Change behavior for the network-address and broadcast-address.
Users should not specify such addresses, but if they do, generate
something more sensible.
Also, if the address was in network larger then /24, the
generated address range was rather unexpected. Change behavior
here.
There are no particularly strong reasons for the chosen range.
It just seems suitable. The decision to hand out at most a /24
is because it is likely to be plenty, and because that is what
the previous code did -- at least, if the address was in the
first /24 of the subnet. See how the result for 192.168.0.1/20
is unchanged, but 192.168.1.1/20 changes.
Changes:
- merge reserve_shared_ip() into shared4_new_config().
shared4_new_config() needs to register release_shared_ip(). However, it
wrongly would always register release_shared_ip(), even for user-supplied
addresses. To fix that, we would need yet another argument to
reserve_shared_ip() and coupling it even more with shared4_new_config().
At that point, it's cleaner to just merge the two functions.
- only create the shared_ips hash when needed, and delete it when
it's empty. The idea is, that NetworkManager possibly runs for a long
time, and most of the time no shared connection is active. Just clean
up the empty hash while we don't need it.
This is useful because bond gets its mac from first slave and order of
slave attachement is not guaranteed.
For example if we have bond with wifi and ethernet we want mac from
wifi to be use by bond interface because many wifi defices do now
allow one to send data with spoofed macs.
Signed-off-by: Nikolay Martynov <mar.kolya@gmail.com>
Ethernet settings for VLAN are always created later in settings
normalization. We might as well always created them.
This fixes a user visible problem: currently if user specifies cloned
mac when connection is created this value doesn't get saved. User has
to go, edit saved connection to add it again.
Similar problem existed for wireless security and wifi
Signed-off-by: Nikolay Martynov <mar.kolya@gmail.com>
nmtstp_env1_add_test_func() allows to register test functions in a
particular test environment ("env1", for lack of a better name).
It will be reused for "test-route.c"
We use init_if() as handler for g_list_foreach(). Since we already
cast the function pointer because it doesn't have (GFunc) signature,
we can just as well use the correct types right away.
Otherwise a device which was set as unmanaged (updated to the REMOVED
internal sys-state) will never update its own sys-state if later set
back as managed.
Manage either when setting explictly the device to managed either when
just upping a connection on an unmanaged device.
In the previous code it is not clear to me that there won't be
a situation where we don't react on a state change, waiting for a
device-state-change that never comes.
Now, only wait for a better device-state reason if:
- we have a device
- and the ac-reason is unspecific (NM_ACTIVE_CONNECTION_STATE_REASON_DEVICE_DISCONNECTED)
- and the device still references the current active connection
- and the device state is not yet FAILED or DISCONNECTED.
The most important change is that we only wait longer, if the device's
active-connection is still the same as our current active connection.
I don't think this commit is really necessary, but I cannot understand
the previous logic.
See-also: 57a26fd2aa
Commits 39d0559d9a ("platform: sort links by name instead of
ifindex") and 529a0a1a7f ("manager: sort slaves to be autoconnected
by device name") changed the order of activation of slaves. Introduce
a system-wide configuration property to preserve the old behavior.
https://bugzilla.redhat.com/show_bug.cgi?id=1452585
With address sanitizer, the call to setrlimit() fails by default,
because the core dump would be huge. That could be overwritten via
ASAN_OPTIONS=disable_core=0
But just don't try to enable core-dumps with asan.
On cleanup, unconditionally release a device from its master if the
link is missing or it doesn't have a master, otherwise the master
would later try to release the slave, hitting the following assertion:
"nm_platform_link_release: assertion 'slave > 0' failed"
#0 g_logv
#1 g_log
#2 g_return_if_fail_warning
#3 nm_platform_link_release
#4 release_slave
#5 nm_device_master_release_one_slave
#6 slave_state_changed
#7 ffi_call_unix64
#8 ffi_call
#9 g_cclosure_marshal_generic
#10 g_closure_invoke
#11 signal_emit_unlocked_R
#12 g_signal_emit_valist
#14 _set_state_full
#15 nm_device_state_changed
#16 nm_device_unrealize
#17 _platform_link_cb_idle
#18 g_main_context_dispatch
#19 g_main_context_dispatch
#20 g_main_context_iterate
#21 g_main_loop_run
#22 main
Fixes: 9e8218f99ahttps://bugzilla.redhat.com/show_bug.cgi?id=1448907
Since commit 1afbf948a0,
"build: use different defaults for snapshot builds",
configure would enable debugging options if the version
number is odd.
Hence, on the master branch it was no longer possible to
build an RPM without debugging enabled. Especially,
./contrib/fedora/rpm/build_clean.sh -g -W debug
would not work as one would expect.
This way, we get tab completion for the enum values, and
can reuse existing code.
This requires a pre-set-notify hook, that is invoked before
setting the property.
