add support for ipv6 slaac #259
Conversation
|
ping @julienfortin for review ^_^ |
|
@aderumier thanks for the ping, i just started some tests, i'll update once i have the results |
|
@aderumier when running test i'm seeing the following: I don't have the config used for the test only the output right now.
Also i'm not a fan of using
|
mmm, that's mean with accept_ra/autoconf are enabled in sysctl (they are disabled by default in kernel, and address module set them to 0 by default too when not defined, like on debian ifupdown1). That's strange that it's not intentionnaly enabled... could be great if you could have the config to reproduce on my side. But I'll look to fix this "bool object" error.
Well, I'm not a big fan too, but this is how debian defined it in ifupdown1, and I have a lot of proxmox users, complain with error on upgrade from ifupdown1 to ifupdown2 when "inet6 auto" is defined in their config. (almost everybody with a clean default install of debian). ifupdown2 can't be restart with "inet6 auto" ,and apt install is dying too with ifupdown2 package install in error state. I'm currently on holiday, I'll rework on this in 2 weeks. |
aderumier
force-pushed
the
ipv6auto
branch
4 times, most recently
from
1a4067c
to
e56ffd9
Compare
ok, I was able to reproduce. I have also do a fix for dotted vlan interfaces name for sysctl apply. Could you give it a try again ? |
|
(Note that I'm still working on it for default accept_ra value. Currently I'm forcing it to 0, this is wrong, it should use the net.ipv6.conf.all.accept_ra value) |
aderumier
force-pushed
the
ipv6auto
branch
3 times, most recently
from
825be38
to
2fd8fdb
Compare
fixed in |
|
improvement: I check ipv6 attribute flags (permanent 0x80), to known if ip is dynamic (slaac attributed). (Like this, it's possible to add/remove static ipv6 when autoconf is enabled) |
The current code is buggy if an interface only use ipv6.
ipv6 don't have primary and are not ordered, so change can randomly
remove/re-add all ipv6 address
```
auto eno4
iface eno4 inet6 static
address 2001:db8:a0b:12f0::1/64
info: eno4: netlink: ip addr add 2001:db8:a0b:12f0::1/64 dev eno4
auto eno4
iface eno4 inet6 static
address 2001:db8:a0b:12f0::1/64
address 2001:db8:a0b:12f0::2/64
info: eno4: netlink: ip addr add 2001:db8:a0b:12f0::2/64 dev eno4
auto eno4
iface eno4 inet6 static
address 2001:db8:a0b:12f0::1/64
address 2001:db8:a0b:12f0::2/64
address 2001:db8:a0b:12f0::3/64
info: eno4: primary ip changed (from 2001:db8:a0b:12f0::1/64 to 2001:db8:a0b:12f0::2/64) we need to purge all ip addresses and re-add them
info: eno4: netlink: ip addr del 2001:db8:a0b:12f0::2/64 dev eno4
info: eno4: netlink: ip addr del 2001:db8:a0b:12f0::1/64 dev eno4
info: reading '/proc/sys/net/ipv6/conf/eno4/disable_ipv6'
info: eno4: netlink: ip addr add 2001:db8:a0b:12f0::1/64 dev eno4
info: eno4: netlink: ip addr add 2001:db8:a0b:12f0::2/64 dev eno4
info: eno4: netlink: ip addr add 2001:db8:a0b:12f0::3/64 dev eno4
```
only first attributes used in any kernels
we want to get IFA_FLAGS to known if an ip is permanent (0x80), or dynamic (slaac)
simply sysctl accept_ra=1 && autoconf=1
Currently it's only available in dhcp method. Set 0 by default (if interface was previously method auto).
we manage them in address now
user can defined both static ips && enable accept_ra 1 in as or auto method exemple1: ---------- iface eth0 inet auto iface eth0 inet6 static address .... exemple2: -------- iface eth0 inet6 static address .... accept_ra 1 autoconf 1 We need to process configured addresses, but we won't purge ipv6 on reload
…pty netlink cache factorize with adding_get_netlink_cache_accept_ra && _get_netlink_cache_auto and correctly test if cache exist or not the netlink cache can have empty value for an interface, if the interface was not existing when we have populate the cache. for example, a vlan interface created in pre-up by vlan module. In this case, we return an empty string
accept_ra is only to get default gw, autoconf is for generate ip Also skip only dynamic ipv6 (!ipflag permanent 0x80)
|
I think it's ok for a second review, the bugs should be fixed. |
Signed-off-by: Julien Fortin <jfortin@nvidia.com>
Replace indexing by variables named start/end and prefix/suffix.
ifquery excluded the last digit of interfaces range given. ex: eth[1-2] would give only eth1 instead of eth1 + eth2. This commit fix this behavior by increasing the range in expand_iface_range.
The commit make the auto_ifaces container pointing to the allow_classes['auto'] list. (since it's a mutable object, we get the same instance)
This change goal is to make auto behave like allow-auto. This commit will also provide the interfaces range capability to any other allow-class names.
The only real change is the creation of a ifaceobj before testing it's name. (The ifaceobj will still not be added if deemed invalid)
This commit fix the following by making an understandable error msg: * the 'allow eth0' would make a IndexError shown to the user. * the 'allow-' would be valid and use an empty classname.
some nic like mellanox connectx don't work well with vlan aware bridge && rx-vlan-filter. (They are limited in number of vlans in hardware filtering, and break with big number of vlans like bridge-vids 2-4096) Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
if a script in /etc/network/ifup.d/ is segfaulting,
on config loading (ifup -a), ifupdown2 is stopping
ifup -a
"
info: executing2 /etc/network/if-up.d/postfix
debug: lo: up : running script /etc/network/if-up.d/resolved
info: executing2 /etc/network/if-up.d/resolved
error: name 'traceback' is not defined
debug: saving state ..
info: exit status 1
"
with this fix:
debug: lo: up : running script /etc/network/if-up.d/resolved
info: executing2 /etc/network/if-up.d/resolved
File "/usr/share/ifupdown2/ifupdown/scheduler.py", line 325, in run_iface_list
cls.run_iface_graph(ifupdownobj, ifacename, ops, parent,
File "/usr/share/ifupdown2/ifupdown/scheduler.py", line 315, in run_iface_graph
cls.run_iface_list_ops(ifupdownobj, ifaceobjs, ops)
File "/usr/share/ifupdown2/ifupdown/scheduler.py", line 188, in run_iface_list_ops
cls.run_iface_op(ifupdownobj, ifaceobj, op,
File "/usr/share/ifupdown2/ifupdown/scheduler.py", line 150, in run_iface_op
ifupdownobj.log_error('%s: %s %s' % (ifacename, op, str(e)))
File "/usr/share/ifupdown2/ifupdown/ifupdownmain.py", line 226, in log_error
raise Exception(str)
error: lo : lo: up cmd '/etc/network/if-up.d/resolved' failed: returned -11
debug: vmbr0: found dependents ['bond0']
debug: bond0: found dependents ['enp65s0d1', 'enp65s0']
info: enp65s0d1: running ops ...
...
...
|
I just send a fix for bridge interface where accept_ra|autoconf was not applied. (also resync to last master to master) |
Hi
This patch series add support for 'inet6 auto" method,
add support for accept_ra && autoconf on "inet6 static" interfaces (previously if was only working for inet6 dhcp)
This fix
#249
#122
#178
example:
up again (accept_ra,autoconf are in cache, so no need to reapply them again)
switch to static ((accept_ra && autoconf are reset to 0)
up again (accept_ra,autoconf are in cache, so no need to reapply them again)
add accept_ra=1/autoconf=1 on eno3
remove ip from interface (here we can't purge ip as slaac is still enable, and we can't known which ip is coming from slaac)
testing dhcp + accept_ra/autoconf
mixing both auto && static
Mixing both auto && static, with accept_ra=0 in static (will be skipped as auto need accept_ra=1)
ifquery check is also implemented