The systemd DHCPv6 client requires a hardware address only to
determine the IAID; NM always overrides the IAID with its own and
therefore the hwaddr is not used.
Removing such requirement allows DHCPv6 to run over PPP, which is
useful with DHCPv6-PD to get a prefix from the ISP.
To test this, I set up a server with pppoe-server, radvd and the Wide
DHCPv6 server providing an address and a prefix. On the client, NM was
able to obtain a prefix using both dhcp=dhclient and dhcp=systemd.
Note that if there is no hardware address and you specify
ipv6.dhcp-duid=ll or ipv6.dhcp-iaid=mac, a warning will be emitted and
NM will use a random DUID/IAID.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/478
NML3ConfigData is supposed to be used as immutable, ref-counted type.
You create it once, initialize it, seal it, and pass (immutable) references
around.
In such a scheme, having ref/unref functions not operate on const pointers
is a major inconvenience.
NML3ConfigData tracks IP addresses and routes. In their current form, these
types (NMPObject) always have an ifindex and there is no sensible way to have
an NMPObject (for routes or addresses) that have a wildcard ifindex.
Honor that by also tying NML3ConfigData to an ifindex. In most cases, the
user knows the ifindex before and can create it. On the unlikely case where
the user doesn't know the ifindex, we should add a new nm_l3_config_data_clone()
function, which allows migrating the setting from one ifindex to another.
This code is not specific to "nm-ip4-config.h"/"nm-ip6-config.h".
It applies to everybody who wants to iterate over a dedup-multi-index of
certain NMPObjects. Move it.
They ensure to consistently return -1, 0, 1. Also, I think they are
easier to understand.
What is in general hard to understand, whether a comparison sorts
ascending or descending. The macros maybe make that easier too, but it's
still confusing. That's why we have a test.
First of all, the entire nm_device_generate_connection() and
nm_ip._config_create_setting() approach is fundamentally flawed. You
cannot generate sensible configuration by reading IP addresses from
an interface. Anyway, that's what we still sometimes do, and we possibly
should do it less and less.
It's ugly that nm_ip6_config_capture() would read the "disable_ipv6"
sysctl value and cache it in NMIP6Config. Only so that it can be use
much later during nm_ip6_config_create_setting().
Instead, read the sysctl value shortly before it's needed.
If IPv6 is disabled, changing the IPv6 MTU fails and NM complains with
a warning. Since this error is expected and doesn't do any harm,
downgrade the logging level to DEBUG.
Since IPv6 kernel support can be built as a module, we have to check
the existence of /proc/sys/net/ipv6 every time. Instead of checking it
and then setting the MTU (adding one /proc access for everyone), just try
to set the MTU; in case of failure, determine the reason for the error.
https://bugzilla.redhat.com/show_bug.cgi?id=1840989https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/585
Currently NMIP4Config and NMIP6Config both track the data to be
configured, they expose properties on D-Bus, and they have logic for
capturing and applying settings to platform.
We will split that.
- NMIP4Config and NMIP6Config will expose data on D-Bus.
- NML3Cfg will have the logic for handling IP configuration.
- NML3ConfigData will track data to be configured.
NML3ConfigData mirrors NMIP4Config/NMIP6Config in many aspects. For now,
this duplicates a lot of code. More will be done later. Eventually,
NMIP4Config/NMIP6Config will drop the duplicated functionality.
The NML3Cfg instance tracks and prepares the IP configuration.
However, that is also partly exposed on other objects, like
NMIP4Config's "route-data" property.
Add an API, so that NMIP4Config can register itself to be notified
when something relevant changes.
This is an alternative to standard GObject properties and signals. They
often seem more effort than worth. That is, because in this case,
NMIP4Config.route-data has no other task then to re-emit the signal.
So, to implement that with GObject properties/signals, we would have to
add a property/signal to NML3Cfg, subscribe to it from NMIP4Config,
and remit the signal. An alternative is to bind properties, but that
would still be quite some extra code, and unclear that it would be
simpler. Not to mention the overhead, as bindings are themself full
GObject instances, that register to and emit signals by name.
We have several fields in the header file, so that the frequently used accessors
can be inlined. However, we also want some private data. Add a structure for that.
We need to react to platform changes. Also, we usually want to delay the
reaction to an idle handler.
Instead of subscribing each NML3Cfg instance itself to platform changes,
let only NMNetns do that. The goal is of course that each platform event
only needs to notify the NML3Cfg instance, which collects the events and
schedules them on the idle handler.
_ip_iface_update() only had one caller. The code is simpler to
understand by inlining it.
Also, it is relevant where and how we set ip_iface_ and ip_ifindex_
fields. Keep the places few and easily understandable.
and rename to nm_utils_ip_route_attribute_to_platform(). The function is independent
from NMIP4Config. We also will use it outside of NMIP4Config. Also, "NetworkManagerUtils.c"
already has similar functions that parse libnm structures to internal structures.
_nm_ip_config_best_default_route_set() doesn't really do anything
special. Use the generic helper function for the same job.
Also because NMIP4Config in the current form will be replaced by
something else, and this code needs to change.
Currently, we would not mark non-unicast routes with their type, so they
would wrongly appear as unicast routes in the D-Bus API.
That is wrong. For now, just hide them.
Fixes: 5d0d13f570 ('platform: add support for local routes')
A connection that fails due to dependency-failed is not able to
reconnect until the master connection activates again; when this
happens, the master clears the blocked reason for all its slaves in
activate_slave_connections() and tries to reconnect them. For this to
work, the slave should be marked as blocked when it fails with
dependency-failed.
If the device state change (to disconnected or unmanaged) triggered by
a sleep event happens after the wake, the devices becomes wrongly
unmanaged and it's necessary to manually manage it again, or restart
NM.
During the wake event we should disconnect the device_sleep_cb()
callback for all devices because we don't want to react to state
changes anymore; in particular we don't need to detect when the device
becomes disconnected to unmanage it.
"nm-device.c" is large and complicated. It's hard to find relevant places
that modify the ifindex,ip_ifindex,iface,ip_iface fields.
Mark them as const, to make that easier.
When NM fails to connect to teamd during an activation, it sets the
device state to FAILED. Eventually the device will become DISCONNECTED
and will call the ->deactivate() method that will perform the cleanup
of timers, teamd process and teamdctl instance.
However, in this way, when the device is DISCONNECTED timers are still
armed and can be triggered in the wrong state. Instead, perform the
cleanup immediately on failure.
https://bugzilla.redhat.com/show_bug.cgi?id=1856723
When using VRF devices we must pre-generate dependent local
routes in the VRF's table otherwise they will be incorrectly added
to the local table instead.
https://bugzilla.redhat.com/show_bug.cgi?id=1857133
Fixes: a199cd2a7d ('core: add dependent local routes configured by kernel')
Kernel will reject setting "active_slave", if the interface is not enslaved or not
up. We already handle that by setting the option whenever we enslave an interface.
However, we also must not set it initially, otherwise we get an ugly error log message:
NetworkManager[939]: <debug> [1594709143.7459] platform-linux: sysctl: setting net:/sys/class/net/bond99/bonding/active_slave to eth1 (current value is )
NetworkManager[939]: <error> [1594709143.7459] platform-linux: sysctl: failed to set bonding/active_slave to eth1: (22) Invalid argument
NetworkManager[939]: <warn> [1594709143.7460] device (bond99): failed to set bonding attribute active_slave to eth1
...
kernel: bond99: (slave eth1): Device is not bonding slave
kernel: bond99: option active_slave: invalid value (eth1)
See-also: https://bugzilla.redhat.com/show_bug.cgi?id=1856640https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/577
