We no longer need a class method for reading the hardware address
length, for a couple reasons:
1) Using the IP interface hardware address for IP operations now makes
NMDevice's priv->hw_addr_len constant. So there's no reason to re-read
it all the time.
2) get_hw_address_length() is now only used for determining whether the
hardware address is permanent, and that only mattered for Bluetooth.
Since Bluetooth interfaces have a bogus interface name, they will never
have a valid ifindex, and thus nm_platform_link_get_address() would be
useless. So instead of using the 'permanent' stuff, just don't bother
updating the hardware address if the NMDevice's ifindex isn't valid,
and let subclasses pass the initial hardware address at device creation.
This also works correctly for NMDevice classes that previously
implemented get_hw_address_length() like ADSL and WWAN, since those
too will never have a valid ifindex or a valid hardware address.
3) Reading the device's hardware address length just ended up calling
nm_platform_link_get_address() for most devices anyway, so
nm_device_update_hw_address() would effectively read the link address
twice (once to read the length, the second time to read the actual
address). Let's just read the address once.
The IP interface may have its own hardware address (like the net
port for WWAN devices) and that's the hardware address that must be
used for DHCP and IPv6 SLAAC, not the hardware address (if any) of
the NMDevice itself.
This patch does change the NMDevice hardware address property to
always be the Device's hardware address, instead of the IP interface
hardware address. This means that ADSL and WWAN will no longer
change their hardware address to the hardware address of their
IP interface. But in all these cases, the hardware address is
non-existent (PPP) or transient and meaningless (WWAN/ADSL).
NMConfigDevice was added because in the 0.9.8 days, when each subdir
of src/ was compiled separately, it was impossible to make src/config/
depend on src/devices/ because of circular dependencies.
Since now everything gets compiled into a single libNetworkManager.la,
this is no longer a problem, and so NMConfigDevice is just an
unnecessary complication.
NMConfig accepted no-auto-default and ignore-carrier lists with
untagged specs (ie, interface names not prefixed with
"interface-name:" and hardware addresses not prefixed with "mac:").
Move that handling into nm_match_spec_interface_name() and
nm_match_spec_hwaddr() instead.
We were linking libndp into the NetworkManager binary, but it ought to
be marked as a dependency of libNetworkManager, in case a test
exercises that code.
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 $(POLKIT_CFLAGS)) by moving all
GObject struct definitions for src/ and src/settings/ into nm-types.h
(which already existed to solve the NMDevice/NMActRequest circular
references).
Update various .c files to explicitly include the headers they used to
get implicitly, and remove some now-unnecessary -I options from
Makefiles.
Several plugins were using -I$(top_srcdir)/libnm-glib, which is bad
since libnm-glib has its own nm-types.h which is different from src's.
Worse yet, some were actually linking against libnm-glib (which
presumably only worked at all because they weren't calling any
functions in it and so the linker just ignored the request). Fix both
problems.
5b4be27f moved test-dhcp-options from src/tests/ to
src/dhcp-manager/tests/, but src/tests/Makefile.am still included it
in "TESTS" (which would still work until you tried to build from a
clean tree...)
The goal is to remove generic logic from nm_dhcp_client_new_options(),
since library-ized DHCP clients won't ever call that function since
their native data format isn't a hash table of strings.
The options hash is never used except for BOUND events, so don't
bother caching it in the DHCP client object. Just pass it along
with the BOUND state change, like the IP configuration object.
Previously the tests required that a NMDHCPClient object was
created, but all that's really being tested is the DHCP option
parsing and that doesn't need a client object now that the
option parsing has been split out.
Also, the options test wasn't in src/dhcp-manager/, so put it there.
Lastly, the test ran for both dhclient and dhcpcd, but the code
executed for both cases was the same, so there's no reason to keep
running the test for both clients.
More robust against ifname changes, plus we save some strcmps().
For the most part, the interface name should be informational
only and unused for actual control logic.
Instead of using a separate signal for it, just drop and release
the client object when it reaches terminate states like TIMEOUT,
FAIL, and DONE. The 5 second removal signal timeout appears to
no longer be necessary, since the running child process is killed
synchronously.
Modify code so that listeners remove state change signal handlers
before stopping the client. Which means we can remove the 'emit_state'
argument from nm_dhcp_client_set_state().
DHCP failure should just clean up the client in all cases. This
also has the benefit of removing the signal handler for the DHCP
client's state-change signal before telling the client to terminate,
which will simplify some DHCP code later.
No reason to have two signals for the same thing. Previously, the
TIMEOUT signal was used for the internal overall DHCP transaction
bound, while DHCP_STATE_TIMEOUT/DHC_TIMEOUT was a signal from
the DHCP client itself that something had timed out. But in both
cases the results should be the same, so just collapse the
stand-alone TIMEOUT signal into the DHCP_STATE_TIMEOUT state.
The existing DHC_* states are pretty specific to dhclient, and aren't
useful for more generalized DHCP. NetworkManager wasn't using many
of the states anyway, and doesn't need to differentiate between
states like REBOOT/REBIND/RENEW anyway. So simplify the DHCP states
into the ones we really care about.
'address' was not initialized, so using address.ifindex was not
going to work.
==6777== Conditional jump or move depends on uninitialised value(s)
==6777== at 0x44A6DB: check_cache_items (nm-linux-platform.c:1510)
==6777== by 0x44AA4F: announce_object (nm-linux-platform.c:1586)
==6777== by 0x44A59A: refresh_object (nm-linux-platform.c:1666)
==6777== by 0x44A676: check_cache_items (nm-linux-platform.c:1517)
==6777== by 0x44A812: announce_object (nm-linux-platform.c:1566)
==6777== by 0x44D957: handle_udev_event (nm-linux-platform.c:3964)
==6777== by 0x3BAFC125A7: g_closure_invoke (gclosure.c:777)
==6777== by 0x3BAFC2465C: signal_emit_unlocked_R (gsignal.c:3586)
==6777== by 0x3BAFC2C3E1: g_signal_emit_valist (gsignal.c:3330)
==6777== by 0x3BAFC2C69E: g_signal_emit (gsignal.c:3386)
==6777== by 0x3D93A03BE3: ??? (in /usr/lib64/libgudev-1.0.so.0.1.3)
==6777== by 0x3BAF8521D5: g_main_context_dispatch (gmain.c:3066)
Always requesting this appears to cause some devices to ignore
DHCP requests, like the ZTE 823D, ZTEMF93E, Alcatel W800Z.
dhclient probably knows better than NM when to request the
server's ID, so leave that up to dhclient. We don't do anything
interesting with it anyway.
nm-version.h was getting disted, making srcdir!=builddir work for
tarball builds, but not for git builds.
Also, remove "-I${top_builddir}/include" from all Makefile.ams, since
there's nothing generated in include/ any more.
NetworkManager.h, NetworkManagerVPN.h, and nm-version.h are part of
the libnm-util API, so move them to libnm-util.
include/ still contains headers that are strictly NM-internal (eg,
nm-glib-compat.h).
Even if administrator-configured hostname (/etc/hostname) takes precedence
over other hostname configurations, we don't take "localhost", "localhost6",
"localhost.localdomain", "localhost6.localdomain6" as such. These values might
be set by some tools (like installer). But that's not right and we compensate
for that. It doesn't make much sense that an admimistrator would set these
values manually (intentionally), because leaving /etc/hostname empty will
result in "localhost" hostname anyway (set by systemd).
https://bugzilla.redhat.com/show_bug.cgi?id=1110436
The kernel adds these for various operations; they are short-lived,
added often, and not useful to NetworkManager. Ignore them. This
prevents NetworkManager from continuously updating the IPv6 config
and emitting state changes to clients via D-Bus for useless changes.
Quite possibly these logging lines hit before we setup nm-logging
initially and before NM main() calls nm_logging_syslog_openlog().
This is because NM first needs to read the configuration, before setting
up logging.
Note however, that nm-logging is robust against both.
If nm_logging_setup() was not yet called, it will automatically setup
"DEFAULT","INFO".
If nm_logging_syslog_openlog() was not yet called, it will redirect
logging to g_log() -- which basically ends up at g_warning() again.
Still this is useful in case when hitting those lines *afterwards*.
Especially, later when we support reloading of configuration.
Signed-off-by: Thomas Haller <thaller@redhat.com>
