The onlink flag is part of each next hop.
When NetworkManager configures ECMP routes, we won't support that. All
next hops of an ECMP route must share the same onlink flag. That is fine
and fixed by this commit.
What is not fine, is that we don't track the rtnh_flags flags in
NMPlatformIP4RtNextHop, and consequently our nmp_object_id_cmp() is
wrong.
Fixes: 5b5ce42682 ('nm-netns: track ECMP routes')
(cherry picked from commit 6ed966258c)
nm_setting_diff() ends up calling the compare_fcn() hook. Previously,
the hook for "dns" was _nm_setting_property_compare_fcn_default()
and the hook for "dns-data" was _nm_setting_property_compare_fcn_ignore().
That's wrong. _nm_setting_property_compare_fcn_default() converts
the property to D-Bus and compares the GVariant. However, "dns" has
to_dbus_only_in_manager_process set, so it wouldn't
Fixes: 63eaf168d1 ('libnm: add "dns-data" replacement for "ipv[46].dns" properties on D-Bus')
- nm_setting_ip_config_add_dns() and nm_setting_ip_config_remove_dns_by_value()
used to assert that the provided input is valid. That is not
documented and highly problematic.
Our parsing code for keyfile, ifcfg-rh and GVariant rightly just call
add. Likewise, nmcli. We cannot reasonably expect them to pre-validate
the input. Why would we anyway?
This is wrong in particular because we usually want the user to be
able to construct invalid settings. That is often necessary, because
whether a value is valid depends on other values. So in general, we
can only validate when all properties are set. We have
nm_connection_verify() for that, and asserting/validating during add
is very wrong. Note that "add" still filters out duplicates, which
may be an inconsistency, but well.
Also, the user could set any bogus value via the NM_SETTING_IP_CONFIG_DNS
property. Those should be allowed to be removed, and the same values
should be allowed to be added via the add method.
- add() does a normalization, presumably so that the values look nice.
Do the same normalization also when using the NM_SETTING_IP_CONFIG_DNS
property setter.
- previously, the setter could also set unnormalized values. As
nm_setting_ip_config_remove_dns_by_value() looked for the normalized
value, you couldn't remove such values anymore. That is fixed now,
by letting the property setter do the same normalization.
- don't allocate a GPtrArray unless we need it. No need for the extra
allocation.
- in the property setter, first set the new value before destroying the
previous GPtrArray. It might not be possible, but it's not clear to me
whether the strv argument from the GValue is always deep-copied or
whether it could contain strings to the DNS property itself.
On D-Bus, the properties "ipv[46].dns" are of type "au" and "aay",
respectively.
Btw, in particular "au" is bad, because we put there a big-endian
number. There is no D-Bus type to represent big endian numbers, so "u"
is bad because it can cause endianess problem when trying to remote
the D-Bus communication to another host (without explicitly
understanding which "u" properties need to swap for endinness).
Anyway. The plain addresses are not enough. We soon will also support
the DNS-over-TLS server name, or maybe a DoT port number. The previous
property was not extensible, so deprecate it and replace it by
"dns-data".
This one is just a list of strings. That is unlike "address-data" or
"route-data", which do a similar thing but are "a{sv}" dictionaries.
Here a string is supposed to be sufficient also for the future. Also,
because in nmcli and keyfile will will simply have a string format for
representing the extra data, not a structure (unlike for routes or
addresses).
We have 4 legacy properties ("ipv[46].addresses", "ipv[46].routes") that
got replaced by newer variants ("ipv[46].address-data", "ipv[46].route-data").
When the client side of libnm (_nm_utils_is_manager_process) serializes
those properties, it must only serialize the newer version. That is so
that the forward/backward compatibility works as intended.
Previously, there was the NM_SETTING_PARAM_LEGACY GObject property flag.
That was fine, but not very clear.
For one, the legacy part of those properties is only about D-Bus. In
particular, they are not deprecated in libnm, keyfile, or nmcli. Thus
the name wasn't very clear.
Also, in the meantime we have more elaborate property meta data, that
goes beyond the meta data of the GObject property.
Move NM_SETTING_PARAM_LEGACY to NMSettInfoProperty.to_dbus_only_in_manager_process.
I think, this is a better name. It's also right at
```
_nm_properties_override_gobj(
properties_override,
g_object_class_find_property(G_OBJECT_CLASS(setting_class), NM_SETTING_IP_CONFIG_ROUTES),
NM_SETT_INFO_PROPERT_TYPE_DBUS(NM_G_VARIANT_TYPE("a(ayuayu)"),
.to_dbus_fcn = ip6_routes_to_dbus,
.compare_fcn = _nm_setting_ip_config_compare_fcn_routes,
.from_dbus_fcn = ip6_routes_from_dbus, ),
.to_dbus_only_in_manager_process = TRUE,
.dbus_deprecated = TRUE, );
```
that is, directly at the place where we describe how the D-Bus property behaves.
The G_TYPE_INSTANCE_GET_CLASS() macro is just one pointer dereference
(self)->g_class, plus additional assertions with debug builds.
As such, it is as fast as it gets. Embed the address family there, and
implement NM_SETTING_IP_CONFIG_GET_ADDR_FAMILY() that way.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/1395
Introduction of a new setting ipv4.link-local, which enables
link-local IP addresses concurrently with other IP address assignment
implementations such as dhcp or manually.
No way is implemented to obtain a link-local address as a fallback when
dhcp does not respond (as dhcpd does, for example). This could be be
added later.
To maintain backward compatibility with ipv4.method ipv4.link-local has
lower priority than ipv4.method. This results in:
* method=link-local overrules link-local=disabled
* method=disabled overrules link-local=enabled
Furthermore, link-local=auto means that method defines whether
link-local is enabled or disabled:
* method=link-local --> link-local=enabled
* else --> link-local=disabled
The upside is, that this implementation requires no normalization.
Normalization is confusing to implement, because to get it really
right, we probably should support normalizing link-local based on
method, but also vice versa. And since the method affects how other
properties validate/normalize, it's hard to normalize that one, so that
the result makes sense. Normalization is also often not great to the
user, because it basically means to modify the profile based on other
settings.
The downside is that the auto flag becomes API and exists because
we need backward compatibility with ipv4.method.
We would never add this flag, if we would redesign "ipv4.method"
(by replacing by per-method-specific settings).
Defining a default setting for ipv4.link-local in the global
configuration is also supported.
The default setting for the new property can be "default", since old
users upgrading to a new version that supports ipv4.link-local will not
have configured the global default in NetworkManager.conf. Therefore,
they will always use the expected "auto" default unless they change
their configuration.
Co-Authored-By: Thomas Haller <thaller@redhat.com>
IPv4:
routes
A list of IPv4 destination addresses, prefix length, optional IPv4
next hop addresses, optional route metric, optional attribute. The
valid syntax is: "ip[/prefix] [next-hop] [metric]
[attribute=val]...[,ip[/prefix]...]". For example "192.0.2.0/24
10.1.1.1 77, 198.51.100.0/24".
Various attributes are supported:
• "cwnd" - an unsigned 32 bit integer.
• "initcwnd" - an unsigned 32 bit integer.
• "initrwnd" - an unsigned 32 bit integer.
• "lock-cwnd" - a boolean value.
• "lock-initcwnd" - a boolean value.
• "lock-initrwnd" - a boolean value.
• "lock-mtu" - a boolean value.
• "lock-window" - a boolean value.
• "mtu" - an unsigned 32 bit integer.
• "onlink" - a boolean value.
• "scope" - an unsigned 8 bit integer. IPv4 only.
• "src" - an IPv4 address.
• "table" - an unsigned 32 bit integer. The default depends on
ipv4.route-table.
• "tos" - an unsigned 8 bit integer. IPv4 only.
• "type" - one of unicast, local, blackhole, unavailable,
prohibit. The default is unicast.
• "window" - an unsigned 32 bit integer.
For details see also `man ip-route`.
Format: a comma separated list of routes
IPv6:
routes
A list of IPv6 destination addresses, prefix length, optional IPv6
next hop addresses, optional route metric, optional attribute. The
valid syntax is: "ip[/prefix] [next-hop] [metric]
[attribute=val]...[,ip[/prefix]...]".
Various attributes are supported:
• "cwnd" - an unsigned 32 bit integer.
• "from" - an IPv6 address with optional prefix. IPv6 only.
• "initcwnd" - an unsigned 32 bit integer.
• "initrwnd" - an unsigned 32 bit integer.
• "lock-cwnd" - a boolean value.
• "lock-initcwnd" - a boolean value.
• "lock-initrwnd" - a boolean value.
• "lock-mtu" - a boolean value.
• "lock-window" - a boolean value.
• "mtu" - an unsigned 32 bit integer.
• "onlink" - a boolean value.
• "src" - an IPv6 address.
• "table" - an unsigned 32 bit integer. The default depends on
ipv6.route-table.
• "type" - one of unicast, local, blackhole, unavailable,
prohibit. The default is unicast.
• "window" - an unsigned 32 bit integer.
For details see also `man ip-route`.
Format: a comma separated list of routes
(cherry picked from commit 7b1e9a5c3d)
We use clang-format for automatic formatting of our source files.
Since clang-format is actively maintained software, the actual
formatting depends on the used version of clang-format. That is
unfortunate and painful, but really unavoidable unless clang-format
would be strictly bug-compatible.
So the version that we must use is from the current Fedora release, which
is also tested by our gitlab-ci. Previously, we were using Fedora 34 with
clang-tools-extra-12.0.1-1.fc34.x86_64.
As Fedora 35 comes along, we need to update our formatting as Fedora 35
comes with version "13.0.0~rc1-1.fc35".
An alternative would be to freeze on version 12, but that has different
problems (like, it's cumbersome to rebuild clang 12 on Fedora 35 and it
would be cumbersome for our developers which are on Fedora 35 to use a
clang that they cannot easily install).
The (differently painful) solution is to reformat from time to time, as we
switch to a new Fedora (and thus clang) version.
Usually we would expect that such a reformatting brings minor changes.
But this time, the changes are huge. That is mentioned in the release
notes [1] as
Makes PointerAligment: Right working with AlignConsecutiveDeclarations. (Fixes https://llvm.org/PR27353)
[1] https://releases.llvm.org/13.0.0/tools/clang/docs/ReleaseNotes.html#clang-format
Certain properties need to release memory when destroying the NMSetting.
For "direct" properties, we have all the information we need to do that
generically in the NMSetting base class. In practice, this only concerns
string properties.
See _finalize_direct() in "nm-setting.c".
However, if the NMSetting base class takes care of freeing the strings,
then the subclasses must not also unref the variable (to avoid double free).
Previously, subclasses had to opt-in for the base class to indicate that
they are fine with that.
Now, let the base class always handle it. We only need to make sure that
classes that implement direct string properties don't also try to free
the values during destruction.
For IPv4, the order is not like for IPv6. Of course not.
Fixes: 7aa4ad0fa2 ('nmcli/docs: better describe ipv[46].addresses in `man nm-settings-nmcli`')
Note that most implementations use g_object_set(), and it's not
easy to detect modification. In those cases, we assume that modification
happened -- just like also the GObject setter will emit a notification
(as none of our properties use G_PARAM_EXPLICIT_NOTIFY).
These functions tend to have many arguments. They are also quite som
boilerplate to implement the hundereds of properties we have, while
we want that properties have common behaviors and similarities.
Instead of repeatedly spelling out the function arguments, use a macro.
Advantages:
- the usage of a _NM_SETT_INFO_PROP_*_FCN_ARGS macro signals that this
is an implementation of a property. You can now grep for these macros
to find all implementation. That was previously rather imprecise, you
could only `git grep '\.to_dbus_fcn'` to find the uses, but not the
implementations.
As the goal is to keep properties "similar", there is a desire to
reduce the number of similar implementations and to find them.
- changing the arguments now no longer will require you to go through
all implementations. At least not, if you merely add an argument that
has a reasonable default behavior and does not require explicit
handling by most implementation.
- it's convenient to be able to patch the argument list to let the
compiler help to reason about something. For example, the
"connection_dict" argument to from_dbus_fcn() is usually unused.
If you'd like to find who uses it, rename the parameter, and
review the (few) compiler errors.
- it does save 573 LOC of boilerplate with no actual logic or useful
information. I argue, that this simplifies the code and review, by
increasing the relative amount of actually meaningful code.
Disadvantages:
- the user no longer directly sees the argument list. They would need
cscope/ctags or an IDE to jump to the macro definition and conveniently
see all arguments.
Also use _nm_nil, so that clang-format interprets this as a function
parameter list. Otherwise, it formats the function differently.
This is a normalization employed by NMSettingIPConfig.gateway.
Also rework NMSettingIPConfig.set_property() to no longer assert against
valid input. We want to pass there untrusted strings from D-Bus,
asserting is a horrible idea. Instead, either normalize the string or
keep the invalid text that will be rejected by verify().
When looking at a property, it should always be clear how it is handled.
Also the "default" action should be an explicit hook.
Add _nm_setting_property_from_dbus_fcn_gprop() and set that as
from_dbus_fcn() callback to handle the "default" case which us
build around g_object_set_property().
While this adds lines of code, I think it makes the code easier to
understand. Basically, to convert a GVariant to a property, now all
properties call their from_dbus_fcn() handler, there is no special casing.
And the gprop-hook is only called for properties that are using
_nm_setting_property_from_dbus_fcn_gprop(). So, you can reason about
these two functions at separate layers.
The goal is to get rid of gprop_to_dbus_fcn() uses.
Note that there is a change in behavior. The "dns" GPtrArray in
NMSettingIPConfig is never NULL (the default of the boxed property),
thus the previous code always serialized the property, even the
empty list.
Now, empty dns properties are omitted from D-Bus.
Also, there is another change in behavior: nm_utils_ip4_dns_to_variant()
will now skip over strings that are not valid IPv4 addresses.
Previously, it would have added 0.0.0.0 (or some undefined address).
So far, we only have NMSettingClass.compare_property() hook.
The ugliness is that this hook is per-setting, when basically
all implementations only compare one property.
It feels cleaner to have a per-property hook and call that consistently.
In step one, we give all properties (the same) compare_fcn() implementation,
which delegates to the existing NMSettingClass.compare_property().
In a second step, this will be untangled.
There is one problem with this approach: NMSettInfoPropertType grows by
one pointer size, and we have potentially many such types. That should
be addressed by unifying types in the future.
Various NMSetting API would accept a property_idx parameter. Together
with the NMSettInfoSetting instance, this was useful to find the actual
NMSettInfoProperty instance.
The idea was, to provide the most of the functionality. That is, if you
might need the property_idx too, you had it -- after all, the
property_info you could lookup yourself.
However,
- literally zero users care about the property_idx. The care about
the property_info.
- if the user really, really required the property_idx, then it
is a given that it can be easily computed by
(property_info - sett_info->property_infos)
For our property meta data handling we require that all the meta data is
associated with one GType.
NMSettingIPConfig is a parent class of NMSettingIP[46]Config. Note that
we already have _nm_sett_info_property_override_create_array_ip_config()
because the meta data must be initialized together at one place.
We will require that we can find the offset for properties based on one
offset per type. That is cumbersome, if NMSettingIPConfigPrivate is
private itself.
Simplify that, by internally sharing NMSettingIPConfigPrivate and let
the subclasses embed the private data in their own private data.
Optimally we would simply embed the private struct as field into
NMSettingIPConfig. But that would be an ABI change as that struct
was public before 1.32. Don't change ABI for now, so we have to
awkwardly place it into the subclasses private data.
NMSetting instances either have no private data, they use
g_type_add_class_private(), or they embed the private data in the
NMSetting struct.
In all cases, we can find the private data at a fixed offset. Track that
offset in the NMSettInfoSetting meta data.
This will be useful, because properties really are stored in simple
fields, like a boolean property can be stored in a "bool" field. We will
extend the property meta data to track the offset of this property
field, but we also need to know where the offset starts.
For each property we have meta data in form of NMSettInfoProperty.
Each meta data also has a NMSettInfoProperty.property_type
(NMSettInfoPropertType).
The property type is supposed to define common behaviors for properties,
while the property meta data has individual properties. The idea is that
several properties can share the same property-type, and that
per-property meta data is part of NMSettInfoProperty.
The distinction is not very strong, but note that all remaining uses
of NMSettInfoPropertType.gprop_to_dbus_fcn were part of a property
type that was only used for one single property. That lack of
reusability hints to a wrong use.
Move gprop_to_dbus_fcn to the property meta data as a new field
NMSettInfoProperty.to_dbus_data.
Note that NMSettInfoPropertType.gprop_from_dbus_fcn still suffers from
the same problem. But the from-dbus side is not yet addressed.
If a property can be converted to D-Bus, then always set the
to_dbus_fcn() handler. The only caller of to_dbus_fcn() is
property_to_dbus(), so this means that property_to_dbus()
has no more default implementation and always delegates to
to_dbus_fcn().
The code is easier to understand if all properties implement
to_dbus_fcn() the same way.
Also, there is supposed to be a split between NMSettInfoProperty (info about
the property) and NMSettInfoPropertType (the type). The idea is that
each property (obviously) requires its distinct NMSettInfoProperty, but
they can share a common type implementation.
With NMSettInfoPropertType.gprop_to_dbus_fcn that is often violated because
many properties that implement NMSettInfoPropertType.gprop_to_dbus_fcn
require a special type implementation. As such, gprop_to_dbus_fcn should
be part of the property info and not the property type. The first step towards
that is unifying all properties to use to_dbus_fcn().
When subclassing a GObject type, the class and object structs
must be available and defined in the header.
For libnm, and in particular for NMSetting classes, we don't want
users to subclass NMSetting. It also doesn't work, because libnm
has internal code that is necessary to hook up the NMSetting class.
You cannot define your own type and make it work together with
libnm.
Having the structs in public headers limits what we can do with them.
For example, we could embed the private data directly in the structures
and avoid the additional indirection.
This is an API break, but for something that most likely nobody cares
about. Or better, nobody should care about. API is not what is
accidentally defined in a header, API was the library provides to
meaningfully use. Subclassing these types is not meaningful and was
only accidentally possible so far.
Only hide the structs for now. More cleanup is possible later. We shall
however aim to keep the padding and struct layout to not also break ABI.
(cherry picked from commit e46d484fae)
nm-settings-connection.c has code similar to this in two places:
/* FIXME: improve NMConnection API so we can avoid the overhead of cloning the connection,
* in particular if there are no secrets to begin with. */
connection_cloned = nm_simple_connection_new_clone(new);
/* Clear out unwanted secrets */
_nm_connection_clear_secrets_by_secret_flags(connection_cloned,
NM_SETTING_SECRET_FLAG_NOT_SAVED
| NM_SETTING_SECRET_FLAG_AGENT_OWNED);
secrets = nm_g_variant_ref_sink(
nm_connection_to_dbus(connection_cloned, NM_CONNECTION_SERIALIZE_ONLY_SECRETS));
It seems the secrets filtering can be done by nm_connection_to_dbus() if
the NM_CONNECTION_SERIALIZE_* flags are extended. The current set of
flags contains flags that start with NO, ONLY and WITH prefixes, which
makes it useless for combining the flags because most combinations of
more than one flag don't have a clear interpretation. So they're mostly
useful when used alone, i.e. you'd need to add a new enum value for
each new subset of settings to be serialized.
To get the most flexibility from a small set of flags they should
either all be of the WITH_* type or NO_* type. In the former case they
could be combined to extend the subset of properties serialized, in the
latter case each flag would reduce the subset. After trying both
options I found it's easier to adapt the current set of flags to the
WITH_* schema while keeping binary and source compatibility. This
commit changes the set of flags in the following way:
NM_CONNECTION_SERIALIZE_ALL is kept for compatibility but is equivalent
to a combination of other flags.
NM_CONNECTION_SERIALIZE_WITH_NON_SECRET is added with the same value as
NM_CONNECTION_SERIALIZE_NO_SECRETS, it implies that non-secret
properties are included but doesn't prevent including other properties.
Since it couldn't be meaningfully combined with any other flag this
change shouldn't break compatibility.
Similarly NM_CONNECTION_SERIALIZE_WITH_SECRETS is added with the same
value as existing NM_CONNECTION_SERIALIZE_ONLY_SECRETS with the same
consideration about compatibility.
NM_CONNECTION_SERIALIZE_WITH_SECRETS_AGENT_OWNED and the new
NM_CONNECTION_SERIALIZE_WITH_SECRETS_SYSTEM_OWNED and
NM_CONNECTION_SERIALIZE_WITH_SECRETS_NOT_SAVED add only subsets of
secrets and can be combined. For backwards compatibility
NM_CONNECTION_SERIALIZE_ONLY_SECRETS is basically ignored when either of
these three is present, so that the value:
..ONLY_SECRETS | ..AGENT_OWNED works as previously.
