Instead of looking up the signal-id every time, cache it.
g_signal_lookup() requires a g_quark_try_string() and a
lock a lock on a global mutex.
Downside is that the InterfaceData structure grows.
NMExportedObject is the center of every D-Bus exported object in
NetworkManager's core. It makes sense to optimize it.
Transform the GSList of interfaces to be a array. The array is still
allocated via the slice allocator (as we expect that there are only few
types in the list). This saves the overhead to allocate a GSList item
for each entry.
Another advantage is that the interfaces list is now strongly typed
instead of an opaque data pointer.
This makes sure that devices like bond get their dhcp renewed
[thaller@redhat.com: original patch modified to rename
now-public function update_dynamic_ip_setup()]
https://bugzilla.gnome.org/show_bug.cgi?id=764398
If replace_and_commit() found existing route files (and the callback
has potentially already been invoked), it is wrong to chain up to
parent class and continue the update.
Fixes: f79d62692e
The link_stats structure grew between 4.5 and 4.6 and this would cause
the messages to me ignored when compiling with 4.6 headers and running
on 4.5.
https://bugzilla.gnome.org/show_bug.cgi?id=764332
Make sure there's always the device and connection as well as the reason
when a slave activation fails. The slave connection could in fact be
chosen automatically on "nmcli d connect" and the user might not be
aware activation of which connection was attempted:
$ nmcli d connect enp0s25
Error: Device activation failed: Master connection not found or invalid
GVariantBuilder doesn't care if you give it multiple dict entries with
the same key. But that causes duplicate dict entries in the legacy
D-Bus PropertiesChanged signals that NM emits. So instead go back to
a hash table to ensure that any previous value is dropped in favor of
the new one.
When porting to GDBus property change notifications were converted from a
hash table to a GVariantBuilder. GVariantBuilder doesn't care about
duplicated properties in the dict so each g_object_notify() will add
an additional item with possibly different values:
signal time=1458571005.592811 sender=:1.10 -> destination=(null destination) serial=64451 path=/org/freedesktop/NetworkManager; interface=org.freedesktop.NetworkManager; member=PropertiesChanged
array [
dict entry(
string "ActiveConnections"
variant array [
object path "/org/freedesktop/NetworkManager/ActiveConnection/19"
object path "/org/freedesktop/NetworkManager/ActiveConnection/18"
object path "/org/freedesktop/NetworkManager/ActiveConnection/15"
object path "/org/freedesktop/NetworkManager/ActiveConnection/0"
]
)
dict entry(
string "ActiveConnections"
variant array [
object path "/org/freedesktop/NetworkManager/ActiveConnection/24"
object path "/org/freedesktop/NetworkManager/ActiveConnection/19"
object path "/org/freedesktop/NetworkManager/ActiveConnection/18"
object path "/org/freedesktop/NetworkManager/ActiveConnection/15"
object path "/org/freedesktop/NetworkManager/ActiveConnection/0"
]
)
]
Fix that by not emitting notify events for the manager's ActiveConnections
property until the property has actually been updated in active_connection_add().
The unexport also isn't required for VPN connections since it will get
unexported when it's disposed after _internal_activation_failed() gets called.
When a value of a TYPE_BOTH option is read back from kernel it
contains both string and numeric values ("balance-rr 0"), so we must
chop off the number before adding the option to the setting. Also
change the default values of options to the string form so that the
option matching logic works.
This can regularly happen when a virtual device depends on a parent/master
that is not yet created. We will retry later when the parent is ready, so
logging a warning about it is wrong and confusing.
If we want to ensure that we create only one single instance of
NMPolicy, just don't create multiple instances. The nm_policy_new()
method should not be restriced and behave like other *new() functions
and create a new object as requested.
There is no excuse for clients to send connections to NetworkManager
that have invalid/unknown fields. Just reject them.
This is a dangerous change, because we might now reject connections
that we were accepting previously. Who know what clients were sending
and it used to work.
AddAndActivateConnection is allowed to provide an incomplete connection
that will be completed by NetworkManager. That is, a connection that
does not verify.
But we still want to catch invalid properties or unknown setting types.
Thus, we want to reject invalid partial connections.
This possibly rejects invalid requests from clients that were accepted
before. Thus this change has the potential to break misbehaving clients.
There is no excuse for clients to send connections to NetworkManager
that have invalid/unknown fields. Just reject them.
As Reapply() is new API in nm-1-1, there is no problem with backward
compatibility.
NM_UTILS_ERROR is our way to say, that we don't care about
the GError domain and code. nmcli sometimes passes domain "1"
and code "0" to g_set_error(), which could be considered
a bug.
We usually don't care about the error but only about the error
message, so let's have a universally available error quark around.
Don't try to realize our device when the parent device is not real.
Instead, enqueue the activation and wait until it is active before
realizing our device and progressing the device to DISCONNECTED so that
it can get connected.
If it's traversing from unavailable to disconnected (e.g. realizing
of the device was delayed because it was awaiting the parent connection),
then we just want to progress the activation.
We'll need to share the best conneciton logic and it's the only caller
of nm_device_get_available_connections(). Let's just move it all to
NMDevice and provide the best connection from there instead.
The root-test "/general/netns/bind-to-path" mounts a tmpfs
over /var/run for the duration of the test and unmounts it at
the end.
Due to a bug, we first mount the tmpfs, then switch to another
namespace, and finally unmount /var/run on that other namespace.
That means, when you run the tests as root, it will bork /run
on your system.
Fixes: 46f5c07643
If the user specifies an invalid 'dhcp' option in configuration,
currently we disable DHCP. Instead, fall back to other available
clients, as we do for other options.
