Note that in NetworkManager API (D-Bus, libnm, and nmcli),
the features are called "feature-xyz". The "feature-" prefix
is used, because NMSettingEthtool possibly will gain support
for options that are not only -K|--offload|--features, for
example -C|--coalesce.
The "xzy" suffix is either how ethtool utility calls the feature
("tso", "rx"). Or, if ethtool utility specifies no alias for that
feature, it's the name from kernel's ETH_SS_FEATURES ("tx-tcp6-segmentation").
If possible, we prefer ethtool utility's naming.
Also note, how the features "feature-sg", "feature-tso", and
"feature-tx" actually refer to multiple underlying kernel features
at once. This too follows what ethtool utility does.
The functionality is not yet implemented server-side.
Add a new way how NMSetting subclasses can be implemented.
Currently, most NMSetting implementations realize all their properties
via GObject properties. That has some downsides:
- the biggest one, is the large effort to add new properties.
Most of them are implemented on a one-by-one basis and they come
with additional API (like native getter functions).
It makes it cumbersome to add more properties.
- for certain properties, it's hard to encode them entirely in
a GObject property. That results in unusable API like
NM_SETTING_IP_CONFIG_ADDRESSES, NM_SETTING_BOND_OPTIONS,
NM_SETTING_USER_DATA. These complex valued properties only
exist, because we currently always need GObject properties
to even implement simple functionality. For example,
nm_setting_duplicate() is entirely implemented via
nm_setting_enumerate_values(), which can only iterate
GObject properies. There is no reason why this is necessary.
Note also how nmcli badly handles bond options and VPN
data. That is only a shortcoming of nmcli and wouldn't
need to be that way. But it happend, because we didn't
keep an open mind that settings might be more than just
accessing GObject properties.
- a major point of NMSetting is to convert to/from a GVariant
from the D-Bus API. As NMSetting needs to squeeze all values
into the static GObject structure, there is no place to
encode invalid or unknown properties. Optimally,
_nm_setting_new_from_dbus() does not loose any information
and a subsequent _nm_setting_to_dbus() can restore the original
variant. That is interesting, because we want that an older
libnm client can talk to a newer NetworkManager version. The
client needs to handle unknown properties gracefully to stay
forward compatible. However, it also should not just drop the
properties on the floor.
Note however, optimally we want that nm_setting_verify() still
can reject settings that have such unknown/invalid values. So,
it should be possible to create an NMSetting instance without
error or loosing information. But verify() should be usable to
identify such settings as invalid.
They also have a few upsides.
- libnm is heavily oriented around GObject. So, we generate
our nm-settings manual based on the gtk-doc. Note however,
how we fail to generate a useful manual for bond.options.
Also note, that there is no reason we couldn't generate
great documentation, even if the properties are not GObject
properties.
- GObject properties do give some functionality like meta-data,
data binding and notification. However, the meta-data is not
sufficient on its own. Note how keyfile and nmcli need extensive
descriptor tables on top of GObject properties, to make this
useful. Note how GObject notifications for NMSetting instances
are usually not useful, aside for data binding like nmtui does.
Also note how NMSettingBond already follows a different paradigm
than using GObject properties. Nowdays, NMSettingBond is considered
a mistake (related bug rh#1032808). Many ideas of NMSettingBond
are flawed, like exposing an inferiour API that reduces everything
to a string hash. Also, it only implemented the options hash inside
NMSettingBond. That means, if we would consider this a good style,
we would have to duplicate this approach in each new setting
implementation.
Add a new style to track data for NMSetting subclasses. It keeps
an internal hash table with all GVariant properies. Also, the
functionality is hooked into NMSetting base class, so all future
subclasses that follow this way, can benefit from this. This approach
has a few similiarties with NMSettingBond, but avoids its flaws.
With this, we also no longer need GObject properties (if we would
also implement generating useful documentation based on non-gkt-doc).
They may be added as accessors if they are useful, but there is no
need for them.
Also, handling the properties as a hash of variants invites for a
more generic approach when handling them. While we still could add
accessors that operate on a one-by-one bases, this leads to a more
generic usage where we apply common functionality to a set of properties.
Also, this is for the moment entirely internal and an implementation
detail. It's entirely up to the NMSetting subclass to make use of this
new style. Also, there are little hooks for the subclass available.
If they turn out to be necessary, they might be added. However, for
the moment, the functionality is restricted to what is useful and
necessary.
NMSetting internally already tracked a list of all proper GObject properties
and D-Bus-only properties.
Rework the tracking of the list, so that:
- instead of attaching the data to the GType of the setting via
g_type_set_qdata(), it is tracked in a static array indexed by
NMMetaSettingType. This allows to find the setting-data by simple
pointer arithmetic, instead of taking a look and iterating (like
g_type_set_qdata() does).
Note, that this is still thread safe, because the static table entry is
initialized in the class-init function with _nm_setting_class_commit().
And it only accessed by following a NMSettingClass instance, thus
the class constructor already ran (maybe not for all setting classes,
but for the particular one that we look up).
I think this makes initialization of the metadata simpler to
understand.
Previously, in a first phase each class would attach the metadata
to the GType as setting_property_overrides_quark(). Then during
nm_setting_class_ensure_properties() it would merge them and
set as setting_properties_quark(). Now, during the first phase,
we only incrementally build a properties_override GArray, which
we finally hand over during nm_setting_class_commit().
- sort the property infos by name and do binary search.
Also expose this meta data types as internal API in nm-setting-private.h.
While not accessed yet, it can prove beneficial, to have direct (internal)
access to these structures.
Also, rename NMSettingProperty to NMSettInfoProperty to use a distinct
naming scheme. We already have 40+ subclasses of NMSetting that are called
NMSetting*. Likewise, NMMetaSetting* is heavily used already. So, choose a
new, distinct name.
Previously, each (non abstract) NMSetting class had to register
its name and priority via _nm_register_setting().
Note, that libnm-core.la already links against "nm-meta-setting.c",
which also redundantly keeps track of the settings name and gtype
as well.
Re-use NMMetaSettingInfo also in libnm-core.la, to track this meta
data.
The goal is to get rid of private data structures that track
meta data about NMSetting classes. In this case, "registered_settings"
hash. Instead, we should have one place where all this meta data
is tracked. This was, it is also accessible as internal API,
which can be useful (for keyfile).
Note that NMSettingClass has some overlap with NMMetaSettingInfo.
One difference is, that NMMetaSettingInfo is const, while NMSettingClass
is only initialized during the class_init() method. Appart from that,
it's mostly a matter of taste, whether we attach meta data to
NMSettingClass, to NMMetaSettingInfo, or to a static-array indexed
by NMMetaSettingType.
Note, that previously, _nm_register_setting() was private API. That
means, no user could subclass a functioning NMSetting instance. The same
is still true: NMMetaSettingInfo is internal API and users cannot access
it to create their own NMSetting subclasses. But that is almost desired.
libnm is not designed, to be extensible via subclassing, nor is it
clear why that would be a useful thing to do. One day, we should remove
the NMSetting and NMSettingClass definitions from public headers. Their
only use is subclassing the types, which however does not work.
While libnm-core was linking already against nm-meta-setting.c,
nm_meta_setting_infos was unreferenced. So, this change increases
the binary size of libnm and NetworkManager (1032 bytes). Note however
that roughly the same information was previously allocated at runtime.
Properties that are backed by a GObject property are fundamentally
different.
I think it's clearer to rework the check, to first check whether
we have a param_spec, and then implement different checks.
_nm_utils_init() is a __attribute__((constructor)) function,
that is, it runs during dlopen().
On the other head, g_module_open() itself calls dlopen().
It is prone to deadlock. Don't do it.
The check is only an aggressive assertion to crash the application
if it wrongly loads libnm and libnm-util/libnm-glib at the same time.
If that happens, all is lost already. We can just as well call the
assertion later. It's not supposed to fail anyway.
https://bugzilla.gnome.org/show_bug.cgi?id=796804
Utilize _nm_setting_to_dbus() to serialize the setting. The main reason
is that this way we can also print the more complicated values
g_strdup_value_contents() can't grok, e.g. the GArrays and GHashTables.
Some effort was spent on tidying up the results in a manner it was done
previously, instead of reducing this to a plain g_variant_print(). It
looks good that way:
Before:
vpn
service-type : "org.freedesktop.NetworkManager.VPN.Novpn" (s)
user-name : NULL (sd)
persistent : FALSE (sd)
data : ((GHashTable*) 0xc61060) (s)
secrets : ((GHashTable*) 0xdda640) (s)
timeout : 0 (sd)
After:
vpn
service-type : 'org.freedesktop.NetworkManager.VPN.Novpn'
data : {'gateway': 'novpn.example.com', 'username': 'hello'}
secrets : {'password': 'world'}
Note that no effort was spent on printing the defaults. There are
multiple ways that could be achieved, but I'm not sure it would be all
that necessary given this is really just a quick'n'dirty debugging facilty.
We commonly don't use the glib typedefs for char/short/int/long,
but their C types directly.
$ git grep '\<g\(char\|short\|int\|long\|float\|double\)\>' | wc -l
587
$ git grep '\<\(char\|short\|int\|long\|float\|double\)\>' | wc -l
21114
One could argue that using the glib typedefs is preferable in
public API (of our glib based libnm library) or where it clearly
is related to glib, like during
g_object_set (obj, PROPERTY, (gint) value, NULL);
However, that argument does not seem strong, because in practice we don't
follow that argument today, and seldomly use the glib typedefs.
Also, the style guide for this would be hard to formalize, because
"using them where clearly related to a glib" is a very loose suggestion.
Also note that glib typedefs will always just be typedefs of the
underlying C types. There is no danger of glib changing the meaning
of these typedefs (because that would be a major API break of glib).
A simple style guide is instead: don't use these typedefs.
No manual actions, I only ran the bash script:
FILES=($(git ls-files '*.[hc]'))
sed -i \
-e 's/\<g\(char\|short\|int\|long\|float\|double\)\>\( [^ ]\)/\1\2/g' \
-e 's/\<g\(char\|short\|int\|long\|float\|double\)\> /\1 /g' \
-e 's/\<g\(char\|short\|int\|long\|float\|double\)\>/\1/g' \
"${FILES[@]}"
constructor functions are ugly, because code is running before
main() starts. Instead, as the registration code for NMSetting types
is insid the GType constructor, we just need to ensure at the
right place, that the GType was created.
The right place here is _register_settings_ensure_inited(), because
that is called before we need the registration information.
_nm_register_setting() and _nm_register_setting_impl() are called from within
the GType constructor for the NMSetting subtype. As such, at that point it
runs inside a g_once_init_enter() block. However, each implementation
for initializing the GType has a separate g_once_init_enter() variable, hence,
if two threads create GType instances for different NMSetting subclasses, there
is a race.
libnm is not thread safe. However, it should be at least thread safe
with respect to constructing the GType instances.
For NMSetting subtypes, we need the static dictionaries "registered_settings" and
"registered_settings_by_type" to keep track of existing NMSetting types.
Initialize these dictionaries inside NMSetting's type initialization code.
This is guaranteed to run before any use of NMSetting type, and is also
guarded by a mutex.
Also, drop the __attribute__((constructor)) function to initialize the
hash tables. They are not needed, and it's ugly to run code before
main().
We also do this for libnm and libnm-core, where it causes visible changes
in behavior. But if somebody would rely on the hashing implementation
for hash tables, it would be seriously flawed.
Normalizing can be complicated, as settings depend on each other and possibly
conflict.
That is, because verify() must exactly anticipate whether normalization will
succeed and how the result will look like. That is because we only want to
modify the connection, if we are sure that the result will verify.
Hence, verify() and normalize() are strongly related. The implementation
should not be spread out between NMSettingOvsInterface:verify(),
NMSettingOvsPatch:verify() and _normalize_ovs_interface_type().
Also, add some unit-tests.
Previously, nm_setting_diff() would return !(*results), that means,
if the caller passed in a hash table (empty or not), the return value
would always be FALSE, indicating a difference.
That is not documented, and makes no sense.
The return value, should solely indicate whether some difference was
found. The only convenience is, if nm_setting_diff() created a hash
table internally and no difference was found, it would destroy
it again, without returning it to the caller.
NMSettingGeneric has no properties at all. Hence, nm_connection_diff() would report that
a connection A with a generic setting and a connection B without a generic setting are
equal.
They are not. For empty settings, let nm_setting_diff() return also empty difference
hash.
Functions call each other, like
nm_connection_get_id()
nm_connection_get_setting_connection()
nm_connection_get_setting()
Along the way, each function asserts that the input argument
is of type NMConnection via
g_return_val_if_fail (NM_IS_CONNECTION (connection), NULL);
Avoid such duplicate assertions when we already verifyied the
input argument.
For example, in case of nm_connection_get_id(), don't check just call
nm_connection_get_setting_connection() right away. It already
asserts.
The downside is, that the assertion no longer fails in the function
that immediately called it. But these are assertions after all.
Using plain numbers make it cumbersome to grep for
setting types by priority.
The only downside is, that with the enum values it
is no longer obvious which value has higher or lower
priority.
Also, introduce NM_SETTING_PRIORITY_INVALID. This is what
_nm_setting_type_get_base_type_priority() returns. For the moment
it still has the same numerical value 0 as before. Later, that
shall be distinct from NM_SETTING_PRIORITY_CONNECTION.
When the two base settings are present, use one of higher priority.
This will pick the "bridge" setting when both "bridge" and "bluetooth" are
present for a Bluetooth NAP connection.
We'll need two "base" settings for Bluetooth NAP connections: bridge to set up
the actual link and bluetooth to identify the HCI to register the network
server with.
Let's use two priorities for base setting, with "1" marking one of higher
priority and "2" of lower priority when both are present.
In some situations, we want strict checking of errors, for example when
NetworkManager receives a new connection from a client, the connection
must make sense as a whole (and since NetworkManager service is backward
compatible to the clients and not the other way around, there is no
excuse for sending invalid data to the server).
In other situations, we want a best-effort behavior. Like when
NetworkManager sends a connection to its clients, those clients
want to extract as many properties as they understand, but in order
to be forward compatible against newer server versions, invalid
or unknown properties must be accepted.
Previously, a mixture of both was done. Some issues caused a failure
to create a new NMSetting, other invalid parts were just silently
ignored or triggered a g_warning() in glib.
Now allow for both. When doing strict-validation, be more strict and
reject all unknown properties and catch when the user sets an invalid
argument. On the other hand, allow for a best-effort mode that
effectively cannot fail and will return a new NMSetting instance.
For now, add NMSettingParseFlags so that the caller can choose the
old behavior, strict parsing, or best effort.
This patch doesn't have any externally visible change except that
no more g_warnings will be emitted.
- All internal source files (except "examples", which are not internal)
should include "config.h" first. As also all internal source
files should include "nm-default.h", let "config.h" be included
by "nm-default.h" and include "nm-default.h" as first in every
source file.
We already wanted to include "nm-default.h" before other headers
because it might contains some fixes (like "nm-glib.h" compatibility)
that is required first.
- After including "nm-default.h", we optinally allow for including the
corresponding header file for the source file at hand. The idea
is to ensure that each header file is self contained.
- Don't include "config.h" or "nm-default.h" in any header file
(except "nm-sd-adapt.h"). Public headers anyway must not include
these headers, and internal headers are never included after
"nm-default.h", as of the first previous point.
- Include all internal headers with quotes instead of angle brackets.
In practice it doesn't matter, because in our public headers we must
include other headers with angle brackets. As we use our public
headers also to compile our interal source files, effectively the
result must be the same. Still do it for consistency.
- Except for <config.h> itself. Include it with angle brackets as suggested by
https://www.gnu.org/software/autoconf/manual/autoconf.html#Configuration-Headers
- "gsystem-local-alloc.h" and <gio/gio.h> are already included via
"nm-default.h". No need to include them separately.
- include "nm-macros-internal.h" via "nm-default.h" and drop all
explict includes.
- in the modified files, ensure that we always include "config.h"
and "nm-default.h" first. As second, include the header file
for the current source file (if applicable). Then follow external
includes and finally internal nm includes.
- include nm headers inside source code files with quotes
- internal header files don't need to include default headers.
They can savely assume that "nm-default.h" is already included
and with it glib, nm-glib.h, nm-macros-internal.h, etc.
for verifying the secrets, because it is not done in plain nm_setting_verify().
For simple verification of free-form string secrets,
_nm_setting_verify_secret_string() helper is used.
The localization headers are now included via "nm-default.h".
Also fixes several places, where we wrongly included <glib/gi18n-lib.h>
instead of <glib/gi18n.h>. For example under "clients/" directory.
Rather than randomly including one or more of <glib.h>,
<glib-object.h>, and <gio/gio.h> everywhere (and forgetting to include
"nm-glib-compat.h" most of the time), rename nm-glib-compat.h to
nm-glib.h, include <gio/gio.h> from there, and then change all .c
files in NM to include "nm-glib.h" rather than including the glib
headers directly.
(Public headers files still have to include the real glib headers,
since nm-glib.h isn't installed...)
Also, remove glib includes from header files that are already
including a base object header file (which must itself already include
the glib headers).
Before, get_property_for_dbus() would @ignore_defaults.
That is for example wrong for properties of type G_TYPE_STRV.
In this case, if one operand has the property at its default
(NULL) and the other has it to an empty string list, both would
compare equal.
This has the effect that different settings might compare equal.
It yields completely unpredictable results on Ubuntu 12.04 (the global variable
successfully comparing to NULL despite demonstrably not NULL). Possibly a
toolchain bug.
The sort order of nm_setting_enumerate_values() affects the
order in which keyfile writer serializes the properties.
Have a defined, stable sort order by sorting the properties
by name (with prefering id,uuid,type for NMSettingConnection).
