pppd restores the previous settings for the serial port it uses right
before exiting. It is especially important to do so because otherwise
ModemManager is not able to recover the port as it can receive a hangup
event from the port due to CLOCAL not being restored. However, there is
currently a race condition that produces this issue. This is because
when PHASE_DEAD is notified, pppd still has not restored the port
settings - it does that a bit later, in the die() function.
This patch delays notifying PHASE_DEAD until when the exitnotify() hook
is called by pppd: when this happens the port settings have already been
restored.
There were previously efforts to fix this in commit fe090c34b7, so
PHASE_DEAD was used instead of PHASE_DISCONNECT to notify MM that the
port was disconnected, but that still early to ensure that the port
settings are restored.
The MM traces seen when the bug is triggered are:
ModemManager[2158]: <warn> (ttyACM1): could not re-acquire serial port lock: (5) Input/output error
ModemManager[2158]: <warn> Couldn't load Operator Code: 'Cannot run sequence: 'Could not open serial device ttyACM1: it has been forced close'
https://mail.gnome.org/archives/networkmanager-list/2019-June/msg00014.html
This doesn't make any difference in practice, but it seems more correct.
It would cause issues if we decided to remove an interface from the
signal handler.
When an interface (other OVS device types can not fail) encounters an error
it indicates it by changing the error column. Watch for those changes so
that we can eventually communicate them to the OVS factory to deal with
them.
Don't crash in situations, where the bridge or a port has a child with
UUID we don't know. This could happen if we mess up the parsing of
messages from OVSDB, but could also theoretically happen in OVSDB sends
us bad data.
Note that now the empty list will be represented as %NULL instead of an
empty strv array.
That makes no difference in pratice. The main use of this property is as
glue for NMDBusManager to expose the property on D-Bus. Thereby it uses
g_dbus_gvalue_to_gvariant() which handles %NULL just fine.
NMManager and NMSettings both may have multiple authorization requests
ongoing. They need to keep track of them, at the very least to be able
to cancel them on shutdown.
Since NMAuthChain is not ref-countable, it always has only one clear
user/owner. It makes little sense otherwise. Since most callers already
want to track their NMAuthChain instances, let NMAuthChain help with that.
Embed a "parent" CList field inside NMAuthChain. This avoids requiring
an additional GSList allocation to track the element. Also, it allows to
link and append an element without iterating the list.
This ties the caller and the NMAuthChain a bit tighter together (making them
less indepdendent). Generally that is not desirable. But here it seems the
logic (of tracking the NMAuthChain) is still trivial and well separated.
It's just that NMAuthChain instances now can be linked in a CList.
VPN settings (for openconnect) can only be handled by the keyfile settings
plugin.
In any case, such special casing belongs to the settings plugin and not
"nm-settings.c". The reason is that the settings plugin already has an
intimate understanding of the content of connections, it knows which fields
exist, their meaning, etc. It makes sense special handling of
openconnect is done there.
See also commit 304d0b869b ('core: openconnect migration hack').
Unfortunately it's not clear to me why/whether this is still the
right thing to do.
nm_device_check_connection_compatible() is potentially expensive.
Check first whether the connection candidate is of a relevant type,
hoping that this check is cheaper and thus shortcuts other checks
early.
NMSettings is complicated enough. We should try to move independent code out
of it, so that there is only logic that is essential there.
While at it, rework how we copy the GSList items. I don't like GSList as
a data structure, but there really is no need to allocate a new list.
Just unlink the list element and prepend it in the other list.
As nm_settings_plugin_initialize() could not fail (it returned no value indicating
failure), there is no reason to explicitly call this. Instead just
initialize the plugin when needed.
Also, we don't need the plugin to initialize early before nm_settings_plugin_get_connections().
Instead of
<info> [1558284380.2045] settings: Loaded settings plugin: SettingsPluginIfcfg ("/usr/lib64/NetworkManager/1.19.2/libnm-settings-plugin-ifcfg-rh.so")
log
<info> [1558284380.2045] settings: Loaded settings plugin: ifcfg-rh ("/usr/lib64/NetworkManager/1.19.2/libnm-settings-plugin-ifcfg-rh.so")
Note how `man NetworkManager.conf` documents "main.plugins" configuration
option where settings plugins have names like "keyfile" and "ifcfg-rh".
It's not helpful to log the GObject type name, which is an implementation
detail.
It was only kept to compare whether we loaded the same
plugin multiple times.
Note that load_plugins() already checks for duplicate plugin names,
so it actually could not happen that we tried to load the same file
more than once.
The values cached in the device may be stale when we start a new
activation because in a disconnected state we might have called
ip_config_merge_and_apply() which cached the main table value.
We no longer add these. If you use Emacs, configure it yourself.
Also, due to our "smart-tab" usage the editor anyway does a subpar
job handling our tabs. However, on the upside every user can choose
whatever tab-width he/she prefers. If "smart-tabs" are used properly
(like we do), every tab-width will work.
No manual changes, just ran commands:
F=($(git grep -l -e '-\*-'))
sed '1 { /\/\* *-\*- *[mM]ode.*\*\/$/d }' -i "${F[@]}"
sed '1,4 { /^\(#\|--\|dnl\) *-\*- [mM]ode/d }' -i "${F[@]}"
Check remaining lines with:
git grep -e '-\*-'
The ultimate purpose of this is to cleanup our files and eventually use
SPDX license identifiers. For that, first get rid of the boilerplate lines.
While at it, rename the "addr" field to "l_address". The term "addr" is
used over and over. Instead we should use distinct names that make it
easier to navigate the code.
nmtst_get_rand_int() was originally named that way, because it
calls g_rand_int(). But I think if a function returns an uint32, it
should also be named that way.
Rename.
Maybe DHCP plugins should be configurable per address family and be
re-loadable via SIGHUP. But that just adds complexity.
Nowadays we always have the "internal" DHCP plugin, which is known to
support both IPv4 and IPv6. One day, we should get rid of all plugins
and only use one implementation (that works well). The "internal" plugin
is supposed to be(come) that.
That also means, that we are not going to add more (external) DHCP
plugins and we are not going to invest work in the existing plugins
(except the "internal" plugin).
Some DHCP plugins are known to not support IPv6. If the user selects
"dhcpcd" we should just fallback to the "internal" plugin. What's the
point of letting the activation fail? Probably users shouldn't use
"dhcpcd" plugin anyway, but that's a different story. Doing such fallback
could be a problem with forward compatibility if we ever would add IPv6
support to "dhcpcd". But we won't.
Also, we are going to add "n-dhcp4" as replacement for the systemd based
code. For a time, there will be an experimental plugin "nettools" that
eventually will become the new "internal" plugin. Until that happens,
we want for IPv6 automatically fallback to systemd based "internal"
plugin. This patch will make that simple.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/merge_requests/173
NM_MODEM_OPERATOR_CODE property is construct-only. Add our common
code comment to the property setter.
Construct-only setters are pretty simple. They run before the object is
constructed, hence their scope is clearer. As such, there is no need to
emit property changed notifications (also because that is already taken
care by the GObject property setter). Don't call _nm_modem_set_operator_code(),
just directly set the property.
Usually we aim to have only one place where we set state (_nm_modem_set_operator_code()).
But a construct-only property setter is trivial enough that we can affort having two
places to modify the property. In particular, because the property setter does not "modify"
the property, it merely initializes it before the object is fully
constructed.
The early boot tooling gets the root-path from our state file due to a
lack of a better way to do that. However, when booting with NFS root,
the root path alone is not sufficient; the server address is communicated
via the next-server option. Save that one in the state file as well.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/merge_requests/168
Clang (3.4.2-9.el7) on CentOS 7.6 fails related to nm_hash_update_vals().
Clang seems to dislike passing certain complex arguments to typeof().
I'd prefer to fix nm_hash_update_vals() to not have this problem,
but I don't know how.
This works around the issue.
clang (3.4.2-9.el7) on CentOS 7.6 fails related to nm_hash_update_vals().
I am not even quoting the error message, it's totally non-understandable.
nm_hash_update_vals() uses typeof(), and in some obscure cases, clang dislikes
when the argument itself is some complex macro. I didn't fully understand why,
but this works around it.
I would prefer to fix nm_hash_update_vals() to not have this limitation.
But I don't know how.
There is probably no downside to have this an inline function instead of
a macro.
This code is now unused.
Also, it does not seem state of the art to me
anymore.
Drop it, it could always be resurrected if need by, but maybe
GFileMonitor could be used instead.
It's deprecated and off by default for a long time.
It is bad to automatically reload connection profiles. For example, ifcfg
files may consist of multiple files, there is no guarantee that we
pick up the connection when it's fully written.
Just don't do this anymore.
Users should use D-Bus API or `nmcli connection reload` or `nmcli
connection load $FILENAME` to reload profiles from disk.
If the AddAndActivate() caller didn't explicitely a MAC address, default
to pinpointing the connection to the device by the means of an interface
name. This makes more sense than a MAC address with stable device names.
