We've previously been just watching for state changes into UNMANAGED state. No
state change is emitted upon removal of a device which is already unmanaged.
https://bugzilla.gnome.org/show_bug.cgi?id=737659
The previous commit made NM enforce the default route on interfaces for
which NM manages a default route.
For interfaces that are configured never-default, NM will now pick up
any externally configured default route, as if it was managed by NM.
This is important, because NMDefaultRouteManager needs a notion of which
is the best device. Without this change, it was agnostic to default routes
on managed, never-default interfaces.
Signed-off-by: Thomas Haller <thaller@redhat.com>
When quitting, the Manager asks each device to spawn the interface helper,
which persists and manages dynamic address on the interface after NetworkManager
is gone. If the dynamic address cannot be maintaned, the helper quits and
the interface's address may be removed when their lifetime runs out.
To keep the helper as simple as possible, NetworkManager passes most of the
configuration on the command-line, including some properties of the device's
current state, which are necessary for the helper to maintain DHCP leases
or IPv6 SLAAC addresses.
Up to now, NMPolicy would iterate over all devices to find the "best"
device and assign the default route to that device.
A better approach is to add a default route to *all* devices that
are never-default=no. The relative priority is choosen according to
the route metrics.
If two devices receive the same metric, we want to prefer the device
that activates first. That way, the default route sticks to the same
device until a better device activates or the device deactivates.
Hence, the order of activation is imporant in this case (as it is
already now).
Also, if several devices have identical metrics, increment their
metrics so that every metric is unique.
This makes the routing deterministic according to what we choose as best
device.
A special case is assumed devices. In this case we cannot adjust the metric
in face of equal metrics.
Add a new singleton class NMDefaultRouteManager that has a list of all
devices and their default routes. The manager will order the devices by
their priority and configure the routes using platform.
Also update the metric for VPN connections. Later we will track VPN
routes also via NMDefaultRouteManager. For now, fix the VPN metric because
otherwise VPNs would always get metric 1024 (which is usually much larger then the
device metrics).
https://bugzilla.gnome.org/show_bug.cgi?id=735512
Signed-off-by: Thomas Haller <thaller@redhat.com>
Before, we would always call unanimously nm_device_get_priority()
to get the default route metric for a device. Add new functions
nm_device_get_ip4_route_priority() and nm_device_get_ip6_route_priority()
and use them at the proper places.
Also add new function nm_vpn_connection_get_ip4_route_metric() and
nm_vpn_connection_get_ip6_route_metric().
Signed-off-by: Thomas Haller <thaller@redhat.com>
Merge libnm's NMDeviceError and the daemon's NMDeviceError into a
single enum (in nm-errors.h). Register the domain with D-Bus, and add
a test that the client side decodes it correctly.
The daemon's NM_DEVICE_ERROR_CONNECTION_INVALID gets absorbed into
libnm's NM_DEVICE_ERROR_INVALID_CONNECTION, and
NM_DEVICE_ERROR_UNSUPPORTED_DEVICE_TYPE gets dropped, since it was
only returned from one place, which is now using
NM_DEVICE_ERROR_FAILED, since (a) it ought to be a "can't happen", and
(b) the only caller of that function just logs error->message and then
frees the error without ever looking at the code.
nm_device_get_best_auto_connection() was only used at one place.
It was a very simple function, just iterated over a list finding
the first can_auto_connect() connection. At the very least, the name
was misleading, because it did not return the 'best', but the 'first'
connection.
Get rid of the function altogether.
Signed-off-by: Thomas Haller <thaller@redhat.com>
Instead of creating it in NMSettings, where we must use
NM_IS_DEVICE_ETHERNET() (not NM_DEVICE_TYPE_ETHERNET because various generic
devices masquerade as NM_DEVICE_TYPE_ETHERNET too), push knowledge
of which device types create default wired connections into the device
types themselves. This solves a problem with testcases where
libNetworkManager.a (which testcases link to) requires the symbol
nm_type_device_ethernet().
The IPv4 pre-commit hook was called right before the config was
committed, while the IPv6 one was called before commit in only
one case (from nm_device_activate_ip6_config_commit). The IPv4
behavior is the intended behavior.
Note that this doesn't have any actual effect yet, since nothing
actually implements the IPv6 pre-commit hook
Previously, src/nm-ip4-config.h, libnm/nm-ip4-config.h, and
libnm-glib/nm-ip4-config.h all used "NM_IP4_CONFIG_H" as an include
guard, which meant that nm-test-utils.h could not tell which of them
was being included (and so, eg, if you tried to include
nm-ip4-config.h in a libnm test, it would fail to compile because
nm-test-utils.h was referring to symbols in src/nm-ip4-config.h).
Fix this by changing the include guards in the non-API-stable parts of
the tree:
- libnm-glib/nm-ip4-config.h remains NM_IP4_CONFIG_H
- libnm/nm-ip4-config.h now uses __NM_IP4_CONFIG_H__
- src/nm-ip4-config.h now uses __NETWORKMANAGER_IP4_CONFIG_H__
And likewise for all other headers.
The two non-"nm"-prefixed headers, libnm/NetworkManager.h and
src/NetworkManagerUtils.h are now __NETWORKMANAGER_H__ and
__NETWORKMANAGER_UTILS_H__ respectively, which, while not entirely
consistent with the general scheme, do still mostly make sense in
isolation.
Now that we have nm_utils_hwaddr_matches() for comparing addresses
(even when one is a string and the other binary), there are now places
where it's more convenient to store hardware addresses as strings
rather than binary, since we want them in string form for most
non-comparison purposes. So update for that.
In particular, this also changes nm_device_get_hw_address() to return
a string.
Also, simplify the update_permanent_hw_address() implementations by
assuming that they will only be called once. (Since they will.)
"NetworkManager.h"'s name (and non-standard capitalization) suggest
that it's some sort of high-level super-important header, but it's
really just low-level D-Bus stuff. Rename it to "nm-dbus-interface.h"
and likewise "NetworkManagerVPN.h" to "nm-vpn-dbus-interface.h"
GLib/Gtk have mostly settled on the convention that two-letter
acronyms in type names remain all-caps (eg, "IO"), but longer acronyms
become initial-caps-only (eg, "Tcp").
NM was inconsistent, with most long acronyms using initial caps only
(Adsl, Cdma, Dcb, Gsm, Olpc, Vlan), but others using all caps (DHCP,
PPP, PPPOE, VPN). Fix libnm and src/ to use initial-caps only for all
three-or-more-letter-long acronyms (and update nmcli and nmtui for the
libnm changes).
Ethernet-like interfaces aren't the only type of interfaces that can
run IPv6 but the rdisc code only returns an address if the interface's
hardware address is 6 bytes.
Interface types like PPP (rfc5072) and IPoIB (rfc4391) have their own
specifications for constructing IPv6 addresses and we should honor
those.
So instead of expecting a MAC address, let each device subclass
generate an Interface Identifier and use that for rdisc instead.
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.
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.
The NMDevice dispose() function contained some badly-duplicated logic
about when to deactivate a device on its last ref. This logic should
only run when the device is removed by the manager, since the manager
controls the device's life-cycle, and the manager knows best when to
clean up the device. But since it was tied to the device's refcount,
it could have run later than the manager wanted, or not at all.
It gets better. Dispose duplicated logic that was already done in
nm_device_cleanup(), and then *called* nm_device_cleanup() if the
device was still activated and managed. But the manager already
unmanages the device when removing it, which triggers a call to
nm_device_cleanup(), takes the device down, and resets the IPv6
sysctl properties, which dispose() duplicated too. So by the time
dispose() runs, the device should already be unmanaged if the
manager wants to deconfigure it, and most of the dispose() code
should be a no-op.
Clean all that up and remove duplicated functions. Now, the flow
should be like this:
1) manager decides to remove the device and calls remove_device()
2) if the device should be deconfigured, the manager unmanages
the device
3) the NMDevice state change handler tears down the active connection
via nm_device_cleanup() and resets IPv6 sysctl properties
4) when the device's last reference is finally released, only internal
data members are freed in dispose() because the device should
already have been cleaned up by the manager and be unmanaged
5) if the device should be left running because it has an assumable
connection, then the device is not unmanaged, and no cleanup
happens in the state change handler or in dispose()
For any function in nm-device.h which is not used outside of
nm-device.c, remove the public prototypes. Functions that
are actually used get moved above their caller, and functions
that have no callers are removed.
If a link's "master" property changes unexpectedly (ie, from outside
NM), update the master and slave NMDevices to reflect it, without
making any changes to them.
If the initial attempt to assume a connection on a device fails, and
the device remains un-activated, but then something changes its
configuration externally, try to generate a new connection and assume
that.
Add a parameter to nm_device_add_pending_action() to silently
accept adding duplicate actions.
Same for nm_device_remove_pending_action(), to silently ignore
removing non-pending actions.
Signed-off-by: Thomas Haller <thaller@redhat.com>
We'll want to track internal management separately in the future, so split out
user management (eg, whether the device has been explicitly marked unmanaged
by the user).
Instead of tracking unmanaged-ness in a couple variables (and because
I'd like to add one for user-unmanaged later) let's do it in a single
flags variable, and consolidate setting of the unmanaged states in one
place.
Because not all clients set the 'hidden' property in a connection for
hidden/non-SSID-broadcasting networks, they may not show up in
the device's available-connections property. After the
PendingActivation object removal, all activations require the
connection to be in available-connections, and thus hidden SSID
networks could not be activated.
Unfortunately check_connection_available() is used both during
activation and to populate the available-connections array, but we
only want to special-case activation paths, and still ensure that
SSIDs not found in the scan list are not in available-connections.
To make it clear this is a WiFi only hack, and that we should
remove it at some point in the future, create another class method
specifically for hidden WiFi and use that in activation paths to
special-case hidden WiFi connection activation.
Add a generic signal that devices can use to indicate that something
material in the network situation changed, and that auto-activation
may now be possible. This reduces specific knowledge of device types
in the policy.
In preparation for making WWAN and Bluetooth plugins, rework
the device plugin interface to meet those plugins' needs and
port WiMAX over in the process.
In reality the connection provider (NMSettings) is always the same
object, and some device plugins need access to it. Instead of
cluttering up the device plugin API by passing the provider into
every plugin regardless of whether the plugin needs it, create
a getter function.
If a device is already activated, queue the new activation to allow
the transition through the DEACTIVATING state.
---
Also remove the "HACK" bits in nm_device_deactivate(). This hack was
added on 2007-09-25 in commit 9c2848d. At the time, with user settings
services, if a client created a connection and requested that NM
activate it, NM may not have read the connection from the client over
D-Bus yet. So NM created a "deferred" activation request which waited
until the connection was read from the client, and then began activation.
The Policy watched for device state changes and other events (like
it does now) and activated a new device if the old one was no longer
valid. It specifically checked for deferred activations and then
did nothing. However, when the client's connection was read, then
nm-device.c cleared the deferred activation bit, leading to a short
period of time where the device was in DISCONNECTED state but there
was no deferred activation, because the device only changes state to
PREPARE from the idle handler for stage1. If other events happened
during this time, the policy would tear down the device that was
about to be activated. This early state transition to PREPARE
worked around that.
We need to remove it now though, because (a) the reason for its
existence is no longer valid, and (b) _device_activate() may now
be called from inside nm_device_state_changed() and thus it cannot
change to a new state inside the function.
This fixes a regression introduced in 5074898591.
The while loop did only refetch the cached value (because the glib main loop
was blocked and only the cached device flags were checked).
Also, instead on relying of g_usleep(), wait until a maximum time of waiting
is expired. The duration of g_usleep() might not be very accurate.
Also, do no longer check the cached device state before setting the
device flag. The cache might be out of date, so we just set the flag.
https://bugzilla.gnome.org/show_bug.cgi?id=724363
Signed-off-by: Thomas Haller <thaller@redhat.com>
Add a property to NMDevice that can be used to tell whether the device
is enslaved, and if so what its master is.
This is currently internal-only, but it could be exported later
perhaps.
Assumed slave connections need to be added to their master devices,
which didn't used to happen because the devices activating assumed
connections jumped directly to stage3, bypassing all the master/slave
handling stuff.
Instead, make all assumed connections go through all activation stages,
but make sure that things which touch the device don't get done for
assumed connections. This requires moving the master/slave code out
of the override-able class methods because we need to call the
master/slave code for assumed connections, but we don't want to call
the override-able class activation methods.
Previously, ignore-carrier devices were always in the unavailable state
until they were activated. This required some complicated code to keep
track of whether the device was available or not based on what connections
existed, whether those connections were static-IP, and whether the device
was ignore-carrier. Various bits of the code used nm_device_can_activate()
for two different purposes: (1) to determine if the device was available
on an L2 basis, which nm_device_can_activate() wasn't well-suited to, and
(2) whether a specific connection could be activated at a given time
based on ignore-carrier and whether the connection was static IP or not.
Remove that complexity and confusion by making ignore-carrier devices
always move to DISCONNECTED state, and simply refuse to activate
connections that require connectivity, but allow connections that don't
require connectivity. Also, when the device has no carrier, don't
add connections that require connectivity to the AvailableConnections
device property.
