We need to know whether we can create interfaces of any given
NMDevice subclass or not. So don't rely on just the NMPlatformLink
for that information, because we won't have a platform link for
software devices before we create them.
Instead of hacky stuff in the Manager, let plugins themselves indicate
which links should be ignored (because they are really child links that
are controlled by a different device that the plugin handles).
Ethernet, WiFi, and VLAN used the same implementation for initial address.
Ethernet and WiFi used the same implementation (and duplicated code) for
permanent MAC address, plus they both used ethtool in what should be
generic code, which is better done in the platform.
Instead of having a bunch of logic in the Manager for determining the
VLAN and Infiniband virtual interface names, move the type-specific
logic into the plugins themselves.
Instead of looping over all plugins and asking each plugin whether it
can handle a link or a connection, have them advertise the link and
connection types they support, and use that when creating new devices.
For existing devices, depending on the order that netlink sends interfaces to
us, the parent may be found after the VLAN interface and not be available when
the VLAN interface is constructed. Instead of failing construction, when a
NMDeviceVlan has no parent keep it unavailable for activation. Then have
the Manager notify existing devices when a new device is found, and let
NMDeviceVlan find the parent later and become available via that mechanism.
This doesn't apply to VLANs created by NM itself, because the kernel requires
a parent ifindex when creating a VLAN device. Thus this fix only applies to
VLANs created outside NetworkManager, or existing when NM starts up.
Most nm_platform_*() functions operate on the platform
singleton nm_platform_get(). That made sense because the
NMPlatform instance was mainly to hook fake platform for
testing.
While the implicit argument saved some typing, I think explicit is
better. Especially, because NMPlatform could become a more usable
object then just a hook for testing.
With this change, NMPlatform instances can be used individually, not
only as a singleton instance.
Before this change, the constructor of NMLinuxPlatform could not
call any nm_platform_*() functions because the singleton was not
yet initialized. We could only instantiate an incomplete instance,
register it via nm_platform_setup(), and then complete initialization
via singleton->setup().
With this change, we can create and fully initialize NMPlatform instances
before/without setting them up them as singleton.
Also, currently there is no clear distinction between functions
that operate on the NMPlatform instance, and functions that can
be used stand-alone (e.g. nm_platform_ip4_address_to_string()).
The latter can not be mocked for testing. With this change, the
distinction becomes obvious. That is also useful because it becomes
clearer which functions make use of the platform cache and which not.
Inside nm-linux-platform.c, continue the pattern that the
self instance is named @platform. That makes sense because
its type is NMPlatform, and not NMLinuxPlatform what we
would expect from a paramter named @self.
This is a major diff that causes some pain when rebasing. Try
to rebase to the parent commit of this commit as a first step.
Then rebase on top of this commit using merge-strategy "ours".
nm_device_get_hw_address() may return NULL and nm_platform_link_get_type may
return NM_LINK_TYPE_NONE. While it might be a good idea to check for such cases
at the init time it seems easier to just ignore it and prevent blowing up in
subsequent deactivation.
A quick test case:
# while :; do ip link add moo0 type veth peer moo1; ip link del moo0 ; done
Yields:
NetworkManager:ERROR:devices/nm-device-ethernet.c:268:constructor:
assertion failed: (link_type == NM_LINK_TYPE_ETHERNET ||
link_type == NM_LINK_TYPE_VETH)
nm_device_set_hw_addr: assertion 'addr != NULL' failed
https://bugzilla.gnome.org/show_bug.cgi?id=740992
Add nm-core-types.h, typedefing all of the GObject types in
libnm-core; this is needed so that nm-setting.h can reference
NMConnection in addition to nm-connection.h referencing NMSetting.
Removing the cross-includes from the various headers causes lots of
fallout elsewhere. (In particular, nm-utils.h used to include
nm-connection.h, which included every setting header, so any file that
included nm-utils.h automatically got most of the rest of libnm-core
without needing to pay attention to specifics.) Fix this up by
including nm-core-internal.h from those files that are now missing
includes.
Most NMDevice types defined their own error domain but then never used
it. A few did use their errors, but some of those errors are redundant
with NMDeviceError, and others can be added to it.
Make all mac-address properties (including NMSettingBluetooth:bdaddr,
NMSettingOlpcMesh:dhcp-anycast-addr, and NMSettingWireless:bssid) be
strings, using _nm_setting_class_transform_property() to handle
translating to/from binary form when dealing with D-Bus.
Update everything accordingly for the change, and also add a test for
transformed setting properties to test-general.
NMDeviceBond, NMDeviceBridge, and NMDeviceTeam all used basically the
same code to generate a default interface name. Move it into
nm_utils_complete_generic().
The virtual :interface-name properties (eg,
NMDeviceBond:interface-name) are deprecated in favor of
NMSettingConnection:interface-name, and nm_connection_verify() ensures
that their values are kept in sync. So (a) there is no need to set
those properties when we can just set
NMSettingConnection:interface-name instead, and (b) we can replace any
calls to the setting-specific get_interface_name() methods with
nm_connection_get_interface_name() or
nm_setting_connection_get_interface_name().
Since we enforce the fact that bond, bridge, team, and vlan
interface-name properties match NMSettingConnection:interface-name,
nm_connection_get_virtual_iface_name() can be replaced with
nm_connection_get_interface_name() basically everywhere.
The one place this doesn't work is with InfiniBand partitions (where
get_virtual_iface_name() was actually computing the name), but for the
most part we only need to care about the interface names of InfiniBand
partitions in places where we also already need to do some other
InfiniBand-specific handling as well, so we can use an
InfiniBand-specific method
(nm_setting_infiniband_get_virtual_interface_name()) to get it.
(Also, while updating nm_device_get_virtual_device_description(), fix
it to handle InfiniBand partitions too.)
For NMDeviceWifi and NMDeviceWimax, the printf format string for
nm_utils_complete_generic() was created based on ssid/nsp. Since
these input strings are untrusted, this is a serious bug.
Signed-off-by: Thomas Haller <thaller@redhat.com>
Now that we have nm_utils_hwaddr_matches() for comparing addresses
(even when one is a string and the other binary), there are now places
where it's more convenient to store hardware addresses as strings
rather than binary, since we want them in string form for most
non-comparison purposes. So update for that.
In particular, this also changes nm_device_get_hw_address() to return
a string.
Also, simplify the update_permanent_hw_address() implementations by
assuming that they will only be called once. (Since they will.)
Add nm_utils_hwaddr_matches(), for comparing hardware addresses for
equality, allowing either binary or ASCII hardware addresses to be
passed, and handling the special rules for InfiniBand hardware
addresses automatically. Update code to use it.
Include <linux/if_ether.h> and <linux/if_infiniband.h> from
nm-utils.h, to get ETH_ALEN and INFINIBAND_ALEN, and remove those
includes (as well as <net/ethernet.h> and <netinet/ether.h>, and
various headers that had been included to get the ARPHRD_* constants)
from other files where they're not needed now.
Drop the arptype-based nm_utils_hwaddr funcs, and rename the
length-based ones to no longer have _len in their names. This also
switches nm_utils_hwaddr_atoba() to using a length rather than an
arptype, and adds a length argument to nm_utils_hwaddr_valid() (making
nm_utils_hwaddr_valid() now a replacement for nm_utils_hwaddr_aton()
in some places, where we were only using aton() to do validity
checking).
Clean up some of the cross-includes between headers (which made it so
that, eg, if you included NetworkManagerUtils.h in a test program, you
would need to build the test with -I$(top_srcdir)/src/platform, and if
you included nm-device.h you'd need $(POLKIT_CFLAGS)) by moving all
GObject struct definitions for src/ and src/settings/ into nm-types.h
(which already existed to solve the NMDevice/NMActRequest circular
references).
Update various .c files to explicitly include the headers they used to
get implicitly, and remove some now-unnecessary -I options from
Makefiles.
Remove all remaining GParamSpec name and blurb strings (and fix
indentation while we're there), and add G_PARAM_STATIC_STRINGS to all
paramspecs that were lacking it.
In reality the connection provider (NMSettings) is always the same
object, and some device plugins need access to it. Instead of
cluttering up the device plugin API by passing the provider into
every plugin regardless of whether the plugin needs it, create
a getter function.
These are (most likely) only warnings and not severe bugs.
Some of these changes are mostly made to get a clean run of
Coverity without any warnings.
Error found by running Coverity scan
https://bugzilla.redhat.com/show_bug.cgi?id=1025894
Co-Authored-By: Jiří Klimeš <jklimes@redhat.com>
Signed-off-by: Thomas Haller <thaller@redhat.com>
Currently, ethernet-based VLANs can specify the hardware address of
the parent device (and, in theory, the cloned hardware address and MTU
of the VLAN device) by using an NMSettingWired in addition to the
NMSettingVlan.
The theory was that non-ethernet-based VLANs, when we eventually
supported them, would likewise use the setting type corresponding to
their parent device. However, this turns out to be both complicated
(the settings plugins and connection editor would have a
hard-to-impossible time figuring out which setting type to use in some
cases) and incorrect (for most L2 settings [eg, BSSID, bond mode,
etc], the VLAN can't have its own values separate from the parent
device).
What we should have done was just have :mac-address,
:cloned-mac-address, and :mtu properties on NMSettingVlan. However, at
this point, for backward-compatibility, we will just stick with using
a combination of NMSettingVlan and NMSettingWired, but we will use
NMSettingWired regardless of the underlying hardware type.
Rather than having NMManager know how to parse various settings to
create each kind of software device, add a _new_for_connection()
constructor to each of them and let them call NMPlatform to create the
device correctly themselves.
Rather than setting the VLAN maps when the device is created, set them
at activation time, which is more in line with how other device types
work.
Like the old code, this doesn't attempt to reset any existing
ingress/egress mappings on the device.
Software devices don't have a UDI until udev finds them, and since we need
to know about the software devices before udev finds them the UDI will be
missing. Instead of requiring a UDI on NMDevice creation, update the
property from the NMPlatform link change signal when udev does find the
device.
Now that a UDI is no longer required for device creation, software devices
added by NM would be created in the platform_link_added_cb() signal
handler triggered by the various software device creation methods in
system_create_virtual_device() (eg nm_platform_bridge_add() etc). Then
the NMDevice created in system_create_virtual_device() would be a duplicate
and cause problems when it was added. Since system_create_virtual_device()
needs to do setup on some devices, suppress the device creation from the
platform link added handler in this function.
Much of this is a hack which should be cleaned up later.
This is really, really old 2007-era code. Any NMDevice that gets
created is already supported, so there's no reason to have every
device set NM_DEVICE_CAP_NM_SUPPORTED. For those subclasses that
only set that capability, we can remove the subclass method
entirely. Next, it turns out that the "type capabilities" code
wasn't used anywhere, so remove that too. Lastly, "cipsec"
interfaces haven't been used on linux in about 5 years (they
were created by the Cisco binary-only IPSec kernel module for
Cisco VPNs long before vpnc and openswan came around) so we can
remove that code too.
Change the way that nm-properties-changed-signal works, and parse the
dbus-binding-tool-generated info to get the exact list of properties
that it's expected to export.
This makes NM_PROPERTY_PARAM_NO_EXPORT unnecessary, and also fixes the
problem of properties like NMDevice:hw-address being exported on
classes where it shouldn't be.
