When logging, messages include the interface name to specify what
device they refer to. In most case the interface name is unique.
There are some devices that don't have a kernel link associated, and
their interface name is not guaranteed to be unique. This is currently
the case for OVS bridges and OVS ports. When reading a log with
duplicate interface names, it is difficult to understand what is
happening. And this is made worse by the fact that it is common
practice to assign the same name to all devices in a OVS hierarchy
(bridge, port, interface).
To make logs unambiguous, we want to print the device type together
with the name; however we don't want to *always* print the type
because in most cases it's not useful and it would consume valuable
real estate on the screen. Adopt a simple heuristic of showing the
type only for OVS devices.
This commit adds a helper function to return the device type to show
in logs, when it is needed.
Hook the information for tracking the activation of a device, to the
NMDevice itself. Sure, that slightly couples the NMPolicy closer to
NMDevice, but the result is still simpler code because we don't need a
separate ActivateData.
It also means we can immediately tell whether the auto activation check
for NMDevice is already scheduled and don't need to search through the
list.
It's the better name. Especially since there is no more signal involved,
the term "emit" doesn't match.
Note also how the previous approach using a signal tried to abstract
what is happening. So we were no longer rechecking-autoconnect, instead,
we were emitting-a-signal-to-recheck-autoconnect. Just be plain about
what it is doing and don't go through a layer of signal.
GObject signals don't make the code easier to understand, on the
contrary. They may have their purpose, when objects truly must/should
not be aware of each other, and need to be composed very loosely. That
is not the case here.
There really is only one subscriber to NM_DEVICE_RECHECK_AUTO_ACTIVATE
signal, and it only makes sense this way. Instead of going through a
signal invocation, just call the well known method directly. It becomes
clearer who calls this code (and it has a lower overhead).
When using cscope/ctags it also is easier to follow the code because the
tools understand function calls.
This changes the signature of detach_port() to be asynchronous,
similarly to attach_port(). The implementation can return TRUE/FALSE
on immediate completion.
Current implementations return immediately and so there is no change
in behavior for now.
Improve logging:
- log only when something changes
- print the new resolver state, instead of the old one
- rename state "in-progress" to "started"
- log when the resolver state is reset due to DNS changes
With multi-connect enabled, this can cause infinite retries to autoconnect,
see [1].
That has bad consequences for example in initrd, where
nm-wait-online-initrd.service would wait up to one hour before failing
and blocking boot.
This reverts commit 1656d82045.
[1] https://bugzilla.redhat.com/show_bug.cgi?id=2039734#c5
Fixes: 1656d82045 ('policy: track the autoconnect retries in devices for multi-connect')
G_TYPE_CHECK_INSTANCE_CAST() can trigger a "-Wcast-align":
src/core/devices/nm-device-macvlan.c: In function 'parent_changed_notify':
/usr/include/glib-2.0/gobject/gtype.h:2421:42: error: cast increases required alignment of target type [-Werror=cast-align]
2421 | # define _G_TYPE_CIC(ip, gt, ct) ((ct*) ip)
| ^
/usr/include/glib-2.0/gobject/gtype.h:501:66: note: in expansion of macro '_G_TYPE_CIC'
501 | #define G_TYPE_CHECK_INSTANCE_CAST(instance, g_type, c_type) (_G_TYPE_CIC ((instance), (g_type), c_type))
| ^~~~~~~~~~~
src/core/devices/nm-device-macvlan.h:13:6: note: in expansion of macro 'G_TYPE_CHECK_INSTANCE_CAST'
13 | (G_TYPE_CHECK_INSTANCE_CAST((obj), NM_TYPE_DEVICE_MACVLAN, NMDeviceMacvlan))
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
Avoid that by using _NM_G_TYPE_CHECK_INSTANCE_CAST().
This can only be done for our internal usages. The public headers
of libnm are not changed.
In some scenarios, autoconnect should not be blocked if the device is
activated on the external connection (e.g. autoconnect on the loopback
device).
Adding the `allow_autoconnect_on_external` flag to support such
behavior.
Support managing the loopback interface through NM as the users want to
set the proper mtu for loopback interface when forwarding the packets.
Additionally, the IP addresses, DNS, route and routing rules are also
allowed to configure for the loopback connection profiles.
https://bugzilla.redhat.com/show_bug.cgi?id=2060905
For connections with multi-connect property set to "multiple", the
autoconnect-retries should be tracked per device and not per connection.
That means, if autoconnect-retries is set to 2, each device using that
connection should retry to autoconnect 2 times.
The device autoconnect retries is -2 by default. This is a special
value, in NMPolicy context, if the connection used is multi-connect the
device value will be set to match the connection retries. Each time the
device picks a different connection, it will reset the device
autoconnect retries to -2 and if needed, sync. with the connection
retries.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/1387https://bugzilla.redhat.com/show_bug.cgi?id=2039734
For some device types the attach-port operation doesn't complete
immediately. NMDevice needs to wait that the operation completes
before proceeding (for example, before starting stage3 for the port).
Change attach_port() so that it can return TERNARY_DEFAULT to indicate
that the operation will complete asynchronously. Most of devices are
not affected by this and can continue returning TRUE/FALSE as before
without callback.
When we have a bridge interface with ports attached externally (that is,
not by NetworkManager itself), then it can make sense that during
checkpoint rollback we want to keep those ports attached.
During rollback, we may need to deactivate the bridge device and
re-activate it. Implement this, by setting a flag before deactivating,
which prevents external ports to be detached. The flag gets cleared,
when the device state changes to activated (the following activation)
or unmanaged.
This is an ugly solution, for several reasons.
For one, NMDevice tracks its ports in the "slaves" list. But what
it does is ugly. There is no clear concept to understand what it
actually tacks. For example, it tracks externally added interfaces
(nm_device_sys_iface_state_is_external()) that are attached while
not being connected. But it also tracks interfaces that we want to attach
during activation (but which are not yet actually enslaved). It also tracks
slaves that have no actual netdev device (OVS). So it's not clear what this
list contains and what it should contain at any point in time. When we skip
the change of the slaves states during nm_device_master_release_slaves_all(),
it's not really clear what the effects are. It's ugly, but probably correct
enough. What would be better, if we had a clear purpose of what the
lists (or several lists) mean. E.g. a list of all ports that are
currently, physically attached vs. a list of ports we want to attach vs.
a list of OVS slaves that have no actual netdev device.
Another problem is that we attach state on the device
("activation_state_preserve_external_ports"), which should linger there
during the deactivation and reactivation. How can we be sure that we don't
leave that flag dangling there, and that the desired following activation
is the one we cared about? If the follow-up activation fails short (e.g. an
unmanaged command comes first), will we properly disconnect the slaves?
Should we even? In practice, it might be correct enough.
Also, we only implement this for bridges. I think this is where it makes
the most sense. And after all, it's an odd thing to preserve unknown,
external things during a rollback -- unknown, because we have no knowledge
about why these ports are attached and what to do with them.
Also, the change doesn't remember the ports that were attached when the
checkpoint was created. Instead, we preserve all ports that are attached
during rollback. That seems more useful and easier to implement. So we
don't actually rollback to the configuration when the checkpoint was
created. Instead, we rollback, but keep external devices.
Also, we do this now by default and introduce a flag to get the previous
behavior.
https://bugzilla.redhat.com/show_bug.cgi?id=2035519https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/ # 909
Names in header files should have an "nm" prefix. We do that pretty
consistently. Fix the offenders RfKillState and RfKillType.
Also, rename the RfKillState enums to follow the type name. For example,
NM_RFKILL_STATE_SOFT_BLOCKED instead of RFKILL_SOFT_BLOCKED.
Also, when we camel-case a typedef (NMRfKillState) we would want that
the lower-case names use underscore between the words. So it should be
`nm_rf_kill_state_to_string()`. But that looks awkward. So the right solution
here is to also rename "RfKill" to "Rfkill". That make is consistent
with the spelling of the existing `NMRfkillManager` type and the
`nm-rfkill-manager.h` file.
GObject Properties are flexible and powerful. In practice, NMDevicePrivate.rfkill_type
was only set once via the construct-only property NM_DEVICE_RFKILL_TYPE.
Which in turn was always set to a well-known value, only depending on the device
type.
We don't need this flexibility. The rfkill-type only depends on the
device type and doesn't change. Replace the property by a field in
NMDeviceClass.
For one, construct properties have an overhead, that the property setter is
called whenever we construct a NMDevice. But the real reason for this
change, is that a property give a notion as this could change during the
lifetime of a NMDevice (which it in fact did not, being construct-only).
Or that the type depends on something more complex, when instead it only
depends on the device type. A non-mutated class property is simpler,
because it's clear that it does not depend on the device instance,
only on the type/class.
Also, `git grep -w rfkill_type` now nicely shows the (few) references to
this variable and its easier to understand.
We don't run glib-mkenums for certain sources like "core" and
"libnm-glib-aux".
These annotations have no effect. Drop them.
They also mess with the automated formatting.
When the NM_UNMANAGED_PLATFORM_INIT flag is cleared last in
device_link_changed(), a recheck-assume is scheduled and then the
device goes immediately to UNAVAILABLE. During the state transition,
addresses and routes are removed from the interface. Then,
recheck-assume finds that the device can be assumed but it's too late
since the device was already deconfigured.
This is a problem as the whole point of assuming a device is to
activate a connection while leaving the device untouched.
In the NMCI "dracut_NM_vlan_over_bridge and dracut_NM_vlan_over_bond"
test, NM in real root tries to assume a vlan device that was activated
in initrd. When the interface gets deconfigured in UNAVAILABLE, the
connection to the NFS server breaks and the rootfs becomes
inaccessible.
The fix to this problem is to delay state transitions in
device_link_changed() to a idle handler, so that recheck-assume can
run before.
Fixes-test: @dracut_NM_vlan_over_bridge
Fixes-test: @dracut_NM_vlan_over_bond
https://bugzilla.redhat.com/show_bug.cgi?id=2047302
We use clang-format for automatic formatting of our source files.
Since clang-format is actively maintained software, the actual
formatting depends on the used version of clang-format. That is
unfortunate and painful, but really unavoidable unless clang-format
would be strictly bug-compatible.
So the version that we must use is from the current Fedora release, which
is also tested by our gitlab-ci. Previously, we were using Fedora 34 with
clang-tools-extra-12.0.1-1.fc34.x86_64.
As Fedora 35 comes along, we need to update our formatting as Fedora 35
comes with version "13.0.0~rc1-1.fc35".
An alternative would be to freeze on version 12, but that has different
problems (like, it's cumbersome to rebuild clang 12 on Fedora 35 and it
would be cumbersome for our developers which are on Fedora 35 to use a
clang that they cannot easily install).
The (differently painful) solution is to reformat from time to time, as we
switch to a new Fedora (and thus clang) version.
Usually we would expect that such a reformatting brings minor changes.
But this time, the changes are huge. That is mentioned in the release
notes [1] as
Makes PointerAligment: Right working with AlignConsecutiveDeclarations. (Fixes https://llvm.org/PR27353)
[1] https://releases.llvm.org/13.0.0/tools/clang/docs/ReleaseNotes.html#clang-format
Completely rework IP configuration in the daemon. Use NML3Cfg as layer 3
manager for the IP configuration of an interface. Use NML3ConfigData as
pieces of configuration that the various components collect and
configure. NMDevice is managing most of the IP configuration at a higher
level, that is, it starts DHCP and other IP methods. Rework the state
handling there.
This is a huge rework of how NetworkManager daemon handles IP
configuration. Some fallout is to be expected.
It appears the patch deletes many lines of code. That is not accurate, because
you also have to count the files `src/core/nm-l3*`, which were unused previously.
Co-authored-by: Beniamino Galvani <bgalvani@redhat.com>
The property `PROP_PORTS` should be of type g_param_spec_variant() with
variant 'ao'. This way the variant can be cached.
The deprecated property 'device::slaves' in
'src/core/devices/nm-device.c' must have the same getter-implementation,
returning the same GVariant instance.
Signed-off-by: Fernando Fernandez Mancera <ffmancera@riseup.net>
dhcp-anycast-address is only set by OLPC mesh device. It's ugly to have
this in form of a nm_device_set_dhcp_anycast_address() method, because
that means to cache the address in NMDevice. Meaning, we have more state
in NMDevice, where it's not clear where it comes from.
Instead, whenever we need to DHCP anycast address, as the subclass to
provide it (if any). This way, it gets extracted from the currently
applied connection at the moment when it is needed. Beyond that, the
setting is not duplicated/cached in NMDevice anymore.
"nm-device.c" is huge, and it does complicated things like handling the
state of the device and IP configuration.
It also contains simpler, individual functions, like converting enums to
strings. Let's move those trivial functions to a new module, so that the
remaining part is smaller.
"nm-device-utils.[ch]" should only contain simpler functions that have
no complex behavior or state.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/840
Currently "src/" mostly contains the source code of the daemon.
I say mostly, because that is not true, there are also the device,
settings, wwan, ppp plugins, the initrd generator, the pppd and dhcp
helper, and probably more.
Also we have source code under libnm-core/, libnm/, clients/, and
shared/ directories. That is all confusing.
We should have one "src" directory, that contains subdirectories. Those
subdirectories should contain individual parts (libraries or
applications), that possibly have dependencies on other subdirectories.
There should be a flat hierarchy of directories under src/, which
contains individual modules.
As the name "src/" is already taken, that prevents any sensible
restructuring of the code.
As a first step, move "src/" to "src/core/". This gives space to
reorganize the code better by moving individual components into "src/".
For inspiration, look at systemd's "src/" directory.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/743
