Moving the PPP manager to a separate plugin that is loaded when needed
has the advantage of slightly reducing memory footprint and makes it
possible to install the PPP support only where needed.
https://bugzilla.gnome.org/show_bug.cgi?id=773482
nm_platform_ip4_address_sync() tries to apply the new configuration
with the minimum effort and doesn't delete addresses if they are
already present on the interface. This can break the ordering, as an
existing address would be promoted by kernel to primary, even if it
was last in our configuration.
Add some logic to ensure the correct order of addresses is always
enforced. This fixes situations like:
# nmcli connection add type ethernet ifname eth0 con-name t \
ipv4.method manual \
ipv4.addresses "1.1.1.1/24,1.1.1.2/24,1.1.1.5/24"
# nmcli connection up t
=> addresses are applied in the right order:
inet 1.1.1.1/24 brd 1.1.1.255 scope global eth0
inet 1.1.1.2/24 brd 1.1.1.255 scope global secondary eth0
inet 1.1.1.5/24 brd 1.1.1.255 scope global secondary eth0
# nmcli connection mod t ipv4.addresses "1.1.1.5/24,1.1.1.2/24,1.1.1.1/24"
# nmcli device reapply eth0
=> order is wrong:
inet 1.1.1.2/24 brd 1.1.1.255 scope global eth0
inet 1.1.1.5/24 brd 1.1.1.255 scope global secondary eth0
inet 1.1.1.1/24 brd 1.1.1.255 scope global secondary eth0
Co-Authored-By: Thomas Haller <thaller@redhat.com>
When multiple address are assigned to an interface and the kernel must
decide which one should be used to communicate with a given IP, it
chooses the most specific one in the same subnet as the
destination. In case there are multiple addresses in the same subnet,
the primary address is choosen, which is basically the first one that
was added.
With commit 7197425137 ("device: expose NMIP4Config:addresses in
stable/defined sort order") we sorted all the addresses before
committing the configuration, with the side effect that the order no
longer respected the one in the user configuration.
Instead, change the sort function to keep the subnet order unchanged.
Since commit 9fafb382db, we would
explicitly set libnl's socket buffer size to 4*getpagesize().
That is also the default of libnl itself. Additionally, we would
workaround too small buffers by increasing the buffer size up to 512K.
A too small buffer causes messages to be lost. Usually, that only
results in a cache-resync, which isn't too bad. Lost messages are however
a problem if the lost message was an ACK that we were waiting for.
However, it is rather unlikely to happen, because it's expected that
the buffer size gets adjusted already when the cache is filled initially,
before any other requests are pending.
Still, let's increase the default buffer size to 32K, hoping that this
initial value is already large enough to avoid the problem altogether.
Note that iproute2 also uses a buffer size of 32K [1] [2].
Alternatively, we could use MSG_PEEK like systemd does [3]. However,
that requires two syscalls per message.
[1] https://patchwork.ozlabs.org/patch/592178/
[2] https://git.kernel.org/cgit/linux/kernel/git/shemminger/iproute2.git/tree/lib/libnetlink.c?id=f5f760b81250630da23a4021c30e802695be79d2#n274
[3] cd66af2274/src/libsystemd/sd-netlink/netlink-socket.c (L323)
We don't want to enable MSG_PEEK due to the overhead. But when we detect
that we just lost a message due to MSG_TRUNC, increase the buffer size and
retry.
See-also: 55ea6e6b6c
A new DNS may have different data for us. So tell dnsmasq to
automatically clear the cache when servers change.
Signed-off-by: Christian Hesse <mail@eworm.de>
When going to sleep, we unmanage devices setting the unmanaged flags
immediately but delaying the state transition (because we do it from
another state transition). The signal handler can be executed after
the wake and, especially, after we have already re-managed the device,
making the device unmanaged again.
Detect such situation and force the state to UNMANAGED (which will
also clear any pending state change), so that later we manage the
device again and it will try to activate any available connection.
Fixes: 81ea812362https://bugzilla.redhat.com/show_bug.cgi?id=1382526
Also, ifnet plugin would read the configuration value, which is just wrong
because:
- the configuration might not be set and ifnet would fail to fallback
to the compile time default.
- the configuration only is in effect if the plugin is also available.
Otherwise, we fallback to the next plugin.
Only the dhcp-manager knows which DHCP plugin is in use.
Since commit fb2ca0ce3d we would no longer pre-set the main.plugins
value in NMConfig's keyfile to recognize unset default settings.
This breaks with
[main]
plugins+=foo
which now results in
main.plgin=foo
while previously it would have extended the compile time default.
https://bugzilla.redhat.com/show_bug.cgi?id=1397938
Fixes: fb2ca0ce3d
Instead of having the caller do the fallback to the compile time default
plugins, let it be handled by nm_config_get_plugins().
The knowledge of fallback to a compile time default (and how to do that
properly) should be inside NMConfig/NMConfigData alone.
Also, as this function is only called once, let NMConfig not cache
the string list but create it once as needed.
This makes it easier to install the files with proper names.
Also, it makes the makefile rules slightly simpler.
Lastly, the documentation is now generated into docs/api, which makes it
possible to get rid of the awkward relative file names in docbook.
NOTE: changed the default value for auto-negotiate from TRUE to FALSE.
Normalization enforces that no values for speed and duplex are there
when autonegotiation is on. This is required as autoneg on with specific
speed and duplex set means to ethtool to use autonegotiation but
advertise that specific speed and duplex only.
autoneg off, speed 0 and duplex NULL means to ignore link negotiation.
Added platform functions to retrieve device link mode status and to
switch from auto to manual link negotiation:
nm_platform_ethtool_get_link_settings
nm_platform_ethtool_set_link_settings
Keep the include paths clean and separate. We use directories to group source
files together. That makes sense (I guess), but then we should use this
grouping also when including files. Thus require to #include files with their
path relative to "src/".
Also, we build various artifacts from the "src/" tree. Instead of having
individual CFLAGS for each artifact in Makefile.am, the CFLAGS should be
unified. Previously, the CFLAGS for each artifact differ and are inconsistent
in which paths they add to the search path. Fix the inconsistency by just
don't add the paths at all.
Plugins could be already failed or disconnected when the helper fires.
E.g. they could send in an invalid IP4Config that would cause them to
fail and then follow with an IP6Config before they realize it's of no
use. We'd hit an assertion failure in that case, because the
NMVpnConnection would already be cleaned up.
With systemd < 219, restarting the journald service closes the stdout
and stderr streams associated with services.
The NM process has SIGPIPE ignored, but g_spawn_sync()/g_spawn_async()
re-enable it and so any child executed with those functions will
terminate by default if it tries to log anything to stdout/stderr.
The teamd instance launched by NM is affected by this problem since it
writes debug messages before actually ignoring SIGPIPE.
To fix this, use the @child_setup callback of g_spawn() to ignore
again SIGPIPE in the child process.
https://bugzilla.redhat.com/show_bug.cgi?id=1393853
Otherwise its path remains visible on D-Bus despite the object is gone,
making libnm sad and grumpy:
libnm-WARNING **: no object known for /org/freedesktop/NetworkManager/AccessPoint/666
On some architectures, it seems we don't properly expose
the symbol of these static variables from NetworkManager
binary.
Just avoid that and don't instead use a static array
inside the device plugin itself.
While at it, make the arrays all const, which possibly allows
the linker to put those symbols in the read-only section.
