NMTernary is part of libnm's public API. It thus cannot be used by code
without libnm/libnm-core dependency.
Add another enum with the same purpose.
The name "NMTernary" is already taken, and we should not use some macro
trickery to use (effectively) different types under the same name.
Another possible name would be "NMTern", but for no strong reasons
we choose NMOptionBool. The naming reminds of rust's std::option::Option.
We want to use this by "shared/nm-platform", which should have
no dependency on "libnm-core".
Move "libnm-core/nm-ethtool-utils.h" to "libnm/nm-ethtool-utils.h" so
that it is only used by libnm. This file contains the defines for
the option names.
Also, symlink "libnm/nm-ethtool-utils.h" as "shared/nm-base/nm-ethtool-utils-base.h".
We want to use the same defines also internally. Since they are both
public API (must be in libnm) and should be in "shared/nm-base", this
is the way.
We want to use these defines for option names also in "shared/nm-base"
(and in turn in "shared/nm-platform), which cannot include "libnm-core".
However, they are also public API of libnm.
To get this done, in a first step, move these defines to a new header
"libnm-core/nm-ethtool-utils.h".
Since now the name "nm-ethtool-utils.h" is taken, also rename
nm-libnm-core-intern files.
There should be a clear hierarchie of dependency. That is,
"nm-platform.h" may use "nm-platform-utils.h", but not the
other way around.
Move nm_platform_link_duplex_type_to_string().
Currently src/platform depends on libnm-core. libnm-core is large
optimally we have a better separation between our code. That means
libnm-core does not depend on platform and vice versa.
However, nm-platform re-uses some enums from libnm-core for internal code.
To avoid that dependency, add _NMSettingWiredWakeOnLan as a duplicate to
nm-base/nm-base.h. nm-base can both be used by libnm-core, nm-platform
and src/platform.
The only problem is that NMSettingWiredWakeOnLan is also part of public
API of libnm. It means, we must duplicate the enum. But with several
static assertions in unit tests I think that is not a problem to do.
Our dependencies are complicated.
Currently "src/platform" uses parts of libnm-core and is relatively
strongly entangled with core. It would be nice to have that part
clearly independent from "src" and from "libnm-core".
Also, "src/platform/nm-platform-utils.h" uses NMEthtoolID enum, which
previously was defined in "libnm-core/nm-libnm-core-intern/nm-ethtool-utils.h".
Move that to a new place "shared/nm-base/nm-base.h".
Note that we have "libnm-core/nm-libnm-core-intern", which is
libnm/core related code which uses and is used by libnm-core.
There is a need for a library which is used by libnm-core, but
does not depend on libnm-core itself. Here comes "shared/nm-base".
Yes, many libraries. But the goal is to entangle the dependencies
and have a clear hierarchy of includes. And to have "shared/nm-platform"
independent of libnm-core.
We want to move platform code to "shared/nm-platform". However, platform
code uses the logging infrastructure from the daemon, there is thus
an odd circular dependency.
Solve that by moving the "src/nm-logging.[hc]" to a new helper library
in "shared/nm-log-core".
"src/nm-logging.c" should be independent of libnm-core. It almost
is, except the error domain and code.
Move NM_MANAGER_ERROR to "nm-glib-aux/nm-shared-utils.h" so that
"nm-logging.c" is independent of libnm-core.
NetworkManager core is huge. We should try to split out
parts that are independent.
Platform code is already mostly independent. But due to having it
under "src/", there is no strict separation/layering which determines
the parts that can work independently. So, while the code is mostly
independent (in practice), that is not obvious from looking at the
source tree. It thus still contributes to cognitive load.
Add a shared library "shared/nm-platform", which should have no
dependencies on libnm-core or NetworkManager core.
In a first step, move the netlink code there. More should follow.
This is the same as libnm's nm_utils_hwaddr_aton(), which however
is public API.
We want to use this function also without libnm(-core). Hence add
the helper to "shared/nm-glib-aux".
Enums can also be negative (contrary to Flags). Fix the parsing.
$ nmcli connection modify "$PROFILE" connection.llmnr -1
Error: failed to modify connection.llmnr: invalid option '-1', use one of [default,no,resolve,yes].
In the vast majority of cases is the string for _nm_utils_enum_from_str_full()
short. As we duplicate it for stripping, prefer to clone it on the stack
with nm_strdup_maybe_a().
The BOOTIF MAC address can be prefixed with a hardware address
type. Typically it is 01 (for ethernet), but the legacy network module
accepts (and strips) any byte value.
It seems wrong to take any address type without validation. In
addition to "01", also accept a zero type which, according to the
bugzilla below, is used in some configurations to mean "undefined".
While at it, also accept ':' as separator for the first byte.
https://bugzilla.redhat.com/show_bug.cgi?id=1904099https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/713
This assert sometimes fails during copr builds. But the way
the assert was, it was hard to see what the actual problem
was.
Restructure the assert (again) to get the errno in the
test logs.
RFCs actually expect to honor the lifetime. See for example [1].
This is just not right, and totally arbitrary. It was added
when our libndp based implementation was added, but unclear
why this was done (beyond the code comment).
[1] page 204, v6LC.2.2.25: Processing Router Advertisement DNS (Host
only) at https://ipv6ready.org/docs/Core_Conformance_5_0_0.pdf
There is no actual change in behavior, because "struct nd_opt_hdr"
as two uint8_t, so in practice this struct was always packed already.
But make it explicit, because it's clear that we use these structs
to set the binary message and they need a well defined (packed) memory
layout.
The endpoints of WireGuard peers can be configured as DNS name, which
NetworkManager will resolve.
Since activating a profile might affect now names get resolved, we must
first resolve names before completing the activation of the WireGuard
device (and before reconfiguring DNS accordingly).
For example, if you configure exclusive DNS resolution via the WireGuard
device, and if the peer needs to be resolved via DNS, then resolving the
peer name must happen before the reconfiguration of DNS. Otherwise the
new DNS configuration will be broken due to being unable to reach the
WireGuard peer.
Fix that by waiting.
There is still an unfixed problem. If resolving any peers fails,
activation silently proceeds -- again possibly breaking the network
setup. Of course, NetworkManager will repeatedly try to re-resolve
the name, but that may never succeed if DNS would be resolved via
the VPN itself.
That is different from `wg set` which resolves hostnames and fails.
Consequently `wg-quick up` would also fail. But these are both one shot
applications, they are not around and basically let the user handle the
error (by reading the log and invoking the command again). NetworkManager
can do something different and proceed activation (as it will also
periodically re-resolve the hostnames again). Note that it's also valid
to activate a WireGuard device without any peers (and to modify the
activated device later with Reapply()). As such, having no peers (or
being unable to resolve a hostname) may be a valid configuration.
I think we should add an option/flag that when enabled will cause
the activation to fail of names cannot be resolved.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/535https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/616https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/721
