Ok, I changed my mind.
The new behavior seems to make more sense to me. Not that it matters,
because we always use nm_utils_strbuf*() API with buffers that we expect
to be large enough to contain the result. And when truncation occurs,
we usually don't care much about it. That is, there is no code that
uses nm_utils_strbuf*() API and handles string truncation in particular.
NMConnection is an interface, which is implemented by the types
NMSimpleConnection (libnm-core), NMSettingsConnection (src) and
NMRemoteConnection (libnm).
NMSettingsConnection does a lot of things already:
1) it "is-a" NMDBusObject and exports the API of a connection profile
on D-Bus
2) it interacts with NMSettings and contains functionality
for tracking the profiles.
3) it is the base-class of types like NMSKeyfileConnection and
NMIfcfgConnection. These handle how the profile is persisted
on disk.
4) it implements NMConnection interface, to itself track the
settings of the profile.
3) and 4) would be better implemented via delegation than inheritance.
Address 4) and don't let NMSettingsConnection implemente the NMConnection
interface. Instead, a settings-connection references now a NMSimpleConnection
instance, to which it delegates for keeping the actual profiles.
Advantages:
- by delegating, there is a clearer separation of what
NMSettingsConnection does. For example, in C we often required
casts from NMSettingsConnection to NMConnection. NMConnection
is a very trivial object with very little logic. When we have
a NMConnection instance at hand, it's good to know that it is
*only* that simple instead of also being an entire
NMSettingsConnection instance.
The main purpose of this patch is to simplify the code by separating
the NMConnection from the NMSettingsConnection. We should generally
be aware whether we handle a NMSettingsConnection or a trivial
NMConnection instance. Now, because NMSettingsConnection no longer
"is-a" NMConnection, this distinction is apparent.
- NMConnection is implemented as an interface and we create
NMSimpleConnection instances whenever we need a real instance.
In GLib, interfaces have a performance overhead, that we needlessly
pay all the time. With this change, we no longer require
NMConnection to be an interface. Thus, in the future we could compile
a version of libnm-core for the daemon, where NMConnection is not an
interface but a GObject implementation akin to NMSimpleConnection.
- In the previous implementation, we cannot treat NMConnection immutable
and copy-on-write.
For example, when NMDevice needs a snapshot of the activated
profile as applied-connection, all it can do is clone the entire
NMSettingsConnection as a NMSimpleConnection.
Likewise, when we get a NMConnection instance and want to keep
a reference to it, we cannot do that, because we never know
who also references and modifies the instance.
By separating NMSettingsConnection we could in the future have
NMConnection immutable and copy-on-write, to avoid all unnecessary
clones.
1) the command line gets shorter. I frequently run `make V=1` to see
the command line arguments for the compiler, and there is a lot
of noise.
2) define each of these variables at one place. This makes it easy
to verify that for all compilation units, a particular
define has the same value. Previously that was not obvious or
even not the case (see commit e5d1a71396
and commit d63cf1ef2f).
The point is to avoid redundancy.
3) not all compilation units need all defines. In fact, most modules
would only need a few of these defines. We aimed to pass the necessary
minium of defines to each compilation unit, but that was non-obvious
to get right and often we set a define that wasn't used. See for example
"src_settings_plugins_ibft_cppflags" which needlessly had "-DSYSCONFDIR".
This question is now entirely avoided by just defining all variables in
a header. We don't care to find the minimum, because every component
gets anyway all defines from the header.
4) this also avoids the situation, where a module that previously did
not use a particular define gets modified to require it. Previously,
that would have required to identify the missing define, and add
it to the CFLAGS of the complation unit. Since every compilation
now includes "config-extra.h", all defines are available everywhere.
5) the fact that each define is now available in all compilation units
could be perceived as a downside. But it isn't, because these defines
should have a unique name and one specific value. Defining the same
name with different values, or refer to the same value by different
names is a bug, not a desirable feature. Since these defines should
be unique accross the entire tree, there is no problem in providing
them to every compilation unit.
6) the reason why we generate "config-extra.h" this way, instead of using
AC_DEFINE() in configure.ac, is due to the particular handling of
autoconf for directory variables. See [1].
With meson, it would be trivial to put them into "config.h.meson".
While that is not easy with autoconf, the "config-extra.h" workaround
seems still preferable to me.
[1] https://www.gnu.org/software/autoconf/manual/autoconf-2.63/html_node/Installation-Directory-Variables.html
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' \
"${FILES[@]}"
It's useless and redundant noise.
The original motivation seems to have been compatibility with ancient
versions uClibc (2011), but given CLOCK_BOOTTIME definition is shipped with
kernel headers, the libc version shall not matter anyway.
Even if it was the case, uClibc has shipped the definition for over
7 years now and been superseded by uClibc-ng that always had the
definition.
Originally, we used "nm-utils/siphash24.c", which was copied
from systemd's source tree. It was both used by our own NetworkManager
code, and by our internal systemd fork.
Then, we added "shared/c-siphash" as a dependency for n-acd.
Now, drop systemd's implementation and use c-siphash also
for our internal purpose. Also, let systemd code use c-siphash,
by patching "src/systemd/src/basic/siphash24.h".
Use two common defines NM_BUILD_SRCDIR and NM_BUILD_BUILDDIR
for specifying the location of srcdir and builddir.
Note that this is only relevant for tests, as they expect
a certain layout of the directories, to find files that concern
them.
Add new stable-id specifier "${DEVICE}" to explicitly declare that the
connection's identity differs per-device.
Note that for settings like "ipv6.addr-gen-mode=stable" we already hash
the interface's name. So, in combination with addr-gen-mode, using this
specifier has no real use. But for example, we don't do that for
"ipv4.dhcp-client-id=stable".
Point being, in various context we possibly already include a per-device
token into the generation algorithm. But that is not the case for all
contexts and uses.
Especially the DHCPv4 client identifier is supposed to differ between interfaces
(according to RFC). We don't do that by default with "ipv4.dhcp-client-id=stable",
but with "${DEVICE}" can can now be configured by the user.
Note that the fact that the client-id is the same accross interfaces, is not a
common problem, because profiles are usually restricted to one device via
connection.interface-name.
No longer rely on nm_connection_get_path() being meaningful in server.
It also was wrong. During update, nm_settings_connection_update()
would call
nm_utils_log_connection_diff (replace_connection, NM_CONNECTION (self), ...
where replace_connection has no path set, and nothing was logged.
Fix it, by explicitly passing the D-Bus path. Also, because
nm-core-utils.c should be independent of nm-dbus-object.h.
There are multiple tests with the same in different directories; add a
unique prefix to test names so that it is clear from the output which
one is running.
An asynchronous request should either be cancellable or not keep
the target object alive. Preferably both.
Otherwise, it is impossible to do a controlled shutdown when terminating
NetworkManager. Currently, when NetworkManager is about to terminate,
it just quits the mainloop and essentially leaks everything. That is a
bug. If we ever want to fix that, every asynchronous request must be
cancellable in a controlled way (or it must not prevent objects from
getting disposed, where disposing the object automatically cancels the
callback).
Rework the asynchronous request for connectivity check to
- return a handle that can be used to cancel the operation.
Cancelling is optional. The caller may choose to ignore the handle
because the asynchronous operation does not keep the target object
alive. That means, it is still possible to shutdown, by everybody
giving up their reference to the target object. In which case the
callback will be invoked during dispose() of the target object.
- also, the callback will always be invoked exactly once, and never
synchronously from within the asynchronous start call. But during
cancel(), the callback is invoked synchronously from within cancel().
Note that it's only allowed to cancel an action at most once, and
never after the callback is invoked (also not from within the callback
itself).
- also, NMConnectivity already supports a fake handler, in case
connectivity check is disabled via configuration. Hence, reuse
the same code paths also when compiling without --enable-concheck.
That means, instead of having #if WITH_CONCHECK at various callers,
move them into NMConnectivity. The downside is, that if you build
without concheck, there is a small overhead compared to before. The
upside is, we reuse the same code paths when compiling with or without
concheck.
- also, the patch synchronizes the connecitivty states. For example,
previously `nmcli networking connectivity check` would schedule
requests in parallel, and return the accumulated result of the individual
requests.
However, the global connectivity state of the manager might have have
been the same as the answer to the explicit connecitivity check,
because while the answer for the manual check is waiting for all
pending checks to complete, the global connectivity state could
already change. That is just wrong. There are not multiple global
connectivity states at the same time, there is just one. A manual
connectivity check should have the meaning of ensure that the global
state is up to date, but it still should return the global
connectivity state -- not the answers for several connectivity checks
issued in parallel.
This is related to commit b799de281b
(libnm: update property in the manager after connectivity check),
which tries to address a similar problem client side.
Similarly, each device has a connectivity state. While there might
be several connectivity checks per device pending, whenever a check
completes, it can update the per-device state (and return that device
state as result), but the immediate answer of the individual check
might not matter. This is especially the case, when a later request
returns earlier and obsoletes all earlier requests. In that case,
earlier requests return with the result of the currend devices
connectivity state.
This patch cleans up the internal API and gives a better defined behavior
to the user (thus, the simple API which simplifies implementation for the
caller). However, the implementation of getting this API right and properly
handle cancel and destruction of the target object is more complicated and
complex. But this but is not just for the sake of a nicer API. This fixes
actual issues explained above.
Also, get rid of GAsyncResult to track information about the pending request.
Instead, allocate our own handle structure, which ends up to be nicer
because it's strongly typed and has exactly the properties that are
useful to track the request. Also, it gets rid of the awkward
_finish() API by passing the relevant arguments to the callback
directly.
- consistently set options, searches, domains fields to %NULL,
if there are no values.
- in nm_global_dns_config_update_checksum(), ensure that we uniquely
hash values. E.g. a config with "searches[a], options=[b]" should
hash differently from "searches=[ab], options=[]".
- in nm_global_dns_config_to_dbus(), reuse the sorted domain list.
We already have it, and it guarantees a consistent ordering of
fields.
- in global_dns_domain_from_dbus(), fix memleaks if D-Bus strdict
contains duplicate entries.
Previously, we used the generated GDBusInterfaceSkeleton types and glued
them via the NMExportedObject base class to our NM types. We also used
GDBusObjectManagerServer.
Don't do that anymore. The resulting code was more complicated despite (or
because?) using generated classes. It was hard to understand, complex, had
ordering-issues, and had a runtime and memory overhead.
This patch refactors this entirely and uses the lower layer API GDBusConnection
directly. It replaces the generated code, GDBusInterfaceSkeleton, and
GDBusObjectManagerServer. All this is now done by NMDbusObject and NMDBusManager
and static descriptor instances of type GDBusInterfaceInfo.
This adds a net plus of more then 1300 lines of hand written code. I claim
that this implementation is easier to understand. Note that previously we
also required extensive and complex glue code to bind our objects to the
generated skeleton objects. Instead, now glue our objects directly to
GDBusConnection. The result is more immediate and gets rid of layers of
code in between.
Now that the D-Bus glue us more under our control, we can address issus and
bottlenecks better, instead of adding code to bend the generated skeletons
to our needs.
Note that the current implementation now only supports one D-Bus connection.
That was effectively the case already, although there were places (and still are)
where the code pretends it could also support connections from a private socket.
We dropped private socket support mainly because it was unused, untested and
buggy, but also because GDBusObjectManagerServer could not export the same
objects on multiple connections. Now, it would be rather straight forward to
fix that and re-introduce ObjectManager on each private connection. But this
commit doesn't do that yet, and the new code intentionally supports only one
D-Bus connection.
Also, the D-Bus startup was simplified. There is no retry, either nm_dbus_manager_start()
succeeds, or it detects the initrd case. In the initrd case, bus manager never tries to
connect to D-Bus. Since the initrd scenario is not yet used/tested, this is good enough
for the moment. It could be easily extended later, for example with polling whether the
system bus appears (like was done previously). Also, restart of D-Bus daemon isn't
supported either -- just like before.
Note how NMDBusManager now implements the ObjectManager D-Bus interface
directly.
Also, this fixes race issues in the server, by no longer delaying
PropertiesChanged signals. NMExportedObject would collect changed
properties and send the signal out in idle_emit_properties_changed()
on idle. This messes up the ordering of change events w.r.t. other
signals and events on the bus. Note that not only NMExportedObject
messed up the ordering. Also the generated code would hook into
notify() and process change events in and idle handle, exhibiting the
same ordering issue too.
No longer do that. PropertiesChanged signals will be sent right away
by hooking into dispatch_properties_changed(). This means, changing
a property in quick succession will no longer be combined and is
guaranteed to emit signals for each individual state. Quite possibly
we emit now more PropertiesChanged signals then before.
However, we are now able to group a set of changes by using standard
g_object_freeze_notify()/g_object_thaw_notify(). We probably should
make more use of that.
Also, now that our signals are all handled in the right order, we
might find places where we still emit them in the wrong order. But that
is then due to the order in which our GObjects emit signals, not due
to an ill behavior of the D-Bus glue. Possibly we need to identify
such ordering issues and fix them.
Numbers (for contrib/rpm --without debug on x86_64):
- the patch changes the code size of NetworkManager by
- 2809360 bytes
+ 2537528 bytes (-9.7%)
- Runtime measurements are harder because there is a large variance
during testing. In other words, the numbers are not reproducible.
Currently, the implementation performs no caching of GVariants at all,
but it would be rather simple to add it, if that turns out to be
useful.
Anyway, without strong claim, it seems that the new form tends to
perform slightly better. That would be no surprise.
$ time (for i in {1..1000}; do nmcli >/dev/null || break; echo -n .; done)
- real 1m39.355s
+ real 1m37.432s
$ time (for i in {1..2000}; do busctl call org.freedesktop.NetworkManager /org/freedesktop org.freedesktop.DBus.ObjectManager GetManagedObjects > /dev/null || break; echo -n .; done)
- real 0m26.843s
+ real 0m25.281s
- Regarding RSS size, just looking at the processes in similar
conditions, doesn't give a large difference. On my system they
consume about 19MB RSS. It seems that the new version has a
slightly smaller RSS size.
- 19356 RSS
+ 18660 RSS
The next commit will completely rework NMBusManager and replace
NMExportedObject by a new type NMDBusObject.
Originally, NMDBusObject was added along NMExportedObject to ease
the rework and have compilable, intermediate stages of refactoring. Now,
I think the new name is better, because NMDBusObject is very strongly related
to the bus manager and the old name NMExportedObject didn't make that
clear.
I also slighly prefer the name NMDBusObject over NMBusObject, hence
for consistancy, also rename NMBusManager to NMDBusManager.
This commit only renames the file for a nicer diff in the next commit.
It does not actually update the type name in sources. That will be done
later.
While the numerical values of IFA_F_SECONDARY and IFA_F_TEMPORARY
are identical, their meaning is not.
IFA_F_SECONDARY is only relevant for IPv4 addresses, while
IFA_F_TEMPORARY is only relevant for IPv6 addresses.
IFA_F_TEMPORARY is automatically set by kernel for the addresses
that it generates as part of IFA_F_MANAGETEMPADDR. It cannot be
actively set by user-space.
IFA_F_SECONDARY is automatically set by kernel depending on the order
in which the addresses for the same subnet are added.
This essentially reverts 8b4f11927 (core: avoid IFA_F_TEMPORARY alias for
IFA_F_SECONDARY).
Some targets are missing dependencies on some generated sources in
the meson port. These makes the build to fail due to missing source
files on a highly parallelized build.
These dependencies have been resolved by taking advantage of meson's
internal dependencies which can be used to pass source files,
include directories, libraries and compiler flags.
One of such internal dependencies called `core_dep` was already in
use. However, in order to avoid any confusion with another new
internal dependency called `nm_core_dep`, which is used to include
directories and source files from the `libnm-core` directory, the
`core_dep` dependency has been renamed to `nm_dep`.
These changes have allowed minimizing the build details which are
inherited by using those dependencies. The parallelized build has
also been improved.
Tests are commonly created via copy&paste. Hence, it's
better to express a certain concept explicitly via a function
or macro. This way, the implementation of the concept can be
adjusted at one place, without requiring to change all the callers.
Also, the macro is shorter, and brevity is better for tests
so it's easier to understand what the test does. Without being
bothered by noise from the redundant information.
Also, the macro knows better which message to expect. For example,
messages inside "src" are prepended by nm-logging.c with a level
and a timestamp. The expect macro is aware of that and tests for it
#define NMTST_EXPECT_NM_ERROR(msg) NMTST_EXPECT_NM (G_LOG_LEVEL_MESSAGE, "*<error> [*] "msg)
This again allows the caller to ignore this prefix, but still assert
more strictly.
Note that:
- we compile some source files multiple times. Most notably those
under "shared/".
- we include a default header "shared/nm-default.h" in every source
file. This header is supposed to setup a common environment by defining
and including parts that are commonly used. As we always include the
same header, the header must behave differently depending
one whether the compilation is for libnm-core, NetworkManager or
libnm-glib. E.g. it must include <glib/gi18n.h> or <glib/gi18n-lib.h>
depending on whether we compile a library or an application.
For that, the source files need the NETWORKMANAGER_COMPILATION #define
to behave accordingly.
Extend the define to be composed of flags. These flags are all named
NM_NETWORKMANAGER_COMPILATION_WITH_*, they indicate which part of the
build are available. E.g. when building libnm-core.la itself, then
WITH_LIBNM_CORE, WITH_LIBNM_CORE_INTERNAL, and WITH_LIBNM_CORE_PRIVATE
are available. When building NetworkManager, WITH_LIBNM_CORE_PRIVATE
is not available but the internal parts are still accessible. When
building nmcli, only WITH_LIBNM_CORE (the public part) is available.
This granularily controls the build.
The logging macros already prepend a "config: " prefix. Don't
repeat that in the message, otherwise we get
config: config: signal SIGHUP (no changes from disk)
Now:
config: signal: SIGHUP (no changes from disk)
Source files for enum types are generated by passing segments of the
source code of the files to the `glib-mkenums` command.
This patch removes those parameters where source code is used from
meson build files by moving those segmeents to template files.
https://mail.gnome.org/archives/networkmanager-list/2017-December/msg00057.html
meson spawns the tests in a way that the test process is
a session leader. Since the test wants to create a new
process group to kill a group of processes that it starts,
it failed.
$ meson test -C build test-general-with-expect
The test would have succeed when wrapping the test for example
by strace:
$ meson test -C build --wrap='strace' test-general-with-expect
Fix that, by forking once more.
There are some tests located in different directories which are
using the same name. To avoid any confussion a prefix was used to
name the test and the target.
This patch uses the prefix just for the target, to avoid any
collision that may happen, and uses the `test-` pattern as the
name.
https://mail.gnome.org/archives/networkmanager-list/2017-December/msg00051.html
The matching works fuzzy and is not reliable. That is why we store
which connection should be assumed after restart in the state file
of NetworkManager.
In that case, we don't need to do a full check (with the possibility
of a false-reject). Just check for the minimum required properties:
the type and slave-type.
Yes, if the user modifies the connection while restarting NM, then
we might wrongly assume a connection that no longer would match.
But NM should not read minds, it should do as indicated.
We also do this for libnm, where it causes visible changes
in behavior. But if somebody would rely on the hashing implementation
for hash tables, it would be seriously flawed.
Instead of having 3 properties @gateway, @never_default and @has_gateway
on NMIP4Config/NMIP6Config that determine the default-route, track the
default-route as a regular route.
The gateway setting is the configuration knob for the default-route.
Since an NMIP4Config/NMIP6Config instance only has one gateway property,
it cannot track more then one default-routes (see related bug rh#1445417).
Especially with policy routing, it might be interesting to configure a
default-route in multiple tables.
Also, later it might be interesting to allow adding default-routes as
regular static routes in a connection, so that the user can configure additional
route parameters for the default-route or add default-routes in multiple tables.
With this patch, default-routes now have a rt_source property according to their
origin.
Also, the previous commits of this branch broke handling of the
default-route :) . That should be working now again.
The MSS is only set for VPN connections (by merging it in the respective
NMIP4Config/NMIP6Config).
It is also only used when setting the MSS of the default route.
Don't track that property in NMIP4Config/NMIP6Config, instead, set the
mss of the route directly in nm_vpn_connection_ip4_config_get() and
nm_vpn_connection_ip6_config_get().
There is a potential change in behavior here: NMDevice also consisdered
the MSS for the default route, but that would only be set if the MSS
gets merged from an vpn-ip-config. Which at most is the case for
iterface-less VPN types (libreswan). But even in that case, it doesn't
seem right to me to use the VPN's MSS for the device's default-route.
- merge the IPv4 and IPv6 implementations. They are for the most
part identical. Also, they are independent of NMIP4Config/NMIP6Config.
- parse the entire file at once. Don't parse it twice, once for the
name servers and once for the options. This also avoids loading
/etc/resolv.conf twice, as it would be done before.
For kernel, route ID compare identical according to NM_PLATFORM_IP_ROUTE_CMP_TYPE_ID.
Well, mostly. In practice, NM ignores several route properties that
kernel considers part of the ID too. This leaves the possibility that
kernel allows addition of two routes that compare idential for
NetworkManager.
Anyway, NMIP4Config/NMIP6Config should use the same equality as platform
cache. Otherwise, there is the odd situation that ip-config merges routes
that are treated as different by kernel.
For IP addresses the ID operator already corresponded to what kernel
does. There is no change for addresses.
Note that NMSettingIPConfig also uses a different algorithm for
comparing routes. But that doesn't really matter here, it it differed
before too.
