_phys_port_name,
  --

  # On other more specific kernels the regression potential can be
  expanded to virtio-net driver as well.

  # cloud-init may taint the result as it renames switchdev(s) after
  systemd at first initialization (even without the phys_port_name
  support) as follows:

  (Testing on a sfc card / server deployed by MAAS)

  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 36.199674] sfc 0000:08:00.1 
ens1f1: renamed from eth1 ## systemd
  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 37.128236] sfc 0000:08:00.0 
ens1f0: renamed from eth0 ## systemd

  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 84.653460] sfc 0000:08:00.0 
ens1f0np0: renamed from ens1f0 ## cloud-init
  syslog:May 20 22:12:43 OBFUSCATED_HOST kernel: [ 84.697177] sfc 0000:08:00.1 
ens1f1np1: renamed from ens1f1 ## cloud-init

  cloud-init.log:2020-05-20 22:04:53,980 - util.py[DEBUG]: Running command 
['ip', 'link', 'set', 'ens1f0', 'name', 'ens1f0np0'] with allowed return codes 
[0] (shell=False, capture=True)
  cloud-init.log:2020-05-20 22:04:54,016 - util.py[DEBUG]: Running command 
['ip', 'link', 'set', 'ens1f1', 'name', 'ens1f1np1'] with allowed return codes 
[0] (shell=False, capture=True)

  systemd:
  ... OBFUSCATED_HOST systemd[1]: sys-subsystem-net-devices-ens1f1.device: Dev 
sys-subsystem-net-devices-ens1f1.device appeared twice with different sysfs 
paths /sys/devices/pci0000:00/0000:00:03.0/0000:08:00.1/net/ens1f1 and 
/sys/devices/pci0000:00/0000:00:03.0/0000:08:00.1/net/ens1f1np1

  # 1 concern was raise here by ddstreet:
  https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1875927/comments/1

  Here's the result of the test I have performed:
  https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1875927/comments/5

  This change indeed rename existing switchdev interfaces to add the
  'phys_port_name' (if any) into it. I'm afraid this will be a non-
  acceptable behaviour change in stable release and won't be SRU'able.

  More discussion on this topic to come ....

  [Other informations]
  
https://github.com/systemd/systemd/commit/4887b656c22af059d4e833de7b56544f24951184
  https://github.com/systemd/systemd/pull/4506

  [Original Description]

  It has been brought to my attention that systemd in Xenial/16.04LTS
  doesn't have support for phys_port_name[0] attribute.

  The support has been first introduced in systemd version "232" via:
  
https://github.com/systemd/systemd/commit/4887b656c22af059d4e833de7b56544f24951184
  https://github.com/systemd/systemd/pull/4506

  Bionic and late have the necessary bits ( systemd >232), but not
  Xenial (229)[1]

  Support for "phys_port_name" has been first introduced in the kernel
  with v4.1[2]

  [0]
  - https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-class-net
  - 
https://www.freedesktop.org/software/systemd/man/systemd.net-naming-scheme.html
  - https://www.kernel.org/doc/Documentation/networking/switchdev.txt

  [1]
  # git systemd/systemd
  git describe --contains 4887b656c22af059d4e833de7b56544f24951184
  v232~15

  # rmadison
   => systemd | 229-4ubuntu21.27 | xenial-updates
   systemd | 237-3ubuntu10.39 | bionic-updates
   systemd | 240-6ubuntu5.8   | disco-updates
   systemd | 242-7ubuntu3.7   | eoan-updates
   systemd | 245.4-4ubuntu3   | focal
   systemd | 245.4-4ubuntu3   | groovy

  [2]
  https://github.com/torvalds/linux/commit/db24a9044ee1

  $ git describe --contains db24a9044ee1
  v4.1-rc1

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1875927/+subscriptions

-- 
Mailing list: https://launchpad.net/~touch-packages
Post to     : touch-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~touch-packages
More help   : https://help.launchpad.net/ListHelp

Reply via email to